Adaptive stimulation device with integrated feedback mechanism and proxy measurement

Abstract

An adaptive stimulation system optimizes pain relief, pleasure enhancement, and healing through closed-loop physiological effectiveness monitoring. The system comprises a stimulation device, sensors measuring physiological responses indicative of stimulation effectiveness, and a processor implementing real-time adaptive control. The system measures unconscious physiological responses that users cannot voluntarily control and involuntary reactions to objectively assess whether stimulation achieves desired outcomes. Through proxy learning, system sensors correlate with sophisticated clinical measurements via machine learning, enabling clinical-grade optimization using accessible devices. The system continuously explores multidimensional parameter spaces to discover optimal stimulation patterns personalized to individual users, automatically adapting parameters based on measured physiological effectiveness rather than subjective user control or preset patterns. Applications include medical devices, wellness products, and intimate stimulation systems.

Description

Attorney Docket No. SETH-005 AWOADAPTIVE STIMULATION DEVICE WITH INTEGRATED FEEDBACK MECHANISM AND PROXY MEASUREMENTCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Application No. 63 / 714,992, filed November 1, 2024, the entire contents of which are hereby incorporated by reference in their entirety.BACKGROUND

[0002] Stimulation devices span therapeutic and wellness domains, from transcutaneous electrical nerve stimulation (TENS) units to intimate stimulators. Most existing systems operate as open loop devices that deliver fixed stimulation without measuring or adapting to effectiveness. Recent innovations in closed loop neuromodulation have shown that adaptive control can improve therapeutic consistency, but these systems typically rely on implanted electrodes or single channel feedback, limiting their practicality for everyday or consumer use. Because human physiological responses vary widely across individuals and over time, this static and invasive architecture continues to produce inconsistent therapeutic and experiential results.

[0003] This variability is fundamental to human physiology7. Users must manually adjust device settings such as intensity, frequency, waveform, and duration in response to how they feel, placing the full burden of optimization on subjective perception and moment to moment judgment.

[0004] To enable stimulation devices to respond adaptively, they must address a challenge: how to measure internal physiological and affective states such as pain, pleasure, and stress. Measurement modalities span direct and proxy approaches. Direct measurement using technologies such as functional magnetic resonance imaging (fMRI), functional near infrared spectroscopy (fNIRS), or deep cortical stimulation remains the gold standard for research but is impractical for routine or in home use due to cost, size, and invasiveness.

[0005] Responsive systems that reduce manual adjustment and adapt to physiological feedback could transform healthcare, wellness, and intimacy applications. However, most commercially available stimulators still rely on open loop control, requiring users to adjust parameters manually. Even when feedback features exist, they are limited to user initiated inputs and lack the continuity and real time responsiveness required for true adaptive operation.

[0006] Feedback allows systems to respond dynamically to changing conditions, enabling real time optimization, correction, and personalization through iterative refinement rather than predefined programs. In stimulation, where subjective states are fluid and individual variability is high, feedback offers the only scalable path from static operation to adaptive intelligence. Importantly, feedback doesAttorney Docket No. SETH-005 AWO not require perfect accuracy, only signals that are consistently correlated with improvement or decline. While various feedback approaches have been explored, most rely on invasive neural interfaces. Recent closed loop research has focused on implantable or intracranial systems that record electrical or optical brain signals. These approaches remain invasive, complex, and costly, reinforcing the need for non invasive, wearable architectures such as those disclosed herein.SUMMARY

[0007] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.

[0008] The present disclosure introduces a generalized feedback-driven architecture capable of incorporating proxy measurements, navigating multidimensional parameter spaces, and applying outcome-driven adjustment logic for optimization of different stimulation modalities in a wide range of applications. The system detects when stimulation is trending toward or away from desired effects — such as pain reduction, pleasure enhancement, or stress relief — and adapts accordingly across diverse use cases, sensors, devices, and user goals. The present disclosure is based on the recognition that feedback control does not depend on perfect physiological accuracy but rather on the consistent correlation of measurable signals with the intended effect. When interpreted through an adaptive framework, even approximate indicators can support meaningful adjustment of stimulation.

[0009] Aspects of this architecture provide an adaptive stimulation system that automatically optimizes therapeutic and hedonic outcomes through real-time physiological effectiveness monitoring and closed-loop parameter adjustment. The system comprises a stimulation device configured to deliver adjustable stimulation, one or more physiological sensors measuring responses indicative of stimulation effectiveness, and a processor implementing closed-loop feedback control. Unlike open-loop devices that operate with fixed parameters or brain-computer interfaces that translate conscious user intent into device commands, the disclosed system measures unconscious physiological responses to determine whether stimulation achieves desired therapeutic or hedonic effects and autonomously adjusts stimulation parameters to maximize effectiveness. It continuously monitors physiological indicators that users cannot consciously control — such as autonomic responses, muscle activity, and behavioral reactions — providing objective assessment of effectiveness rather than relying on subjective input or pre-authored stimulation patterns.

[0010] In certain implementations, machine-learning models relate accessible physiological signals to higher-fidelity reference data to enable adaptive control using consumer-grade sensors. TheAttomev Docket No. SETH-005 AWO controller computes a scalar effectiveness index, E(t) (or P(t) for hedonic embodiments), from multimodal, non-invasive proxy inputs such as HRV, EDA, PPG, respiration, temperature, motion, facial, or vocal features. No direct neural input is required, and no single biomarker is necessary for operation. The system supports training with any subset of available signals and continues functioning even when one or more previously used biomarkers are absent, maintaining robust performance through graceful degradation.

[0011] The effectiveness index may be obtained by any suitable estimation or inference method — including statistical models, control-theoretic observers, optimization frameworks, machinelearning models, or generative neural networks — without limitation to a single computational paradigm.

[0012] Control is formulated as a search over the space of all permissible stimulation settings, including amplitude, frequency, waveform, spatial distribution, or other adjustable parameters, to maximize the effectiveness index. The optimization proceeds within an explicitly configurable multitimescale safety hierarchy, coordinating fast-safety (< 100 ms), in-session (seconds to minutes), and cross-session (hours to weeks) adaptation loops to ensure safe, stable, and personalized operation.BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Other aspects, features, and advantages of the claimed invention will become more fully apparent from the following detailed description, the appended claims, and the accompanying drawings in which like reference numerals identify similar or identical elements. Reference numerals that are introduced in the specification in association with a drawing figure may be repeated in one or more subsequent figures without additional description in the specification in order to provide context for other features.

[0014] FIG. 1 is a block diagram of an adaptive stimulation system, according to one or more aspects of the disclosure;

[0015] FIG. 2 is a flow diagram of an in-session closed-loop method, according to one or more aspects of the disclosure;

[0016] FIG. 3 is a flow diagram of an example cross-session learning process, according to one or more aspects of the disclosure;

[0017] FIG. 3A is a table showing a mapping of the learning and sensing scenarios, according to aspects of the disclosure;

[0018] FIG. 3B is a table showing the corresponding training and runtime flows of the scenarios in FIG. 3A, according to aspects of the disclosure;

[0019] FIG. 3C is a table showing the shared multi-timescale control structure that coordinates fast-safety, in-session, and cross-session learning loops, according to aspects of the disclosure;Attorney Docket No. SETH-005 AWO

[0020] FIG. 3D is a flow diagram of a generalized multi-timescale feedback architecture used by the adaptive stimulation system, according to aspects of the disclosure;

[0021] FIG. 3E is a flow diagram of a hybrid audio and physiological control architecture, according to aspects of the disclosure;

[0022] FIGS. 4-6 are block diagrams of exemplary adaptive stimulation systems, according to aspects of the disclosure;

[0023] FIG. 6A is a flow diagram of a proxy-model training and deployment process, according to aspects of the disclosure;

[0024] FIG. 7 is a block diagram of a TENS-Based system with sensor feedback, according to aspects of the disclosure;

[0025] FIG. 8 is a diagram of the layers of the flexible multi-electrode pad, according to aspects of the disclosure;

[0026] FIG. 8A is a block diagram showing an example signal processing, according to aspects of the disclosure;

[0027] FIG. 8B is block diagram showing a controller data path and optimization layers, according to aspects of the disclosure;

[0028] FIGS. 9 -10 are block diagrams of diagnostic-probing and adaptive-targeting sequences, according to aspects of the disclosure;

[0029] FIG. 11 is a flow diagram of illustrating the hierarchical control structure comprising a fast-safety7loop, an in-session optimization loop, and a cross-session learning loop, according to aspects of the disclosure;

[0030] FIG. 12 is a block diagram of a population-level federated-learning framework, according to aspects of the disclosure;

[0031] FIG. 13 is a flow diagram of a supervisory7safety7and automatic shutdown subsystem, according to aspects of the disclosure;

[0032] FIGS. 14 - 15A are tables showing representative sensor modalities and comparative performance characteristics, according to aspects of the disclosure;

[0033] FIG. 16 is a flow diagram of a robotic massage system, according to aspects of the disclosure;

[0034] FIG. 16A is a schematic block diagram of the robotic massage system of FIG. 16, according to aspects of the disclosure;

[0035] FIG. 16B is a diagram showing a sequence of neurophysiological events according to a therapeutic response, according to aspects of the disclosure;Attorney Docket No. SETH-005 AWO

[0036] FIG. 16C is a diagram of a hierarchical feedback and control loops of the adaptive robotic stimulation system, according to aspects of the disclosure;

[0037] FIG. 16D is a diagram of a calibration and archetype mapping, according to aspects of the disclosure;

[0038] FIG. 16E is a diagram of a end-effector and sensor integration, according to aspects of the disclosure;

[0039] FIG. 17 is a diagram of a universal closed-loop control framework for an adaptive- stimulation system, according to aspects of the disclosure;

[0040] FIG. 18 is a diagram of a hierarchical organization of sensing modalities used across the adaptive stimulation system, according to aspects of the disclosure;

[0041] FIG. 19 is a flow diagram of a somato-autonomic feedback network, according to aspects of the disclosure;

[0042] FIG. 20 is a flow diagram of an adaptive feedback loop for autonomic regulation, according to aspects of the disclosure;

[0043] FIG. 20A is a flow diagram of a data flow between the modules of a somato-autonomic feedback network, according to aspects of the disclosure;

[0044] FIG. 21 is a flow diagram of an iterative closed-loop process for stimulation parameter refinement, according to aspects of the disclosure;

[0045] FIG. 22 is flow diagram of three interacting timescales, according to aspects of the disclosure;

[0046] FIG. 22A is flow diagram of an adaptive exploration loop, according to aspects of the disclosure;

[0047] FIG. 23 is a block diagram of an adaptive pleasure system architecture, according to aspects of the disclosure;

[0048] FIG. 23A is a diagram of an adaptive pleasure system, according to aspects of the disclosure;

[0049] FIG. 23B is a flow diagram of a pleasure detection and feedback computation process, according to aspects of the disclosure;

[0050] FIG. 23C is a diagram of a laboratory calibration system, according to aspects of the disclosure;

[0051] FIG. 23D is a diagram of a proxy-model and population-level training hierarchy, according to aspects of the disclosure; andAttorney Docket No. SETH-005 AWO

[0052] FIG. 24 is a block diagram illustrating selective components of an example computing device in which various aspects of the disclosure may be implemented, according to aspects of the disclosure.DETAILED DESCRIPTION

[0053] Aspects of the present disclosure relate to physiological adaptive stimulation systems — devices that alter the physiological state of a person or body part for medical, wellness, or intimate purposes and that adapt stimulation in response to the user's physiological or affective state. These systems span disciplines including neurotechnology, wearable biosensing, affective computing, physiological signal processing, and personalized robotics.

[0054] System Overview

[0055] The present disclosure generally relates to devices and methods used for relieving pain or creating pleasure through adaptive stimulation systems that incorporate real-time (on the timescale of seconds) and / or near-real-time (on the timescale of tens of seconds to minutes, but during a stimulation session) multimodal feedback. An adaptive stimulation system and device may include one or more components that work together to provide an enhanced user experience through a feedback mechanism that employs both sophisticated neuro- or hemodynamic sensing and accessible proxy measurements. While the architecture supports optional use of high-fidelity neural or hemodynamic sensors — such as EEG, functional near-infrared spectroscopy (INIRS), or functional ultrasound (fUS) for calibration or research purposes, it is not limited to those modes and does not require continuous direct-neural feedback. Instead, the system operates through multimodal feedback integrating physiological, autonomic, behavioral, and linguistic inputs (see FIG. 14 for representative sensing modalities). Aspects of the disclosure address the critical limitations and gaps identified in existing stimulation technologies. As described above, known devices suffer from several architectural deficiencies that prevent optimal therapeutic and hedonic outcomes.

[0056] Unlike some existing stimulation devices that operate in open-loop mode w ith fixed or manually -adjusted parameters, the present system implements true closed-loop feedback control. Prior physiology -responsive devices operate by selecting or replaying pre-authored stimulation patterns when particular physiological events or thresholds are detected. In contrast, the embodiments described herein compute a fused effectiveness index E(t) / P(t) from multimodal proxy signals and continuously optimize stimulation over the full parameter space under explicit safety constraints. To prevent the brittleness inherent in pattern-based feedback, the control framework is designed for generalized and resilient learning — actively inferring transferable relationships between stimulation and physiological response rather than memorizing or replaying specific patterns. Patterns, if present, may be used as non-bindingAttorney Docket No. SETH-005 AWO priors for initialization; control is not limited to pattern selection and does not require any single biomarker. While devices such as spinal cord stimulators may adjust for posture or movement, they do not measure therapeutic effectiveness or adapt parameters based on physiological indicators of pain relief or pleasure enhancement. Aspects of the present disclosure provide for continuously monitoring objective physiological responses including autonomic nervous system activity, brain signals, behavioral indicators, and vocal stress patterns to determine whether stimulation achieves desired outcomes and automatically optimizes parameters accordingly.

[0057] The present disclosure provides concepts, techniques and structures that differ from brain-computer interfaces such as EEG-controlled devices that translate conscious user intent into device commands. Such systems detect imagined movements or conscious commands and use these as control inputs, but do not assess whether the stimulation produces beneficial physiological effects. In contrast, the present system measures unconscious physiological responses that users cannot voluntarily control — such as heart rate variability patterns, galvanic skin response characteristics, EEG phaseamplitude coupling, and autonomic balance indicators — to objectively determine stimulation effectiveness independent of conscious user assessment. High-fidelity and proxy sensing relationships are further detailed in the Proxy -Measurement Framework section.

[0058] Aspects of the present disclosure include systematic exploration of vast multidimensional parameter spaces rather than relying on predetermined patterns or manual user adjustment. For a typical advanced transcutaneous electrical nerve stimulation (TENS) unit with 20 amplitude settings, 50 frequency options, 10 waveforms, 4 electrode pairs, and temporal pattern variations, over 400,000 parameter combinations exist before considering the flexible electrode array innovations described herein, which expand the space to millions of possibilities. The system employs sophisticated optimization algorithms including genetic algorithms, Bayesian optimization, reinforcement learning, and multi-armed bandit approaches to efficiently navigate these high-dimensional spaces and discover optimal configurations for individual users under specific conditions.

[0059] Integrated Multi-Modal Architecture: The embodiment provides a comprehensive framework integrating multiple innovations into a cohesive system architecture, addressing the fragmentation in known systems. The system combines: advanced physiological sensing across multiple modalities; natural language interface for intuitive user control; proxy learning for accessible deployment; sophisticated optimization algorithms; flexible electrode array technology for spatial optimization; diagnostic stimulation protocols for precise targeting; safety monitoring and override systems; and longitudinal learning for continuous improvement.Attorney Docket No. SETFI-005AWO

[0060] The concepts, techniques and structures disclosed herein provide for creating the a comprehensive framework that transforms stimulation devices from passive tools into intelligent systems capable of autonomous optimization based on physiological feedback. Unlike existing approaches that either stimulate without awareness of effectiveness or require manual user adjustment, the present disclosure provides an integrated ecosystem where real-time biological signals drive systematic parameter optimization across different stimulation sessions and / or diverse stimulation modalities.

[0061] As described herein, the adaptive stimulation system provides an integrated framework for physiological, behavioral, and affective feedback, distinguishing it from prior open-loop stimulation and neuromodulation systems. The architecture combines multimodal sensing, proxy -based physiological modeling, and real-time optimization to create a closed-loop ecosystem that continuously leams the relationship between delivered stimulation and measured response. The systems support electrical, mechanical, thermal, pneumatic, and hybrid modalities, enabling a single control platform for pain-relief, relaxation, and pleasure applications rather than the narrow neural or spinal focus of existing devices. It transforms stimulation from a fixed output into an adaptive process that links waveform, frequency, amplitude, and location to measurable outcomes and extends beyond pain suppression to include enhancement of positive human experience. Through its proxy framework, the system maps high-fidelity neural or hemodynamic data (e.g., EEG, fNIRS) to accessible biosignals such as HRV, GSR, and vocal or facial features, allowing low-cost devices to achieve clinical-grade performance. The architecture defines multiple operating modes — Direct-Neural, Proxy-Direct, Population-Model, and Archetype-Assisted so that each embodiment (e.g., TENS, massage, vibratory, thermal) can select the feedback mode suited to its sensor configuration. It further integrates human-in-the-loop feedback, treating the user’s voice or behavior as an additional sensor stream, and coordinates feedback latencies across fast-safety, in-session, and cross-session learning loops. Learning and optimization policies balance exploration and stability within defined safety limits, while personal calibration within each proxy mode produces individualized mappings that improve automatically over successive sessions. A comprehensive safety and regulatory framework with current, temperature, and impedance limits, redundant processors, and standards-based risk controls ensures reliability. Collectively, these features establish a closed-loop adaptive stimulation system capable of autonomously learning, optimizing, and personalizing therapeutic and experiential outcomes beyond the capability of existing open-loop or intent-driven devices.

[0062] Field and Application OverviewAttorney Docket No. SETH-005 AWO

[0063] Aspects of the present disclosure encompass adaptive bioelectronic and sensory stimulation systems, particularly those that operate in a closed loop by measuring a user's physiological state and modifying stimulus output accordingly. In one aspect, the system operates across multiple domains simultaneously: clinical applications including chronic pain management, neurological and musculoskeletal rehabilitation, and autonomic nervous system regulation; consumer wellness applications including stress reduction, recovery enhancement, and performance optimization; and intimacy wellness applications including arousal enhancement, sexual health improvement, and relationship therapy support.

[0064] Two pillars of human biology, pain and pleasure, form the experiential foundation for this field. Pain, particularly chronic pain, affects more than 50 million Americans and incurs annual costs that exceed those of cancer, heart disease, and diabetes combined. Beyond its physical toll, untreated or undertreated pain is associated with long-term psychological consequences including depression, social isolation, substance use disorder, and functional disability.

[0065] The systems described herein may address multiple therapeutic and experiential domains through corresponding embodiments. In the pain relief embodiment, the system may provide direct neural modulation through electrical stimulation, activation of endogenous pain relief pathways through targeted sensory input, and optimization of autonomic balance to reduce pain related stress responses. In the robotic or mechanical embodiment, the system may deliver adaptive massage or pressure stimulation that responds to physiological feedback to relieve muscular tension and enhance circulation. In the somato autonomic embodiment, gentle surface or sub surface stimulation may modulate autonomic outflow to promote healing, visceral regulation, and systemic recovery. In the pleasure embodiment, the system may employ multimodal sensing and feedback to optimize stimulation patterns associated with arousal and reward while maintaining safety and comfort. Across all embodiments, the system can monitor relevant physiological biomarkers and adapt stimulation parameters in real time to maximize effectiveness while minimizing adverse effects.

[0066] Adaptive Stimulation System Architecture

[0067] FIG. 1 is a block diagram of an adaptive stimulation system 100. According to one or more aspects, an adaptive stimulation device 102 may be configured to provide sensory stimulation 116 to a user 114. The adaptive stimulation device 102 represents a paradigm shift from static stimulation devices to intelligent, learning systems that can adapt to individual physiology and optimize outcomes over time.

[0068] The adaptive stimulation device 102 may include one or more sensor systems 104, an integrated feedback mechanism 106, a proxy measurement generator 108, and a controller 110. A userAttorney Docket No. SETH-005 AWO interface 112, configured to receive inputs from a user 114, may be integrated with or connected to the adaptive stimulation device 102. This user interface 112 includes the multi-modal control architecture described below, supporting natural language voice commands, traditional manual controls, and physiological monitoring integration.

[0069] This architecture enables the system to operate across multiple domains: including clinical pain management through devices such as transcutaneous electrical nerve stimulation (TENS) units, spinal cord stimulators (SCS), vagus nerve stimulators (VNS); consumer wellness tools such as percussive massagers, robotic massage systems, and heating pads; and intimacy devices for sexual wellness and arousal enhancement. The modular design allows the same core technology to be adapted for diverse applications while maintaining consistent learning and optimization capabilities.

[0070] As shown in FIG. 2, and further described below) the adaptive stimulation framework highlights the distinction between traditional transcutaneous electrical nerve stimulation (TENS) and the proposed closed-loop architecture. Conventional TENS systems operate in an open loop, delivering fixed stimulation patterns that require manual user adjustment. The adaptive system instead forms a feedback-driven loop in which physiological measurements inform real-time control, allowing stimulation to evolve automatically in response to the user’s state.

[0071] Multi-Modal Control Architecture - Hybrid Intelligence Through Conscious And Unconscious Signal Integration

[0072] Foundational Distinction: Conscious Intent vs. Unconscious Physiological Response

[0073] Aspects of the present disclosure implement a control paradigm that distinguishes itself from existing brain-computer interface (BCI) technologies through its integration of both conscious volitional commands and unconscious involuntary physiological responses. This hybrid intelligence architecture represents a departure from known systems that rely exclusively on either conscious user control or autonomous algorithmic control, but not both in integrated fashion.

[0074] Existing EEG-controlled stimulation devices employ brain-computer interfaces that detect and translate conscious user intent into device commands. These systems monitor imagined movements, focused attention, or deliberate cognitive tasks — voluntary neural activities that users consciously generate to control device behavior. The user must actively think specific thoughts or imagine particular movements to modulate stimulation parameters. This approach suffers from several limitations: (1) cognitive load burden requiring sustained conscious attention during device operation,(2) inability to determine whether the commanded stimulation achieves desired physiological outcomes,(3) susceptibility to distraction and attentional lapses that disrupt control, and (4) requirement for training periods where users learn to generate recognizable mental commands.Attorney Docket No. SETH-005 AWO

[0075] The systems disclosed herein operate on different principles. Rather than detecting conscious intent to control device parameters, the system measures unconscious physiological responses that objectively indicate whether stimulation produces desired therapeutic or hedonic effects. The system monitors autonomic nervous system activity, hemodynamic responses, electrodermal changes, and other involuntary physiological signals that users cannot consciously manipulate or suppress. These signals provide objective ground truth about internal states — pain intensity, pleasure magnitude, stress levels, arousal states — independent of user reporting or conscious awareness.

[0076] The Hybrid Control Framework: Conscious Voice Commands Plus Unconscious Autonomic Signals

[0077] Aspects of the present disclosure may provide for the integration of two distinct and complementary control channels operating simultaneously: conscious control channels and unconscious measurement channels.

[0078] Conscious Control Channel — Natural Language Voice Interface: Users provide conscious high-level commands through natural spoken language, expressing therapeutic intent ("reduce the sharp pain in my shoulder"), preference specifications ("make it gentler"), spatial targeting ("move the stimulation lower"), or safety overrides ("stop immediately"). This conscious channel enables users to communicate goals, preferences, and real-time adjustments without requiring specialized training, complex motor sequences, or sustained cognitive effort. The natural language interface interprets conversational speech to extract user intent, translating semantic content into actionable device instructions.

[0079] Unconscious Measurement Channel — Involuntary Physiological Monitoring:Simultaneously, the system continuously monitors unconscious physiological responses that objectively quantify stimulation effectiveness and internal state changes. These measurements include, for example: heart rate variability patterns reflecting autonomic nervous system balance; galvanic skin response indicating sympathetic arousal; respiratory' rate and depth changes associated with pain or relaxation states; electromyographic signals revealing muscle tension or release; hemodynamic responses (via fNIRS or photoplethysmography) indicating neural activity in affect-processing regions; electroencephalographic patterns including phase-amplitude coupling associated with pain intensity; facial micro-expressions revealing affective responses users cannot voluntarily control; and vocal stress analysis detecting unconscious changes in speech characteristics.

[0080] These unconscious measurements provide feedback that the user cannot fake, suppress, or consciously manipulate. They reveal objective truth about whether stimulation reduces pain, enhancesAttorney Docket No. SETH-005 AWO pleasure, alleviates stress, or produces adverse effects regardless of what users consciously report or believe.

[0081] Arbitration Logic: Synthesizing Conscious and Unconscious Information Streams

[0082] The system, in one aspect, implements sophisticated arbitration algorithms that synthesize information from both conscious and unconscious channels to determine optimal stimulation parameters. This arbitration addresses situations where conscious user commands conflict with unconscious physiological responses:

[0083] Scenario 1 Conscious Request Contradicted by Physiological Distress: User verbally requests "increase intensity" while physiological monitoring reveals elevated pain indicators, sympathetic arousal suggesting distress, or protective reflex activation. The system prioritizes physiological safety signals, implementing gradual intensity increases with continuous monitoring, or refusing unsafe parameter adjustments while explaining the physiological rationale to the user.

[0084] Scenario 2 User Reports Ineffectiveness While Physiology Shows Improvement: User states "this isn't working" while objective measurements indicate pain reduction, improved autonomic balance, or muscle relaxation. The system continues current parameters while providing feedback to the user about measured physiological improvements, recognizing that conscious perception may lag behind physiological changes or that users may lack accurate interoceptive awareness.

[0085] Scenario 3 Converging Signals Enabling Confident Optimization: Conscious feedback aligns with physiological measurements user reports "that feels better" concurrent with improved HRV, reduced electrodermal activity, and relaxed facial expressions. The system recognizes convergent evidence and prioritizes maintaining or refining the current parameter configuration.

[0086] Scenario 4 Unconscious Adaptation in Absence of Conscious Feedback: User provides no verbal feedback but physiological monitoring reveals changing response patterns. The system autonomously explores parameter space to optimize unconscious response metrics, operating effectively even when users are unable or unwilling to provide conscious input.

[0087] Advantages Over Motor Imagery -Based Brain-Computer Interfaces

[0088] In one aspect, the systems described herein provide an approach offering numerous advantages over motor imagery-based BCIs and research systems for prosthetic control:

[0089] Motor imagery BCIs detect imagined movements or conscious cognitive tasks through EEG pattern recognition, translate these voluntary mental activities into device control commands, and require users to actively generate specific neural signatures through focused attention. Such systems also function as input mechanisms translating intent into action, provide no information about whether commanded actions achieve desired outcomes, are susceptible to cognitive fatigue, distraction, andAttorney Docket No. SETH-005 AWO individual variation in imagery capability, and require training periods for users to leam reliable signal generation.

[0090] The hybrid control approach described herein differs from motor-imagery or intent- driven brain-computer interfaces known in existing systems. Conventional BCIs rely on the user’s deliberate generation of distinct electroencephalographic signatures through imagined movements or specific cognitive tasks. They translate consciously produced neural activity into device commands, functioning as input mechanisms analogous to switches or joysticks operated by thought. Such systems demand training, sustained attention, and consistent mental rehearsal, yet they provide no indication of whether the commanded action achieves its intended physiological effect. Their operation deteriorates when user concentration lapses, and their accuracy varies considerably among individuals.

[0091] Aspects of the present disclosure may provide for unconscious physiological monitoring. In one aspect, the system may measure involuntary autonomic responses users cannot consciously control, quantify objective physiological correlates of pain, pleasure, stress, and arousal, operates continuously without requiring sustained user attention or effort, and functions as outcome measurement determining stimulation effectiveness. The systems disclosed herein may also provide ground truth feedback independent of conscious user perception, maintain accuracy regardless of user attention state, cognitive load, or awareness, and require no special training — physiological responses occur automatically. This physiological monitoring provides continuous, objective feedback that guides stimulation in real time, as described below,

[0092] A distinction arises in the direction of information flow and the nature of measured signals. Motor imagery’ BCIs translate conscious user intentions into device commands (input). The present disclosure includes systems that measure unconscious physiological outcomes to assess stimulation effectiveness (output assessment). Motor imagery BCIs detect voluntary neural activities users deliberately generate (controllable signals). The systems disclosed herein monitor involuntary’ autonomic responses users cannot suppress or fake (uncontrollable ground truth).

[0093] Existing closed-loop spinal cord and responsive neurostimulation (RNS) systems measure internal electrical activity such as local field potentials or evoked potentials to control stimulation. These systems adapt to mechanical posture or seizure thresholds, but do not evaluate therapeutic effectiveness in real-time. In contrast, the present adaptive stimulation devices, described herein, interpret autonomic and affective indicators such as heart-rate variability, galvanic skin response, and cortical hemodynamic signals to adjust stimulation based on the user’s physiological outcome rather than on neural spike activity alone. This distinction allows optimization across pain, pleasure, and autonomic domains beyond the scope of existing neural feedback devices.Attorney Docket No. SETH-005 AWO

[0094] This hybrid intelligence foundation supports multiple operating configurations, later illustrated in FIG. 3E, enabling physiological-only, voice-guided, or collaborative control as appropriate to context and hardware.

[0095] Advantages of Multi-Modal Architecture

[0096] As described herein, the multi-modal control architecture may provide several critical advantages over known stimulation and feedback technologies:

[0097] Closed-Loop Systems Without Effectiveness Measurement: Existing adaptive devices such as posture-responsive spinal cord stimulators implement closed-loop control but measure only mechanical context (body position, movement) rather than therapeutic outcomes (pain relief, symptom improvement). The systems described herein may close the loop around actual therapeutic effectiveness, measuring physiological indicators of pain, pleasure, and internal state changes.

[0098] Brain-Computer Interfaces Without Outcome Assessment: EEG-controlled devices translate conscious intent into commands but provide no information about whether those commands achieve desired effects. The systems described herein may measure objective outcomes, enabling iterative refinement toward optimal parameter configurations rather than merely executing user- specified commands.

[0099] Manual Control Requiring Sustained User Attention: Traditional stimulation devices require continuous conscious monitoring and manual adjustment, imposing cognitive burden during periods when users seek relief or relaxation. The hybrid architecture described herein may enable users to specify high-level goals through brief verbal commands while automated physiological monitoring handles continuous optimization, reducing cognitive load while maintaining user agency.

[0100] Single-Modality Feedback Susceptible to Artifact and Ambiguity: Prior attempts at physiological feedback typically rely on single measurement modalities susceptible to motion artifacts, environmental interference, or ambiguous interpretation. The multi-modal architecture of the present disclosure fuses information across multiple physiological channels and conscious verbal feedback, implementing robust sensor fusion that maintains accuracy despite individual channel limitations.

[0101] Fixed Control Paradigms Ignoring Individual Differences: Existing systems enforce uniform control interfaces regardless of user capability, preferences, or context. The multi-modal architecture described herein adapts control paradigms to individual users enabling fully autonomous operation for users unable to provide conscious input, conversational interaction for users preferring active control, or balanced hybrid operation combining both channels.

[0102] Technological Enablement Through Integrated Natural Language and Physiological ProcessingAttorney Docket No. SETH-005 AWO

[0103] According to one or more aspects, the multi-modal control architecture may utilize integration of several sophisticated technologies operating in coordination, including natural language understanding systems, multi-channel physiological signal processing, arbitration and decision logic, and bidirectional communication interfaces.

[0104] Natural Language Understanding Systems: Systems described herein may include advanced speech recognition engines capable of handling varied acoustic environments, speaking styles, and the altered speech characteristics of users experiencing pain or arousal. They may also include intent classification algorithms specifically trained to extract therapeutic and hedonic intent from conversational speech as well as medical and anatomical vocabulary recognition for symptom descriptions and spatial targeting. Emotional prosody analysis may be used to supplement semantic content with affective state information.

[0105] Multi-Channel Physiological Signal Processing: Such processing may include real-time acquisition and processing of diverse physiological signals with appropriate sampling rates, noise filtering, and artifact rejection. Feature extraction algorithms may compute meaningful physiological metrics from raw sensor data HRV frequency domain analysis, electrodermal response peak detection, EMG amplitude envelope computation, and respiratory pattern analysis. Sensor fusion algorithms may combine information across modalities to compute robust effectiveness metrics less susceptible to individual channel artifacts.

[0106] Arbitration and Decision Logic: Sophisticated algorithms may weigh evidence from conscious verbal feedback against unconscious physiological measurements, implementing safety prioritization while respecting user preferences. Conflict resolution strategies may determine appropriate actions when conscious requests contradict physiological responses. Learning systems may adapt arbitration strategies to individual users based on historical correlations between verbal feedback and physiological measurements.

[0107] Bidirectional communication interfaces may include audio, visual, and haptic channels that provide information to and receive information from the user. Audio output capabilities may enable the system to explain decisions, report physiological findings, request clarification, and provide feedback about measured effectiveness. Visual interfaces may display real time physiological metrics, effectiveness indicators, and system status information. Haptic feedback options may convey information through tactile channels when visual or auditory feedback may be distracting. In some aspects, the system may also use these communication channels to influence the feeling or physiological state of the subject by presenting sound, light, or tactile patterns that promote relaxation, focus, or arousal consistent with the intended therapeutic or experiential outcome.Attorney Docket No. SETH-005 AWO

[0108] Learning the Connection Between Stimulation and Sensation

[0109] Conventional adaptive stimulation devices have focused on recognizing or classifying physiological states detecting pain, arousal, or stress without learning how different patterns of stimulation alter those states. Their use of machine learning typically ends at recognition, rarely extending to calibration of stimulation itself, particularly for non-invasive or non-neural modalities. The adaptive stimulation systems described herein may be built around a different principle, namely learning the relationship between what is applied and what is experienced. The system treats every stimulation event as a data point, exploring and mapping the range of possible frequencies, amplitudes, waveforms, pressures, rhythms, and thermal or mechanical patterns in relation to what can be measured and what is felt. This approach transforms stimulation from a fixed prescription into a process of continual discovery7in which the device leams empirically how stimulation modifies perception and physiology. In one aspect, the learning process is designed to develop generalized relationships rather than narrow responses, allowing the system to adapt to variations in user state, condition, and environment without losing stability7. By abstracting patterns of response across multiple sessions, the system builds a resilient model that can guide individualized therapy for a particular complaint or circumstance while maintaining reliable performance over time. These adaptive discoveries accumulate across sessions, enabling progressive refinement of stimulation strategies as the system leams how specific parameter changes influence measured physiological and affective outcomes.

[0110] While the architecture allows direct connection to sophisticated neuro-sensory instruments such as EEG, fNIRS, or functional ultrasound, practical deployment often relies on accessible proxy sensors. To accommodate this reality, the framework defines a tiered learning architecture that functions across multiple levels of measurement sophistication from direct neural sensing to proxy-only operation using population- and archetype-trained models. Each tier preserves the same capacity to explore, learn, and optimize stimulation based on measured effectiveness.

[0111] This design allows the adaptive stimulation system to operate not only as a controller but also as a continuous experimenter, learning through use how different stimulations influence internal state and subjective experience. It unites the measurement of effectiveness with the discovery of effective stimulation a capability absent from prior closed-loop or Al-based devices.

[0112] According to one or more aspects, this framework may be formalized or implemented according to four structured learning and sensing modes that combine the source of runtime sensing with the method of model initialization. These modes, illustrated in FIGS. 3A-3C, demonstrate how the adaptive stimulation system can learn across laboratory, clinical, and consumer environments while maintaining a consistent feedback and control logic.Attorney Docket No. SETH-005 AWO

[0113] Multiple Proxy Modes for Different Embodiments

[0114] Existing “Al-driven” stimulation devices typically apply machine learning to tune stimulation parameters frequency, amplitude, waveform after a clear performance metric is already available. In such systems the model learns how to stimulate, but assumes that what is being measured (pain relief, arousal, relaxation) is already a direct and reliable signal. The adaptive stimulation system described here, therefore, expands the learning objective itself. Rather than presuming that the measurement of effectiveness is fixed or known, the system first learns which signals accurately represent effectiveness and then learns how stimulation parameters influence those signals. To accomplish this, the architecture supports four complementary proxy modes of sensing and learning: (1) Direct-Neural, (2) Proxy -Direct Calibration, (3) Population-Model Calibration, and (4) Archetype- Assisted Quick Calibration. High-fidelity neural or hemodynamic measurements such as EEG, fNIRS, fUS, or photoacoustic imaging are used for calibration only and not for continuous feedback, after which the system operates entirety on accessible proxy sensors. Each embodiment TENS, Robotic Massage, Somato- Autonomic, and Pleasure operates in whichever proxy mode fits its sensors and deployment her. These proxy modes replace the purely direct-neural feedback loops described in earlier neural-interface art with a functional, multimodal feedback hierarchy. Calibration with high-fidelity references (e.g., EEG, fNIRS, functional ultrasound) is optional and confined to training; routine operation uses consumer-grade proxies only, with no invasive or direct-neural sensing required.

[0115] According to one or more aspects of the disclosure these relationships may be implemented or formalized as four structured learning and sensing modes, each representing a distinct balance between sensing fidelity, personalization, and deployment scalability.

[0116] The present disclosure addresses a deeper prerequisite problem: learning how to measure and interpret effectiveness itself. Because physiological correlates of pain, pleasure, and autonomic balance vary among individuals and cannot be captured by any single biomarker, the system must first establish the mapping between accessible biosignals and validated, high-fidelity measurements before parameter optimization can occur.

[0117] As shown in FIG. 3, and described further below, the adaptive stimulation framework organizes learning and optimization across sessions through a closed-loop process. Feedback data from each stimulation cycle are analyzed, model parameters are updated, and refined settings are applied in subsequent sessions. This cross-session learning structure provides the foundation for the multi-mode hierarchy illustrated in FIGS. 3A-3C.

[0118] To accomplish this, the disclosed systems and architectures employ a hierarchical learning framework that bridges direct neurophysiological sensing with cost-effective proxyAttomev Docket No. SETH-005 AWO measurement. As shown in FIG. 3 and detailed in FIGS. 3A-3C, the system progressively links high- fidelity' neural and hemodynamic data to accessible biosignals through structured learning and deployment modes. This scalable architecture enables the same adaptive feedback and control logic to operate consistently across research, clinical, and consumer environments. Unlike prior machine learning control systems that optimize stimulation relative to a predefined outcome metric, the present framework first learns the meaning of effectiveness itself by establishing its physiological basis from ground-truth neural data and extending that understanding to proxy-only operation. The controller operates across four learning and sensing modes that combine the source of runtime sensing with the basis of model initialization.

[0119] Latency -linked operation across these layers is depicted in FIG. 3A through FIG. 3C, which collectively illustrate the hierarchical feedback architecture used by the adaptive stimulation system. According to one aspect, FIG. 3A shows a mapping of the learning and sensing scenarios, FIG. 3B illustrates the corresponding training and runtime flows, and FIG. 3C shows the shared multitimescale control structure that coordinates fast-safety, in-session, and cross-session learning loops. A summary' table of the same modes is provided in FIG. 3D, which consolidates the definitions and operational relationships described below. Latency -linked operation across these layers is illustrated in FIG. 3D. which shows a generalized multi-timescale feedback architecture used by the adaptive stimulation system. The architecture includes three concurrent control loops that operate on different time scales. A fast-safety' loop (375) monitors immediate physiological feedback and prevents adverse responses. An in-session optimization loop (376) adjusts stimulation parameters in real time to maintain effectiveness during use. A cross-session learning loop (377) aggregates effectiveness data over hours or weeks to update user profiles and population models. These loops form the hierarchical framework that supports adaptive control across all embodiments described herein..

[0120] According to one aspect, FIG. 3 A is a two-by-two map of learning and sensing modes used by the adaptive stimulation system. Columns indicate runtime sensing (direct-neural sensing in the loop 315 versus proxy-only sensing 316). Rows indicate model initialization (population-informed model plus per-user calibration 317 versus per-user calibration only 318). Each quadrant defines its runtime sensing method, training approach, and primary strengths and weaknesses. Modes 3A-l(a / b) represent direct-neural closed-loop operation with or without population priors. Mode 3A-2 represents the proxy-direct user-paired mode. Mode 3A-3 represents the population-informed proxy mode. Mode 3A-4 represents the archetype-assisted quick-calibration variant of mode 3A-3.F1G. 3B shows illustrative training and runtime pathways for the four learning and sensing modes of the adaptive stimulation system. The direct-neural mode (3A-1) operates with neural sensing in the loop. The proxy-Attorney Docket No. SETH-005 AWO direct mode (3A-2) learns a user-specific mapping of proxies to user state from paired neural and proxy sessions and then runs on proxies. The population-model and archety pe-assisted modes (3A-3 and 3A-4) initialize from cohort or archetype models, complete short calibrations, and operate with proxy sensing for runtime adaptation and cross-session learning. According to one aspect, the direct neural loop mode (3A-1) uses direct neural or hemodynamic measurements (e.g., EEG, fNIRS, fUS), shown at block 330, in the control loop for immediate optimization. An effectiveness metric (E) may be computed to quantify pain / pleasure states from neural features, shown in block 332. Stimulation parameters may be adapted by updating frequency, amplitude, waveform, and location in real time to minimize pain or maximize pleasure, shown in block 334. Shown in block 336, log session data and / or an update policy may store the effectiveness metric versus parameter pairs for cross-session learning. As shown in block 338, cross-session optimization may refine the model using aggregated sessions.

[0121] According to one aspect, the proxy-direct mode (3A-2) may collect direct neural and proxy signals on the same user during stimulation, shown in block 340. A neuro-ground-truthed model may map proxy features to user state (pain or pleasure) as persistent training, shown in block 342, and subsequently operate as proxy -only, shown in block 344. The mode may further infer the user state by computing the current effectiveness metric, shown in block 346. As shown in block 348, stimulation may be adapted by adjusting the parameters to improve the effectiveness metric.

[0122] In one aspect, the Population Model with Individual Calibration mode (3A-3) initializes from population-trained models using neuro + proxy datasets, shown in block 350, and personalizes through short voice- or task-based calibration, shown in block 352. As shown in block 354, the mode may deploy a personal proxy model that may run proxy-only monitoring in real-time. The mode may infer a user state and adapt stimulation according to the controller using the effectiveness metric to tune the system parameters, shown in block 356. As a function of cross-session learning, a personal policy may be updated and encry pted gradients may be uploaded to population model (e.g., federated learning).

[0123] In one aspect, an archetype-assisted quick calibration mode (3A-4) may start, as shown in block 360, from choosing an archetype cluster matched to user features, loading the archetype model, shown in block 362, and confirms fit with a short lock-in sequence that applies a brief stimulus sw eep, shown in block 364. As shown in block 366, in run-time, the mode may operate similarly to the population model with individual calibration mode (3A-3) with continuous optimization. Shown in block 368, if measured responses deviate from an expected level, a better archetype may be assigned, or alternatively, a full individual calibration may be performed.

[0124] Referring now to FIG. 3C, a schematic of the multi-timescale control architecture is shown. Three nested loops operate concurrently: a fast-safety loop 375 that monitors immediateAttorney Docket No. SETH-005 AWO biosignals and prevents adverse responses; an in-session optimization loop 376 that adjusts stimulation parameters in real time to maximize the effectiveness metric E; and a cross-session learning loop 377 that aggregates session data to update personal and population models through federated learning. This architecture applies to all four modes (3A-1 through 3A-4) and underlies the embodiments described herein, including the TENS, robotic massage, somato-autonomic, and pleasure systems. Multi-Modal Control and Natural Language Interface

[0125] Despite pleasure being one of the most sought-after aspects of life — it is poorly measured, regulated, or therapeutically integrated. Experiences of pleasure play a critical role in recovery, bonding, intimacy, and emotional resilience. Whether in the form of arousal, comfort, satisfaction, or meditative focus, the capacity to generate and sustain pleasurable states has profound consequences for mental and physical well-being.

[0126] Aspects of the present disclosure include systems that provide pleasure enhancement through sophisticated understanding of neurochemical and physiological pleasure mechanisms: dopaminergic reward pathway activation, oxytocin release through appropriate stimulation patterns, endogenous opioid system engagement, and autonomic nervous system optimization for pleasure states. Yet, the tools intended to stimulate or enhance pleasure, such as massage devices, vibratory interfaces, and sexual wellness products, operate without reference to any physiological indicators of success.

[0127] Aspects of the present disclosure include bridge this gap by providing real-time (seconds) and / or near-real-time (tens of seconds to minutes, but during stimulation) measurement and optimization of pleasure-related and not-pleasure-related physiological responses, enabling devices to automatically discover and maintain optimal stimulation patterns for individual users. Not-real-time (> minutes) analyses can be used to predict stimulus conditions in future stimulation sessions.

[0128] Multi-Modal Control Architecture and Natural Language Interface

[0129] Real-Time Audio Feedback Loop Architecture

[0130] In one aspect, the system incorporates a multi-modal control paradigm that combines traditional physiological monitoring with advanced natural language audio interface capabilities. This enables users to interact with stimulation devices through natural speech while the system simultaneously monitors physiological responses to create an optimal feedback loop that responds to both conscious intent and unconscious biological responses.

[0131] In one aspect, high-quality audio input is provided through integrated microphones, external microphone arrays, or smartphone-based audio capture. The audio subsystem applies noisereduction, echo-cancellation, and automatic gain-control algorithms to maintain reliable signal quality across a wide range of acoustic environments and user conditions. These safeguards ensure accurateAttorney Docket No. SETH-005 AWO voice capture even when the user is in pain, under stress, or in intimate contexts where speech patterns may deviate from normal cadence or intensity7.

[0132] Spoken input is then processed by an automatic speech-recognition (ASR) engine that converts voice data into structured text for interpretation by the controller. The ASR module employs robust algorithms capable of adapting to different accents, intonations, and the speech characteristics associated with pain, pleasure, or emotional arousal. To enhance accuracy, the recognition vocabulary7includes medical terminology, anatomical references, and language patterns relevant to therapeutic and pleasure-enhancement applications. This combination of precise audio capture and domain-specific language modeling allows the system to interpret user intent quickly and reliably, ensuring seamless interaction between human instruction and physiological feedback within the adaptive stimulation framework.

[0133] Aspects of the present disclosure include a natural language processing (NLP) system specifically7designed to interpret therapeutic and experiential intent from conversational speech during active stimulation sessions. The system employs multiple classification approaches to extract key information from user statements:

[0134] Pain Characteristics Classification: The system identifies pain descriptors including intensity indicators ("mild," "severe." "excruciating"), quality descriptors ("sharp," "dull," "aching," "burning," "tingling"), temporal patterns ("constant," "intermittent," "throbbing"), and location specifications ("shoulder," "lower back," "neck," "deeper," "surface").

[0135] Pleasure Enhancement Recognition: The NLP system categorizes pleasure-related requests including intensity preferences ("gentler," "stronger," "more pressure"), rhythm adjustments ("slower," "faster," "steady"), spatial targeting ("move left," "right there," "lower"), and emotional preferences ("relaxing," "exciting," "build up").

[0136] Real-Time Parameter Commands: The system interprets immediate adjustment requests including "harder," "softer," "stop," "that's perfect," "too much," "not enough," "move it," and userspecific vocabulary learned through interaction history.

[0137] Multi-Modal Operation Modes

[0138] Aspects of the present disclosure include providing one or more multi-modal operation modes include, a physiological-only mode, an audio-only mode, and a hy brid mode.

[0139] In one aspect, the physiological-only mode may provide sensor-based feedback operation where the system relies entirely on measured physiological responses (heart rate variability, galvanic skin response, facial expressions, EEG patterns) to optimize stimulation parameters without audio input.Attorney Docket No. SETH-005 AWOThis mode is suitable for users who prefer silent operation or situations where voice interaction is impractical.

[0140] In an audio-only mode: voice command control without physiological monitoring, enabling users to maintain complete conscious control over their experience while benefiting from natural language interaction. This mode is particularly valuable for users who prefer to maintain agency over their treatment or intimate experience while enjoying hands-free control.

[0141] In a hybrid mode, the most sophisticated operating mode, combining real-time audio feedback with continuous physiological monitoring. The system employs intelligent arbitration algorithms to resolve conflicts between audio commands and physiological indicators, learning individual user patterns and preferences for optimal integration.

[0142] Hybrid Audio-Physiological Arbitration Logic

[0143] According to one or more aspects, when a conflict between audio commands and physiological feedback arises, the system may rely on a conflict resolution strategy. For example, when audio commands conflict with physiological feedback (e.g., user says "harder" while stress indicators suggest overstimulation), the system employs learned user-specific patterns to determine appropriate responses. Default behavior prioritizes user agency through voice commands while maintaining safety overrides for potentially harmful situations.

[0144] As shown in FIG. 3E, the hybrid audio and physiological control architecture integrates conscious user intent with unconscious physiological feedback within a unified adaptive framework. Audio input 386 provides high level user commands and preferences, while physiological inputs 388 convey involuntary biosignals that reflect the user’s actual state. These inputs are first evaluated through a safety gate 384 that filters unsafe or contradictory instructions and ensures operation within defined limits. The filtered data are then processed by an arbitration engine 382 that determines final control actions based on context, historical patterns, and safety constraints. The controller 380 executes the resulting commands and communicates with a learning module 390 that provides long term personalization and federated model updates. This configuration enables the system to maintain user agency through voice interaction while preserving physiological safety and adaptive intelligence. The arbitration system considers the current application context (pain management vs. pleasure enhancement), user historical patterns, session duration, and safety parameters when resolving audio- physiological conflicts. For pain management, the system may prioritize physiological safety indicators, while for pleasure enhancement, user voice commands may take precedence.

[0145] The system continuously leams from successful audio-physiological combinations, building user-specific models that improve arbitration decisions over time. Machine learning algorithmsAttorney Docket No. SETH-005 AWO identify patterns where users' verbal feedback aligns with or contradicts their physiological responses, enabling more accurate future arbitration.

[0146] Caregiver and Third-Party Control Implementation

[0147] The system also provides secure mechanisms for caregiver and third-party control. It incorporates voice-recognition and authorization functions that allow caregivers, healthcare providers, or intimate partners to issue audio commands on behalf of users who cannot communicate effectively. Biometric voice authentication ensures that only authorized individuals can assume control or override user settings, preserving both safety and privacy.

[0148] Wireless communication systems enable caregiver control from remote locations, particularly valuable for elderly or disabled users who require assistance with device operation. Healthcare providers can monitor and adjust treatments remotely while maintaining communication with patients.

[0149] For pleasure enhancement applications, the system supports partner voice input to guide stimulation based on observed responses and intimate communication. This creates a three-way interaction between user, partner, and device, similar to natural intimate interactions where partners respond to both verbal and non-verbal cues.

[0150] Application-Specific Audio Interaction Patterns

[0151] The system supports application-specific audio interaction patterns that adapt voice control and feedback to different therapeutic and experiential contexts. In pain-management applications, spoken commands convey spatial targeting, intensity adjustment, and qualitative feedback, allowing the user to say things such as "The pain is more toward my shoulder blade,'’ "‘That's helping but 1 need it stronger,” or “The sharp pain is becoming more of an ache.” As treatment progresses, the user can report improvement through natural phrases like “It’s getting better, keep doing what you’re doing,” enabling the system to correlate subjective reports with physiological changes.

[0152] For pleasure-enhancement applications, the audio interface interprets expressive, rhythmic, and spatial guidance cues such as “Slower build-up, then faster,” “More pressure but gentler rhythm,” or “Perfect spot, stay right there.” Progression feedback like “I’m getting close, don’t change anything” allows the device to synchronize stimulation patterns with the user’s evolving physiological state, optimizing timing and intensity dynamically.

[0153] In clinical or caregiver-assisted contexts, the voice interface supports multi-party operation and safety oversight through commands including “Patient appears uncomfortable, reduce intensity,” “Focus on the lower lumbar region,” or “Patient reports fifty' percent pain reduction.” Immediate intervention commands such as “Stop stimulation immediately” override all automatedAttorney Docket No. SETH-005 AWO control, ensuring safety while maintaining seamless verbal coordination between user, caregiver, and device.

[0154] Safety Integration and Override Systems

[0155] The system integrates several safety and override mechanisms to ensure reliable operation under all circumstances. All audio commands are processed through safety-validation routines that prevent harmful stimulation levels regardless of user requests. Upper and lower bounds are continuously maintained for every stimulation parameter, and emergency stop commands receive the highest processing priority to guarantee immediate response.

[0156] Automatic detection of user non-responsiveness triggers safety protocols, including stimulation cessation and emergency notification systems. The system monitors for verbal responses and can detect when users become unable to provide feedback.

[0157] Specific voice commands ("stop." "emergency," "help") trigger immediate cessation of all stimulation with rapid response times. These commands are recognized even in emergency situations where users may have difficulty speaking clearly.

[0158] Integrated Feedback Mechanism and Real-Time Adaptation

[0159] Returning to FIG. 1, and as previously described, aspects of the present disclosure provide an architecture that enables the system to operate across multiple domains, for example: including clinical pain management through devices such as transcutaneous electrical nerve stimulation (TENS) units, spinal cord stimulators (SCS), vagus nerve stimulators (VNS); consumer wellness tools such as percussive massagers, robotic massage systems, and heating pads; and intimacy devices for sexual wellness and arousal enhancement. The modular design allows the same core technology to be adapted for diverse applications while maintaining consistent learning and optimization capabilities.

[0160] As detailed further herein, the integrated feedback mechanism 106 may adjust the applied stimulation 116 based on real-time and / or near-real-time and predictive effectiveness measurements obtained from the sensor systems 104. employing both sophisticated neuroimaging modalities and cost- effective proxy measurements to achieve scalable deployment while maintaining clinical-grade accuracy in affective state detection.

[0161] The feedback mechanism 106 operates on multiple timescales simultaneously: immediate safeN monitoring (milliseconds to seconds) to detect and prevent adverse responses; short-term optimization (seconds to minutes) to adjust current session parameters for maximum effectiveness; medium-term adaptation (sessions to weeks) to learn individual patterns and preferences; and long-term population learning (months to years) to improve system performance across all users.Attorney Docket No. SETH-005 AWO

[0162] The feedback mechanism 106 enables the device to explore the multidimensional parameter space of stimulation variables — including amplitude, frequency, waveform characteristics, spatial location, and temporal patterns — to discover optimal configurations for individual users and adapt to changing physiological conditions over time.

[0163] Closed-Loop Framework and Real-Time Operation

[0164] Known clinical and consumer products lack an integrated framework that: (1) measures the user's real-time and / or near-real-time physiological response(s) to stimulation using validated biomarkers that can reliably distinguish between therapeutic effectiveness and ineffectiveness; (2) learns over time what responses correspond to desirable outcomes (such as pain reduction or increased arousal) through machine learning algorithms that can identify subtle patterns across multiple physiological channels and sessions; and (3) adjusts stimulation parameters — such as frequency, waveform, intensify, or spatial location — accordingly using optimization algorithms that can navigate complex, highdimensional parameter spaces while maintaining safety constraints and user preferences.

[0165] Control follows the multi -timescale hierarchy previously defined (FIGS. 3D and 10), coordinating fast-safety, in-session, and cross-session adaptation.

[0166] Comprehensive Sensor Systems and Measurement Hierarchy

[0167] According to one aspect, the sensor system 104 may obtain and collect data across a hierarchy of measurement sophistication, accessibility, and cost-effectiveness. This hierarchical approach enables the system to achieve clinical-grade accuracy using consumer-grade sensors by leveraging training data from high-fidelity measurement systems.

[0168] High fidelity measurements serve as the foundation for proxy model training. Functional magnetic resonance imaging (fMRI) provides high spatial resolution mapping of brain region activation related to pain circuits and reward systems but is limited by size, cost, and latency. Functional near infrared spectroscopy (fNIRS) offers greater portability with moderate spatial resolution and has demonstrated feasibility for detecting certain types of pain. For example, it may be known that using fNIRS with deep learning models may give approximately ninety one percent accuracy in distinguishing pain conditions. The systems described herein integrate fNIRS or equivalent high fidelity measurements within a closed loop stimulation system for real time parameter optimization. Recent advances in fNIRS technology have produced wearable systems capable of continuous monitoring with wireless data transmission and real-time signal processing, making them suitable for integration into consumer devices.Attorney Docket No. SETH-005 AWO

[0169] Recent advances in fNIRS technology have produced wearable systems capable of continuous monitoring with wireless data transmission and real time signal processing, making them suitable for integration into consumer devices.

[0170] Referring now to FIGS. 4-6 representative architectures for implementing proxy measurement and learning systems used in adaptive stimulation devices are provided. These figures illustrate how accessible physiological sensors can generate proxy signals, how those signals are correlated with high fidelity reference data during training, and how resulting models are deployed within scalable system configurations.

[0171] FIG. 4 illustrates a proxy measurement control architecture integrating proxy sensors, inference model, and controller for real time adaptation. FIG. 5 illustrates a proxy model training configuration in which high fidelity measurements such as EEG or fNIRS are paired with proxy sensors during supervised learning to create predictive mappings. FIG. 6 illustrates an alternative system configuration showing how these trained models can be deployed across different processing environments including on-device, edge, and cloud based implementations. Various sensor modalities suitable for use in the adaptive stimulation device exhibit different characteristics, making them more or less appropriate for particular embodiments described herein. Some sensors provide high-fidelity neural or hemodynamic data, while others offer accessible physiological or behavioral measures better suited for consumer deployment. More detailed examples of these modalities and their corresponding measurement parameters are presented later in this specification.

[0172] EEG Integration and Neural Biomarkers

[0173] Electroencephalography (EEG) v\i th phase amplitude coupling (PAC) analysis offers excellent temporal resolution. Prior research has reported accuracies exceeding eighty five percent for classifying certain pain states under controlled conditions using EEG derived features. In this embodiment, EEG derived signals may be incorporated as optional inputs to a closed loop adaptive stimulation system that adjusts therapeutic parameters in real time, distinguishing this approach from measurement only biomarker applications.

[0174] PAC analysis provides a way to quantify how lower frequency oscillations modulate the amplitude of higher frequency activity, which has been associated with cognitive and affective processing. In some aspects, dry electrode headbands or caps can provide usable EEG for inference when signal quality is sufficient and appropriate artifact handling and calibration are applied. Performance may vary by use case and environment. The system therefore treats EEG as a supplemental or calibration modality rather than a sole clinical diagnostic source.Attorney Docket No. SETH-005 AWO

[0175] Advanced EEG analyses implemented by the system may include time frequency analysis of spectral power across multiple bands and inter regional coherence, event related potential detection for stimulus response correlation, and machine learning classification of multi channel patterns to distinguish among pain, pleasure, and neutral states. Emerging Neuroimaging Modalities

[0176] Emerging modalities include functional ultrasound (fUS), which presents a compromise offering better spatial accuracy than EEG and better latency than fMRI at lower cost, with prototypes suggesting portable fUS scanners could be developed for field or clinic use. Recent advances in ultrasound technology have produced handheld devices capable of brain hemodynamic monitoring with spatial resolution approaching that of fMRI and temporal resolution suitable for real-time and / or near- real-time feedback applications.

[0177] Photoacoustic imaging has been shown in early -stage studies to localize and quantify nociceptive responses with high spatial resolution, providing accurate mapping of pain-related blood oxygenation patterns. This technique combines the depth penetration of ultrasound with the contrast sensitivity of optical imaging, potentially enabling precise measurement of pain-related physiological changes in both superficial and deep tissue structures.

[0178] Functional interferometric diffusing wave spectroscopy (fiDWS) represents an emerging technology with particular promise for real-time and / or near-real-time monitoring of intracellular dynamics associated with neural activity, offering millisecond temporal resolution with high sensitivity to functional changes.

[0179] Autonomic and Physiological Monitoring

[0180] The sensor system 104 may further include autonomic nervous system monitoring through heart rate variability (HRV) analysis, which is a well-established marker of parasympathetic nervous system engagement and is often used to infer states of relaxation or stress. Advanced HRV analysis implemented in the system includes: time-domain measures (RMSSD, pNN50), frequencydomain analysis (LF / HF ratio, total power), and nonlinear measures (sample entropy, detrended fluctuation analysis) that can distinguish between different types of autonomic responses associated with pain, pleasure, stress, and relaxation.

[0181] Galvanic skin response (GSR) tracks changes in skin conductance that increase in response to heightened sympathetic activity, which may reflect pain, anxiety, or arousal. The system implements sophisticated GSR analysis including: tonic level measurement for baseline autonomic state, phasic response detection for event-related changes, and pattern recognition algorithms that can distinguish between different types of emotional and physiological arousal based on response characteristics and temporal patterns.Attorney Docket No. SETH-005 AWO

[0182] Additional measurements include respiratory pattern detection using strain gauges or impedance measurement, blood pressure monitoring with beat-to-beat variability' analysis, skin temperature measurement with spatial and temporal pattern recognition, muscle tension assessment via electromyography (EMG) with multi-channel recording and pattern classification, body temperature regulation monitoring, electrocardiogram (ECG) patterns with advanced arrhythmia detection and morphology analysis, breathing patterns with depth and rhythm analysis, and blood composition analysis for biomarkers such as epinephrine and endorphins using non-invasive optical techniques.

[0183] Behavioral and Expression-Based Measurements

[0184] Behavioral and expression-based measurements may include facial expression analysis through computer vision systems trained on validated facial action unit databases such as the Facial Action Coding System (FACS), with real-time recognition of pain-related expressions (brow lowering, eye closure, mouth opening) and pleasure-related expressions (genuine smiles, relaxed features, positive micro-expressions).

[0185] Measurements may also be made using vocal pattern recognition and tremor detection using machine learning algorithms trained on large databases of vocal stress indicators, including fundamental frequency analysis jitter and shimmer measurement, and spectral analysis of vocal patterns associated with pain, pleasure, and emotional states. This vocal analysis capability integrates seamlessly with the natural language interface to provide both conscious intent interpretation and unconscious stress / pleasure detection from vocal characteristics.

[0186] Body posture and movement tracking may be measured through inertial measurement units (IMUs) with multi-axis accelerometry and gyroscopy, enabling detection of protective posturing, movement restrictions associated with pain, and postural changes associated with relaxation or pleasure states.

[0187] Pupil dilation measurement may also be made using standard cameras w ith appropriate image processing algorithms, providing a reliable indicator of autonomic arousal that can distinguish between different types of stimulation responses. Sympathetic activity leads to pupillary dilation and parasympathetic activity leads to pupillary constriction.

[0188] Measurements may also be made using micro-expression detection for subtle emotional state changes using high-speed cameras and advanced computer vision algorithms capable of detecting facial movements lasting as little as 1 / 25th of a second.

[0189] Facial expression measurement and analysis hether through surface EMG, camerabased landmark tracking, or infrared microexpression detection may correlate with both hedonic and aversive responses with accuracy exceeding 90% for basic emotional state classification.Attorney Docket No. SETH-005 AWO

[0190] Sensor Integration and Flexible Electrode Systems

[0191] The sensor system 104 may be designed and configured to be non-intrusive or non- invasive and integrate seamlessly with the device 102 for optimal user comfort, using leads, electrode pads, wireless sensors, wearable devices, smartphone-integrated sensors, and ambient monitoring technologies. Many of these signals can be gathered unobtrusively using sensors already embedded in smartphones, wearables, or lightweight headgear, making scalable deployment feasible without requiring specialized hardware.

[0192] The system supports multiple sensor integration approaches: embedded sensors within stimulation devices themselves, external wireless sensors that communicate with the main device, smartphone and smartwatch integration using existing consumer hardware, and ambient monitoring systems that can operate without direct user contact.

[0193] Sensor-fusion algorithms combine data from multiple sources to create robust measurements that are less susceptible to individual sensor failures or artifacts, while privacy-protection mechanisms ensure that sensitive physiological data is processed securely and user preferences for data sharing are respected.

[0194] Representative sensors and their corresponding measurement parameters are discussed below in connection with FIG. 14. The table of FIG. 14 provides exemplary autonomic, somatic, neural, behavioral, and environmental sensing modalities (for example, sensor types 1402-1422) and identifies the physiological or contextual signals they measure, their primary role in the adaptive system, and how each contributes to real-time or near-real-time feedback for stimulation optimization.

[0195] Enhanced Electrode Array System and Spatial Optimization

[0196] Flexible Multi-Array Electrode Pad System

[0197] Building upon foundational sensor systems, aspects of the present disclosure incorporate an innovative flexible multi-array electrode pad system that addresses limitations in traditional stimulation device electrode placement. This system enables dynamic spatial optimization without requiring physical repositioning, representing a significant advancement over static four-pad configurations.

[0198] Design Objectives and Functional Requirements

[0199] The electrode array surface is designed to overcome limitations of traditional electrode placement by providing a flexible, intelligent stimulation surface that can dynamically select optimal electrode pairs without requiring physical repositioning. The primary objectives are: (1) enabling discovery of optimal stimulation locations within a large anatomical area; (2) providing multiple independent electrode pairs that can operate simultaneously; (3) allowing dynamic reconfiguration ofAttorney Docket No. SETH-005 AWO electrode pairs based on real-time and / or near-real-time effectiveness feedback; and (4) maintaining electrical isolation and safety characteristics required for transcutaneous electrical stimulation.

[0200] Multi-Array Architecture and Pair-Based Operation

[0201] The flexible multi-array electrode pad comprises a biocompatible substrate measuring typically 15cm x 20cm or larger, capable of covering major anatomical regions such as the shoulder complex, lumbar region, or upper trapezius muscles. The substrate incorporates a grid-based architecture with multiple stimulation points arranged in a regular matrix pattern, typically implementing an 8x10 or 10x12 array yielding 80-120 potential stimulation locations.

[0202] While the given dimensions and locations described above provide one example implementation without limitation, one skilled in the art will recognize that the sizes and number of stimulation locations provide are merely illustrative and additional sizes and quantities are possible without deviating from the scope of the disclosure.

[0203] In one aspect, the system operates on a pair-based architecture where stimulation occurs between two points to complete the electrical circuit, consistent with TENS operational principles. Multiple electrode pairs can operate simultaneously, with each pair independently controlled for current amplitude, waveform characteristics, and temporal patterns. The central control system can dynamically select any combination of points to form active pairs, enabling systematic exploration of different electrode configurations within the coverage area.

[0204] Diagnostic Stimulation for Precise Targeting

[0205] The system incorporates a diagnostic stimulation capability that enables precise identification of treatment areas before therapeutic intervention. This approach addresses the common challenge in conditions such as lateral epicondylitis (tennis elbow), where the precise location of pathology may vary from the typical anatomical landmarks.

[0206] The system can deliver diagnostic pulses at intensities above therapeutic levels to systematically probe anatomical regions and identify areas of maximum sensitivity or therapeutic response. These diagnostic pulses are carefully controlled within safety parameters while providing sufficient stimulus intensity to generate detectable physiological responses through the comprehensive sensor array.

[0207] Using the flexible electrode array, the system systematically stimulates different spatial locations while monitoring physiological responses including autonomic reactions, facial expressions, vocal responses (when the natural language interface is enabled), and other pain-related biomarkers. This process creates a detailed map of response sensitivity across the treatment area.Attorney Docket No. SETH-005 AWO

[0208] Once optimal locations are identified through diagnostic stimulation, the system automatically configures electrode pairs for therapeutic stimulation, ensuring precise targeting that may differ from standard anatomical placement guidelines. This approach enables personalized treatment protocols that account for individual anatomical variation and specific pathology patterns.

[0209] A diagnostic probing mode using the same flexible electrode array may be employed to map effectiveness across the stimulation surface. This process, described below in connection with FIGS. 9 and 10, systematically activates electrode pairs to identify regions that elicit the strongest physiological responses and selects those regions for subsequent closed loop optimization.

[0210] Stimulation Component Architecture and Parameter Control

[0211] The adaptive stimulation device 102 may further include a stimulation component 116 that generates the desired stimulation by manipulating muscles, nerves, or other tissues through direct or electrical stimulation applied through pads, electrodes, leads, or other interfaces in contact with the user’s skin or body. The stimulation component 116 represents a significant advance over existing devices by providing precise, programmable control over multiple stimulation parameters simultaneously, with real time adjustment capabilities based on multimodal feedback and natural language commands. The stimulation component 116 may include, without limitation, motors, oscillators, actuators, and other mechanical or electronic elements configured to produce the stimulation across diverse modalities and applications. The component architecture supports the techniques developed through the adaptive optimization process.

[0212] Electrical (TENS) Applications

[0213] For transcutaneous electrical nerve stimulation (TENS) applications, the stimulation component provides multiple adjustable parameters with precision that exceeds conventional TENS units, including for example: waveform shape (e g., pulsed, biphasic, triangular, sinusoidal, and complex custom patterns), pulse frequency ranging from about 1-200 Hz with about 0.1 Hz resolution, amplitude control from about 0-80 mA with 0. 1 mA precision, pulse duration from about 50-400 microseconds with microsecond-level control, and spatial location via the placement of adhesive pads with support for up to sixteen independent channels through the flexible electrode array system.

[0214] The parameter space is vast, with each device defining a control surface: amplitude x frequency x waveform x location! x locatio x time. For a advanced TENS implementation, this creates millions of possible parameter combinations. The ideal setting for one user under one condition may be very different from another user, or from that same user at a later time, requiring sophisticated optimization algorithms to navigate this complex space efficiently.Attorney Docket No. SETH-005 AWO

[0215] Advanced TENS capabilities include, without limitation: waveform modulation with realtime parameter sweeping, spatial stimulation patterns that can move across multiple electrode sites through the flexible array system, temporal patterns that vary stimulation over time scales from seconds to hours, and adaptive amplitude control that maintains consistent perceived intensity despite changes in skin impedance or user sensitivity.

[0216] Mechanical Stimulation Applications

[0217] For mechanical stimulation applications, the stimulation component 116 may include various stimulation modalities, including, for example: variable-frequency vibration with programmable patterns and intensities ranging from gentle tactile stimulation to deep tissue massage; percussive massage with adjustable intensity, frequency, and spatial targeting (such as systems like the percussion massagers or robotic massage beds) with force feedback control and pressure monitoring; pneumatic pressure modulation with precise pressure control and spatial distribution; thermal therapy with heating and cooling elements capable of precise temperature control and thermal pattern generation; and robotic massage systems with force feedback control, Al-guided positioning, and adaptive pressure adjustment based on tissue response.

[0218] Advanced mechanical stimulation features may include, for example: multi-modal stimulation combining vibration, pressure, and thermal elements; adaptive force control that adjusts pressure based on tissue compliance and user response; spatial pattern generation that can create complex stimulation sequences across multiple body regions; and texture variation devices that can simulate different tactile sensations.

[0219] Hybrid and Emerging Stimulation Modalities

[0220] The stimulation component 116 may further or also include modalities such as focused ultrasound for non-invasive deep tissue stimulation with millimeter-precision targeting, electromagnetic field generation with controllable field patterns and intensities, targeted sensory7stimulation through controlled air flow that moves hair on the skin for subtle tactile effects, texture variation devices with programmable surface patterns, temperature change mechanisms with rapid heating and cooling capabilities, and other emerging sensory input technologies including haptic feedback systems and virtual reality integration.

[0221] The stimulation component operates under control of the adaptive feedback system, with real-time parameter adjustment based on measured user response and natural language commands, safety monitoring with automatic stimulation cessation if adverse responses are detected, and learning algorithms that optimize stimulation patterns over time while maintaining user safety and comfort.Attorney Docket No. SETH-005 AWO

[0222] Stimulation approaches generated through the adaptive process may include, for example: complex temporal patterns that vary' multiple parameters simultaneously, spatial-temporal stimulation sequences that move across different body regions through the flexible electrode array, personalized waveforms optimized for individual physiological characteristics, and hybrid approaches that combine multiple stimulation modalities for synergistic effects.

[0223] Somato-Autonomic Reflex and Therapeutic Applications

[0224] According to one or more aspects, the system supports stimulation for diverse therapeutic applications beyond pain and pleasure, leveraging the somato-autonomic reflex where gentle skin stimulation causes changes in autonomic activity. Examples include: sexual arousal and orgasm enhancement through skin and genital stimulation with real-time optimization based on arousal measurement; bladder function improvement through inguinal area stimulation to reduce hyperactive bladder activity’, or back stimulation above the rectum to improve bladder emptying, with effectiveness monitoring through ultrasound bladder measurement and autonomic response tracking; nausea reduction through specific acupressure methods or gentle abdominal stroking that increases parasympathetic activity, monitored through HRV analysis and gastrointestinal motility measurement; and pain relief through skin stimulation that enhances natural pain relief through physiological release of endogenous opioids, monitored through pain-related brain activity and autonomic responses.

[0225] Each therapeutic application may include specialized monitoring and optimization protocols tailored to the specific physiological mechanisms involved, with safety7constraints appropriate to the application domain and effectiveness metrics validated through clinical research.

[0226] Proxy Measurement Framework and Machine Learning Integration

[0227] A proxy measurement component 108 may be included to leverage more accessible and affordable measurements that can be correlated with sophisticated measurements. The proxy measurement component 108 may allow the device 102 to achieve effective feedback, using proxy measurements in lieu of some sophisticated and expensive measurements, while keeping production costs low and enabling scalable deployment across diverse user populations.

[0228] As shown in FIG. 4, the proxy-measurement control architecture integrates proxy sensors 404, an inference model 410, and controller 416. The proxy sensors provide indirect physiological inputs that are processed by the inference model to estimate stimulation effectiveness, allowing the controller to adapt stimulation output in real time even when high-fidelity’ sensing is unavailable.

[0229] The proxy measurement framework represents a breakthrough in making sophisticated neuroscience accessible for practical applications. By creating validated mappings between expensive,Attorney Docket No. SETH-005 AWO high-fidelity sensors and affordable consumer-grade sensors, the system can achieve clinical-grade accuracy at consumer-friendly price points.

[0230] Proxy Modeling Challenges and Deployment Context

[0231] The challenge of implementing real-time adaptive stimulation systems at scale has historically been hindered by the need for expensive or intrusive sensing systems — such as fMRI, PET, or high-resolution EEG arrays. A practical system for adaptive stimulation may reconcile the need for meaningful physiological insight with the constraints of consumer-grade hardware, regulatory requirements, and real-world deployment environments.

[0232] This tension underscores the importance of proxy modeling: the use of accessible, lower- fidelity sensors that have been trained against higher-cost or higher-resolution sources. According to one aspect, as described herein, sophisticated machine learning algorithms can extract clinically relevant information from simple, affordable sensors when properly trained on high-quality data.

[0233] Proxy Modeling System Architecture

[0234] FIG. 5 illustrates the proxy-model training configuration in which high-fidelity measurements (e.g., EEG, fNIRS) are correlated with proxy-sensor inputs during supervised learning. A training engine validates the mapping function f (proxies state) to produce models that infer effectiveness from accessible sensors during runtime.

[0235] The mapping from multimodal inputs to E(t) / P(t) may be implemented by any estimation or inference technique, including statistical models, observers, optimization-based estimators, classical control framew orks, or data-driven approaches such as deep neural networks, convolutional or recurrent architectures, transformer networks, and other machine-learning or generative- Al models capable of learning temporal and contextual dependencies.

[0236] As shown in FIG. 6, the adaptive stimulation system may be implemented through multiple processing configurations, including on-device, edge-assisted, and cloud-enabled architectures. Each configuration supports secure communication links, privacy boundaries, and encrypted model updates. In some aspects, real-time control and safety monitoring may occur locally on the device, while long-term learning and population-model refinement are performed on edge or cloud servers. This flexible arrangement enables scalable deployment of adaptive stimulation systems across research, clinical, and consumer environments while maintaining data security and consistent performance.

[0237] According to one or more aspects, the proxy modeling system operates through a sophisticated tiered approach: (1) Ground truth data collection: expensive but accurate modalities such as fMRI, photoacoustic imaging, or functional ultrasound are used in controlled environments to establish validated measurements of pain, pleasure, and related affective states with large participantAttorney Docket No. SETH-005 AWO pools to ensure robust training data; (2) Parallel proxy data collection: simultaneously, more scalable sensors (e.g., optical pulse monitoring, skin temperature, facial video, consumer EEG headbands) are used to collect parallel data streams with precise temporal synchronization; (3) Advanced model training: algorithms including deep neural networks, ensemble methods, and transfer learning are trained to map the proxy sensor input to the high-fidelity signal output using supervised learning with cross- validation and performance testing on held-out datasets; (4) Deployment and continuous learning: once trained and validated, these algorithms are ported to local device processors or cloud-based platforms for real-time inference with ongoing performance monitoring and model updates.

[0238] A representative proxy-model training and deployment process is summarized later in connection with FIG. 6A, which depicts how high-fidelity measurements are correlated with accessible proxy signals and used to train models for real-time inference. This architecture enables the deployment of intelligent, adaptive systems using more cost-effective devices (e.g., devices that cost under $500) and can operate in non-climcal settings while achieving accuracy approaching that of systems costing tens of thousands of dollars. With the increasing prevalence of multimodal wearable sensors, it is feasible for a stimulation device to continuously gather multiple proxy signals — combining cardiovascular, electrodermal, and musculoskeletal data — to create robust estimates of pain relief or pleasurable engagement.

[0239] The system implements advanced machine learning techniques including: deep convolutional neural networks for processing time-series physiological data, recurrent neural networks for modeling temporal patterns in user responses, ensemble methods that combine multiple proxy measurements for improved accuracy, transfer learning approaches that enable rapid adaptation to new users based on population data, and federated learning protocols that enable collective improvement while preserving individual privacy.

[0240] Demographic and Contextual Adaptation

[0241] Proxies may work differently across different cohorts and to different degrees, requiring population-specific model training and validation. According to one or more aspects, the system includes demographic-aware modeling that can adapt proxy relationships based on age, gender, health status, and other relevant factors, with continuous validation ensuring that proxy models maintain accuracy across diverse user populations.

[0242] The proxy measurement approach also enables contextual adaptation. For instance, the same heart rate change may reflect excitement in one context or anxiety in another. By layering proxy signals with metadata — such as time of day, user activity level, and prior session outcomes — a feedbackAttorney Docket No. SETH-005 AWO system can develop increasingly accurate models of user state. Furthermore, these models can be refined through user-specific data, supporting individualized learning curves and reducing calibration time.

[0243] As described herein, advanced contextual adaptation includes, for example: environmental awareness using ambient sensors, activity recognition through motion pattern analysis, circadian rhythm modeling for time-of-day effects, stress level assessment through multi-modal physiological monitoring, and social context recognition for appropriate response modulation.

[0244] Advanced Control Algorithms and Parameter Optimization

[0245] According to one or more aspects, a feedback mechanism 106 and control algorithm (i.e., controller 110) may work in conjunction to receive and process the user's response to the stimulation and adapt the stimulation according to the response. The user's response may be processed to determine an effectiveness of the applied stimulation, with the potential for direct user / caregiver feedback for automated or semi-automated changes in parameters, natural language voice command integration for real-time user direction, and additional learning to improve models through processed, analyzed and potentially modified data against historically learned modeled data from a conditional user pool.

[0246] The control system represents a sophisticated integration of real-time optimization, machine learning, natural language processing, and safety monitoring that can navigate complex, highdimensional parameter spaces while maintaining user safety’ and satisfaction.

[0247] Optimization Strategies and Temporal Adaptation

[0248] According to one or more aspects, the system treats the configuration space — waveform, amplitude, frequency, location — as a dynamic optimization problem. Using physiological signals such as HRV, GSR, or EEG-derived affective estimates, combined with natural language commands and vocal stress analysis, the system could infer whether a given stimulation protocol is yielding a positive or negative effect. It could then adjust stimulation either incrementally or explore alternative configurations to identify better ones. Over time, this would enable the system to converge on settings that work for a specific user under specific conditions.

[0249] The optimization process operates on multiple levels, including for example: immediate parameter adjustment for real-time optimization, session-level pattern optimization for improved effectiveness, and long-term learning for user-specific adaptation and population-level improvement.

[0250] Closed-Loop Framework and Real-Time Operation

[0251] A closed-loop control framework for real-time adaptive operation will be described later in connection with FIG. 17. In general, the controller applies stimulation, receives physiological feedback, and updates parameters continuously under supervision of the safety subsystem to maintain effectiveness and user safety.Attorney Docket No. SETH-005 AWO

[0252] Optimization and Learning Algorithms

[0253] According to one or more aspects, the control algorithms implement multiple optimization strategies including gradient descent for continuous parameter spaces with adaptive step sizes and momentum terms, genetic algorithms for discrete parameter exploration with mutation and crossover operations tailored to stimulation parameters, reinforcement learning for session-to-session improvement using reward signals derived from physiological effectiveness measures and user feedback, Bayesian optimization for efficient parameter space exploration with limited training data and uncertainty quantification, and exploration protocols to discover optimal patterns in highly individualized affective responses using techniques such as upper confidence bound sampling and Thompson sampling.

[0254] The system balances exploration of new parameter combinations with exploitation of known effective settings while implementing safety constraints and user preference boundaries. Advanced optimization features include: multi-objective optimization that can simultaneously optimize for multiple goals (e g., pain relief and user comfort), constraint handling for safety and user preference boundaries, adaptive exploration strategies that adjust exploration rate based on current performance and user feedback, and meta-leaming approaches that learn how- to optimize more effectively for individual users over time.

[0255] Control follows the multi-timescale hierarchy previously defined (FIGS. 3C and 10), coordinating fast-safety, in-session, and cross-session adaptation. Each timescale employs appropriate algorithms and data sources: immediate responses use simple threshold-based safety monitoring and voice command processing, short-term adaptation uses online optimization algorithms with natural language integration, medium-term learning uses batch machine learning on session data, and long-term learning uses population-scale data analysis and model updating.

[0256] The system supports longitudinal personalization where a device that records and leams from past sessions can identify patterns — for example, that low-frequency pulsed stimulation on the shoulders improves relaxation in the evening but not in the morning, or that certain massage locations become ineffective after repeated use. These insights allow the system not only to adapt in the moment but to recommend session planning, preventative use, or context-aware stimulation choices. The natural language interface enables users to provide rich contextual information such as "I'm feeling more stressed today" or "the usual setting isn't working," which the system can incorporate into its optimization algorithms.

[0257] A federated-learning architecture that enables collective improvement of adaptive- stimulation models while preserving user privacy will be described later in connection with FIG. 12. InAttorney Docket No. SETH-005 AWO this framework, individual devices contribute encrypted model-update data for aggregation and refinement without sharing raw physiological information. According to one or more aspects, advanced longitudinal learning capabilities may include, for example: circadian pattern recognition for time-of- day optimization, habituation detection and countermeasures for maintaining long-term effectiveness, progression tracking for monitoring changes in user condition and treatment needs, predictive modeling for anticipating optimal treatment timing and parameters, and natural language pattern analysis to understand user preferences and communication styles over time.

[0258] In one aspect, the system maintains comprehensive user profiles that include physiological baselines, response patterns, preference indicators, treatment history, and natural language interaction patterns while implementing privacy protection measures and user control over data sharing.

[0259] Effectiveness Measurement & Affective State Detection

[0260] The effectiveness of stimulation may be determined from one or more physiological or neuro-imaging measurements, individually or in multi-modal combination, as described later in connection with FIG. 14. In certain implementations, the system fuses multiple sensing modalities to generate a composite effectiveness metric that reflects autonomic, neural, and behavioral responses. The system implements sophisticated signal processing and pattern recognition algorithms to extract meaningful information from these diverse measurement modalities, with fusion algorithms that combine multiple measurements for robust effectiveness assessment.

[0261] Some of these measurements, such as data from imaging modalities like brain scans or photoacoustic imaging, can be expensive and difficult to implement, limiting their applicability in cost- effective devices. However, experiments that correlate photoacoustic imaging with autonomic signals — like heart rate variability or skin conductance — can enable supervised machine learning systems to identify meaningful relationships that can then be deployed as classifiers or predictors in embedded devices equipped with low-cost sensors.

[0262] According to one or more aspects, the proxy modeling approach described herein enables clinical-grade accuracy using consumer-grade sensors through sophisticated machine learning algorithms that have been trained on high-quality data from research environments.

[0263] Pain is generally associated with higher breathing rate, cardiac rate, and muscle activity, while pain relief is associated with bringing those parameters down. The autonomic nervous system is tied to both pain and pleasure states, with simple measures like galvanic skin response and HRV providing straightforward assessment tools. HRV analysis reveals that low frequency components reflect parasympathetic activity, while high frequency components reflect sympathetic activity. TheAttorney Docket No. SETH-005 AWO sympathetic nervous system represents the fight-or-flight response, so transitions between these states provide indicative measures of changing pain or pleasure states.

[0264] Advanced physiological analysis implemented in the system may include, for example: multi-parameter pain assessment using validated pain-related biomarker combinations, pleasure detection through reward-related physiological patterns including dopaminergic and oxytocinergic response indicators, stress assessment through comprehensive autonomic nervous system monitoring, arousal detection through multi-modal physiological pattern recognition, and integration of vocal stress analysis with conscious natural language communication to provide comprehensive user state assessment.

[0265] Referring back to FIG. 1, aspects of the present disclosure may incorporate the feedback mechanism 106 to measure the effectiveness of the stimulation provided by the adaptive stimulation device 102, driving the device 102 to adjust its outputs and settings in real-time based on the user's response. The feedback mechanism 106 may enhance the efficacy and user experience of the device 102, tailoring the stimulation to the user's unique needs and preferences. Furthermore, aspects of the present disclosure may leverage more accessible proxy measurements as biomarkers that correlate with sophisticated measurements, allowing cost-effective implementation and enhanced device efficacy.

[0266] The feedback mechanism represents a paradigm shift from reactive to proactive device behavior, enabling devices to anticipate user needs and optimize treatment approaches before problems arise, while incorporating natural language understanding to respond to both conscious user direction and unconscious physiological signals.

[0267] Somato-Autonomic Reflex Applications and Physiological Mechanisms

[0268] Many devices exist that apply stimuli to the body for causing pleasure or reducing pain. Stimulation can also improve other body functions, such as bladder control, leveraging the somato- autonomic reflex where ty pically gentle skin stimulation causes a change in autonomic activity7. Several examples are well-known in clinical practice, while others represent emerging therapeutic opportunities that can be explored through the adaptive optimization capabilities of the system.

[0269] The somato-autonomic reflex represents a physiological mechanism that the system can leverage across multiple therapeutic domains, with real-time monitoring enabling optimization of reflex activation for specific therapeutic goals.

[0270] One example is skin (including vaginal) stimulation leading to sexual arousal and orgasm, which are primarily dependent upon autonomic nervous system activity. The physiological mechanisms involve complex interactions between sympathetic and parasympathetic nervous systems, with measurable changes in heart rate variability7, skin conductance, respiratory7patterns, and neuralAttorney Docket No. SETH-005 AWO activity that can be detected and used for adaptive stimulation control. As described herein, the natural language interface enables users to provide real-time guidance during intimate experiences while maintaining the flow and naturalness of the interaction.

[0271] According to one or more aspects, the system implements specialized arousal monitoring and optimization protocols including multi-phase arousal tracking, personalized stimulation pattern development, partner synchronization capabilities for couples' devices with coordinated natural language interfaces, and safety monitoring for appropriate stimulation intensity and duration.

[0272] Additionally, gentle skin stimulation applied to the inguinal area has been shown to improve bladder function by reducing the activity of a hyperactive bladder. Gentle skin stimulation to the back, above the rectum, can improve bladder emptying by increasing bladder contractility. These responses involve sympathetic nerve activity changes that help preserve sleep by reducing nocturia episodes, and can be monitored through autonomic nervous system measurements and ultrasound bladder monitoring.

[0273] In one aspect, the system may include specialized bladder function monitoring using wearable ultrasound sensors for non-invasive bladder volume measurement, autonomic response tracking for parasympathetic and sympathetic activity balance, and sleep quality monitoring to assess treatment effectiveness for nocturia management.

[0274] Nausea has been reported to be decreased with specific acupressure methods. It is also commonly experienced that gentle abdominal stroking reduces feelings of nausea. Both categories of stimulation may increase parasympathetic activity. The parasympathetic outflow to the gastrointestinal (GI) tract is significant, and increasing parasympathetic tone may be the mechanism for decreasing nausea, which can be measured through HRV analysis and gastrointestinal motility monitoring.

[0275] Advanced nausea management capabilities may include, for example: real-time parasympathetic monitoring through HRV analysis, gastrointestinal motility tracking using non-invasive sensors, personalized acupressure point identification through effectiveness mapping, and integration with motion sickness prevention protocols.

[0276] Pain relief, while not technically autonomic, may be modulated similarly to nausea by both changing perception and through direct physiological mechanisms. Skin stimulation has been shown to enhance natural pain relief through the physiological release of endogenous opioids, with measurable changes in pain-processing brain regions that can be detected through neuroimaging and correlated with accessible proxy measurements.

[0277] The system may implement comprehensive pain management protocols including multimodal pain assessment using validated pain scales and physiological indicators, endogenous opioidAttorney Docket No. SETH-005 AWO release monitoring through indirect physiological measures, pain circuit modulation through targeted stimulation approaches, and chronic pain management with long-term adaptation and progression tracking.

[0278] Each of these examples (sexual response, bladder function, GI sensations, pain) demonstrates that skin stimulation, when applied to the correct location on the body and wi th the correct stimulus features, can have a significant impact on diverse body functions and perception. The adaptive feedback system enables automatic discovery of optimal stimulation locations and parameters for individual users and specific therapeutic goals, enhanced by natural language communication that allows users to guide the discovery process through conversational interaction.

[0279] The system's ability to leam and optimize across multiple therapeutic domains represents a significant advancement over single-purpose devices, enabling comprehensive wellness management through integrated physiological monitoring and adaptive intervention.

[0280] Measurement Processing and Biosignal Feedback Training

[0281] FIG. 2 is a flow diagram of a method 200 for adaptive stimulation according to aspects of the present disclosure. According to one aspect, an adaptive stimulation device may be used to stimulate a subject, or user, as shown in block 202.

[0282] Stimulation of a subject may be accomplished using any number of devices, including, but not limited to, a handheld massager with adaptive pressure control, STIM machines with advanced waveform generation, transcutaneous electrical nerve stimulation (TENS) units which use electrical pulses to modulate nerve activity' for pain relief with real-time parameter optimization, nerve stimulation devices for various therapeutic applications including vagus nerve stimulation and peripheral nerve modulation, massage chairs with programmable pressure and movement patterns that adapt to user response, male or female sex toys with variable stimulation patterns optimized through multimodal feedback and natural language guidance, robots with sophisticated manipulation capabilities and force feedback systems, and skin stimulation devices using various modalities including mechanical, electrical, thermal, and other approaches discovered through adaptive optimization.

[0283] The system supports techniques discovered through the optimization process, enabling continuous innovation in therapeutic and enhancement applications while maintaining safety and effectiveness standards.

[0284] Feedback Measurement and Signal Processing

[0285] As shown in block 204, results of the stimulation on the subject may be measured as feedback. Results of the stimulation may be measured according to a number of metrics including, but not limited to, heart rate and cardiac patterns yvith advanced variability7analysis and arrhythmiaAttorney Docket No. SETH-005 AWO detection, across-skin resistance and galvanic skin response with temporal pattern recognition and event detection, sound and vocal analysis using machine learning algorithms trained on emotional stress indicators that can distinguish between natural language communication and unconscious vocal stress patterns, self-recording on validated scales including visual analog scales and standardized pain and pleasure assessment instruments, blood pressure monitoring with beat-to-beat analysis and trend detection, muscle contractions and EMG patterns with multi-channel recording and pattern classification, temperature regulation monitoring with spatial and temporal analysis, EEG with advanced signal processing including spectral analysis, coherence measurement, and machine learning classification of brain states, fNIRS and other neurological feedback systems for cortical blood flow monitoring with real-time hemodynamic analysis, photoacoustic imaging for hemodynamic responses with high spatial and temporal resolution, brain scans including MRI, CAT, FMRI for structural and functional assessment with real-time processing capabilities where feasible, functional ultrasound imaging with array transducers for deep brain monitoring, coded facial expressions using computer vision analysis wi th emotion recognition algorithms, and other emerging measurement modalities including optical, acoustic, and electromagnetic sensing approaches.

[0286] The measurement system implements sophisticated signal processing algorithms to extract meaningful information from noisy physiological signals, with artifact rejection, signal enhancement, and multi-modal fusion capabilities that ensure robust and accurate assessment of user state.

[0287] According to one aspect, one or more measurements may be used as proxies for other measurements that the stimulation device cannot make directly due to cost, size, or practical constraints. For example, a feedback control system may incorporate feedback signals that can be trained with user input via machine learning algorithms, where the user input serves as supervisory signals for training physiological inference models that can distinguish between effective and ineffective stimulation patterns. Natural language feedback provides an additional rich source of training data, enabling the system to correlate spoken descriptions of sensation with physiological measurements.

[0288] The proxy measurement approach enables democratization of sophisticated physiological monitoring by making clinical-grade insights accessible through consumer-grade sensors and advanced machine learning algorithms.

[0289] The biosignal feedback recording system may be initially used and trained by correlating biosignals with user experience of pleasure or symptom relief using validated assessment scales and objective physiological measures. Initial training may be supplemented and continued by closed-loop optimization of stimulus with biosignal measures to seek optimal stimulation patterns throughAttorney Docket No. SETH-005 AWO systematic parameter exploration and effectiveness assessment. This training permits automatic adjustment of stimuli without user input to optimize the function of the stimulus device and to permit use when the individual does not or cannot control the stimulus manually, such as in cases of physical disability, cognitive impairment, or during medical procedures where manual control is not feasible. The natural language interface provides an additional pathway for users to contribute to system training through conversational interaction, even when other forms of feedback may be limited.

[0290] The training process implements sophisticated machine learning approaches including supervised learning for initial model development, reinforcement learning for ongoing optimization, transfer learning for rapid adaptation to new users and conditions, and natural language processing for incorporating conversational feedback into optimization algorithms.

[0291] More sophisticated measures such as EEG with advanced signal processing, neuroimaging including functional magnetic resonance imaging (fMRI) and functional near-infrared spectroscopy (fNIRS), and other neural recording modalities can be added to the training program to refine and select the biosignal feedback measures that are most efficient for closed-loop training. This approach is particularly valuable for severe chronic pain relief where precise parameter optimization can significantly improve quality of life, and adaptation to regular physiological changes over time where the system must continuously adjust to changing baseline conditions and treatment responses.

[0292] The integration of high-fidelity measurement systems enables validation and continuous improvement of proxy -based models while providing gold-standard reference data for system calibration and performance assessment.

[0293] Parameter Space Exploration and Optimization Strategies

[0294] As shown in block 206, one or more of these feedback measurements may be analyzed and optimized to adapt the applied stimulation to the user. According to one aspect, the feedback measurements and their resulting changes to stimulation may be used to generate an intelligent system, for example, through machine learning techniques including supervised learning for pattern recognition, reinforcement learning for optimal policy development, deep neural networks for complex pattern analysis, meta-leaming approaches for rapid adaptation to new users, and natural language processing for incorporating conversational feedback and commands, to generate stimulation profdes for specific devices, individual users, treatment conditions, and therapeutic goals.

[0295] The optimization system represents a sophisticated integration of multiple Al and machine learning approaches that can discover therapeutic approaches while maintaining safety and effectiveness standards.Attorney Docket No. SETH-005 AWO

[0296] According to one or more aspects, the system addresses the challenge of navigating the multidimensional space of possible stimulation parameters. In mathematical terms, each device defines a control surface: amplitude x frequencyxwaveform x locationi x locatior x time. For example, an advanced TENS unit with 20 amplitude settings, 50 frequency options. 10 waveforms, and 4 electrode pairs, may provide 400,000 possible configurations before even considering temporal patterns. The ideal setting for one user under one condition may be very different from another user, or from that same user at a later time. The system explores this space in real time, tracking which combinations are associated with improved physiological or affective outcomes rather than relying on user trial-and-error or static presets. The flexible electrode array system exponentially increases this parameter space, with, for example, 80-120 potential stimulation points enabling millions of possible electrode pair combinations.

[0297] Advanced parameter space exploration includes: multi-dimensional optimization algorithms that can efficiently search high-dimensional spaces, constraint handling for safely and user preference boundaries, adaptive exploration strategies that balance discovery of new approaches with exploitation of known effective methods, meta-optimization approaches that optimize the optimization process itself based on user characteristics and response patterns, and natural language-guided exploration where user descriptions can direct the search toward promising parameter regions.

[0298] The system implements structured exploration algorithms to avoid local maxima and discover globally optimal stimulation patterns. This includes techniques such as simulated annealing for escaping local optima with temperature scheduling adapted to physiological response characteristics, multi-armed bandit algorithms for balancing exploration and exploitation with upper confidence bound and Thompson sampling approaches, evolutionary’ algorithms for population-based parameter optimization with mutation and crossover operations tailored to stimulation parameters, and Bayesian optimization for sample-efficient parameter space exploration with uncertainty quantification and acquisition function optimization.

[0299] These advanced optimization techniques enable the system to discover stimulation approaches that might not be found through conventional parameter adjustment, potentially leading to breakthrough therapeutic protocols and enhanced user experiences.

[0300] From a feedback perspective, the critical need for adaptive control may be demonstrated in cases such as migraine, where continuous skin stroking remains pleasant in normal people but stops being pleasant in migraine sufferers. Pain causes increased sympathetic activity’, which can be assessed from heart rate and galvanic skin response, or through direct user reporting of pain levels using validated pain assessment scales or natural language descriptions of changing sensations. Bladder filling and nausea also cause increased sympathetic activity’ and can be inferred from the same autonomicAttorney Docket No. SETH-005 AWO measures, with bladder filling alternatively assessed using simple wearable ultrasound devices that periodically capture bladder size with non-invasive measurement techniques.

[0301] The system's ability- to detect and respond to these physiological changes enables automatic adjustment of stimulation approaches before effectiveness is lost, maintaining therapeutic benefit and preventing user frustration with treatment failure.

[0302] In one aspect, if skin stimulation to a particular location on the body is the system output, specific biosignal feedback and user input for the particular condition being targeted (pain, bladder function, arousal, etc.) may be acquired and used to monitor the efficacy of the output. Many of the physiological examples described herein can be monitored using biosignal feedback from direct measures of sympathetic tone (such as heart rate, heart rate variability, galvanic skin response) and arousal indicators (skeletal muscle tone, respiratory' rate). User input feedback comes from the user indicating through a simple interface the level of effectiveness of each set of stimulus conditions, enabling hybrid human-machine optimization where automated physiological monitoring is combined with subjective user assessment to achieve optimal treatment outcomes. The natural language interface provides an intuitive and expressive channel for user feedback, allowing detailed descriptions of sensations and preferences that can guide optimization algorithms.

[0303] The integration of objective physiological measures with subjective user feedback creates a robust optimization framework that can achieve both measurable physiological improvements and high user satisfaction.

[0304] Longitudinal Learning and Population-Level Intelligence

[0305] Referring now to FIG. 3, a flow diagram illustrating a method 300 of cross-session learning and personalization is shown. According to one or more aspects, data collected from each stimulation session are processed to extract features that update a personalized model. The refined parameters are used to initialize subsequent sessions, enabling progressive optimization over time. Population-level or federated learning aggregates patterns across users to further enhance individual and collective performance.

[0306] As shown in block 302, an adaptive stimulation system or device may learn, through machine-learning techniques including deep learning wi th convolutional and recurrent neural networks, federated learning for privacy-preserving population-level improvement, and population-based optimization using genetic algorithms and evolutionary strategies, various pain and pleasure measurement patterns and responses across diverse user populations and therapeutic conditions. Those learned patterns may, as shown in block 304, be used to implement profiles and settings on new devices,Attorney Docket No. SETH-005 AWO enabling rapid initialization and reducing calibration time for new users through transfer learning and meta-leaming approaches.

[0307] “The adaptive-learning framework illustrated in FIG. 3 supports both individual and population-level refinement. Cross-session updates from each device may contribute to the federated- learning architecture described later in connection with FIG. 12, enabling privacy -preserving collective improvement while maintaining personalized adaptation. The effectiveness of the adapted profiles and settings may be measured using the comprehensive feedback systems, shown in block 306, with continuous monitoring and assessment of therapeutic outcomes and user satisfaction through validated outcome measures and long-term tracking of treatment effectiveness. Block 306 performs model inference, integrating multimodal physiological and behavioral signals to estimate stimulation effectiveness. The inference process is designed for graceful degradation, functioning reliably with any available subset of sensors and without dependence on a single biomarker. If beneficial changes are identified, they may be incorporated into the controls and operations of the device, shown in block 308, through automated parameter updates and profile refinement using online learning algorithms that can adapt to changing user needs and physiological responses over time. As shown in block 310 of FIG. 3, data relating to measurements, settings, and profiles may be logged for long-term adaptation. In certain implementations, these records participate in the privacy-preserving federated-learning process described later in connection with FIG. 12, allowing collective model improvement while maintaining user data protection. This process enables devices and users — including self-users, caregivers, and clinicians — to leam new profiles and techniques to treat conditions more quickly and effectively through collective intelligence while maintaining individual privacy protections and user control over data- sharing preferences. The privacy -preserving approach provides the benefits of population-scale learning while respecting user privacy rights and regulatory requirements, supporting a sustainable ecosystem for continuous improvement and innovation.

[0308] Cold Start Problem and New User Initialization

[0309] According to one or more aspects, the system addresses the "cold start" problem, where new users lack historical data for personalized optimization. In one aspect, the solution may employ a three-tier initialization strategy that balances safety, effectiveness, and rapid personalization.

[0310] Tier 1 - Population-Based Initialization: New users, according to one aspect, are initialized using parameter distributions derived from demographically similar users with matching therapeutic objectives. The system collects basic demographic and clinical information (age, sex, condition category, anatomical region, chronicity) and queries population databases to identify matchingAttorney Docket No. SETH-005 AWO user cohorts. Initial parameters are set conservatively within the effective range observed in the matched population, ensuring safety while providing reasonable therapeutic potential.

[0311] Tier 2 - Intelligent Exploration: Initial sessions, according to one aspect, implement accelerated exploration using optimization algorithms with population-derived priors. The system prioritizes exploration of high-variance parameters showing substantial inter-individual variation (electrode location, frequency preferences) while constraining low-variance parameters to narrow ranges around population consensus. Exploration maintains strict safety constraints, including limited parameter space coverage per session, mandatory comfort validation, and conservative amplitude limits to provide safety margins for unexpected individual sensitivity.

[0312] Tier 3 - Progressive Personalization: Following initial exploration, the system, according to one aspect, progressively transitions from population-prior dependence to personalized individual models. Early sessions rely heavily on population priors while emphasizing safety; middle sessions balance population and individual data; later sessions emphasize individual response patterns; eventual full personalization bases parameter selection predominantly on individual history with population priors serving as regularization.

[0313] Failure Mode Handling: The system, according to one aspect, implements protocols for cold start failures, including persistent ineffectiveness (alerting users to potential non-responsiveness). excessive sensitivity (employing extended gradual titration with reduced session frequencies), and erratic responses (extending data collection phase before aggressive optimization). As the system accumulates data across growing populations, cold start performance improves through increasingly specific demographic matching, richer parameter distributions, and faster convergence to optimal settings for new users.

[0314] The system enables cloud storage and Al machine learning protocols that work for specific conditions and user populations. By know ing what data users contribute for crowd learning, the system can help people reach effective stimulation parameters more quickly. This includes learning optimal stimulation patterns for specific medical conditions such as chronic pain syndromes, fibromyalgia, and neuropathic pain, demographic groups including age-specific and gender-specific optimization approaches, and individual user characteristics such as pain sensitivity, medication status, and comorbid conditions, as well as discovering therapeutic applications through population-level pattern analysis and identification of previously unknown effective treatment approaches.

[0315] The collective intelligence approach accelerates innovation while ensuring that new discoveries benefit all users through validated, safety-tested protocol updates and evidence-based treatment recommendations.Attorney Docket No. SETH-005 AWO

[0316] System Implementation Architectures

[0317] Referring now to FIGS. 4-6, block diagrams of exemplary' adaptive stimulation systems 400, 500, 600, are shown, where like reference numbers indicate corresponding components, illustrating different implementation approaches and deployment models ranging from basic proxy-only systems suitable for consumer applications to sophisticated research-grade systems with comprehensive multimodal sensing capabilities. As shown in FIG. 4, a user 402 may be connected to a stimulation device, for example TENS unit 418, by one or more leads 422a, 422b with advanced electrode systems that provide precise stimulation delivery and impedance monitoring. According to one aspect the user may have leads 422a attached to a front portion of their body in addition to leads 422b attached to a back portion of their body, enabling comprehensive coverage and spatial selectivity for stimulation delivery' with support for complex stimulation patterns and multi-site coordination through the flexible electrode array system.

[0318] The modular architecture enables flexible deployment across different use cases and cost points while maintaining consistent core functionality7and learning capabilities.

[0319] The TENS unit 418 may supply or apply stimulation to the user 402 according to a number of parameters, generally labeled 420, through a control API 416 that provides real-time parameter adjustment and safety monitoring capabilities with natural language command integration. According to one aspect, the parameters 420 may include, without limitation, waveform time duration with precise temporal control down to microsecond resolution, amplitude modulation with safety limits and automatic adjustment for impedance changes, frequency sw eeping and harmonic generation with real-time optimization based on physiological response and user voice commands, spatial selectivity through selectively active pad arrays with support for up to 16 independent channels through the flexible electrode array, and complex waveform generation including biphasic, monophasic, and custom patterns designed for specific therapeutic applications and discovered through adaptive optimization processes.

[0320] The advanced parameter control sy stem enables precise, real-time adjustment of all stimulation characteristics based on continuous physiological monitoring and optimization algorithms integrated with natural language understanding.

[0321] As the stimulation is applied to the user 402, one or more sensors 404 may measure the user's response to the stimulation according to a comprehensive array of measurements, including without limitation, pulse rate and advanced cardiac pattern analysis with arrhythmia detection and heart rate variability assessment, blood pressure monitoring with beat-to-beat variability analysis and trend detection for autonomic assessment, skin resistance and galvanic skin response with temporal pattern analysis and event-related response detection, sound analysis including vocal stress indicators andAttorney Docket No. SETH-005 AWO emotional state recognition using machine learning algorithms that integrate with the natural language processing system, muscle contractions with EMG pattern recognition and tension assessment across multiple muscle groups, temperature regulation and thermal comfort assessment with spatial and temporal pattern analysis, and other physiological indicators detailed throughout this disclosure including respiratory patterns, pupil dilation, and micro-movement detection.

[0322] The comprehensive sensor array provides multiple independent measures of user state and treatment effectiveness, enabling robust assessment even when individual sensors experience artifacts or failures.

[0323] In addition to sensory responses to the stimulation, other devices, such as a camera 406 for facial expression analysis using computer vision algorithms trained on validated emotion recognition databases with real-time processing capabilities, or an fNIRS device 408 for cortical blood flow monitoring with spatial pattern analysis and hemodynamic response modeling, may observe and quantify the user's physiological and emotional responses with high temporal and spatial resolution.

[0324] The integration of advanced sensing modalities enables sophisticated assessment of user state that goes beyond simple physiological measures to include emotional and cognitive responses to treatment.

[0325] Feedback information from one or more of the sensors 404, camera 406, and fNIRS device 408 may be collected, processed, and transmitted to a pain / pleasure detection and processing module 410 using secure, real-time data transmission protocols. The processing module 410 may analyze the various response signals and data from the user using advanced signal processing techniques including digital filtering, artifact rejection, and noise reduction, machine learning algorithms including deep neural networks and ensemble methods, pattern recognition systems that can distinguish between different types of physiological responses, and natural language processing algorithms that interpret user voice commands and vocal stress patterns to determine how to adapt the applied stimulation for optimal therapeutic or enhancement outcomes.

[0326] The processing module represents the intelligent core of the system, integrating multiple data streams and applying sophisticated algorithms to optimize treatment effectiveness in real time.

[0327] A memory 412 may store and maintain a comprehensive library of algorithms and data obtained from one or more users, including individual user profiles with physiological baselines and response patterns, population-level patterns and validated therapeutic protocols, validated stimulation protocols for specific conditions and applications, continuously updated machine learning models that improve performance over time, and natural language interaction patterns and vocabulary specific to individual users. The memory 412 may further include learned or programmed profiles that correlateAttorney Docket No. SETH-005 AWO feedback data with optimal stimulation parameters, enabling personalized treatment recommendations and automated parameter optimization based on accumulated experience and validated therapeutic approaches.

[0328] The comprehensive data storage and management system enables both individual personalization and population-level learning while maintaining appropriate privacy protections and user control over data sharing.

[0329] A calculator 414 may translate data from the memory 412 into one or more specific settings to be applied to the stimulation device, e.g., the TENS unit 418, using optimization algorithms including gradient-based methods, evolutionary algorithms, and reinforcement learning approaches, real-time decision-making systems that can adapt to changing user conditions and treatment responses, and natural language command interpretation that can translate user voice instructions into specific parameter adjustments. Those settings may be processed by the API controller 416 resulting in adapted stimulation generated and applied to the user according to the various stimulation parameters 420 with real-time safety monitoring and automatic adjustment capabilities. Accordingly, the adaptive stimulation device, e.g., TENS unit 418, is continuously and in real-time updated to adapt the applied stimulation based on measured effectiveness and user response patterns while maintaining safety constraints and user preferences.

[0330] The closed-loop control system enables autonomous optimization while maintaining appropriate human oversight and safety monitoring to ensure optimal treatment outcomes.

[0331] Advanced Proxy Learning and Machine Learning Integration

[0332] Turning to FIG. 5, the adaptive stimulation system 500 may include a machine learning proxy model 502 connected to, or in communication with, the processing module 410, the memory 412, and the calculator 414 through secure, high-speed data connections that enable real-time model inference and updating. The proxy model 502, as detailed herein, may include sophisticated machine learning techniques including deep neural networks with convolutional and recurrent architectures optimized for physiological signal processing, ensemble methods that combine multiple algorithms for improved accuracy and robustness, transfer learning approaches that enable rapid adaptation to new users and conditions, and natural language processing models that can interpret user speech patterns and integrate conversational feedback into optimization algorithms, to leam and predict user responses from applied stimulation, as well as generate proxy measurements based on direct measurements from devices like the sensors 404, camera 406, and / or fNIRS device 408.Attorney Docket No. SETH-005 AWO

[0333] The proxy learning system represents a breakthrough in making sophisticated neuroscience accessible through consumer-grade technology while maintaining clinical-grade accuracy and effectiveness.

[0334] As detailed throughout this disclosure, certain sophisticated or advanced measurements may not be feasible, financially or structurally, for some device implementations due to cost constraints, size limitations, or regulatory requirements. Accordingly, the proxy model 502 may analyze, generate, and learn certain hybrid combinations of measurements from the physically present sensors that represent or correlate to those sophisticated but unavailable measurements through validated machine learning models trained on high-quality reference data. The use of proxy measurements may reduce implementation costs while generating valuable and effective data representative of measurements typically obtained from more invasive and complicated devices, enabling broader accessibility and deployment across diverse user populations and applications.

[0335] The proxy approach democratizes access to sophisticated physiological monitoring by making laboratory -quality insights available through consumer-grade devices and accessible price points.

[0336] The proxy learning system enables both individual user adaptation and group-level crowd learning, where patterns discovered across multiple users can inform initialization strategies for new users and improve overall system performance through federated learning approaches that preserve privacy. The system can distinguish between universal physiological patterns that apply across all users and individual-specific responses that require personalized adaptation, enabling personalized adaptation while leveraging collective intelligence for improved effectiveness and discovery of therapeutic approaches.

[0337] The combination of individual and population-level learning creates a continuously improving ecosystem that benefits all users while respecting individual privacy and preferences.

[0338] Proxy-Only Implementation and Scalable Deployment

[0339] Referring now to FIG. 6. an adaptive stimulation system 600 is shown in which stimulation feedback is obtained primarily from accessible sensors 404, with the proxy measurement component integrated into the processing module 410 for streamlined implementation and reduced system complexity. In such an exemplary system, the processing module may obtain sensory data from the sensor 404 and the user 402 and determine or generate proxy measurements based on that data as well as historical data and population models stored in the memory 412 using validated machine learning algorithms that have been trained on high-fidelity reference data.Attorney Docket No. SETH-005 AWO

[0340] This implementation approach enables widespread deployment of adaptive stimulation technology7while maintaining clinical-grade effectiveness through carefully trained predictive models that can operate using only consumer-grade sensors.

[0341] Proxy Learning Validation and Operational Transition Protocol

[0342] According to one or more aspects, the proxy learning system implements a structured validation and transition protocol ensuring consumer-grade sensors achieve clinically acceptable accuracy before transitioning from high-fidelity measurement dependence to autonomous proxy -based operation. This protocol addresses the challenge of democratizing sophisticated physiological monitoring by establishing a systematic workflow for validation, transition, and ongoing quality assurance.

[0343] Phase 1: Concurrent High-Fidelity' and Proxy Data Collection: The initial training phase employs simultaneous operation of high-fidelity’ clinical measurement equipment (fNIRS, researchgrade EEG, or other validated neuroimaging modalities) and consumer-grade proxy sensors (heart rate variability monitors, galvanic skin response sensors, facial expression cameras, respiratory' rate monitors, and skin temperature sensors). This concurrent operation continues for a calibration period ensuring adequate training data collection across diverse physiological states and stimulation parameters.

[0344] Proxy-Model Training and Deployment Pipeline (Data Flow)

[0345] FIG. 6A illustrates a proxy -model training and deployment process that supports the adaptive stimulation architectures described in FIGS. 4-6. During the training phase, high-fidelity' measurements 602 from research-grade modalities such as fNIRS or EEG are correlated with simultaneous proxy-sensor data 604 including HRV, GSR, temperature, facial EMG, and voice features. A feature-extraction module 606 filters, normalizes, and synchronizes both data sets before the training engine 608 builds a machine-learning model that maps proxy inputs to target outputs. Validation and testing 610 compare predicted versus true responses, establishing a training-to-deployment boundary' 612 where models transition from supervised generation to runtime autonomy. In the deployment phase, the trained model 614 runs locally or via cloud processing to interpret proxy signals and drive controller 110 and feedback mechanism 106 for real-time stimulation adaptation. A data logger 616 may record anonymized features for periodic federated updates and ongoing model refinement.

[0346] Training Data Requirements: The system collects paired high-fidelity and proxy measurements across multiple stimulation sessions to capture representative physiological response patterns. Data collection spans multiple days to capture day-to-day physiological variability, circadian rhythm effects, and diverse contextual factors. Training sessions explore diverse stimulation parameterAttorney Docket No. SETH-005 AWO combinations including amplitude variations, frequency modulation, and spatial targeting to ensure model generalization across the operational parameter space.

[0347] Temporal Synchronization: High-fidelity and proxy measurements maintain precise temporal alignment to ensure proxy measurements correlate with corresponding high-fidelity states rather than temporally offset physiological responses. The system implements hardware-level timestamp synchronization or software-based cross-correlation alignment for measurements from independent sensor systems.

[0348] Phase 2: Model Training and Validation: Upon completing data collection requirements, the system trains predictive models mapping proxy sensor inputs to high-fidelity measurement outputs using supervised machine learning algorithms. The training process implements rigorous validation methodology7to assess model performance.

[0349] Training Methodology: The collected dataset undergoes partitioning into training, validation, and held-out test subsets with stratification ensuring balanced representation of different physiological states across all subsets. Multiple machine learning architectures are trained including deep neural networks for capturing complex non-linear relationships, ensemble methods combining multiple algorithms, and recurrent neural networks for temporal sequence modeling accounting for physiological response dynamics.

[0350] Performance Assessment: Models undergo evaluation using multiple performance metrics including correlation between proxy -predicted values and ground-truth high-fidelity measurements, classification accuracy for discrete physiological states, and temporal stability' to avoid erratic parameter adjustments. Only models meeting validation criteria proceed to deployment consideration, with failed models undergoing retraining with augmented data collection or alternative architecture exploration.

[0351] Phase 3: Supervised Proxy Transition with Ongoing Validation: Models meeting validation criteria enter a supervised transition phase where proxy -based operation occurs concurrently with periodic high-fidelity validation. This phase ensures real-world proxy performance matches laboratory validation results.

[0352] Transition Protocol: The system operates in proxy -only mode for therapeutic parameter optimization, using trained models to infer physiological states from consumer-grade sensors without real-time high-fidelity measurement. However, high-fidelity sensors remain available and activate periodically for validation, assessing correlation maintenance between proxy predictions and actual high-fidelity measurements. The system tracks rolling performance metrics across validation sessions, comparing against initial validation performance baselines.Attorney Docket No. SETH-005 AWO

[0353] Transition Acceptance: Full transition to autonomous proxy-only operation requires maintained accuracy across multiple validation checkpoints, consistency indicating no catastrophic performance degradation episodes, and stable or improving user-reported therapeutic effectiveness confirming proxy -based optimization maintains clinical utility.

[0354] Phase 4: Autonomous Proxy Operation with Degradation Monitoring: Upon successful transition validation, the system enters fully autonomous proxy-only operation mode where high-fidelity sensors are no longer required for routine operation. However, ongoing quality monitoring continues through scheduled recalibration sessions using high-fidelity sensors at appropriate intervals for ongoing performance monitoring, with recalibration frequency adapted based on degradation indicators.

[0355] Degradation Detection and Automatic Reversion: The system continuously monitors proxy performance indicators detecting potential model degradation through prediction instability7suggesting noise sensitivity or artifact susceptibility, therapeutic effectiveness decline compared to historical baseline, physiologically implausible predictions indicating model failure, or validation failure during scheduled recalibration sessions. Upon detecting degradation indicators, the system automatically reverts to requiring high-fidelity7sensor operation, alerts the user to degraded proxy performance, initiates data collection for model retraining, and prevents autonomous proxy operation until revalidation criteria are met. This fail-safe mechanism ensures the system never operates with insufficient accuracy, maintaining user safety and therapeutic effectiveness.

[0356] User-Specific vs. Population-Based Models: The proxy learning system supports both personalized and population-based operational modes. Personalized models trained exclusively on individual user data achieve optimal accuracy but require full multi-phase training. Population-based transfer learning initializes new users with models pretrained on demographically similar cohorts, requiring abbreviated calibration before transition. Hybrid approaches begin with population priors and progressively personalize through ongoing data collection, balancing rapid deployment w ith ultimate personalization accuracy. This structured validation and transition protocol ensures the proxy learning system maintains clinical-grade accuracy while democratizing access to sophisticated physiological monitoring through consumer-affordable sensor systems. The workflow provides systematic risk mitigation without requiring the high-fidelity sensors to remain permanently connected, enabling cost- effective deployment while maintaining quality assurance through periodic validation checkpoints.

[0357] The proxy-only system represents the most scalable approach for consumer applications, leveraging the benefits of sophisticated research while operating within practical cost and complexity constraints that enable broad market adoption. This approach maintains the core adaptive capabilitiesAttorney Docket No. SETH-005 AWO while reducing hardware requirements and implementation complexity, making the technology accessible to larger user populations and diverse application domains.

[0358] The scalable deployment model enables rapid market penetration while maintaining the sophisticated learning and optimization capabilities that distinguish the system from conventional stimulation devices.

[0359] TENS PLUS

[0360] Introduction and Technical Context

[0361] The following embodiment description, as one of many exemplary practical applications of the concepts, techniques and structures described herein, demonstrates a practical implementation of the adaptive stimulation principles described herein, specifically addressing critical limitations in current transcutaneous electrical nerve stimulation (TENS) technology7while providing a commercially viable pathway for deploying objective pain measurement and automated optimization in clinical and consumer settings.

[0362] Current TENS devices suffer from an architectural constraint: they rely on fixed electrode placement and static stimulation parameters that cannot adapt to the dynamic nature of pain or individual physiological variation. Traditional four-pad configurations require users to manually position electrodes based on general anatomical guidelines rather than individualized optimization, leading to suboptimal therapeutic outcomes and frequent device abandonment.

[0363] The static nature of conventional TENS presents several critical problems, including that optimal stimulation locations vary between individuals and over time due to physiological adaptation; manual electrode repositioning is cumbersome and often imprecise, particularly for users with limited mobility; the constrained four-pad architecture prevents discovery of more effective stimulation sites within a broader anatomical region; and users lack objective feedback about stimulation effectiveness, relying solely on subjective pain perception to guide manual adjustments.

[0364] Integration of Objective Pain Measurement

[0365] The present embodiment addresses these limitations by integrating real-time (seconds) and / or near-real-time (tens of seconds to minutes, but during a stimulation session) objective pain measurement using functional near-infrared spectroscopy (INIRS). The system provides objective quantification of pain states, enabling automated optimization without relying on subjective user reporting or clinical guesswork.

[0366] Technical and Commercial Viability

[0367] The described embodiment represents a convergence of several technological advances: miniaturized control circuits enabling dense electrode arrays; wireless communication protocolsAttorney Docket No. SETH-005 AWO optimized for medical applications; machine-learning algorithms capable of real-time and near-real-time parameter optimization; and cost-effective neuroimaging suitable for clinical deployment. Importantly, the system includes a sophisticated proxy-learning framework that enables transition from expensive neuroimaging to accessible consumer sensors while maintaining clinical effectiveness.

[0368] Embodiment Scope and Applications

[0369] This specific embodiment focuses on TENS applications for pain management, but the underlying adaptive stimulation principles are broadly applicable to other therapeutic and enhancement applications such as muscle stimulation, wound healing, and sensory modulation. The modular architecture supports deployment across clinical, home-healthcare, and consumer markets through configurable hardware and software implementations.

[0370] Referring now to FIG. 7, a TENS-Based Embodiment with sensor feedback is shown. According to one or more aspects, the system 700 includes a TENS unit 702 housing controller 708 that communicates wirelessly through interface 710 with a wearable proxy sensor 712. The wearable proxy sensor 712 acquires physiological proxy signals such as heart-rate variability (HRV) and galvanic skin response (GSR), which are processed by controller 708 to inform stimulation adjustments. A high- fidelity neuroimaging module 706 (functional near-infrared spectroscopy, fNIRS) provides calibration feedback by measuring cortical hemodynamic responses. Controller 708 then modulates stimulation delivered through the adaptive electrode matrix 704, forming a closed-loop adaptive system

[0371] Advanced TENS System with Real-Time Pain Feedback and Adaptive Control

[0372] System Architecture Overview

[0373] FIG. 7 illustrates an embodiment of the adaptive stimulation system 700 implementing an advanced TENS configuration with real-time (seconds) and / or near-real-time (tens of seconds to minutes, but during a stimulation session) pain monitoring and intelligent parameter optimization during a stimulation session. This embodiment demonstrates the practical implementation of the core adaptive- stimulation principles within a clinically viable and commercially scalable framework.

[0374] As illustrated in FIGS. 3A-3C. the adaptive stimulation architecture supports four learning and sensing modes that differ in their balance between direct-neural and proxy feedback as well as in how their models are initialized. Within this framework, the TENS-plus embodiment operates primarily in the Proxy-Direct mode (3A-2). In this configuration, high-fidelity neural or hemodynamic sensing (for example, fNIRS or EEG) may be used temporarily during calibration but is not required for continuous feedback; for cost and convenience, routine operation may rely on proxy-based physiological and behavioral inputs. During paired calibration, the device records direct neural or hemodynamic signals (such as EEG or fNIRS) together with proxy signals (HRV, GSR, facial EMG, voice) on theAttorney Docket No. SETH-005 AWO same user to train a persistent f (proxies — > state) mapping. After training, runtime control relies on proxies only while maintaining the user-specific mapping (see FIG. 3B). Research configurations may employ a Direct Neural Loop (e g., mode 3A-1, described above) using continuous neural feedback, whereas consumer modes may initialize from mode 3A-3 or mode 3A-4 for proxy-only operation. In one aspect, optimization and safety follow the multi -timescale control architecture shown in FIG. 3C, integrating fast-safety, in-session optimization, and cross-session learning loops.

[0375] According to one or more aspects, the system 700 comprises four primary subsystems, including a TENS unit 702 with Bluetooth wireless connectivity for remote control, a flexible multiarray electrode pad 704 providing spatially distributed stimulation points with individual Bluetooth- controlled electrode circuits, a real-time neuroimaging module 706 using functional near-infrared spectroscopy (fNIRS) for direct pain-state measurement, and an intelligent control system 708 that coordinates all subsystems and implements machine-learning algorithms for continuous optimization and exploration of the stimulation parameter space.

[0376] TENS Unit wi th Bluetooth Control

[0377] The TENS unit 702 can utilize standard, commercially available TENS hardware with the addition of Bluetooth wireless communication capability 710, enabling remote control from the intelligent control system. Rather than requiring significant hardware advancement, one modification is the integration of Bluetooth connectivity that allows external control of the unit’s stimulation parameters.

[0378] The Bluetooth interface 710 operates using standard protocols for medical device applications, with encryption to protect patient data and stimulation commands. The unit maintains traditional TENS safety features including current limiting, short-circuit protection, and emergency -stop functionality, while accepting remote commands for waveform generation, frequency modulation, amplitude control, and timing patterns.

[0379] Instead of relying on the typical TENS unit control surface with preset massage types and program shortcuts, the present embodiment implements direct parameter control through the intelligent control system. This enables precise, real-time adjustment of all stimulation characteristics based on feedback rather than limiting users to manufacturer-programmed presets. Complex multi-phase waveform patterns can be implemented as an optional enhancement, with basic monophasic and biphasic patterns sufficient for initial deployment.

[0380] Flexible Multi-Array Electrode Pad System

[0381] In one aspect, the electrode array surface 704 is designed to overcome the limitations of traditional TENS electrode placement by providing a flexible, intelligent stimulation surface that canAttorney Docket No. SETI4-005AWO dynamically select optimal electrode pairs without requiring physical repositioning. The primary objectives are to enable discovery7of optimal stimulation locations within a large anatomical area, to provide multiple independent electrode pairs that can operate simultaneously, to allow dynamic reconfiguration of electrode pairs based on real-time and / or near-real-time effectiveness feedback, and to maintain the electrical isolation and safety characteristics required for transcutaneous electrical stimulation.

[0382] FIG. 8 is a diagram of the layers of the flexible multi-electrode pad 704, according to one or more aspects of the disclosure. The pad 704 comprises a biocompatible flexible substrate 714 with embedded conductive traces 715 connecting individually addressable electrode nodes 716, each containing an impedance and safety circuit 718. A skin-safe conductive adhesive layer 720 ensures electrical coupling to the skin, while insulation layer 722 isolates conductive pathways. Connector edge 724 links the pad to the TENS unit 702 through wireless interface 710. The grid architecture allows dynamic, pair-based selection of stimulation sites (e.g., nodes 716a, 716b) for adaptive transcutaneous electrical stimulation.

[0383] Multi-Array Architecture and Pair-Based Operation

[0384] In one aspect, the flexible multi-array electrode pad 704 comprises a biocompatible substrate 714 measuring approximately 15 cm * 20 cm or larger, capable of covering major anatomical regions such as the shoulder complex, lumbar region, or upper trapezius muscles. The substrate incorporates a grid-based architecture with multiple stimulation points 716 arranged in a regular matrix pattern, ty pically implementing an 8 x 10 or 10 * 12 array yielding about 80 to 120 potential stimulation locations.

[0385] In one aspect, the system operates on a pair-based architecture in which stimulation occurs between tw o points to complete the electrical circuit, consistent with TENS operational principles. Multiple electrode pairs can operate simultaneously, with each pair independently controlled for current amplitude, waveform characteristics, and temporal patterns. The central control system can dynamically select any combination of points to form active pairs, enabling systematic exploration of different electrode configurations within the coverage area.

[0386] Each intersection point in the grid 716 in the grid of the electrode pad 704 incorporates a miniaturized impedance and safety7control circuit 718 capable of independent activation and current regulation under central command. The centrally controlled architecture ensures coordinated operation of all electrode pairs while maintaining individual point control for optimal current distribution and safety monitoring.

[0387] Control Circuit Architecture for Pair-Based OperationAtorney Docket No. SETH-005 AWO

[0388] Each control circuit 718 includes its own Bluetooth communication capability to receive commands directly from the intelligent control system, enabling independent control of each electrode point while maintaining centralized coordination. This architecture allows standard TENS units to be enhanced with the flexible electrode pad system through Bluetooth connectivity without requiring significant modifications to the base TENS hardware.

[0389] In one aspect, the electrical specifications for pair operation include an output current range of approximately 0. 1 mA to 50 mA per electrode point, current matching of about ± 1 percent between paired electrodes, voltage compliance up to approximately 100 volts to accommodate varying skin impedance, and frequency response of about 10 kHz (-3 dB bandwidth). Inter-electrode isolation is greater than 10 M£1 between non-paired points. One skilled in the art will recognize that the values described herein are exemplary and other values for currents, voltages, frequencies and the like are possible.

[0390] Safety features for multi-pair operation include automatic current limiting based on calculated current density per electrode; skin-impedance monitoring with automatic disconnection for values outside safe ranges (approximately 100 Q to 10 kQ); temperature monitoring via an integrated thermistor at each point; cross-pair interference detection and mitigation; and an emergency -stop capability with response time of less than about 10 milliseconds across all active pairs.

[0391] Centralized communication and control are provided through a Bluetooth interface for command and status exchange from each point, with individual node addressing supporting up to 256 control points. The system enables synchronized operation across multiple electrode pairs and provides real-time status reporting for impedance, current, and safety parameters.

[0392] Advanced Substrate Construction for Multi-Array Functionality

[0393] The flexible substrate 714 utilizes a multi-layer construction engineered to support the complex electrical and mechanical requirements of the multi-array electrode system.

[0394] Primary Conductive Layer: Silver-chloride traces (715) provide low-resistance current paths with minimal electrochemical polarization, arranged in a grid patern that enables any-to-any point connectivity while maintaining electrical isolation between independent electrode pairs.

[0395] Insulation and Control Layer: A medical-grade polymer layer (722) incorporates the control circuits and provides electrical isolation between conductive pathways. This layer includes embedded microcontrollers and communication circuits while maintaining substrate flexibility.

[0396] Power Distribution Network: An integrated power-distribution system uses low-voltage DC (approximately 12 volts) distributed via dedicated conductors within the substrate, with local voltage regulation at each control circuit to ensure consistent operation across all electrode points(716).Attorney Docket No. SETH-005 AWO

[0397] Biocompatible Interface Layer: The surface features a biocompatible adhesive layer 720 incorporating conductive hydrogel that provides secure skin contact and electrical conductivity while allowing for multiple applications and removals. The adhesive system is engineered to maintain consistent impedance characteristics across all electrode points.

[0398] Protective and Serviceability Features: A replaceable protective cover 722 enables cleaning and sterilization between uses, extending the operational life of the device while protecting the complex internal circuitry.

[0399] Real-Time and Near-Real -Time Pain Measurement Using fNIRS

[0400] Referring back to FIG. 7, the neuroimaging module 706 employs a wearable functional near-infrared spectroscopy (fNIRS) system designed specifically for pain-state detection and monitoring. The module incorporates multiple light-emitting diode (LED) sources and photodetectors positioned to monitor cortical regions associated with pain processing, including the prefrontal cortex, somatosensory cortex, and anterior cingulate cortex.

[0401] The fNIRS system operates by continuously measuring changes in blood-oxygenation levels (hemodynamic response) that correlate with neural activity in pain-processing brain regions. The system samples at frequencies of about 10 Hz, providing opportunities for real-time (seconds) and / or near-real-time (tens of seconds to minutes, but during a stimulation session) measurement of pain-state changes with temporal resolution sufficient for closed-loop control applications.

[0402] As shown in FIG. 8A, signal -processing algorithms 730 implemented in the module’s embedded processor filter noise, correct for motion artifacts, and extract features 731 specifically associated with nociceptive processing. The system employs machine-learning models 732 trained on validated pain datasets to convert raw hemodynamic signals into standardized pain-intensity scores, providing objective quantification of the subjective pain experience.

[0403] The fNIRS design prioritizes user comfort and compliance. The headset or band uses lightweight, ergonomic materials that maintain consistent optode-skin contact while minimizing motion interference. Wireless data transmission 734 to the intelligent control system 708 eliminates cable- related movement restrictions while maintaining continuous monitoring capability throughout treatment sessions. For research or clinical use, high-fidelity calibration modes may operate with multi-channel fNIRS arrays; for cost and convenience, routine operation may rely on a reduced-channel or proxybased configuration while preserving functional correlation with the full neural measurement.

[0404] Integration of this neuroimaging module enables the adaptive stimulation system to measure treatment effectiveness objectively, allowing the controller to modify stimulation parameters automatically based on detected neural activity patterns rather than relying on user-reported outcomes.Attorney Docket No. SETH-005 AWOThis capability provides the basis for closed-loop optimization and for training proxy models that relate easily measured physiological signals to brain-level indicators of pain relief.

[0405] As shown in FIG.8A , the wearable fNIRS module 706 employs multiple light sources 726 and photodetectors 728 with on-board preprocessing 730 and model inference 732 to compute a pain score P(t) that is transmitted wirelessly 734 to controller 708 for closed-loop control. In one aspect, the head-mounted fNIRS module 706 (LED sources 726 and photodetectors 728) acquires cortical hemodynamic signals from pain-processing regions. Embedded preprocessing 730 and a trained model 732 generate a time-varying pain score P(t) that is sent wirelessly 734 to controller 708, which adapts TENS unit 702 and electrode matrix 704 in closed loop.

[0406] Intelligent Control System and Advanced Algorithms

[0407] The intelligent control system 708 serves as the central coordination hub that integrates data from the fNIRS pain-measurement module 706, controls both the TENS unit 702 and the flexible electrode pad 704 via Bluetooth communications, and implements the machine-learning algorithms necessary for adaptive stimulation optimization. The system operates on a handheld computing device 736 such as a smartphone, tablet, or dedicated medical controller. The device receives real-time and / or near-real-time pain-related blood-flow measurements from the fNIRS module 706, processes this information using advanced algorithms, and transmits optimized stimulation commands to both the TENS unit 702 and the individual electrode control circuits 718 in the flexible pad system.

[0408] FIG. 8B illustrates the controller 708 data path and optimization layers. Inputs include the fNIRS-derived pain score P(t) from module 706 and proxy biosignals (e.g., HRV / GSR) from wearable sensor 712, together with electrical telemetry (impedance / current) from the stimulation path. A signalquality and safety gate 709 filters artifacts and enforces bounds; a state estimator 711 derives a painstate vector and trend; and the optimization engine 738 (fast-safety, in-session, and cross-session learning) generates parameter updates through a command synthesizer 713 — amplitude, frequency, waveform, and spatial electrode-pair selection — for TENS unit 702 and matrix 704.

[0409] FIGS. 9 -10 illustrate the diagnostic-probing and adaptive-targeting sequence implemented through the flexible multi-array electrode pad system described above. During an initial diagnostic phase, the controller sequentially activates electrode pairs across the array while monitoring physiological responses such as autonomic-response index, impedance change, or pain-related biomarkers. As shown in FIG. 9, these data are compiled into a ranked-effectiveness map that identifies regions producing the strongest physiological response (where darker shading indicates higher relative effectiveness, e.g., on a scale from 0-9). FIG. 10 illustrates the transition from diagnostic probing to therapeutic stimulation, in which the controller automatically selects the highest-ranked electrode pairsAttorney Docket No. SETH-005 AWO and configures the TENS unit for closed-loop operation. This diagnostic-to-therapeutic workflow enables automatic discovery' of optimal stimulation sites without manual electrode repositioning and provides precise targeting for subsequent adaptive control.

[0410] An additional feature of the intelligent control system is the integration of audio input capability that allows users to initiate sessions by describing their desired therapeutic outcome. Users can provide natural-language descriptions such as “I want relief from sharp pain in my shoulder” or “I need gentle stimulation for muscle relaxation.” The system employs natural-language-processing algorithms to interpret these requests and initialize stimulation parameters accordingly, providing an intuitive interface that requires no technical knowledge from the user. The audio input system includes speech -recognition processing to convert user voice input into text, natural -language parsing to extract therapeutic intent and target sensations, intent classification to categorize requests into appropriate stimulation strategies, and parameter initialization to configure starting electrode pairs, waveforms, and intensities based on the interpreted user request.

[0411] The pain-imaging and sensing subsystem interprets neuroimaging data from the fNIRS module and converts it into actionable control signals for the adaptive stimulation algorithms. The system leverages validated pain-biomarker research to extract relevant pain-state information while accommodating the practical constraints of real-time control. The processing pipeline acquires and filters raw hemodynamic data to remove noise and motion artifacts, extracts temporal and spatial features associated with nociceptive activity7in key cortical regions, and applies pre-trained machinelearning models to infer quantitative pain-intensity7estimates and classify pain traj ectories as increasing, decreasing, or stable. A multidimensional pain-state vector is produced, incorporating current intensity7, rate of change, and spatial distribution of activation. This pain-state vector is transmitted to the optimization engine 738, where it serves as input for stimulation parameter adjustment. These algorithms operate in real time, within seconds, or in near-real-time intervals on the order of tens of seconds to minutes during a stimulation session, enabling the controller to modify stimulation immediately in response to changing pain levels.

[0412] The parameter-discovery and learning subsystem explores the high-dimensional space of potential TENS configurations to identify those that produce optimal therapeutic outcomes for each user. The search space includes electrode-pair selection, waveform characteristics, pulse width and polarity, frequency parameters, amplitude levels, and temporal stimulation patterns. The discovery algorithms must balance exploration of unknown combinations with exploitation of settings known to be effective. The system employs coordinated learning strategies that integrate Bayesian optimization for sampleefficient search, multi-armed-bandit approaches for adaptive allocation of testing time, evolutionary7Attomev Docket No. SETH-005 AWO computation for population-based exploration of promising parameter sets, and gradient-based or reinforcement-learning techniques for fine-tuning based on accumulated reward signals. Each algorithm operates within safety and comfort constraints enforced by the real-time safety layer of the controller. Over successive sessions, the results of each optimization cycle are aggregated to improve initial parameter estimates, reducing the cold-start time for new users and accelerating convergence toward effective personalized settings. Population-level learning can also be implemented through federated models that allow devices to share encrypted parameter-effectiveness updates without transmitting identifiable physiological data, enabling collective performance improvements while maintaining user privacy.

[0413] The optimization engine 738 therefore implements a multi-objective control framework that balances pain reduction, user comfort, and safety7. The system maintains a dynamic model of the multidimensional parameter space — including waveform characteristics, frequency modulation, amplitude control, and spatial electrode-pair configurations — and treats pain level as the primary7objective function to be minimized. Optimization algorithms operate within the same closed-loop architecture defined in FIG. 3C, integrating fast-safety7adjustments, in-session optimization, and longterm learning loops. Continuous monitoring of signal quality and impedance ensures that all algorithmic adjustments remain within safe electrical and physiological boundaries.

[0414] Throughout operation, the controller continuously evaluates incoming sensor data for consistency, signal quality7, and adherence to safety thresholds. The embedded watchdog logic provides immediate interruption of stimulation when measured impedance, current, or physiological variables exceed safe limits. All data streams are time-stamped, encrypted, and logged for clinical validation and cross-session learning, ensuring both device safety and regulatory traceability. The intelligent control system thus forms the adaptive core of the TENS-plus embodiment, unifying neuroimaging-based sensing, parameter discovery7, proxy inference, and user interaction into a closed-loop framework that continuously learns from every7stimulation cycle.

[0415] As shown in FIG. 11, the adaptive control architecture of the TENS-plus embodiment operates across multiple nested feedback loops that function over distinct time scales to balance immediate safety7, session-level optimization, and long-term personalization.

[0416] FIG. 11 illustrates the hierarchical control structure comprising a fast-safety loop (1002), an in-session optimization loop (1004), and a cross-session learning loop (1006). Immediate safety7control monitors physiological distress signals and can halt stimulation within milliseconds. The in- session optimization layer continuously adjusts waveform, amplitude, and electrode configuration to maintain therapeutic effectiveness using Bayesian or reinforcement-learning algorithms. The cross-Attorney Docket No. SETH-005 AWO session learning layer aggregates effectiveness data across sessions, updates proxy-model parameters, and supports federated or population-level learning. Together these loops form the adaptive hierarchy that governs all embodiments of the stimulation system.

[0417] Pain Imaging and Sensing Algorithm Integration[0041S] The pain-imaging and sensing subsystem interprets complex neuroimaging data from the fNIRS module 706 and converts it into actionable control signals for the adaptive stimulation algorithms. This embodiment builds upon validated neuroimaging research that links cortical hemodynamic activity in pain-processing regions to subjective pain intensity and relief. The system extracts relevant pain-state information in real time and transmits it to the intelligent control system for closed-loop modulation of stimulation parameters.

[0419] During operation, the subsystem continuously acquires raw hemodynamic signals from the fNIRS module and applies embedded preprocessing to remove noise and correct for motion artifacts. The filtered data are analyzed for characteristic features of nociceptive processing, including amplitude and phase changes in oxygenated and deoxygenated hemoglobin across monitored cortical regions. These features are combined into a multidimensional representation of neural activation patterns associated with pain perception. Machine-learning models trained on clinical datasets transform these feature vectors into quantitative pain-intensity estimates and trend indicators that reflect whether pain is increasing, decreasing, or stable.

[0420] The controller generates a real-time pain-state vector comprising current intensity, temporal trajectory, and spatial distribution of cortical activation. This vector is transmitted to the optimization engine 738, where it guides adjustment of stimulation parameters such as amplitude, frequency, waveform shape, and electrode configuration. The feedback loop thereby enables automatic modulation of stimulation strength in response to measured pain-related neural activity7. The process occurs continuously in real time (on the scale of seconds) or near-real-time (tens of seconds to minutes, but during a stimulation session), ensuring that the system can adapt dynamically to the user’s changing physiological state.

[0421] Discovery of Settings and Learning Algorithms

[0422] According to one or more aspects, the discovery’ subsystem identifies stimulation parameter combinations that produce optimal therapeutic outcomes for individual users. Because the TENS architecture supports multiple degrees of freedom — including electrode selection, waveform shape, pulse width, polarity, frequency, amplitude, and temporal pattern — the search space is large and highly individualized. This embodiment employs coordinated algorithmic exploration to efficiently discover effective settings without compromising safety or comfort.Attorney Docket No. SETH-005 AWO

[0423] The system begins each session with initial parameters derived from prior user data or population-level priors. It then explores variations in stimulation parameters while monitoring objective effectiveness metrics derived from fNIRS or proxy sensors. During this process, the controller balances exploration of new combinations with exploitation of previously successful settings to ensure continuous improvement of therapeutic performance.

[0424] A combination of learning strategies is employed. Bayesian optimization provides sample-efficient navigation of the multidimensional parameter space. Multi-armed-bandit frameworks dynamically allocate testing time to parameter sets showing the most promise. Evolutionary algorithms generate and evaluate populations of parameter combinations, promoting diversity and avoiding local minima. Gradient-based methods refine continuous variables such as amplitude and frequency, while reinforcement-learning models accumulate experience across sessions to identify parameter trends yielding the highest effectiveness scores.

[0425] All exploration occurs within hard safety and comfort boundaries enforced by the controller’s real-time protection logic. The system halts or scales back intensity immediately upon detecting adverse physiological indicators such as increased muscle guarding or sympathetic activation. Over successive sessions, the accumulated data are used to update individualized priors, shortening convergence time for future optimization cycles and improving reproducibility of pain-relief outcomes.

[0426] For clinical and research implementations, population-level learning can be achieved through federated model aggregation. Devices upload anonymized, encrypted gradient or parameterperformance data to a remote server where population models are refined. Updated model parameters are then distributed back to local devices, enabling continual global improvement while maintaining user privacy and regulatory compliance.

[0427] Proxy Measurement Development and Integration

[0428] According to one or more aspects, the proxy -measurement subsystem enables accurate inference of therapeutic effectiveness using accessible, lower-cost sensors that correlate with high- fidelity neuroimaging measurements. The training and deployment workflow of the proxy -measurement subsystem is described above in connection with FIG. 6A. The objective of this embodiment is to allow the adaptive stimulation system to achieve clinical-grade performance while eliminating dependence on continuous neuroimaging.

[0429] During calibration, the system records simultaneous data from both high-fidelity modalities such as fNIRS or EEG and accessible proxy sensors that measure physiological variables including heart-rate variability (HRV), galvanic skin response (GSR), skin temperature, electromyography (EMG), facial expression, voice stress, and motion dynamics. These paired datasetsAttorney Docket No. SETH-005 AWO are used to train a mapping function f(proxies — > state) that models the relationship between proxy indicators and neural or hemodynamic reference signals. Once validated, the trained model is stored within the intelligent control system 708 and used for real-time inference of user state without requiring direct neural measurement.

[0430] For cost and convenience, routine operation may rely entirely on these proxy inputs, while high-fidelity sensors can be employed periodically for recalibration or verification. The system can therefore function seamlessly across multiple operating modes — from research-grade configurations using full neuroimaging feedback to consumer-grade versions relying solely on proxy data.

[0431] The proxy-learning process operates through a structured pipeline. During concurrent data acquisition, proxy and reference signals are collected simultaneously to establish ground-truth correlations. Feature-extraction algorithms derive statistical and temporal characteristics from each measurement modality. Correlation analysis determines which proxy features most reliably track with fNIRS-derived pain states. Ensemble and deep-leaming methods are then trained to integrate multiple proxies into a unified predictive model that estimates pain-state variables and therapeutic effectiveness.

[0432] To maintain reliability, the system employs sensor-fusion and confidence-weighting logic that prioritizes modalities with stable signal quality. When certain sensors become unavailable or unreliable, the model performs adaptive weighting and interpolation to preserve feedback continuity. Over time, incremental learning updates the proxy model as new session data accumulate, ensuring that performance improves with continued use.

[0433] At runtime, the proxy -inference module continuously generates an estimated effectiveness metric E(t), which serves as the control variable for the optimization engine 738. This value is integrated into the same closed-loop architecture as the direct fNIRS feedback, allowing adaptive control even when sophisticated sensors are absent. Proxy feedback thus enables scalable, consumer-accessible versions of the adaptive TENS system while maintaining alignment with the high- fidelity physiological benchmarks established during calibration.

[0434] Population and Federated-Learning Architecture

[0435] FIG. 12 illustrates a population-level federated-learning framework that enables collective improvement of adaptive stimulation models while preserving user privacy. Multiple devices (1202-1206) perform local training based on user-specific data and generate encrypted model updates AWi. These updates are transmitted across a privacy’ boundary (1212) to an aggregation server (1210), which combines them into refined global and cohort models (1214, 1216). Updated parameters W' are redistributed to participating devices, allowing each system to benefit from collective experience without sharing raw physiological data. By pooling anonymized insights from many users, theAttorney Docket No. SETH-005 AWO framework accelerates learning, shortens calibration time for new users, and enables discovery of more effective stimulation profiles through collective intelligence. This architecture supports continuous performance improvement, regulatory traceability, and population-scale optimization while maintaining strict privacy and security controls.

[0436] User Feedback Integration and Reinforcement Learning

[0437] The adaptive stimulation system includes a user-feedback integration component that combines subjective reports with objective physiological and proxy measurements to refine the adaptive learning process across time. The feedback architecture ensures that the system learns not only from measured effectiveness but also from user experience, comfort, and therapeutic response. This hybrid framework allows the controller to maintain a consistent balance between autonomous operation and human-validated adjustment.

[0438] During each session, users may provide direct input regarding comfort level, perceived pain relief, or the adequacy of stimulation intensity. Feedback can be captured through natural-language interaction, tactile input devices, or post-session surveys. All subjective responses are time-aligned with physiological data streams such as heart-rate variability, galvanic skin response, and muscle activity7to establish a synchronized record linking user perception with measured outcomes. The combination of these data sources creates a multidimensional representation of session effectiveness that is both quantitative and experiential.

[0439] The reinforcement-learning subsystem interprets this feedback as a reward signal within a continuous optimization framework. The reward function integrates objective physiological indicators — such as decreased cortical activation in pain-related regions or increased parasympathetic activity — with user satisfaction and comfort ratings. The controller seeks to maximize the cumulative reward over time, promoting stimulation patterns that consistently produce both physiological improvement and positive user experience. When user-reported comfort diverges from objective metrics, the algorithm applies weighting factors that preserve safety while gradually steering parameters toward consensus performance.

[0440] The learning framework operates hierarchically across three temporal scales. In the immediate loop, real-time adjustments occur within seconds when user feedback or physiological measurements indicate excessive intensity or diminishing relief. Over the course of a session, rolling averages of comfort and effectiveness metrics guide slow adjustments to frequency, amplitude, and spatial distribution. Across multiple sessions, reinforcement updates refine model parameters, allowing the system to anticipate user-specific responses and initialize future sessions with optimized defaults.Attorney Docket No. SETH-005 AWOThis cumulative process transforms the device into an adaptive system that continuously personalizes therapy through experience.

[0441] User-feedback integration also enhances safety and regulatory transparency. If verbal or manual feedback indicates discomfort, the controller automatically reduces stimulation intensity or pauses the session while maintaining monitoring and logging. Conversely, when subjective feedback confirms relief concurrent with favorable physiological changes — such as improved heart-rate variability or reduced fNIRS pain activation — the controller reinforces those parameter settings for future operation. Each interaction contributes to a growing feedback 1 ibrary that supports clinician review, auditability, and ongoing model validation.

[0442] Population-level reinforcement learning may be implemented through federated aggregation, in which encry pted feedback-performance summaries from multiple devices are combined to improve generalized models without disclosing personal data. This collective learning mechanism allows the system to evolve beyond single-user optimization, incorporating statistical insight from diverse therapeutic scenarios while maintaining strict privacy compliance.

[0443] Clinical Data Integration and Healthcare Connectivity

[0444] The adaptive stimulation system supports integration with clinical information systems and distributed learning frameworks to ensure consistent performance across large populations of users while maintaining individual privacy. This capability extends the closed-loop feedback architecture described above into a population-level learning network that connects individual devices through secure data aggregation and model sharing.

[0445] In clinical or supervised settings, each device records session data including stimulation parameters, physiological responses, and user feedback outcomes. These records are stored locally in encrypted form and may be synchronized with electronic health record (EHR) systems through secure communication interfaces compliant with applicable medical-data standards. Integration with EHR platforms enables clinicians to review longitudinal progress, verify adherence, and remotely adjust permissible parameter ranges or therapy objectives. The same interface allows anonymized research data to be exported for regulatory’ validation or large-scale outcomes analysis.

[0446] Beyond local connectivity, the system implements a federated-learning framework that enables distributed model improvement without direct sharing of user data. Each device performs local training on its stored session records to refine its internal proxy and reinforcement-learning models. The resulting model-parameter updates, represented as encrypted gradients AWi, are transmitted across a defined privacy boundary to an aggregation server. The server aggregates updates from multiple devices to form refined global model w eights W' and cohort-specific submodels tailored to demographic orAttorney Docket No. SETH-005 AWO diagnostic groupings. Updated parameters are then redistributed to participating devices, allowing ever}’ system to benefit from collective learning while ensuring that raw physiological and personal data never leave the originating device.

[0447] This federated-learning process operates asynchronously with respect to therapeutic sessions, ty pically executing during idle or charging periods. Communication channels employ end-to- end encryption and differential-privacy safeguards to prevent re-identification. The architecture allows both global and cohort-level adaptation: global models capture universal relationships between stimulation parameters and physiological effectiveness, while cohort models fine-tune these relationships for specific populations such as chronic pain, post-surgical recovery, or neuropathic conditions. Each device merges incoming updates with its personalized model through weighted averaging, preserving local calibration while assimilating collective improvements.

[0448] The combination of EHR interoperability, remote supervision, and federated learning establishes a multiscale ecosystem in which individual therapy optimization contributes to continuous population-level refinement. Clinicians gain visibility into therapeutic progress, researchers obtain anonymized aggregate performance metrics, and users benefit from faster convergence toward effective settings derived from the collective experience of similar cases. The framework thus transforms the adaptive stimulation system from an isolated device into an evolving, data-driven therapeutic platform capable of improving with even’ use across the installed base.

[0449] Safety and Auto-Shutdown Logic

[0450] FIG. 13 illustrates the supervisory’ safety and automatic shutdown subsystem of the adaptive stimulation device. A safety monitor 1302 continuously evaluates physiological and hardware signals from the sensor system 104. Decision logic compares measured parameters to defined safety thresholds 1304, and when an unsafe condition is detected, the system performs parameter reduction 1306, stimulation pause 1308, or complete shutdown 1310, logging each event for later review'. All safety' events are recorded for later review, and controlled restart procedures (from controller! 10) are applied once conditions return to safe limits (e.g.. recovery 1314). This safety subsystem operates independently of adaptive-control and learning functions to ensure fail-safe operation under all conditions.

[0451] The adaptive stimulation device incorporates a multilayer safety architecture designed to prevent adverse physiological or electrical events under all operating conditions. This subsystem functions independently of the optimization and learning engines, ensuring that protective actions always override adaptive control commands when predefined limits are exceeded.Attorney Docket No. SETH-005 AWO

[0452] Continuous safety monitoring is performed by a dedicated processor that receives both hardware-level and physiological data streams. Electrical parameters such as output current, voltage compliance, and impedance are sampled at millisecond intervals to detect open-circuit, short-circuit, or over-current conditions. Simultaneously, physiological indicators including heart-rate variability, galvanic skin response, and electromyographic activity are evaluated for signs of acute distress. When abnormal readings occur, the system immediately transitions to a safe state, reducing or terminating stimulation within a few milliseconds.

[0453] Safety logic is implemented in three functional layers. The first layer, a hardware safetyloop, enforces hard current and voltage limits using analog protection circuits that disconnect the output stage when thresholds are reached. The second layer, a firmware supervisory loop, monitors device diagnostics and communication integrity- between modules, executing automatic ramp-down or pause routines if data packets are corrupted or latency exceeds acceptable bounds. The third layer, a physiological safety loop, interprets biosignal patterns indicative of user discomfort or autonomic activation; if these patterns persist beyond the defined time window, the controller halts stimulation and issues an audible or visual alert to prompt user acknowledgment.

[0454] All safety events are time stamped and written to a secure, non volatile log accessible to clinicians or service personnel. Logged parameters include the stimulation settings active at the moment of interruption, measured impedance, current density, detected biosignal anomalies, and recovery actions taken. This audit trail supports clinical review, post market analysis, and ongoing system validation. Once conditions return to normal and verification tests are completed, the controller may re enable output in a controlled ramp up sequence or remain in safe idle mode pending user confirmation.

[0455] The safety subsystem also supervises firmware updates and model downloads received through the federated learning network described in FIG. 12. Each update is validated through cry ptographic checksums and digital signatures before activation, preventing unauthorized or corrupted software from executing on the device. Watchdog timers continuously verify that the main processor and optimization engine are responding within allowable time limits; if a timeout occurs, stimulation is shut down automatically and the device reboots into diagnostic mode.

[0456] Arbitration and Safety Logic. The control framework treats physiological feedback and conscious user input as cooperating but independent channels. Under normal operation, physiological safety constraints limit all automated adjustments to prevent unsafe output. However, explicit human override commands including emergency expressions such as “stop,” “help,” or “fire” immediately suspend stimulation, regardless of physiological readings. Conversely, when physiological monitoring detects potential harm (e.g., excessive current, sympathetic distress) the system halts stimulation even ifAttorney Docket No. SETH-005 AWO the user continues to request increased intensity. This bidirectional override structure ensures both user sovereignty' and objective safety.

[0457] Referring now to FIGS. 14 - 15A, representative sensor modalities and comparative performance characteristics suitable for use across all embodiments of the adaptive stimulation system are presented. FIG. 14 lists exemplary sensor categories, measurement parameters, and corresponding adaptive applications used for real time and near real time feedback. Neural, autonomic, muscular, and behavioral sensors together provide the physiological foundation for the feedback and control frameworks described in Figures 10 through 13. FIGS. 15-15A provide a comparative overview of these sensing modalities, illustrating relative spatial and temporal resolution, invasiveness, latency, and suitability for pain, pleasure, and autonomic measurement. The instrumentalities and sensors of FIGS.14 - 15 A define a complete sensing hierarchy that supports system operation from research grade direct- neural calibration through consumer grade proxy feedback.

[0458] Together, these layers form a redundant protective envelope that maintains user safety even in the presence of component failure, communication loss, or algorithmic error. The architecture ensures that all adaptive learning and optimization functions operate within well defined electrical and physiological boundaries while supporting reliable closed loop operation.

[0459] As described herein, the present embodiment demonstrates how the adaptive stimulation architecture can be implemented in a transcutaneous electrical nerve stimulation (TENS) system that integrates real time physiological feedback, proxy measurement, and machine learning to achieve autonomous optimization of therapeutic effectiveness. By combining direct neurophysiological sensing with proxy based inference and population level learning, the TENS plus embodiment establishes a scalable foundation for closed loop neuromodulation across clinical, home health, and consumer environments.

[0460] The framework described herein serves as a reference model for subsequent embodiments that extend the same adaptive principles to mechanical and multi modal stimulation systems. One skilled in the art will recognize that the concepts, techniques, and structures described in connection with the particular embodiment (for example, TENS plus) may be equally applicable to other embodiments, whether explicitly described or not.

[0461] Federated personalization and data security'.

[0462] Feedback computation and inference may occur at any computational layer — on the device, at an edge processor, or in a secure cloud — provided that no raw physiological data or preauthored stimulation patterns are transmitted or required for inference. Only abstracted features, encry pted parameters, or model-delta summaries sufficient for calculation of stimulation adjustments areAttorney Docket No. SETH-005 AWO exchanged. This structure allows scalable deployment without exposing personal physiological data and ensures that adaptive control derives from measured effectiveness rather than pattern retrieval.

[0463] Robotic Massage System

[0464] Aspects of the present disclosure include one or more embodiments in the form of a multi-axis robotic actuator with integrated physiological feedback and adaptive control. Certain robotic massage and rehabilitation systems may incorporate pre-treatment body scanning and anatomical mapping using depth sensors, infrared cameras, or structured-light imaging to generate three- dimensional representations of a user’s body. These systems may identify muscle groups, pressure points, and curvature to guide positioning and motion. Some systems may further include force and position sensors to monitor contact pressure and tissue compliance, and may employ algorithmic or AI- based motion controllers to maintain consistent mechanical interaction with the body surface.

[0465] Aspects of the present disclosure provide a non-invasive robotic platform that extends these capabilities through continuous physiological measurement and adaptive control. The platform integrates multi-modal sensing — including, for example, heart-rate variability, galvanic skin response, electromyography, temperature, or optical perfusion data — to compute an effectiveness metric E(t) representing a user’s ongoing physiological response to stimulation. The controller utilizes this metric to adjust actuator parameters such as pressure, trajectory, vibration frequency, or temperature in real time while maintaining predefined safety limits.

[0466] The embodiment combines geometric sensing, mechanical feedback, and physiological monitoring within a unified closed-loop architecture. This arrangement enables the robotic system to modify stimulation parameters based on directly measured biological signals rather than solely on preset routines or operator input. The system thereby establishes a framework for automated, repeatable, and personalized therapy responsive to measurable physiological change.

[0467] FIG. 16 illustrates a flow diagram of a robotic massage embodiment of the adaptive stimulation system, according to one or more aspects of the present disclosure. Controller 110 drives actuator 1606 with integrated thermal / vibration modules 1608 to operate end-effector 1602 in contact with the user’s body. Force and pressure sensors 1604 provide feedback to mechanism 106 for real-time adaptive control. Safety subsystem 1675 monitors operational thresholds and may override actuator output, while user interface 112 enables manual or voice-command input.

[0468] FIG. 16A illustrates an adaptive robotic massage system integrating a robotic manipulator (1606). multi-modal physiological monitoring suite (1604), intelligent control system (110), and user interface and safety system (1675). Physiological data may guide adaptive control of stimulation parameters, while user feedback and safety constraints ensure effective and secure operationAttorney Docket No. SETH-005 AWO

[0469] As shown schematically in FIG. 16 A, the system integrates a multi -axis robotic actuator with a comprehensive physiological sensing suite and an intelligent controller governed by an effectiveness metric designated E(t). The robotic assembly (1602) provides collaborative, human-safe actuation capable of precise pressure and motion control. Surrounding this mechanical core is a physiological monitoring array (1604) that captures, for example, cardiovascular, neuromuscular, autonomic, and cortical signals reflective of the user’s evolving internal state. These measurements are interpreted by a learning controller (110) that adapts mechanical outputs according to measured effectiveness, while a user interface and safety subsystem (1675) supervises the process, ensuring that even’ command remains within medically safe limits. The resulting architecture, according to one aspect, transforms a static manipulator into a self-optimizing therapeutic system in which biological state directly determines mechanical behavior. According to one aspect, FIG. 16A illustrates dual feedback paths: a rapid mechanical loop maintaining force and trajectory, and a slower physiological loop optimizing effectiveness. Additional signal paths include optional fNIRS, facial, and acoustic inputs converging on the control processor. The diagram visually distinguishes this bi-directional architecture from the unidirectional command structure of conventional robotic systems, emphasizing that biological measurement now governs mechanical action.

[0470] The physiological foundation of the embodiment lies in the interplay of pain and pleasure within the nervous system. Activation of low-threshold mechanoreceptors through rhythmic pressure suppresses nociceptive transmission in the dorsal horn according to the gate-control theory of pain. Sustained stimulation elicits descending inhibitory pathways from the periaqueductal gray and medullary centers, releasing serotonin and norepinephrine that reinforce spinal inhibition. At the same time, moderate pressure and vibration induce [3-endorphin release and engage reward circuitry in the anterior cingulate cortex and nucleus accumbens, transforming mild discomfort into perceived relief. When delivered with consistent rhythm and predictable motion, such stimulation also recruits cognitive reappraisal mechanisms in the prefrontal cortex, encouraging the brain to interpret transient nociception as beneficial work rather than threat. Through these converging mechanisms, the device modulates both sensory and affective dimensions of pain.

[0471] Neurophysiological Cascade Under Therapeutic Deep-Tissue Massage

[0472] Referring now to FIG. 16B, a sequence of neurophysiological events according to a therapeutic response according to one more aspects of the present disclosure illustrated. As shown, a progression occurs from peripheral sensory activation (1601) through spinal and cortical modulation, including dual activation of nociceptive and mechanoreceptive channels(1602), inhibition of painAttorney Docket No. SETH-005 AWO transmission through spinal-gate mechanisms (1603), descending modulation (1604) with endogenous opioid release, and cortical reappraisal (1605) that transforms discomfort into perceived relief.

[0473] The same principles may be applied to muscular trigger-point therapy. Localized regions exhibiting sustained electromyographic activity can be compressed by the robotic end-effector until transient activation is followed by relaxation and a reduction in galvanic skin response. Such patterns may correspond to restoration of perfusion. Because user thresholds vary' w ith hydration, fatigue, and other physiological factors, the system may continuously update baseline references during each session, maintaining consistent comfort and relief across users.

[0474] Physiological activity' may be continuously converted into control decisions. Surface electromyography can indicate muscular guarding and trigger automatic pressure adjustment when excessive contraction appears. Heart-rate variability' and skin conductance may provide complementary indices of autonomic balance; increases in parasympathetic tone can cue gentle intensification of motion, whereas sympathetic spikes may prompt lighter strokes. Optical or thermal sensors can gauge local perfusion and temperature, guiding heating or vibration modules. Optional neuroimaging or affectsensing inputs (such as functional near-infrared spectroscopy, facial expression, or voice analysis) can further refine responsiveness. These multimodal signals are combined into an effectiveness metric E(t) representing the instantaneous state of the user and used to update stimulation parameters in real time.

[0475] In representative embodiments, the mechanical subsystem may employ precision actuators and a multi-axis force-torque transducer to maintain commanded pressure and measure tissue compliance. Embedded thermal elements can modulate temperature within a comfortable range.Physiological sensors distributed across the contact surface and the user's body may transmit data to a controller that executes hierarchical feedback algorithms — such as gradient-based adaptation, Bayesian optimization, or reinforcement learning. A safety controller monitors all outputs and initiates automatic interruption if any predefined thresholds are approached or exceeded. The user interface may accept spoken or manual commands and can display or audibly report the evolving effectiveness metric E(t).

[0476] Real-time adaptation is expected to support reproducible therapeutic benefit. In some examples, the system may reduce muscular tension, improve autonomic balance, or enhance subjective comfort as measured by physiological indicators. Because control decisions arise from sensed physiology rather than operator judgment, outcomes may remain consistent across users and sessions.

[0477] The adaptive robotic stimulation system therefore differs from conventional robotic massage or rehabilitation devices, which typically operate in open loop, repeating fixed trajectories irrespective of user response. It also avoids the invasiveness of implanted closed-loop stimulators thatAttorney Docket No. SETH-005 AWO rely on direct neural electrodes. The described embodiments unite robotic precision with physiological sensing to deliver closed-loop adaptivity through wholly non-invasive means.

[0478] The same principles may be applied to muscular trigger-point therapy. Localized regions exhibiting sustained electromyographic activity can be compressed by the robotic end-effector until transient activation is followed by relaxation and a reduction in galvanic skin response. Such patterns may correspond to restoration of perfusion. Because user thresholds vary with hydration, fatigue, and other physiological factors, the system may continuously update baseline references during each session, maintaining consistent comfort and relief across users.

[0479] Physiological activity may be continuously converted into control decisions. Surface electromyography can indicate muscular guarding and trigger automatic pressure adjustment when excessive contraction appears. Heart-rate variability and skin conductance may provide complementary indices of autonomic balance; increases in parasympathetic tone can cue gentle intensification of motion, whereas sympathetic spikes may prompt lighter strokes. Optical or thermal sensors can gauge local perfusion and temperature, guiding heating or vibration modules. Optional neuroimaging or affectsensing inputs (such as functional near-infrared spectroscopy, facial expression, or voice analysis) can further refine responsiveness. These multimodal signals are combined into an effectiveness metric E(t) representing the instantaneous state of the user and used to update stimulation parameters in real time.

[0480] In representative embodiments, the mechanical subsystem may employ precision actuators and a multi-axis force-torque transducer to maintain commanded pressure and measure tissue compliance. Embedded thermal elements can modulate temperature within a comfortable range. Physiological sensors distributed across the contact surface and the user's body may transmit data to a controller that executes hierarchical feedback algorithms — such as gradient-based adaptation, Bayesian optimization, or reinforcement learning. A safety controller monitors all outputs and initiates automatic intermption if any predefined thresholds are approached or exceeded. The user interface may accept spoken or manual commands and can display or audibly report the evolving effectiveness metric E(t).

[0481] Real-time adaptation is expected to support reproducible therapeutic benefit. In some examples, the system may reduce muscular tension, improve autonomic balance, or enhance subjective comfort as measured by physiological indicators. Because control decisions arise from sensed physiology rather than operator judgment, outcomes may remain consistent across users and sessions.

[0482] The adaptive robotic stimulation system therefore differs from conventional robotic massage or rehabilitation devices, which typically operate in open loop, repeating fixed trajectories irrespective of user response. It also avoids the invasiveness of implanted closed-loop stimulators thatAttorney Docket No. SETH-005 AWO rely on direct neural electrodes. The described embodiments unite robotic precision with physiological sensing to deliver closed-loop adaptivity through wholly non-invasive means.

[0483] Feedback and Control Loops

[0484] According to one or more aspects, the adaptive robotic stimulation system operates through a hierarchical feedback framework that coordinates mechanical precision with physiological intelligence. The embodiment comprises nested control loops that function across distinct temporal scales. A rapid inner loop governs actuator dynamics and contact forces, ensuring that every applied motion remains within predetermined mechanical and safety limits. A slower outer loop interprets physiological and proxy measurements, converting biological feedback into command adjustments that optimize therapeutic effectiveness. Both loops converge within the controller to maintain a continuous balance between efficacy and protection. FIG. 16C illustrates hierarchical feedback and control loops of the adaptive robotic stimulation system.

[0485] A mechanical controller (1610) operates actuator (1606) and force sensors (1608) within a fast inner loop that maintains commanded pressure and position. Physiological sensors (1 04) and feedback mechanism (1616) form a slower outer loop that evaluates multimodal signals — such as heartrate variability, galvanic skin response, and electromy ography — to compute the effectiveness metric E(t).

[0486] An optimization controller (1682) refines stimulation parameters on a millisecond-to- second timescale, while an arbitration engine (1684) merges mechanical and physiological inputs and forwards the results to a cross-session learning loop (1686) that aggregates data over hours or days. Safety limits and an emergency cut-off (1688) supervise all operations to ensure electrical, thermal, and mechanical outputs remain within safe boundaries. The overall closed-loop interaction between the body (1690) and the control subsystems provides continuous real-time adaptation across multiple temporal layers. In one aspect, an effectiveness metric designated E(t) represents the instantaneous degree to which stimulation achieves the intended physiological objective of pain reduction or relaxation. The controller computes E(t) as a weighted fusion of multimodal inputs, including heart-rate variability indices, galvanic skin conductance, electromyographic activity, peripheral temperature, and optional cortical oxygenation or affective indicators derived from fNIRS, facial analysis, and vocal stress patterns. Each input contributes a normalized component wi i to a continuous function of time, producingE(t) = S wi xi - X vj yj,

[0487] w here xi represents positive indicators of therapeutic response such as increased parasympathetic tone, and yj represents adverse markers such as excessive sympathetic activation orAttomev Docket No. SETH-005 AWO mechanical strain. This formulation is provided illustratively to show one possible implementation and does not limit how effectiveness may be computed. The controller continuously differentiates E(t) with respect to time to identify improvement or deterioration and modifies actuation parameters in the direction that maximizes E(t) while respecting safety boundaries. In some embodiments. E(t) may be computed as a weighted sum of normalized physiological inputs. In other embodiments, nonlinear or learned mappings (for example, sigmoidal, polynomial, or neural -network functions) may be used to capture complex relationships among the same input signals

[0488] Within the fast-safety’ loop, the system monitors high-frequency signals such as surface EMG and force-sensor output. When the EMG of a targeted muscle exceeds a threshold -typically then electromyographic activity’ rises significantly above its relaxed baseline, the processor interprets this as protective guarding and commands a rapid pressure reduction within a short interval. The corresponding actuator decelerates, and the contact force returns to a safe plateau. In parallel, skin-temperature and current-density sensors prevent localized overheating or over-stimulation, invoking an automatic rampdown of the contact surface approaches a defined temperature limit or if current density approaches a defined exposure limit. These reflexive responses occur independently of higher-level optimization routines, ensuring that user safety is preserved even under transient control-system latency or data loss.

[0489] The in-session optimization loop operates on slower timescales, typically one to three seconds. This loop analyzes evolving physiological trends and adjusts stimulation parameters to sustain positive movement of E(t). For example, a downward drift in galvanic skin response accompanied by increased high-frequency HRV power indicates parasympathetic dominance and successful relaxation. The controller responds by maintaining the current rhythm and gradually reducing amplitude to avoid habituation. Conversely, if HRV balance shifts toward sympathetic activation or if fNIRS analysis reveals increased cortical activation in pain-processing regions, the controller modifies the stroke velocity’ or spatial trajectory’, substituting lighter, more rhythmic patterns. This level of control allows the device to vary technique dynamically, alternating among effleurage, petrissage, tapotement, friction, or compression patterns as dictated by real-time feedback.

[0490] The cross-session learning loop 1686 in FIG. 16C aggregates data across multiple treatment periods to refine personalization over time. After each session, the controller stores parameter-effectiveness pairs consisting of stimulation settings, measured physiological responses, and resulting E(t) trends. When a new session begins, the controller initializes parameters using the most successful historical combinations, adjusted for any baseline drift detected during the initial calibration phase. Features such as habitual pressure tolerance, ty pical autonomic baseline, and recovery kinetics persist betw een sessions, while transient parameters such as skin impedance or fatigue level are re-Attomev Docket No. SETH-005 AWO measured each time. This persistence-and-update structure allows the device to behave as an adaptive therapist that remembers long-term preferences while responding to immediate conditions.

[0491] All loops operate under a unifying arbitration and safety hierarchy. Mechanical control commands generated by the inner loop and physiological optimization signals from the outer loop are merged by an arbitration engine 1684 (arbitration engine) that resolves conflicts in favor of safety constraints 1688 (safety system). If a physiological command requests increased pressure but the force sensor reports approaching load limits, the mechanical loop prevails until the safety margin widens. The arbitration logic therefore functions as a continuous negotiation between effectiveness and protection. Every control decision is logged with its corresponding E(t) value and the sensor readings that justified it, producing a transparent audit trail suitable for regulatory and clinical review.

[0492] The feedback framework also accommodates stochastic exploration. When E(t) remains stable for a defined interval, the controller introduces controlled micro-variations in parameters such as amplitude or frequency within bounded limits. These exploratory perturbations allow the system to search for potentially superior settings while avoiding abrupt transitions perceptible to the user. Bayesian-optimization algorithms estimate the expected improvement of each new parameter set, balancing exploitation of known effective regions with exploration of uncertain ones. Reinforcementlearning components update policy weights based on cumulative reward, ensuring that effective strategies persist across sessions and that ineffective ones are discarded.

[0493] Quantitatively, the latency of the complete physiological feedback cycle may be on the order of approximately two seconds from sensor acquisition to actuator response. The closed-loop control bandwidth can thus extend from tens of milliseconds for reflexive safety corrections to several seconds for adaptive optimization, spanning the range required for both mechanical stability and biological responsiveness. In representative embodiments, such a feedback configuration is designed to maintain consistent physiological tracking across sessions and users, providing a foundation for improved reliability relative to open-loop systems. The embodiment comprises a multilayer control system that continuously interprets biological signals, converts them into measurable effectiveness values, and adjusts mechanical output in real time. The combination of rapid safety reflexes, mid-term optimization, and cross-session learning yields a closed-loop architecture that remains stable yet dynamically responsive to the user’s changing physiology'.As shown in FIG. 16C, and described above, the hierarchy of concentric loops surrounding actuator 1606 includes an inner mechanical loop governed by controller 1610, which maintains precise force and position control; an outer physiological loop incorporating sensors 1604 and feedback mechanism 1606, which monitor the effectiveness metric E(t) and direct the optimization controller 1682; and a cross-Attorney Docket No. SETH-005 AWO session learning loop 1686, which extends adaptation across multiple treatments. Arrows linking each loop to the arbitration engine 1684 and the safety subsystem 1688 emphasize that every layer remains subordinate to user safety while collectively pursuing consistent and effective therapeutic performance.

[0494] Calibration and Archetype Mapping

[0495] According to one or more aspects, calibration in the adaptive stimulation system establishes the dynamic relationship between a user’s physiology and the machine’s stimulation parameters. The system performs this process as a brief, guided sequence in which standardized low- intensity motions are applied while the system observes physiological responses. Expected calibration completion would be within a short interval, often within minutes, after which the controller determines how force, speed, and frequency should scale for that individual. Calibration is an adaptive dialogue between the user’s body and the system’s optimization model.

[0496] In one typical scenario corresponding to operating mode 3A-3 (Fig. 3A), the device performs a population-model calibration. The robotic manipulator applies gentle exploratory strokes across a representative region while monitoring heart-rate variability, skin conductance, and surface electromyography. These data reveal the baseline autonomic and muscular state from which all future feedback will be referenced. The controller gradually varies amplitude and rhythm, searching for the inflection point where muscle activity begins to release and autonomic measures shift toward relaxation. Once this convergence is achieved, the measured scaling factors define the operating envelope for subsequent sessions. In another mode, representative of mode 3 A-4 (FIG. 3A], a quick archety pe- assisted calibration uses pre-existing population priors to shorten the process to less than a minute. A small number of test pulses confirm that the individual's physiological reactions fall within expected ranges, after which the control model adjusts automatically if any discrepancy is detected. When a research or clinical version operates in proxy-direct pairing mode (3A-2, FIG. 3A), more detailed calibration is performed by correlating direct neural or hemodynamic measurements such as fNIRS or EEG with simultaneous autonomic and muscular signals. This paired data allow the system to train an internal mapping that later substitutes accessible proxy measurements for expensive direct sensors during normal operation.

[0497] Across all modes, calibration follows a consistent progression. The controller first acquires resting baselines for primary biosignals, then delivers incremental test stimuli while observing changes in those signals. It treats the relationship between stimulus parameters and physiological responses as a supervised-leaming problem, using either regression or reinforcement strategies to predict which parameter combinations will maximize the illustrative effectiveness metric E(t). When successive iterations of stimulation produce stable physiological responses within safety limits, the model isAttorney Docket No. SETH-005 AWO considered converged. Verification sequences then confirm that the system can reproduce the same physiological transitions using its newly derived parameter scaling. The entire process ordinarily occupies only a few minutes and is repeated briefly at the beginning of subsequent sessions to account for day-to-day physiological variation.

[0498] The mapping derived from calibration is anchored to a broader archetype database. Each new user is compared to stored archetypes that represent clusters of similar responders. The archetype nearest in physiological feature space supplies prior expectations for stimulation sensitivity, recovery' kinetics, and comfort thresholds. These priors accelerate convergence and maintain operational safety even before sufficient personal data are accumulated. Over time, each user’s calibration outcomes refine both the local and population models through aggregated statistical learning, so that future devices start with progressively more accurate initializations.

[0499] FIG. 16D conceptually depicts this process as a flow from population priors 1620 through the archetype database 1622 into the calibration engine 1624, which in turn informs the optimization controller 1626 governing the robotic manipulator. The diagram illustrates the iterative exchange of data between physiological sensors and control algorithms until convergence criteria are met, at which point a personalized control profile 1628 is established for subsequent operation. FIG. 15D visually links these calibration scenarios to the operating-mode taxonomy — modes 3A-2 through 3A-4 (FIG. 3A) — that define how- feedback is managed during normal operation.

[0500] End-Effector and Sensor Integration

[0501] According to one or more aspects, the end-effector of the adaptive stimulation system functions as both a source of therapeutic stimulus and a sensor of physiological response. Rather than treating sensing and actuation as separate tasks, the embodiment merges them within a single compliant interface that maintains continuous awareness of tissue behavior. The contact surface conforms to the body and measures distributed pressure, shear, and temperature while simultaneously delivering controlled mechanical, vibratory, and thermal energy. Each change in tissue tone or perfusion is captured within milliseconds and immediately influences the control loops as described herein.

[0502] The end-effector serves as the terminal node of the fast safety loop, operating on a rapid cycle time consistent with fast reflex control. When surface electromyography indicates sudden muscular contraction or when force sensors detect a rate of load increase beyond safe limits, the controller reduces applied pressure almost instantaneously. Conversely, gradual declines in muscle activity and steady parasympathetic indicators cause the system to taper intensity and extend rhythm to preserve the relaxation phase. These reflexive actions occur without perceptible delay and ensure that the interface never sustains excessive force or heat. The same surface that delivers stimulation alsoAttorney Docket No. SETH-005 AWO detects the very biological signals that guide its modulation, thereby eliminating latency and calibration errors that arise when external sensors are used.

[0503] The mechanical assembly underlying this interface provides continuous, smooth force in the tens-of-newtons range with positional precision sufficient to reproduce therapeutic techniques.Vibratory modules operate across frequencies from tens to hundreds of hertz to engage different mechanoreceptor populations, while thermal elements maintain comfortable, body-safe warmth to promote circulation. The compliance of the contact surface adjusts dynamically so that the device can alternate between broad, gentle contact and focused compression without repositioning. These features allow the system to perform a repertoire of therapeutic patterns analogous to human manual techniques: long flowing strokes for general relaxation, rhythmic kneading to mobilize deep tissue, rapid tapping to activate fatigued muscles, circular friction to release adhesions, and steady compression to deactivate trigger points. The controller chooses among these motions by interpreting physiological feedback in real time, rather than following any predetermined sequence.

[0504] Sensor integration extends beyond mechanical quantities. Optical modules observe changes in superficial blood flow; skin-temperature sensors capture variations in perfusion; and galvanic skin and EMG channels measure autonomic and muscular activity. All of these signals return through the same data pathway to the controller, forming a complete feedback circuit between stimulation and response. The close proximity of sensing and actuation minimizes lag, enabling immediate recognition of therapeutic milestones such as muscle relaxation or autonomic stabilization.

[0505] Safety supervision is inherent to the physical design. Redundant monitoring within the actuator electronics enforces rapid power cutoff in the unlikely event of sensor disagreement, and the compliant surface geometry distributes force uniformly to prevent localized pressure peaks. Every operational mode of the system whether delivering gentle vibration or deeper compression remains bounded by conservative thresholds verified during calibration.

[0506] FIG. 16E schematically represents this integration, including an exemplary end-effector assembly that combines stimulation and sensing components within a single contact interface. The compliant surface (1652) distributes pressure evenly across tissue (1658) to maintain user comfort and safety. Beneath it, a sensor layer (1654) contains embedded transducers for force, temperature, galvanic skin response, optical, and electromyographic signals. The actuation layer (1656) provides mechanical, electrical, or thermal output under control of the system electronics. A support and coupling layer (1650) structurally links the assembly to the controller via connector or wireless interface (1652). This configuration enables simultaneous delivery of therapy and acquisition of physiological feedback, forming the terminal node of the adaptive control hierarchy described in FIG. 16C. The frameworkAttorney Docket No. SETH-005 AWO described herein serves as a reference model for other embodiments that extend the same adaptive principles to mechanical and multi-modal stimulation systems. One skilled in the art will recognize that the concepts, techniques and structures described in connection with the particular embodiment (e g., robotic massage system) may be equally applicable to other embodiments, whether explicitly described or not.

[0507] Operation Modes and Safety Hierarchy

[0508] During operation, three patterns of human interaction are possible. In the Autonomous Physiological Mode, the system functions entirely on sensor input. All adjustments arise from measured changes in heart-rate variability, galvanic skin response, electromyography, and related signals, while the controller maintains stimulation within its learned comfort envelope. In the Hybrid Voice-Guided Mode, natural-language input supplements the physiological data. The user may speak commands such as “a little softer,” “focus lower,” or “continue there,” which the natural-language processor converts to semantic intents. Each command is parsed and vetted through a pre-execution safety layer that verifies compatibility' with current physiological conditions before any motion or intensity change is executed. The Manual Supervisory' Mode provides clinician or operator oversight through a graphical or tactile interface that allows macro-level adjustments while retaining all safety backstops. Across all modes, the arbitration engine reconciles human instruction with sensor feedback so that neither user preference nor algorithmic decision can exceed predefined physiological or mechanical limits.

[0509] Safety within the embodiment is implemented as three interlocking layers that correspond to the physical, computational, and biological boundaries of operation. The Hardware Safety Layer forms the innermost ring, continuously monitoring mechanical load, temperature, and contact integrity’. Redundant force and torque sensors compare readings from independent channels; disagreement beyond tolerance automatically disengages the actuator within moments. Over-temperature protection and emergency -stop interfaces provide immediate power isolation, while capacitive touch detection ensures that the device halts when unintended contact is sensed. The Software Safety Layer surrounds this core and executes constant constraint verification across all control parameters. A watchdog process validates inter-processor communication and forces the system into a safe-idle state if asynchrony occurs. The Physiological Safety7Layer provides the final envelope of protection by observing user biosignals for indications of distress. Sudden spikes in sympathetic activity or abrupt muscular guarding trigger immediate down-modulation of stimulation, followed by escalation to full stop and audible notification if the abnormal pattern persists. Within this hierarchy, physiological safety has highest authority, hardware and software limits override all optimization or user commands, and conservative actionAttorney Docket No. SETH-005 AWO always prevails in case of conflict. Every safety event generates a time-stamped log entry retained for post-session analysis, allowing clinical traceability7and regulatory audit.

[0510] The operation of the control loops reflects this layered protection. Rapid reflex adjustments occur within fractions of a second when sensors detect unsafe mechanical or biological trends. If surface electromyography shows a sudden contraction, the controller proportionally reduces applied pressure; when galvanic skin conductance rises gradually, suggesting sympathetic activation, the system transitions over a few seconds to a slower, lighter rhythm. Longer-term optimization occurs over minutes through incremental adaptation of amplitude, frequency, and spatial trajectory based on averaged effectiveness metrics. Session-to-session learning refines these behaviors so that each iteration begins closer to the user’s physiological equilibrium.

[0511] Measurement technologies integrated within the platform vary in depth, latency, and practicality, yet they share a unified analytical pipeline. High-fidelity instruments such as functional MRI or magnetoencephalography remain research references for correlating brain activity with affective state, while mid-tier methods such as EEG or functional near-infrared spectroscopy provide portable cortical measurements useful in clinical calibration. Proxies including HRV, GSR, and EMG deliver continuous, non-invasive data streams for everyday operation. These modalities collectively define a hierarchy of sensing quality in which expensive, invasive systems establish gold-standard mappings that train the proxy models governing consumer-grade devices. The comparative performance of these technologies is summarized graphically in FIGS. 15-15A, which relates their response speed, invasiveness, and suitability for detecting pain-pleasure dynamics. The embodiment's control framework accommodates each tier without architectural modification, ensuring future compatibility as new measurement modalities emerge.

[0512] Real-time effectiveness optimization occurs through a multi-layer computational structure. The first layer handles high-frequency reflex control using direct sensor thresholds to maintain stability7and safety. A slower analytical layer integrates multimodal features to update the effectiveness estimate that guides ongoing adjustment. A background predictive layer employs model-based control and machine-learning policies to forecast user response and pre-emptively tune stimulation parameters. Together these layers create an adaptive balance between reactivity and foresight. The algorithms operate qualitatively in two speeds: a fast reflex channel responding within instants and a slow evaluative channel refining strategy over successive cycles. This dual-speed mechanism allows the device to behave both as a responsive instrument and as a learning system that improves with continued use.Attorney Docket No. SETH-005 AWO

[0513] Personalization extends beyond a single session. Reinforcement-learning principles guide the controller to associate particular parameter patterns with measured therapeutic improvement, gradually shaping an individualized policy. Over time, these individual models contribute anonymized statistical updates to population-level archetypes, allowing collective intelligence to evolve without compromising privacy. Each new device thus benefits from prior experience while still tailoring behavior to its current user. The controller maintains awareness of both immediate feedback and historical patterns, enabling it to predict which techniques are likely to produce optimal results under given conditions.

[0514] The embodiment translates these computational insights into mechanical behavior resembling that of an experienced therapist. When physiological indicators suggest relaxation and adequate perfusion, the system sustains long, sweeping effleurage motions. If residual tension persists, it introduces slower kneading movements analogous to petrissage to mobilize deeper tissue layers. Rapid, rhythmic tapotement may follow to stimulate fatigued areas, while gentle circular friction resolves localized adhesions. Static compression is reserved for trigger-point release when electromyographic feedback confirms appropriate muscular relaxation. These technique transitions occur fluidly and autonomously, guided by the continuous dialogue between sensor input and optimization logic rather than by any fixed program.

[0515] Natural -language interaction enhances this adaptive behavior. The user’s spoken instructions are interpreted through embedded semantic models that recognize intent, context, and emotional tone. Before execution, each interpreted command passes through the safety hierarchy to verify that it remains within safe mechanical and physiological boundaries. If a user requests ‘“more pressure” while the autonomic sensors detect stress, the controller responds verbally or visually that limits have been reached and maintains current intensity. When physiological and verbal feedback align, the adjustment proceeds smoothly, reinforcing the perception of human-like understanding while preserving objective safety. The voice interface therefore functions not as direct control but as cooperative guidance within the adaptive feedback ecosystem.

[0516] Closed-Loop Framework and Real-Time Operation

[0517] FIG. 17 illustrates the universal closed-loop control framework used across all adaptive- stimulation embodiments. Controller 1610 applies stimulation to the user; sensors 1604 measure physiological responses; feedback mechanism 106 evaluates effectiveness; and controller 1610 updates parameters such as waveform, amplitude, frequency, and spatial configuration subject to supervision by the safety subsystem 1610. These interactions form a continuous adaptive-control cycle that maintains therapeutic or experiential effectiveness while ensuring operational safety.Attorney Docket No. SETH-005 AWO

[0518] Through these combined mechanisms, the embodiment maintains a continuous triad of protection encompassing hardware and software integrity, physiological well-being, and user comprehension. The overall architecture is agnostic to platform scale — whether implemented as a clinical console or a compact consumer device — and operates under the same closed-loop safety principles. The system design layers hardware protection, software supervision, and physiological or user-interface safeguards so that every operational state remains enveloped by successive levels of monitoring. These nested safety functions collectively ensure conservative, self-protective behavior under all conditions.

[0519] Comprehensive Sensor Systems and Measurement Hierarchy

[0520] FIG. 18 illustrates the hierarchical organization of sensing modalities used across the adaptive stimulation system. High-fidelity modalities (1802) such as fMRI and fNIRS provide validated reference data during research and calibration. Medium-fidelity sensors (1804) including EEG and functional ultrasound support portable clinical applications, while low-fidelity or proxy sensors (1806) such as HRV, GSR, EMG, and temperature enable continuous real-time feedback during normal operation. Feedback mechanism (1806) integrates multi-layer inputs to maintain clinical-grade accuracy while enabling consumer-grade accessibility.

[0521] This embodiment therefore establishes a new paradigm in therapeutic robotics and adaptive stimulation. It unites closed-loop physiological feedback, proxy learning, predictive optimization, and human-interactive control into a single coherent framework. By bridging the research precision of high-fidelity sensing with the accessibility7of consumer operation, it defines a new category of measurable, self-optimizing devices capable of delivering personalized, clinically effective, and inherently safe stimulation experiences

[0522] Somato- Autonomic Reflex System

[0523] According to one or more aspects of the present disclosure, an implementation of an adaptive stimulation system designed specifically for healing, visceral regulation, and systemic recovery is provided. It applies substantially the same underlying adaptive-feedback principles used in other embodiments and systems described herein to promote restoration of internal physiological balance through autonomic modulation.

[0524] According to one or more aspects, the system employs the somato-autonomic reflex, a neurophysiological mechanism in which gentle cutaneous or subcutaneous stimulation modifies autonomic outflow and thereby influences internal organ function and tissue repair. Electrical, mechanical, or thermal stimuli delivered to selected body regions can shift sympathetic and parasympathetic activity in measurable ways that support recovery.Attorney Docket No. SETH-005 AWO

[0525] Existing therapeutic devices such as transcutaneous stimulators or mechanical massage systems operate in an open-loop configuration: they deliver preset patterns of stimulation and rely on the user or clinician to assess effect and make manual adjustments. Such devices cannot detect whether stimulation is producing beneficial parasympathetic activation, reducing excessive sympathetic drive, or improving circulatory or visceral function.

[0526] The present embodiment converts static systems into a closed-loop architecture for autonomic healing. The device continuously acquires multi-modal signals such as heart-rate variability, electrodermal activity, respiration, and electromyographic measures that are indicative of autonomic state. These signals are pre-processed and combined into a composite effectiveness metric that estimates the direction and relative magnitude of autonomic change for the individual user. Based on this estimate, the controller adjusts stimulation parameters to sustain or enhance recovery-aligned responses. Adaptation operates in real time or near real time during a session and is refined across sessions through baseline-referenced and model-updated calibration. By integrating biological feedback directly into the control process, this embodiment enables the device to act as an adaptive regulator of internal physiological balance rather than a passive source of stimulation.

[0527] Somato-Autonomic Reflex Mechanism

[0528] According to one or more aspects, stimulation applied to the skin activates sensory afferents that project to the spinal cord and brainstem nuclei, including the nucleus tractus solitarius and the parabrachial complex. These centers integrate sensory input and modulate autonomic efferent output through both sympathetic and parasympathetic pathways.

[0529] Referring to FIG. 19, the system operates as a somato-autonomic feedback network in which stimulation applied to peripheral tissues (1900) by the stimulation component (1916) initiates afferent activity that converges within spinal and brainstem reflex centers (1902). This interaction produces autonomic output (1904) through sympathetic and parasympathetic pathways that modulate cardiovascular, respiratory, and sudomotor activity. Resulting physiological responses (1906) — including changes in heart-rate variability, skin conductance, temperature, and respiration — are detected by the sensor system (1904) and analyzed by the feedback mechanism (1906) within controller (1910). The controller adjusts stimulation parameters to maintain or restore autonomic balance, completing a closed-loop optimization cycle for somato-autonomic regulation. Referring again to FIG. 10, after identification of the optimal electrode pair (X,Y) 716 on the 8 x 10 flexible electrode grid 116, the controller 108 operates the feedback mechanism 106 in coordination with the EEG calibration reference system 118 to deliver adaptive stimulation in a closed-loop sequence. Real-time sensor data are used toAttorney Docket No. SETH-005 AWO adjust stimulation parameters continuously as part of the real-time feedback loop 104, optimizing autonomic or somatic effectiveness based on the measured physiological responses.

[0530] Gentle stimulation at defined sites can therefore alter systemic function. Increased parasympathetic tone lowers cardiac rate, enhances digestive motility, and promotes tissue perfusion, while attenuation of sympathetic tone reduces stress-related vasoconstriction and inflammatory signaling. Hormonal and neurochemical mediators such as endogenous opioids and oxytocin may also be engaged, reinforcing relaxation and repair processes. Each of these autonomic shifts produces measurable multi-modal signatures that the sensor system detects and that the controller uses to guide subsequent stimulation.

[0531] Therapeutic applications of this mechanism include, for example, modulation of bladder activity through sacral or pelvic stimulation, reduction of nausea by abdominal surface stimulation that enhances vagal tone, improvement of peripheral circulation through thoracic or cervical input, and attenuation of inflammation by stimulating cutaneous afferents linked to inhibitory spinal pathways. In every case, the device monitors changes in the user’s multi-modal state and adapts stimulation to maintain the desired healing response.

[0532] Neural and Multi-Modal Integration for Healing Control

[0533] The somato-autonomic reflex operates through interconnected spinal, brainstem, and peripheral processes that together coordinate visceral and vascular function. At the spinal level, mechanoreceptive and nociceptive afferents converge within dorsal-hom intemeuronal networks that influence sympathetic pre-ganglionic neurons. Signals ascend to medullary and hypothalamic centers, which in turn regulate vagal and adrenal pathways controlling heart rate, digestion, and circulation. Peripheral changes in blood flow, glandular secretion, and smooth-muscle tone form the observable outcomes of this control loop.

[0534] Multi-modal sensing provides continuous observation of these dynamics. Heart-rate variability, electrodermal activity, respiratory rhythm, and surface electromyography collectively describe the user's autonomic balance. By correlating these signals with known markers of parasympathetic activation and recovery, the controller maintains stimulation within a range that favors healing.

[0535] Functional studies show overlapping neural regions such as the insula and anterior cingulate cortex are engaged in pain modulation, affective response, and recovery. This embodiment interprets such distributed neural and peripheral feedback as part of a unified healing process. Unlike prior systems that deliver fixed stimulation sequences without reference to systemic effect, the adaptive controller continuously evaluates multi-modal feedback to adjust amplitude, frequency, waveform, orAttorney Docket No. SETH-005 AWO spatial targeting in real time. Through this closed-loop regulation, the device actively supports autonomic re-balancing and promotes the biological conditions necessary for healing and restoration.

[0536] System Architecture and Feedback Framework

[0537] Referring now to FIG. 20, the adaptive feedback loop for autonomic regulation operates through coordinated interaction among the controller (2010), stimulation component (2016), physiological response (2020), comprehensive sensor system (2004), and proxy module (2008). The controller (2010) adjusts stimulation parameters based on feedback metric E(t) and delivers commands to the stimulation component (2016), which provides electrical, mechanical, or thermal output to the user. The resulting physiological response (2020) produces changes in heart-rate variability, galvanic skin response, respiration, and temperature that are detected by the sensor system (2004). The proxy module (2008) substitutes or augments direct measurements with derived effectiveness estimates, completing the closed-loop cycle by returning feedback E(t) to the controller (2010) for continuous optimization of autonomic regulation... Comprehensive sensor system (2004) acquires physiological biosignals such as heart-rate variability, galvanic skin response, respiration, electromyography, and temperature. The proxy module (2008) interprets these signals to estimate autonomic state and generate an effectiveness metric E(t). The controller (2010) adjusts parameters of the stimulation component (2016). which provides electrical, mechanical, or thermal output, to optimize the physiological response (2020) while maintaining safety boundaries. The loop operates continuously to support real time and cross session adaptive learning. The control objective in this embodiment differs from pain-relief or pleasure-enhancement systems. Rather than optimizing for immediate comfort, the controller optimizes for healing state progression — that is, movement of the body's multi-modal markers toward physiological conditions associated with repair and regulation. Transient discomfort may accompany these transitions, but such responses are interpreted as informative, not adverse, when they align with long-term recovery' patterns.

[0538] During operation, the stimulation component applies a defined waveform or mechanical pattern. The sensors detect changes in heart-rate variability, electrodermal activity, respiration, electromyography, and skin temperature. The feedback mechanism evaluates these signals as evidence of systemic adaptation, computing an effectiveness metric that reflects the direction and magnitude of autonomic change. The controller then adjusts amplitude, frequency, or waveform to maintain stimulation in a range correlated with progressive recovery. Multiple temporal loops function simultaneously: a fast safety sub-loop mitigates acute deviations; an optimization loop refines stimulation during the session; and a cross-session learning process updates parameters over time. InAttorney Docket No. SETH-005 AWO combination, these loops enable a device that dynamically supports the body’s intrinsic healing processes rather than suppressing sensory feedback.

[0539] Multi-Modal Metrics and Signal Acquisition

[0540] The somato-autonomic system employs non-invasive sensing modalities that collectively describe autonomic balance and healing progression. These include electrical, optical, acoustic, and behavioral channels that reveal the interaction between sympathetic and parasympathetic systems.

[0541] For this embodiment, the adaptive healing system primarily uses the accessible subset — heart-rate variability, galvanic skin response, respiration, and surface electromyography — while allowing optional integration of higher-fidelity modalities for calibration (FIG. 14). These concurrent channels provide continuous insight into whether stimulation is promoting a healing trajectory: increasing parasympathetic indices, improving circulatory stability7, and restoring rhythmic coherence across systems. When direct neural measurement is unavailable, trained proxy models map the accessible signals to inferred healing states, ensuring that the control algorithm remains focused on systemic restoration rather than sensory relief.

[0542] Multi-Modal Processing and Control Framework

[0543] FIG. 20A illustrates the data flow7betw een these modules and the closed loop interaction that maintains continuous autonomic optimization. As shown in FIG. 20A, the somato-autonomic system implements a processing and control framework that converts multi-modal sensor data into adaptive decisions for healing optimization. Data from the sensor system (2104) first enter the preprocessing and normalization module (2102), which removes noise and aligns signals to user specific baselines. The feature extraction engine (2104) identifies temporal and spectral features such as HRV pattern stability, GSR reactivity, and temperature slope that correspond to autonomic transitions associated with tissue recovery. These features are combined in the effectiveness computation module (2105) to produce a healing effectiveness metric E(t). The adaptive learning and control engine (2108) interprets E(t), refines its weighting parameters, and updates the stimulation component (2116). The feedback pathway (2106) returns the measured response to the sensor system (2104), closing the loop under supervision of the controller (2110).The optimization objective to advance the user’s multi-modal state tow ard markers of repair parasympathetic balance, thermal normalization, reduced sympathetic volatility, and rhythmic stability rather than merely reducing perceived pain or maximizing pleasurable sensation. This distinction allows the device to deliver stimulation that sometimes elicits transient discomfort while still aligning with the biological conditions that drive healing. By interpreting complex multi-modal data as evidence of restorative progress, the embodiment provides adaptive, data-driven regulation of systemic recovery7.Attomev Docket No. SETH-005 AWO

[0544] Therapeutic Mechanisms and Optimization Objectives

[0545] The somato-autonomic system optimizes stimulation for healing and systemic recovery’ rather than for immediate sensory comfort. Adaptive control algorithms interpret multi-modal physiological signals as indicators of progress along a healing trajectory, adjusting stimulation to maintain autonomic and vascular patterns associated with tissue repair, metabolic restoration, and visceral regulation. The controller (2010) and feedback mechanism (2106) evaluate measures such as heart-rate variability, galvanic skin response, respiration, electromyography, and surface temperature to favor parasympathetic stability-, improved circulation, and reduced sympathetic volatility. Examples include targeted parasympathetic activation to improve bladder function, vagal modulation to enhance gastrointestinal motility and reduce nausea, and thoracic or cervical stimulation that supports vasodilation and circulatory- recovery-. Each implementation uses the closed-loop architecture show n in FIG. 20 and FIG. 20A. differing only in the autonomic markers emphasized during optimization. The controller thereby adapts stimulation parameters to align with biological processes underlying healing rather than suppressing sensation as in conventional TENS devices.

[0546] Integration with Device Hardware

[0547] The stimulation component (2116) and sensor system (2004) are implemented on a flexible, conformable substrate that allows both actuation and sensing across the same anatomical region. Each node within the array can deliver controlled stimulation and record localized autonomic responses, enabling the controller (2010) to map spatial variations in healing effectiveness. This configuration allows the system to identify- specific areas that exhibit stronger coupling between stimulation and beneficial multi-modal responses such as improved perfusion or increased HRV coherence and to concentrate treatment at those sites.

[0548] Because healing optimization may involve transient stress responses, safety7supervision operates independently of subjective user feedback. Dedicated monitoring circuits compare real-time sensor data with reference thresholds derived from the user’s baseline and automatically scale or suspend stimulation if signs of excessive sympathetic activation or vascular strain appear. These protections ensure that stimulation remains within a therapeutic window that supports recovery without imposing harmful load on the autonomic system.

[0549] The hardware architecture is compatible with electrical, mechanical, or thermal stimulation modalities and can interface with external computing or mobile platforms for session management and data storage. In all implementations, the hardware serves the same functional objective: to sustain stimulation patterns that advance the body’s measurable state of healing,Attorney Docket No. SETH-005 AWO distinguishing this embodiment from prior devices whose hardware merely executes fixed programs for pain suppression or relaxation.

[0550] Closed -Loop Feedback Operation

[0551] FIG. 20 and FIG. 20A illustrate how the feedback system operates continuously to maintain the user within an autonomic state favorable to recovery. The controller (2010) repeatedly acquires multi-modal signals from the sensor system (2004), computes an effectiveness value representing the current degree of healing progress, and adjusts stimulation parameters through the stimulation component (2116).The feedback process functions across three overlapping timescales: a fast safety loop, a session-level loop, and a cross-session loop. The fast safety loop monitors rapid changes such as galvanic-skin-response spikes or heart-rate surges and immediately reduces or suspends output to prevent physiological overload. The session-level loop interprets trends in HRV, respiration, and temperature over tens of seconds or minutes, optimizing stimulation during a single treatment. The cross-session learning loop analyzes accumulated data from multiple sessions to refine initialization parameters for future use. The optimization objective remains advancing the user’s measurable state of healing. The system distinguishes between discomfort indicating stress and discomfort accompanying beneficial repair. Unlike TENS or massage systems that equate success with immediate relief or pleasure, this embodiment interprets success as movement of the multi-modal indicators toward physiological restoration and autonomic balance.

[0552] Healing-Effectiveness Metric

[0553] In one aspect, the feedback mechanism (106) computes a composite healing-effectiveness metric E(t) that quantifies the overall direction and magnitude of systemic recover)’. This metric may- integrate normalized contributions from heart-rate variability, galvanic skin response, respiration stability, surface temperature, and electromyographic relaxation. Positive values of E(t) correspond to parasympathetic dominance, circulatory' improvement, or reduced sympathetic volatility7conditions associated with tissue repair while negative values indicate deviation from those targets.

[0554] The controller (2010) uses E(t) to modulate stimulation parameters in real time. When E(t) falls below a threshold, indicating diminished healing response, the controller adjusts amplitude, frequency, or spatial pattern to restore favorable dynamics. When E(t) remains stable or increases, the controller gradually tapers stimulation to prevent habituation.

[0555] Other embodiments may implement alternative mathematical or heuristic formulations of the same concept. The purpose of the metric in every case is to provide an objective, data-driven measure of healing effectiveness, distinct from the sensory endpoints used in prior art.Attorney Docket No. SETH-005 AWO

[0556] In some embodiments, the controller (2010) or feedback mechanism (2106) employs adaptive search or machine learning techniques to refine stimulation parameters in relation to the healing-effectiveness metric E(t). These techniques may include gradient-based estimation, probabilistic exploration, reinforcement learning, or other forms of dynamic optimization that can evaluate how incremental parameter adjustments affect measured recovery indicators. The choice of algorithm is implementation-dependent and not limited to any particular mathematical form. The essential function is to search the parameter space efficiently and autonomously, identifying stimulation patterns that improve objective measures of healing while maintaining safety boundaries. By continually relating stimulation input to observed biological response, the system builds an evolving internal model of how the user’s body heals, allowing it to anticipate beneficial parameter adjustments rather than relying on trial-and-error selection.

[0557] In expanded embodiments, the healing-effectiveness metric E(t) may be generated or refined by a trained machine-learning or generative inference model rather than by a fixed mathematical formula. Such models can include neural -network or transformer-based architectures that learn contextual relationships among the input signals and their temporal evolution, producing an adaptive representation of physiological recovery. By inferring E(t) through a learned generative process, the system can generalize across users and conditions, predictively shaping stimulation responses even when new signal patterns differ from the training set.

[0558] Parameter Space and Adaptive Exploration

[0559] To sustain effective healing stimulation, the controller explores a multi-dimensional parameter space defined by variables such as amplitude, frequency, waveform, temporal pattern, and spatial location. Each combination represents a potential therapeutic condition, and the system continually adjusts these parameters based on feedback from the healing-effectiveness metric E(t).

[0560] Rather than prescribing a specific algorithm, this embodiment allows for any adaptive search or gradient-based adjustment method capable of iteratively improving E(t). The exploration process balances two priorities: (1) maintaining stability within safe operating boundaries, and (2) discovering new parameter regions that yield stronger autonomic-recovery signals. Over time, the controller converges on parameter sets that maximize indicators of systemic healing for the individual user.

[0561] As shown in FIG. 21, the controller (2010) executes an iterative closed-loop process that continuously refines stimulation parameters in real time. Each cycle begins with the current parameter set (2002), applies stimulation to the user (2003), measures the resulting multi-modal physiological response (2006), computes the healing-effectiveness metric E(t) (2008), and adjusts parameters toAttorney Docket No. SETH-005 AWO enhance that metric (2011). By repeating this sequence, the controller adaptively searches the parameter space, identifying and reinforcing configurations that promote autonomic recovery' while remaining within physiological safety limits. The diagram represents the logical control cycle at the core of the healing embodiment. This adaptive-exploration framework enables the system to leam how each individual’s body responds to different stimulation configurations and to autonomously select those that produce sustained physiological repair. The approach distinguishes this embodiment from fixed-pattern stimulation devices, which lack the capability to recognize or optimize for evolving healing dynamics.

[0562] Adaptive Learning Across Sessions

[0563] Healing is a dynamic process that unfolds over time rather than in a single session. In one aspect, an adaptive learning framework that allows the system to recognize how an individual’s responses evolve and to refine its operation accordingly. The goal is to ensure that each new session begins closer to the individual’s optimal therapeutic state, reducing the time required for the device to identify effective stimulation parameters.

[0564] During each session, the controller (2010) records the stimulation parameters applied, the resulting healing-effectiveness metric E(t), and the temporal patterns of multi-modal signals from the sensor system (2004). These data are stored as session records. Between sessions, the system analyzes the stored records to identify trends that reflect recovery such as progressively higher HRV coherence or shorter times to parasympathetic stabilization. The learned relationships form priors that guide initialization for subsequent treatments.

[0565] In this embodiment, learning is used not simply to optimize control performance, but to track the trajectory of healing itself. The system thus functions as both a therapeutic and a diagnostic tool, providing continuous insight into how recovery is proceeding and how the user’s autonomic profile changes as healing advances. This adaptive capability' distinguishes the embodiment from pain-relief devices, which reset to the same baseline conditions for every use because their goal is transient comfort rather than cumulative repair.

[0566] Purpose and Method of Diagnostic Probing

[0567] As illustrated in FIG. 9, the system may initiate a diagnostic probing sequence before therapy to determine which stimulation sites or parameter combinations produce the strongest autonomic responses. Because the mechanisms of healing vary across tissue types and individuals, the system may employ a brief diagnostic probing sequence before initiating therapy. The purpose of this procedure is to locate anatomical regions or stimulation parameters that most effectively trigger measurable autonomic improvement. This probing step is particularly valuable when the underlying condition or stage of healing has changed since the previous session. In one embodiment, the system delivers a series of mild,Attorney Docket No. SETH-005 AWO controlled stimuli across different electrode sites or actuator nodes while the sensor system (104) monitors real-time changes in HRV, GSR, EMG, and temperature. The controller (108) interprets these immediate responses as indicators of autonomic reactivity and generates a spatial or parametric '“response map.” Regions that elicit stronger parasympathetic activation or improved circulation are prioritized for subsequent therapeutic stimulation. The rationale for this diagnostic phase is to align treatment with the body’s current healing potential. Injured or inflamed areas may exhibit distinct autonomic signatures that change as recovery progresses. By identifying where stimulation produces the most constructive physiological shift, the device can focus treatment energy precisely where it supports ongoing repair, avoiding overstimulation of regions that are already normalized or hypersensitive.

[0568] Site Selection and Therapeutic Transition

[0569] Following diagnostic probing, the controller (108) selects one or more stimulation sites or parameter combinations that produced the most favorable healing-effectiveness scores. The system then transitions from diagnostic to therapeutic mode, applying continuous stimulation at those selected locations while maintaining real-time monitoring. As shown in FIGS. 9 and 10, the process may be visualized as an iterative mapping loop in which diagnostic inputs generate measurable autonomic responses that are ranked (e.g. in a ranked effectiveness list 904) by their contribution to systemic recovery. The highest-ranked configurations are automatically carried forward into the treatment phase for closed-loop optimization. FIGS. 9 and 10 illustrate diagnostic stimulation mapping and the progression to therapeutic operation for autonomic healing applications. During diagnostic mode, the flexible electrode grid (116) sequentially activates stimulation sites while the sensor system (104) records biosignal changes (AEi). The controller (108) analyzes these responses to generate an effectiveness heatmap and identify' regions yielding maximal autonomic response. Once the optimal site k = arg max (AE;) is determined, the controller coordinates the stimulation component (116) to deliver adaptive therapy at that site, forming the basis for closed-loop healing control. This two-stage approach — probing followed by targeted therapy — serves two purposes. First, it accelerates convergence toward effective stimulation parameters by relying on empirical physiological data rather than assumptions about anatomy or pathology. Second, it ensures that the therapy remains responsive to changes in the user’s condition. As healing progresses, the body’s most responsive sites may shift; the system can therefore repeat a shortened probing cycle periodically to confirm that treatment remains properly targeted. Through these mechanisms, the embodiment maintains its central focus: to guide stimulation according to the body’s own evolving indicators of healing rather than static preprogrammed settings or transient pain relief.

[0570] Adaptive Filtering and Latency CompensationAttorney Docket No. SETH-005 AWO

[0571] As shown in FIG. 22, the adaptive stimulation system organizes healing control across three interacting timescales — immediate safety (2202), in-session optimization (2204), and cross-session learning (2206) — each operating with its own feedback latency. These temporal layers provide the foundation for the adaptive-filtering and latency-compensation processes described below. Healing- related autonomic responses unfold over multiple timescales, from rapid electrodermal reactions to slower circulatory changes. To ensure that each signal is interpreted correctly, one embodiment employs adaptive filtering and latency compensation. The goal is to synchronize the controller’s interpretation of sensor data with the physiological delays inherent to healing processes.

[0572] Fast responses such as GSR or heart-rate changes appear within seconds, whereas vascular or temperature shifts may require longer intervals. Without latency correction, the controller could over-adjust stimulation before slower responses become evident. To address this, the feedback mechanism includes an adaptive filter that estimates expected response timing for each signal type. The controller uses these estimates to predict future trends in the healing-effectiveness metric E(t) and to adjust stimulation pre-emptively rather than reactively.

[0573] This filtering strategy' reflects a broader design philosophy of the system: to understand why changes occur, not merely that they occur. By modeling the natural temporal dynamics of healing, the embodiment maintains stable control and avoids oscillation or over-correction, which can interrupt the body’s own recovery rhythms.

[0574] Contextual and Voice-Integrated Control

[0575] In some therapeutic settings, the sy stem may incorporate audio or contextual input to complement multi-modal sensing. Voice-based or environmental cues can provide additional insight into the user’s condition, particularly when healing is accompanied by' emotional or behavioral changes that affect autonomic balance.

[0576] In one embodiment, the user can communicate sensations such as “warmer,” “tightening,” or “relaxing,” which the natural-language interface interprets as contextual data rather than direct commands. These qualitative descriptions help the controller distinguish between discomfort that is therapeutically productive — indicating tissue engagement — and discomfort signaling excessive stress. When such input conflicts with sensor data, the arbitration engine weighs both sources according to safety priority.

[0577] The purpose of this integration is not to reintroduce subjective control, but to enrich the interpretation of multi-modal signals with human context. By correlating spoken feedback with measurable physiological trends, the system improves its understanding of how healing feels as well asAttorney Docket No. SETH-005 AWO how it manifests biologically. This capability also supports caregiver or clinician oversight, enabling remote monitoring without compromising the autonomy of the closed loop.

[0578] Hierarchical Learning Structure

[0579] As shown in FIG. 22A the controller organizes its adaptive processes into three nested layers operating on different time scales. The immediate layer 2212 provides millisecond-to-second reflexes for safety and comfort, the session layer 2214 refines stimulation in real time using the adaptive exploration loop of FIG. 22A, and the cross-session layer analyzes cumulative data to update initialization parameters for future treatments.

[0580] Healing is not a single-state event but an evolving trajectory. To manage this complexity, the system employs a hierarchical learning structure in which each layer contributes distinct information about how the body responds to therapy. The immediate layer responds to acute deviations in autonomic signals to maintain safety and comfort boundaries. The session layer optimizes stimulation within a single treatment, refining parameters in real time according to trends in the healing-effectiveness metric E(t). The cross-session layer operates over hours to weeks, analyzing accumulated data to identify longterm recovery patterns and adapt the initial conditions for subsequent sessions.

[0581] These layers interact continuously: rapid physiological responses from the immediate layer inform session-level optimization, while session results feed the long-term learning model. This multi-timescale hierarchy preserves both responsiveness and continuity reacting instantly when required while retaining knowledge of what promotes sustained recovery over time.

[0582] Data Integrity and Safety Supervision

[0583] Because the healing optimization process relies on cumulative learning, data integrity is critical. The system, therefore, includes redundant monitoring and verification at both hardware and software levels to ensure that the signals guiding adaptation truly reflect the body’s state. Each stimulation adjustment is linked to a time-stamped multimodal data record, so the relationship between the input and the healing response can be verified during or after a session.

[0584] In one embodiment, dual microcontrollers cross-validate each other's outputs. If the secondary controller detects inconsistent updates, excessive latency, or data corruption, it immediately suspends stimulation and records the event for diagnostic review. The rationale is to prevent faulty' data from influencing the adaptive model and to protect the user from inappropriate responses. This is particularly important for healing applications, where recovery can be disrupted by excessive or unstable stimulation even if it is not painful.

[0585] The safety architecture is therefore designed not only to prevent harm but to preserve the continuity of the healing signal ensuring that the feedback loop remains trustworthy. By maintainingAttorney Docket No. SETH-005 AWO synchronized records and self-verifying algorithms, the system sustains the reliability required for longterm adaptation and clinical interpretation.

[0586] Computational and Hardware Implementation

[0587] The computational framework distributes processing across several components to support real-time operation while maintaining healing-focused control. The on-device module executes fast-loop safety functions and short-latency signal processing, responding to immediate physiological variations such as electrodermal spikes or cardiac acceleration. The edge module ty pically a connected tablet or mobile device manages session-level optimization and model updates based on the healingeffectiveness metric E(t). A remote or cloud module may perform cross-session analysis to identify long-term recovery patterns and refine proxy models across a population of users.

[0588] This distributed approach mirrors the layered structure of biological regulation: rapid reflexes near the body, intermediate integration at the organ or session level, and slower adaptation at the systemic scale. The architecture ensures that each computation occurs at the timescale most relevant to its purpose — moment-to-moment safety near the user and slower, analytical learning at the network level.

[0589] The design philosophy behind this hierarchy is not computational efficiency alone but biological alignment: the processing structure is meant to parallel how the human system heals, with local reflexes ensuring safety and central integration guiding adaptation.

[0590] System Validation and Functional Evaluation

[0591] In one embodiment, system validation focuses on confirming that adaptive control behaves in a manner consistent with healing objectives rather than on collecting subjective user feedback. Bench and simulation tests evaluate whether the controller correctly interprets simulated multi-modal input as evidence of recovery or imbalance and whether it adjusts stimulation accordingly. These evaluations ensure that, when confronted with realistic patterns of physiological variability, the system’s learning behavior remains stable and convergent.

[0592] Functional verification also examines the system’s ability to distinguish between shortterm discomfort and constructive healing response. For instance, an appropriate validation scenario may involve synthetic data showing increased sympathetic activity followed by parasympathetic rebound patterns often seen in regenerative or rehabilitative processes. The system passes validation when it recognizes this as an effective healing sequence rather than suppressing the initial response as it would in a pain-relief system.

[0593] The underlying purpose of validation is therefore to confirm the fidelity’ of the healingoptimization framework that is, to ensure the algorithms correctly interpret and guide the biologicalAttorney Docket No. SETH-005 AWO processes they aim to support. The embodiment thus demonstrates a shift from evaluating user comfort to evaluating systemic progress, completing the distinction between adaptive healing control and conventional stimulation therapy.

[0594] Therapeutic Applications and Use Cases

[0595] The somato-autonomic system supports a range of therapeutic and restorative applications unified by the same principle: the device adapts stimulation based on objective evidence of healing progress rather than immediate sensation. Each implementation uses the healing-effectiveness metric E(t) and the feedback architecture shown in FIG. 22 and FIG. 22A to guide stimulation toward physiological recovery. Example applications may include, for example: bladder control, gastrointestinal function, circulatory and stress recovery, and musculoskeletal healing. Bladder regulation may include gentle electrical or mechanical stimulation near sacral or pelvic regions to re-establish balanced autonomic control of detrusor muscle activity. The controller monitors HRV and GSR to verily parasympathetic engagement, maintaining stimulation only while those indicators trend toward normalization.

[0596] Gastrointestinal function may include slow rhythmic stimulation of the abdominal wall or lower thorax to enhance vagal tone, improving motility and reducing nausea. The system interprets stable respiration patterns and decreased sympathetic markers as evidence of restored digestive rhythm.

[0597] Circulatory and stress recovery may include localized thoracic or cervical stimulation to promote vasodilation and peripheral warming. Rising skin temperature and increased HRV coherence are used as quantitative markers of recovery from exertion or stress.

[0598] Musculoskeletal healing may include, when applied over injured or inflamed tissue, the array sensing gradual changes in EMG and temperature that reflect perfusion and relaxation. The controller modulates amplitude and frequency to sustain those constructive responses even if mild soreness is reported.

[0599] Across all implementations, the optimization target remains the same — movement of multi-modal signals toward equilibrium and repair — thereby differentiating this system from open-loop stimulators that equate success with immediate pain reduction.

[0600] Human Factors and Usability Design

[0601] Because healing requires repeated engagement rather than a single session, usability is designed to encourage consistent, low-effort participation while preserving the accuracy of data used for learning. The flexible substrate conforms to natural body curvature and can remain in place for multiple treatments. The interface presents only essential controls — Start, Pause, Stop, and Intensity so users do not need to interpret complex settings during recovery.Attorney Docket No. SETH-005 AWO

[0602] Visual or haptic indicators summarize progress using simplified color or tone codes that mirror autonomic state (for example, shifting from red toward blue as parasympathetic dominance increases). These cues help users understand that a transient return of discomfort does not necessarily indicate a problem; rather, it may reflect the body’s adjustment during repair. By minimizing cognitive load, the design ensures that adherence supports physiological recovery instead of distracting from it.

[0603] Safety and Protective Architecture

[0604] Safe mechanisms are integral to maintaining a therapeutic environment conducive to healing. The system monitors multiple parameters simultaneously heart-rate variability, skin conductance, and local impedance to detect when stimulation begins to exceed the body’s adaptive capacity. When this occurs, the controller automatically scales output to within safe limits or pauses the session.

[0605] Independent watchdog processes run parallel to the main control loop, verifying that parameter updates remain synchronized with sensor feedback. If communication is lost or an anomaly is detected, stimulation halts immediately and the event is logged for later review. Because the goal of this embodiment is to support ongoing recovery rather than to override pain, safety functions emphasize physiological stability over user comfort perception.

[0606] Thermal, electrical, and mechanical thresholds are established through calibration and updated adaptively as the system leams individual tolerance profiles. This allows stimulation to remain effective as tissue conditions change during healing while preventing over-exposure. The result is a safety framework that acts not only as protection but as a guardian of the healing process, ensuring that every adaptive adjustment remains within the range that fosters restoration rather than fatigue or irritation.

[0607] Integration with Other Embodiments

[0608] The somato-autonomic healing system operates within the same adaptive stimulation framew ork that governs the other embodiments. Each embodiment uses a shared closed-loop architecture consisting of the stimulation component, the multi-modal sensor system, the feedback mechanism, and the controller. What differentiates them is the optimization target. In pain-relief systems, the control objective is the reduction of nociceptive activity; in pleasure-oriented systems, it is the enhancement of hedonic or rew ard-related patterns; and in this embodiment, it is the advancement of multi-modal indicators of healing and autonomic recovery.

[0609] Figures referenced in this embodiment illustrate specific examples of the healing configuration but also represent generalized control and learning structures applicable across the adaptive-stimulation architecture. Unless context dictates otherwise, all such structures, algorithms, andAttorney Docket No. SETH-005 AWO signal-processing methods should be understood as interoperable among the embodiments and falling within the scope of the claims.

[0610] Pleasure Device

[0611] Another embodiment of an adaptive stimulation architecture, as described herein, includes a system configured to generate, sustain, and modulate sexual pleasure through real-time physiological feedback. The system recognizes that most erotic devices deliver fixed patterns of vibration, suction, or pressure without awareness of their effect on the user’s body. Such open-loop operation can require user effort or delivers inconsistent results, stimulation that the user habituates to, or overstimulation. The present embodiment describes a pleasure device that is a responsive instrument utilizing physiological measures of arousal and orgasmic progression to optimize its behavior automatically to safely maintain or intensify pleasure.

[0612] System Context and Benefit

[0613] The need addressed by this system arises from the gap between human sexual variability and the static design of existing vibratory or pneumatic devices. Female orgasm, in particular, depends on an evolving balance between stimulation intensity, rhythm, and spatial focus, and differs markedly between individuals and even across sessions for the same individual. Male arousal and climax, while typically faster to evoke, exhibit analogous requirements for adaptive pacing and pressure regulation. The inability of current devices to measure physiological feedback results in the need for conscious control and effort, limiting the capabilities of the device.

[0614] The disclosed system can potentially increase the probability, reliability, and duration of orgasm by maintaining appropriate stimulation across a number of variables within a continuously updated zone of physiological optimization. It also allows quantifiable research into sexual function by recording objective markers of arousal, thereby transforming a consumer vibrator into a closed-loop science-driven wellness and therapeutic platform.

[0615] Current State of Pleasure Stimulation Technology

[0616] Known consumer pleasure devices incorporate basic sensing capabilities. A commercially available vibrator incorporating internal pressure sensors and accelerometers includes internal pressure sensors and accelerometers that record data during use, presenting post-session visualizations to users through a smartphone application. A music-responsive device that modulates vibration patterns according to audio amplitude and frequency responds to music input by modulating vibration patterns according to audio amplitude and frequency. Similar products log usage patterns, contact pressure, or movement but do not implement closed-loop physiological feedback.Attorney Docket No. SETH-005 AWO

[0617] These existing sensor-equipped devices share an architectural limitation: they collect data for post-hoc analysis or respond to external audio signals, but they do not adapt stimulation parameters automatically based on the measure of physiological effectiveness in real time and adapt stimulation parameters automatically to optimize arousal or pleasure. Such sensor-based devices, for example, records mechanical signals generated by pelvic floor contractions but do not process autonomic indicators such as heart rate variability, electrodermal activity7, or local blood flow, nor does it adjust vibration parameters dynamically based on those measurements. Music-responsive devices modulate output according to audio characteristics rather than the user's physiological state, resulting in stimulation patterns unrelated to actual arousal progression.

[0618] The present embodiment distinguishes itself through real-time physiological control: the system continuously measures multiple autonomic and local hemodynamic signals, computes a scalar effectiveness metric that quantifies a personalized arousal state, and executes optimization algorithms that adjust frequency, amplitude, spatial distribution, and waveform parameters every 1-3 seconds to maximize that metric. This closed-loop architecture enables the device to detect plateau, prevent habituation through micro-variation, identity7approach to orgasm through characteristic physiological signatures, and personalize stimulation across sessions through reinforcement learning. The resulting system does not simply record what happens during use or respond to external audio cues; it actively steers physiological state toward a user-specific optimum using objective measurement and control theory. These examples are representative of consumer devices available during the 2020-2024 period and are included only to illustrate the current state of the art.

[0619] System Overview

[0620] Referring to FIG. 23, the adaptive pleasure system 2300 comprises a stimulation module 2302, a sensor module 2304, a feedback mechanism 2306, a proxy-measurement generator 2308, and a controller 2310 in communication with a user interface 2312. The stimulation module 2302 delivers mechanical, pneumatic, or electro-tactile energy to external or genital tissue. The sensor module 2304 measures autonomic and local responses including heart-rate variability (HRV), electrodermal activity (EDA), respiration, and local blood flow. The feedback mechanism 2306 computes an effectiveness metric proportional to pleasure or arousal. The proxy generator 2308 converts accessible sensor data into estimates of deeper physiological variables, such as genital vasocongestion or cortical activation. The controller executes a control algorith...

Claims

Attorney Docket No. SETH-005 AWOCLAIMSWhat is claimed is:

1. A system for adaptive therapeutic stimulation comprising: a) a stimulation device configured to apply adjustable stimulation to a user, wherein the stimulation device is controllable to vary at least one stimulation parameter selected from amplitude, frequency, waveform, duration, and spatial location; b) at least one sensor configured to measure a physiological response of the user, wherein the physiological response is indicative of therapeutic effectiveness of the stimulation; c) at least one processor operatively coupled to the stimulation device and the at least one sensor; and d) a non-transitory computer-readable memory operatively coupled to the processor and storing instructions that, when executed by the processor, cause the system to implement closed- loop feedback control that continuously adapts stimulation during a session to optimize a therapeutic outcome.

2. The system of claim 1 comprising a plurality of sensors configured to measure different physiological modalities, including autonomic and neural signals, wherein the processor applies a fusion algorithm combining multimodal data to compute a composite effectiveness metric and explores stimulation parameter space to identify parameter sets maximizing the metric.

3. The system of claim 2, wherein the processor applies a Bayesian optimization algorithm maintaining a probabilistic model of parameter-effectiveness relationships, selecting new parameter combinations to test by maximizing an acquisition function balancing exploration and exploitation.

4. The system of claim 1, wherein the processor implements a genetic algorithm maintaining a population of parameter sets, evaluating fitness based on physiological feedback, and evolving stimulation parameters through selection, crossover, and mutation operations toward optimal therapeutic effectiveness.

5. The system of claim 1, wherein the processor implements a reinforcement-learning algorithm modeling the system as an agent, defining physiological states, actions as parameter changes, and rewards as improvements in effectiveness, thereby learning a policy that initializes future sessions with optimized parameters.Attorney Docket No. SETH-005 AWO6. The system of claim 1, wherein parameter adjustment comprises computing a gradient of the effectiveness metric with respect to continuous stimulation parameters and iteratively modifying the parameters in the gradient direction to increase effectiveness.

7. The system of claim 1. wherein a heart-rate-variability sensor measures inter-beat intervals, the processor computes HRV frequency and time-domain metrics, and increased parasympathetic activity is correlated with therapeutic effectiveness.

8. The system of claim 1, wherein a galvanic skin response sensor measures electrodermal activity, and the processor determines stimulation effectiveness from GSR baseline and event frequency changes indicative of relaxation or controlled arousal.

9. The system of claim 1, wherein a functional near-infrared spectroscopy sensor measures cortical hemodynamic activity7, and the processor adjusts stimulation to minimize pain- related or maximize pleasure-related cortical activation patterns.

10. The system of claim 1, wherein the stimulation component comprises a flexible multiarray electrode pad having individually addressable electrode circuits, each independently controlled to test spatial activation patterns and identify zones maximizing therapeutic effectiveness.

11. The system of claim 1, wherein the processor executes a diagnostic-probing mode that systematically applies higher-intensity pulses across multiple spatial sites, measures physiological response magnitude, identifies optimal response locations, and then transitions to therapeutic stimulation at those sites.

12. The system of claim 1, wherein adaptation occurs on multiple timescales comprising: a) immediate safety adjustment within milliseconds to seconds; b) in-session optimization within seconds to minutes; and c) cross-session learning across hours to weeks for progressive personalization.

13. The system of claim 1, further comprising network communication enabling upload of anonymized data to a remote aggregator implementing federated learning, wherein global model parameters are updated and redistributed to improve local adaptive performance while preserving user privacy.

14. The system of claim 1, further comprising a voice-input interface and arbitration engine configured to interpret natural-language commands and physiological data concurrently, prioritizing safety when conflicts arise and learning user-specific arbitration preferences over time.Attorney Docket No. SETH-005 AWO15. The system of claim 1, wherein stimulation activates somato-autonomic reflexes to modulate visceral or autonomic function for applications including bladder control, nausea reduction, and parasympathetic activation.

16. The system of any preceding claim, wherein a safety subsystem continuously evaluates physiological and electrical parameters, and upon detecting unsafe conditions automatically reduces or halts stimulation and alerts the user before resuming operation.

17. The system of claim 1, further comprising: a) at least one proxy sensor configured to measure a proxy physiological indicator correlated with a sophisticated physiological measurement; b) a non-transitory computer-readable memory storing a trained proxy model comprising a learned relationship between proxy signals and corresponding high-fidelity7physiological measurements; and c) a processor configured to infer a target physiological state using the proxy model and adjust stimulation parameters based on the inferred state.

18. The system of claim 17, wherein the proxy indicator comprises HRV signals correlated to EEG or fNIRS patterns, and the trained model enables inference of brain activity from HRV for feedback control without direct neural measurement.

19. The system of claim 17, wherein proxy indicators include galvanic skin response and skin temperature, and the proxy model correlates these with sympathetic and parasympathetic balance to maintain autonomic equilibrium.

20. A method for adaptive therapeutic stimulation comprising: a) applying a first stimulation with a defined parameter set; b) measuring physiological response during stimulation; c) determining effectiveness via processor analysis; d) automatically adjusting at least one stimulation parameter based on effectiveness to generate a new parameter set; and e) repeating the process in real time to continuously optimize the stimulation outcome.

Images