Context-aware assistive system for behavioral monitoring, memory augmentation, and safety intervention
A distributed context-aware assistive platform with wearable devices and edge computing addresses the limitations of existing technologies by providing real-time behavioral monitoring and adaptive interventions, enhancing safety and memory assistance in aging populations.
Patent Information
- Application Number
- PCT/US2026/027347
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-08
- Filing Date
- 2026-05-08
- Publication Date
- 2026-11-12
AI Technical Summary
Existing assistive technologies lack real-time context awareness, adaptive learning, and closed-loop intervention capabilities for continuous monitoring of user behavior, particularly in aging populations, leading to inadequate safety management and memory assistance.
A distributed, multi-modal context-aware assistive platform that utilizes wearable devices, environmental sensors, and edge computing for real-time behavioral monitoring, adaptive inference models, and hierarchical safety interventions, including haptic, audio, and visual feedback, with closed-loop learning to adapt over time.
Provides continuous, context-aware monitoring and timely interventions, adapting to user behavior and environmental conditions, enhancing safety and memory assistance through personalized and efficient risk management.
Smart Images

Figure US2026027347_12112026_PF_FP_ABST
Abstract
Description
[0001] Title
[0002] Context- Aware Assistive System for Behavioral Monitoring, Memory Augmentation, and Safety Intervention
[0003] Cross-Reference to Related Applications
[0004] This application claims priority to U.S. Provisional Application number 63 / 801,806 filed May 8, 2025.
[0005] Field of the Invention
[0006] The present invention relates generally to assistive technologies, wearable computing systems, artificial intelligence, edge computing, and context-aware monitoring systems and, more particularly, to systems and methods for real-time behavioral monitoring, memory augmentation, adaptive intervention, and safety management using wearable devices, environmental sensors, smart glasses, and distributed intelligence architectures.
[0007] Background
[0008] Cognitive decline, memory impairment, and reduced situational awareness pose significant risks to individuals, particularly in aging populations. Existing assistive technologies provide limited functionality, often relying on static reminders or user-initiated interactions. However, real-world scenarios such as leaving a stove unattended, forgetting medication, or losing track of ongoing tasks require continuous monitoring and dynamic intervention. Current systems lack real-time context awareness, adaptive learning of user behavior, and closed-loop intervention capability. Accordingly, there is a need for a system that:
[0009] • continuously monitors user state,
[0010] • intelligently detects risk conditions,
[0011] • provides timely interventions, and
[0012] • adapts over time.
[0013] Summary of the InventionThe present invention provides a distributed, multi-modal context-aware assistive platform designed for closed-loop behavioral monitoring and hierarchical safety intervention. The system collects multi-modal sensor data from wearable (including smart glasses) and environmental sources, determines a user context state, and identifies behavioral anomalies or risks. By applying adaptive inference models — including edge-based processing for time-critical decisions — the system delivers haptic, audio, or visual feedback. As programmed, the system can remind and alert the user and if ignored inform a family member or caregiver. The platform further incorporates a learning layer that updates behavioral models based on user responses and historical data.
[0014] The system may include:
[0015] • wearable devices (including smart glasses)
[0016] • environmental sensors,
[0017] • edge computing nodes,
[0018] • cloud-based intelligence.
[0019] Applications include:
[0020] • safety monitoring (e.g., cooking hazards),
[0021] • medication adherence,
[0022] • memory assistance,
[0023] • task completion guidance.
[0024] Brief Description of Drawings
[0025] FIG. 1A — System Architecture
[0026] FIG. IB — System Architecture and Logic Flow
[0027] FIG. 2 — Data Flow Diagram
[0028] FIG. 3 — Closed-Loop Feedback Cycle
[0029] FIG. 4 — Safe Cooking Scenario
[0030] FIG. 5 — Memory Recall ScenarioDefinitions
[0031] As used herein:
[0032] • “context state” refers to a dynamically determined representation of a user’s activity, environment, behavioral condition, physiological state, or situational status.
[0033] • “intervention” refers to any notification, feedback, prompt, escalation, or corrective action generated by the system.
[0034] • “multi-modal data” refers to data originating from multiple heterogeneous sensor sources including motion, audio, video, environmental, physiological, or location-based sources.
[0035] • “edge device” refers to a localized computational device configured to perform processing proximate to the sensing source.
[0036] Detailed Description
[0037] 1. System Overview
[0038] Distributed Sensing and Edge Processing Layer
[0039] In one embodiment, the system comprises a sensing layer utilizing low-power microcontrollers (e.g., ESP32-S3) configured for edge-based pre-processing. This layer performs highdimensional feature extraction, such as Visual SLAM (Simultaneous Localization and Mapping) for indoor positioning and Voice Activity Detection (VAD) to preserve privacy and battery life. By performing initial inference at the edge, the system ensures rapid response times for safety-critical events such as fall detection or immediate environmental hazards.
[0040] Hierarchical Decision Engine and Intervention Logic
[0041] The decision engine utilizes a hierarchical approach to intervention. When a deviation from expected behavior is detected, the system calculates a confidence score. Lower-confidence risks may trigger a non-intrusive haptic "nudge," while high-confidence safety risks (e.g., an unattended stove) trigger escalated audio-visual alerts. The system monitors user compliance in real-time, providing a closed-loop mechanism where a "false dismissal" or failure to correct a hazardous state leads to further escalation to designated caregivers or emergency services.
[0042] The system therefore comprises:
[0043] • Sensing Layer
[0044] • Processing Layer
[0045] • Decision Layer
[0046] • Feedback Layer
[0047] • Communication Layer2. Sensing Layer
[0048] The sensing layer may comprise one or more wearable, portable, stationary, or environmental sensing devices configured to continuously or periodically acquire data associated with user behavior, environmental conditions, physiological states, and object interactions.
[0049] In various embodiments, the sensing layer may include:
[0050] • inertial measurement unit (IMU) sensors including accelerometers, gyroscopes, and magnetometers;
[0051] • image acquisition devices including RGB, infrared, depth, or thermal cameras;
[0052] • microphones and audio acquisition circuitry;
[0053] • physiological sensors including heart-rate, blood oxygen, respiration, skin temperature, galvanic response, or sleep monitoring sensors;
[0054] • environmental sensors including temperature, humidity, smoke, gas, appliance state, light level, motion, proximity, or occupancy sensors;
[0055] • geolocation modules including GPS, BLE beacon positioning, Wi-Fi localization, or visual localization systems.
[0056] In certain embodiments, sensing devices may be integrated into:
[0057] • smart glasses,
[0058] • wrist-worn devices,
[0059] • pendants,
[0060] • smartphones,
[0061] • hearing-assistance devices,
[0062] • smart clothing,
[0063] • or distributed smart-home infrastructure.
[0064] The sensing layer may operate continuously, periodically, event-triggered, or adaptively based on power-management policies, detected context states, or user activity.
[0065] Sensor data may be timestamped and synchronized to support temporal correlation across heterogeneous data streams.
[0066] In certain embodiments, the sensing layer performs preliminary feature extraction locally before transmitting processed or partially processed information to downstream processing layers.
[0067] 3. Processing LayerThe processing layer may comprise one or more distributed computational components configured to process, analyze, correlate, compress, and infer information from acquired sensor data.
[0068] In certain embodiments, the processing layer includes an edge processing subsystem implemented using:
[0069] • microcontrollers,
[0070] • mobile processors,
[0071] • smartphones,
[0072] • wearable processors,
[0073] • or embedded Al accelerators.
[0074] The edge subsystem may execute:
[0075] • low-latency event detection,
[0076] • speech activity detection,
[0077] • wake-word detection,
[0078] • object recognition,
[0079] • motion classification,
[0080] • fall detection,
[0081] • appliance-state monitoring,
[0082] • or environmental anomaly detection.
[0083] In some embodiments, the edge subsystem may selectively transmit only high-value events, extracted features, or compressed representations to conserve bandwidth and preserve privacy. The processing layer may additionally include a remote processing subsystem comprising cloudbased or server-based infrastructure configured to execute computationally intensive operations including:
[0084] • long-term behavioral modeling,
[0085] • user habit learning,
[0086] • predictive analytics,
[0087] • historical correlation analysis,
[0088] • large language model (LLM) reasoning,
[0089] • caregiver coordination,
[0090] • or multi-user population analysis.
[0091] The system may dynamically partition workloads between edge and cloud resources based on:
[0092] • latency requirements,
[0093] • bandwidth availability,
[0094] • battery state,
[0095] • privacy policies,
[0096] • or confidence thresholds.4. Context Determination
[0097] The system determines a context state representing a probabilistic assessment of the user’s current activity, condition, environment, and behavioral trajectory.
[0098] The context state may include:
[0099] • user activity classification,
[0100] • current or inferred location,
[0101] • posture,
[0102] • object interaction,
[0103] • environmental condition,
[0104] • emotional indicators,
[0105] • physiological indicators,
[0106] • sleep quality,
[0107] • social interaction state,
[0108] • task progression state,
[0109] • or deviation from historical routines.
[0110] In certain embodiments, context determination utilizes:
[0111] • probabilistic inference,
[0112] • Bayesian estimation,
[0113] • neural -network inference,
[0114] • transformer-based models,
[0115] • recurrent temporal models,
[0116] • hidden Markov models,
[0117] • or sensor fusion algorithms.
[0118] The system may continuously update the context state using temporal correlation and rolling observation windows.
[0119] In certain embodiments, contextual understanding incorporates historical user-specific patterns to distinguish between:
[0120] • normal deviations,
[0121] • temporary anomalies,
[0122] • and high-risk behavioral conditions.
[0123] Therefore as an example user context state may include:
[0124] • activity (walking, cooking, idle)
[0125] • location
[0126] • time-of-day patternsinteraction with objects
[0127] sleep behaviors : breathing and movements
[0128] Therefore context can be derived via:
[0129] • sensor fusion,
[0130] • machine learning models,
[0131] • probabilistic inference.
[0132] 5. Risk Detection
[0133] The system may identify one or more risk conditions based on the determined context state. Risk conditions may include:
[0134] • unattended appliance operation,
[0135] • missed medication,
[0136] • abnormal inactivity,
[0137] • wandering behavior,
[0138] • sleep irregularities,
[0139] • fall events,
[0140] • dehydration risk,
[0141] • unsafe environmental conditions,
[0142] • cognitive confusion,
[0143] • repeated task abandonment,
[0144] • or deviation from expected routines.
[0145] In certain embodiments, the system computes a dynamic risk score using:
[0146] • severity weighting,
[0147] • temporal persistence,
[0148] • historical behavior,
[0149] • environmental state,
[0150] • and user responsiveness.
[0151] Risk detection may incorporate adaptive thresholds personalized to individual users.
[0152] The system may additionally distinguish between:
[0153] • nuisance events,
[0154] • moderate-risk events,
[0155] • and emergency conditions.6. Decision Engine
[0156] The decision engine is configured to determine one or more intervention actions responsive to identified risk conditions or contextual states.
[0157] The decision engine may utilize:
[0158] • deterministic rules,
[0159] • probabilistic reasoning,
[0160] • machine learning inference,
[0161] • reinforcement learning,
[0162] • neural -network models,
[0163] • or large language model (LLM)-assisted reasoning.
[0164] The system may evaluate:
[0165] • intervention urgency,
[0166] • confidence level,
[0167] • user responsiveness,
[0168] • environmental context,
[0169] • time of day,
[0170] • caregiver availability,
[0171] • and prior intervention effectiveness.
[0172] Interventions may be selected to minimize:
[0173] • user annoyance,
[0174] • alert fatigue,
[0175] • unnecessary escalation,
[0176] • and false positives.
[0177] In certain embodiments, the system adaptively escalates intervention intensity when corrective action is not detected within a predefined interval.
[0178] 7. Intervention System
[0179] The intervention system provides one or more corrective, assistive, advisory, or emergencyresponse actions.
[0180] Interventions may include:
[0181] • haptic feedback,
[0182] • spoken prompts,
[0183] • visual notifications,• projected augmented-reality indicators,
[0184] • caregiver notifications,
[0185] • SMS alerts,
[0186] • automated telephone calls,
[0187] • smart-home automation commands,
[0188] • or emergency-service escalation.
[0189] In certain embodiments, interventions are staged hierarchically.
[0190] For example:
[0191] 1. silent haptic cue,
[0192] 2. localized audio reminder,
[0193] 3. repeated alert,
[0194] 4. caregiver escalation,
[0195] 5. emergency escalation.
[0196] The intervention system may further verify whether corrective user action occurred after intervention delivery.
[0197] 8. Closed-Loop Learning
[0198] The system continuously adapts using closed-loop learning mechanisms.
[0199] Following an intervention, the system monitors:
[0200] • user response latency,
[0201] • corrective behavior,
[0202] • intervention effectiveness,
[0203] • false dismissals,
[0204] • and repeated behavioral patterns.
[0205] The system may update:
[0206] • behavioral models,
[0207] • intervention thresholds,
[0208] • escalation timing,
[0209] • context classification parameters,
[0210] • and personalization profiles.
[0211] In certain embodiments, reinforcement-learning techniques optimize intervention selection over time.The system thereby evolves individualized assistance patterns specific to each user. This enables personalization over time.
[0212] 9. Communication Layer
[0213] The communication layer supports secure exchange of data among sensing devices, edge processors, cloud systems, caregivers, and external services.
[0214] Communication technologies may include:
[0215] • BLE,
[0216] . Wi-Fi,
[0217] • LTE,
[0218] • 5G,
[0219] • Zigbee,
[0220] • Thread,
[0221] • NFC,
[0222] . MQTT,
[0223] . HTTPS,
[0224] • Web Socket,
[0225] • or peer-to-peer protocols.
[0226] The communication layer may support:
[0227] • intermittent connectivity,
[0228] • delayed synchronization,
[0229] • edge buffering,
[0230] • priority event transmission,
[0231] • and adaptive bitrate communication.
[0232] In certain embodiments, sensitive information is redacted, encrypted, anonymized, or processed locally prior to transmission.10. Example Use Cases
[0233] Use Case -> Safe Cooking Medication Memory Recall Adherence
[0234] Sensing Layer Detect Presence in kitchen Detect Medication box User prompts: Where did I presence or time of day leave X?
[0235] Processing Layer On Smart phone or local embedded processor
[0236] Context Cooking activity Actual pill box opening Tags visual and audio clips Determination removal and categorizes for later retrieval
[0237] Risk Detection User leaves kitchen with User opens wrong N / A
[0238] burner on pillbox or lack this event
[0239] Decision Engine If time is passed and user Lack of med activity or Based on user demand, will does not return to kitchen wrong activity recover past events recorded on audio of visual Intervention Remind user and if ignored escalate to caregiver N / A
[0240] System
[0241] Communication Notification, sound or haptic feedback to user and SMS Play back to user via Audio of
[0242]
[0243] to Caregiver past events
[0244] 11. Alternative Embodiments
[0245] Include:
[0246] • smart glasses embodiment,
[0247] • smartphone-only embodiment,
[0248] • smart-home embodiment,
[0249] • caregiver dashboard embodiment,
[0250] • vehicle safety embodiment,
[0251] • child monitoring embodiment.
Claims
Claims1. A distributed, multi-modal context-aware assistive platform comprising:o a sensing layer including at least one wearable device and at least one environmental sensor configured to collect multi-modal data;o an edge gateway configured to pre-process the multi-modal data to extract highdimensional features and determine a local context state;o a remote processing system configured to receive the features and apply an adaptive inference model to determine a global user context; ando a decision engine configured to initiate a hierarchical intervention based on the global user context.
2. A method for assisting a user, comprising:
1. collecting multi-modal sensor data from at least one wearable or environmental sensor; 2. determining a user context state using the collected data;3. identifying a deviation from an expected behavior or a safety condition;4. applying an adaptive inference model to determine an intervention;5. delivering the intervention via a feedback mechanism;6. monitoring user response to the intervention; and7. updating the adaptive inference model based on the monitored response,wherein the method provides real-time, context-aware assistance for safety, memory recall, or task completion.
3. A distributed assistive system comprising:• an edge device configured to:o perform real-time inference on sensor data;• a remote processing system configured to:o perform higher-order reasoning using aggregated historical data;• a communication link between the edge device and the remote system;wherein:• time-sensitive interventions are executed at the edge, and• adaptive model refinement is performed at the remote system.
4. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the processors to perform the method of claim 25. The system of claim 1, wherein the sensing devices include IMU sensors, cameras, microphones, or environmental sensors.
6. The system of claim 1, wherein the multi-modal data is fused using probabilistic or neural network-based fusion techniques.
6. The system of claim 1, wherein the user context state includes activity classification, location awareness, and temporal patterns.
7. The system of claim 1, wherein expected behavioral patterns are learned over time.
8. The system of claim 1, wherein the risk condition includes unattended appliance operation.
9. The system of claim 8, wherein the appliance is a stove or heating device.
10. The system of claim 1, wherein intervention actions escalate in multiple stages.
11. The system of claim 10, wherein escalation includes notifying a caregiver or emergency contact.
12. The system of claim 1, wherein the intervention includes providing memory recall assistance.
13. The system of claim 12, wherein memory recall includes object location or recent activity recall.
14. The system of claim 1, wherein the adaptive inference model includes a machine learning or LLM-based model.
15. The system of claim 14, wherein the model adapts based on user-specific behavior patterns.
16. The system of claim 1, wherein the sensing device operates in a low-power mode with periodic wake cycles.
17. The system of claim 16, wherein communication is performed using MQTT or low- bandwidth messaging.
18. The system of claim 1, wherein the sensing layer is configured to redact personal identifiable information (PII) before transmission to the remote processing system.
19. The system of claim 1, wherein the decision engine is configured to escalate an intervention based on a failure to detect a corrective user action following an initial alert.
