A closed-loop control method and system for acupuncture robots based on multi-modal perception fusion and qi arrival state evaluation

Abstract

The application provides a kind of acupuncture robot closed loop control method and system based on multi-modal perception fusion and gas state evaluation, the application obtains acupuncture skill information, acupoint temperature, tissue pressure, ion concentration and electroencephalogram characteristic signal etc. Multi-modal physical signals in real time and carries out space-time coupling analysis, for the first time, acupoint microenvironment multi-parameter is included in robot closed loop control;By adopting layered architecture, the simplified digital twin physical deduction model is decoupled with the four-dimensional traditional Chinese medicine knowledge system that integrates expert experience, classic techniques, operation specifications and ancient literature;Through autonomous micro-motion exploration and personalized model construction, the robot can actively learn and accurately adapt to individual differences in patients during treatment;Through data closed loop and incremental learning mechanism, the system can continuously optimize the knowledge base and twin model parameters as the number of treatments increases, realize the continuous evolution of skills, thereby significantly improving the compliance, safety, intelligence level and clinical applicability of acupuncture robot control.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, and in particular to a closed-loop control method and system for acupuncture robots based on multimodal perception fusion and Qi attainment status assessment. Background Technology

[0002] With the deep integration of medical robots and artificial intelligence technologies, acupuncture robots are gradually evolving from simple automated tools into intelligent agents with autonomous decision-making capabilities. Currently, research on acupuncture robots mainly focuses on identifying acupoints using computer positioning technology and controlling the robot to mimic the techniques of a doctor to perform acupuncture. In terms of automatic acupoint positioning, related technologies have been widely used in medical robots such as puncture robots, moxibustion robots, and massage robots; in terms of path planning and precision operation, some surgical robots and surgical-assisted robots have also demonstrated high technical performance. Furthermore, some studies have attempted to introduce laser rangefinders and six-dimensional force sensors for information perception, designing controllers that dynamically adjust the impedance coefficient to control safe needle insertion, needle speed, and the depth and force of acupuncture techniques. Some solutions have also proposed closed-loop control methods for electronic acupuncture based on multimodal information feedback, using long-term, short-term, and real-time representation information of the subject under the stimulation protocol for control.

[0003] However, existing control methods for acupuncture robots generally suffer from the following bottlenecks. First, most robots' control logic is an "open-loop execution" of a preset program, unable to adjust according to the patient's real-time physiological state, and lacking the ability to perceive and respond to the real-time interaction state of the needle tissue during acupuncture. Second, although some studies have introduced force sensors for closed-loop force control, the feedback signals only come from the force / torque sensor at the end of the robotic arm, failing to perceive the real interaction state of the needle within the tissue and the local physiological response of the acupoint. Third, existing solutions generally lack real-time, objective, and multi-dimensional quantitative evaluation methods for "deqi," a core therapeutic indicator of acupuncture, causing the robot's control objective to remain at the level of physical precision rather than therapeutic efficacy. Fourth, existing digital twin solutions often pursue a large and comprehensive model architecture, mixing macroscopic pathogenesis deduction with real-time physical control, resulting in computational latency that cannot meet the needs of high-frequency physical control. More fundamentally, even if robots in existing technologies achieve closed-loop control, they are merely "conditioned reflex" operations based on real-time sensor signals. They cannot reason and make judgments by combining past knowledge like human experts, nor can they learn from successful or unsuccessful experiences and optimize their skills through continuous iteration.

[0004] Therefore, there is an urgent need in this field for a closed-loop control method and system for acupuncture robots that can be executed autonomously by robots without real-time human intervention, and that possesses knowledge reasoning, personalized adaptation and continuous evolution capabilities, in order to solve the technical problems of open-loop control, single perception, lack of quantitative assessment of Qi attainment and lack of learning and evolution capabilities in the existing technology. Summary of the Invention

[0005] The purpose of this invention is to provide a closed-loop control method and system for acupuncture robots based on multimodal perception fusion and Qi attainment state assessment, so as to solve the problems existing in the prior art.

[0006] To achieve the above objectives, the present invention provides the following solution: This invention provides a closed-loop control method for an acupuncture robot based on multimodal perception fusion. The method is autonomously executed by the robot control unit without real-time human intervention and includes the following steps: Real-time acquisition of multimodal acupoint status parameters: synchronously acquire multimodal physical signals reflecting the real-time interaction status of the needle body-acupoint microenvironment, the multimodal physical signals including acupuncture technique information and at least two of the following: acupoint local temperature signal, tissue pressure signal, ion concentration signal and EEG characteristic signal at the time of obtaining Qi; Multimodal spatiotemporal coupling weighting to generate gas yield confidence: Real-time spatiotemporal coupling analysis is performed on the multimodal physical signals to generate quantified gas yield confidence. Physical drive commands are generated based on the air delivery confidence: The air delivery confidence is compared with the desired physical interaction state to dynamically generate physical control commands for directly driving the motor of the end effector. Online iterative optimization until the physical interaction goal is achieved: continuously monitor and iteratively optimize the physical control commands in real time to form a motion control closed loop in the physical domain until the preset air intake state is confirmed.

[0007] Preferably, the multi-mode spatiotemporal coupling weighting step further includes: The time difference between the abrupt change times of at least two peripheral physical signals and the neural response times of the EEG characteristic signals is calculated. The confidence level of the air intake is corrected only when the time difference falls within the preset physiological delay window.

[0008] Preferably, the step of generating physical drive commands based on the confidence level of air delivery specifically refers to a "speed-force" hybrid control based on multimodal state feedback, including: dynamically adjusting the ratio of needle insertion speed to needle insertion force according to the confidence level of air delivery; when the confidence level of air delivery is lower than a first threshold, speed control is the primary method and force control is the secondary method; when the confidence level of air delivery reaches the first threshold but is lower than a second threshold, force control is the primary method and speed control is the secondary method; when the confidence level of air delivery reaches the second threshold, smooth deceleration or needle insertion is executed.

[0009] Preferably, the step of generating physical driving instructions based on the confidence level of Qi is to integrate predictive inference instructions from the lightweight digital twin subsystem as a feedforward reference, forming a "feedforward-feedback" composite control; the lightweight digital twin subsystem is a simplified physical perception and intention coupling model, specifically used for real-time inference of the physical state of needle-tissue interaction, with a real-time inference cycle of less than 10ms; the model does not include the meridian and viscera state layer for macroscopic pathogenesis analysis.

[0010] Preferably, the TCM prior knowledge required for the multi-modal spatiotemporal coupling weight determination and the generation of the physical driving instructions is provided by an external callable knowledge base independent of the lightweight digital twin subsystem; The externally callable knowledge base includes at least one of the following modules: an expert experience and techniques module for storing clinical techniques and experience data of renowned veteran TCM doctors; a classic techniques module for storing kinematic and dynamic characteristic parameters of standard acupuncture techniques; a standardized operation method module for storing operation specifications for acupoint location, acupuncture depth, angle, needle retention time, and contraindications; and an ancient literature information source module for storing structured text data related to acupoint indications, tonification and sedation principles, Qi attainment signs, and appropriate and inappropriate acupuncture techniques from classic medical books.

[0011] Preferably, before generating the physical drive command, the control unit calls the Qi-de-physiological marker library and various knowledge modules in the external callable knowledge base, matches the real-time feature vector of the multimodal physical signal with the probability mapping model of each Qi-de-state in the knowledge base, and combines the expert experience features retrieved from the expert experience method module to comprehensively assist in generating the physical drive command.

[0012] Preferably, the "feedforward-feedback" composite control step further includes: When the needle reaches the predicted depth but the confidence level of obtaining Qi does not meet the standard, the needle is controlled in a feedback mode to autonomously perform micro-movement exploration within the safety constraints obtained from the standard operation method module, and the characteristic values ​​of the multimodal state parameters obtained when obtaining Qi are recorded in real time to construct a personalized obtaining Qi model for the patient.

[0013] Preferably, the control unit is further configured to, after completing the preset treatment task, encapsulate the three-element data sequence of the twin prediction value, the sensor measurement value and the knowledge base reference value of this operation into an incremental learning sample, which is used to optimize the probability mapping relationship and the content of each knowledge module in the externally callable knowledge base, and to calibrate the model parameters of the lightweight digital twin subsystem.

[0014] This invention also provides a closed-loop control system for an acupuncture robot based on multimodal perception fusion and Qi attainment state assessment, comprising: The multimodal signal acquisition module is used to acquire multimodal physical signals that reflect the real-time interaction status of the needle body-acupoint microenvironment. The spatiotemporal coupling weighting module is used to perform real-time spatiotemporal coupling analysis on the multimodal physical signals and generate quantified air confidence scores. The instruction generation module is used to compare the air confidence level with the expected physical interaction state and dynamically generate physical control instructions. The drive module is used to directly drive the motor of the end effector according to the physical control command; The system also includes a lightweight digital twin subsystem and an externally callable knowledge base; the lightweight digital twin subsystem is used to perform real-time simulation of the physical state of needle-tissue interaction, with a real-time simulation cycle of less than 10ms; the externally callable knowledge base is independent of the lightweight digital twin subsystem and includes at least one of the following modules: expert experience and techniques module, classic techniques module, standardized operation method module, and ancient literature information source module.

[0015] The present invention also provides a body-worn intelligent acupuncture robot, including a multi-axis collaborative robotic arm. The end effector of the multi-axis collaborative robotic arm is equipped with an end effector, which is arranged sequentially from the proximal end to the distal end along its own axis as follows: The push module includes a guide rail extending along the axial direction and a pusher slidably disposed on the guide rail, the pusher having a first working surface facing the end of the needle handle; A clamping module, located at the distal end of the pushing module, includes a clamping body and a first driving component for driving the clamping body to move. The clamping body has a coupling surface that matches a preset structure at the end of the needle handle. The end effector has a first working state and a second working state. In the first working state, the pusher slides along the guide rail to the distal end and applies an axial thrust to the end of the needle handle through the first action surface. In the second working state, the pusher retracts to the proximal end and disengages from the needle handle, and the clamping body engages with the end of the needle handle through the coupling surface. It also includes a control unit, which, when executing a computer program stored therein, implements a closed-loop control method for acupuncture robots based on multimodal perception fusion.

[0016] The present invention achieves the following beneficial technical effects compared to the prior art: This invention provides a closed-loop control method and system for acupuncture robots based on multimodal perception fusion and Qi attainment state assessment. By constructing a motion control closed loop that is completely independent of human physical domain, it achieves a technological leap from "humans who rely on technology" to "machines that understand traditional Chinese medicine, can think, can learn, and can iteratively improve." Specifically, this invention acquires multimodal physical signals such as acupuncture technique information, acupoint temperature, tissue pressure, ion concentration, and EEG characteristic signals in real time and performs spatiotemporal coupling analysis. For the first time, it incorporates multiple parameters of the acupoint microenvironment into the robot's closed-loop control, achieving precise physical interaction with the goal of "deqi" (obtaining qi). By adopting a layered architecture of "lightweight digital twin + externally callable knowledge base," it decouples the simplified physical deduction model from the four-dimensional TCM knowledge system that integrates expert experience, classic techniques, operating procedures, and ancient literature. This ensures a real-time deduction cycle of less than 10ms to meet the requirements of high-frequency physical control, while also endowing the robot with a solid foundation in TCM theory and clinical experience to support its knowledge reasoning ability. Through autonomous micro-motion exploration and personalized model construction, the robot can actively learn and accurately adapt to individual patient differences during treatment. Through data closed-loop and incremental learning mechanisms, the system can continuously optimize the knowledge base and twin model parameters with the increase of treatment sessions, achieving continuous skill evolution. This significantly improves the compliance, safety, intelligence level, and clinical applicability of the acupuncture robot control. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 The flowchart of the closed-loop control method for acupuncture robots based on multimodal perception fusion and Qi attainment state assessment provided by the present invention is shown. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The purpose of this invention is to provide a closed-loop control method and system for acupuncture robots based on multimodal perception fusion and Qi attainment state assessment, so as to solve the problems existing in the prior art.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] Example 1: The following combination Figure 1 The technical solution of this application is described in detail.

[0023] This application provides a closed-loop control method for an acupuncture robot based on multimodal perception fusion. This method is autonomously executed by the robot control unit without real-time human intervention. The method includes the following steps: real-time acquisition of multimodal acupoint state parameters; multimodal spatiotemporal coupling weighting to generate a Qi-de-confidence level; generation of physical drive commands based on the Qi-de-confidence level; and online iterative optimization until the physical interaction goal is achieved. The following provides a detailed description of each step.

[0024] In this embodiment, the acupuncture robot includes a multi-axis collaborative robotic arm, an end effector mounted at the end of the robotic arm, and a control unit. The end effector is generally elongated along its own axis, with its axis aligned with the needle insertion direction; that is, the distal end of the device points towards the patient's skin, and the proximal end points towards the robotic arm. The end effector has a pushing module and a clamping module arranged sequentially from proximal to distal along its own axis. The pushing module is located on the proximal side, and the clamping module is located on the distal side, both arranged in series along the same axis. The pushing module is used to push the needle from the needle storage position to the clamping position axially during the needle feeding stage, and to continue applying axial thrust during the needle insertion stage to insert the needle into the patient's tissue. The pushing module specifically includes a guide rail extending axially along the end effector and a pushing component slidably mounted on the guide rail. The guide rail is fixedly installed inside the housing of the end effector, and its cross-sectional shape is adapted to the shape of the needle. As a specific example, the cross-section of the guide rail can be configured as a V-groove structure, with the two inclined surfaces of the V-groove supporting the two sides of the needle shaft, thus maintaining the stability of the needle shaft in the axial direction during sliding. As another example, the cross-section of the guide rail can also be configured as an arc-shaped groove structure, with the curvature of the arc matching the outer diameter of the needle bar, which can also guide the needle body to slide stably along the axial direction. Preferably, the inner wall surface of the guide rail is provided with a friction-reducing coating, which can be made of a low-friction coefficient material such as polytetrafluoroethylene, to reduce the sliding friction resistance between the needle body and the guide rail during the pushing process and ensure the smoothness of the pushing action. The pushing component is driven by a linear actuator to slide along the guide rail. The linear actuator can be any one of a micro screw motor, a voice coil motor, or a piezoelectric linear motor, and the present invention does not specifically limit it. The pushing component has a first working surface facing the end of the needle handle, the shape of which is adapted to the end face shape of the needle handle. In the first working state, the pushing component slides along the guide rail toward the distal end, and abuts against the end face of the needle handle through the first working surface, thereby applying an axial thrust toward the distal end to the needle handle. In the second working state, the pushing component retracts along the guide rail toward the proximal end and disengages from the needle handle. Preferably, the first working surface of the pusher is set as a concave curved surface, the radius of curvature of which matches the edge contour of the end face of the needle handle, so that the pusher and the end of the needle handle form point contact rather than surface contact or clamping contact. This point contact method allows the pusher to transmit axial thrust to the needle handle without applying radial clamping force. When the needle body is subjected to tissue resistance, it can generate a small amount of free floating in the axial direction, thereby simulating the needle insertion feel of a traditional Chinese medicine practitioner with firm fingers and flexible wrist.

[0025] The clamping module is located at the distal end of the push module and is used to take over the needle after insertion and perform needle manipulation techniques such as twisting and lifting. The clamping module specifically includes a clamping body and a first drive assembly that drives the clamping body. The clamping body is located at the distal center of the end effector and has a coupling surface that matches a preset structure at the end of the needle shank, forming a complementary engagement with the end of the needle shank. In one specific embodiment, the coupling surface of the clamping body is a cross-shaped protrusion, and correspondingly, the preset structure at the end of the needle shank is a cross-shaped groove. The cross-shaped protrusion and the cross-shaped groove are complementary in shape and have a clearance fit. This cross-shaped coupling structure has self-aligning characteristics. When the cross-shaped protrusion moves towards the needle shank under the drive of the axial feed drive mechanism, even with a slight angular deviation, the inclined guiding effect of the protrusion and groove allows the clamping body to automatically rotate and align smoothly under axial thrust, without the need for additional visual or angular alignment steps. The clamping module also includes a second drive assembly, which is connected to the clamping body for driving the clamping body to rotate around its own axis. The second drive component can be implemented using a micro servo motor and reducer, with a rotational angle resolution of not less than 0.1 degrees. Driven by the second drive component, the clamping body reciprocates around its axis, thereby rotating the needle it engages with, thus performing the twisting technique. The first drive component is connected to the clamping body and is used to drive the clamping body to perform a micro-feeding motion along the axial direction. This first drive component can be implemented using a linear drive mechanism independent of the linear actuator of the pusher, with a feed resolution of not less than 0.05 millimeters. Driven by the first drive component, the clamping body reciprocates slightly along the axial direction, thereby driving the needle to perform the insertion and withdrawal technique within the tissue. Preferably, the clamping module also includes a locking component, which is located between the clamping body and the housing of the end effector. During the non-feeding period of the second working state, i.e., when the clamping body does not need to perform the insertion and withdrawal action, the locking component is used to constrain the axial displacement of the clamping body, keeping the needle stable at the current depth and preventing accidental needle displacement due to patient movement or mechanical vibration. The locking assembly can be implemented using an electromagnetic brake structure or a worm gear self-locking structure.

[0026] The end effector has a first working state and a second working state. In the first working state, the pusher slides along the guide rail towards the distal end, applying axial thrust to the end of the needle handle through the first action surface. This thrust drives the needle to move towards the distal end along the guide rail, sequentially completing the action of pushing the needle from the needle storage position to the locking position in the needle feeding stage, and the action of continuing to push the needle body so that its tip breaks through the skin and enters the tissue in the needle insertion stage. In the first working state, the pusher is only responsible for pushing the needle feeding and insertion, and does not participate in the subsequent twisting and lifting actions. In the second working state, after the needle insertion reaches the target depth, the pusher retracts towards the proximal end and disengages from the end of the needle handle. At this time, the clamping body engages with the end of the needle handle through the coupling surface, completing the needle insertion. In the second working state, the pusher remains stationary, and the clamping module independently completes the subsequent needle movement operations. A multi-axis collaborative robotic arm is used to drive the end effector to move in three-dimensional space to realize operations such as needle positioning, needle insertion, needle movement, and needle withdrawal. The control unit is a computer device with data processing and control functions, which includes a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the closed-loop control method of the embodiments of this application.

[0027] The acupuncture robot is also equipped with a multimodal signal acquisition device. Specifically, the system includes a layered integrated quantitative sensing needle, a multi-parameter acupoint sensor, and a brain-computer interface signal acquisition device. The layered integrated quantitative sensing needle is a special type of acupuncture needle, consisting of a handle and a body. The handle collects the practitioner's manipulation techniques, while the body primarily collects acupoint information. Multiple micro-sensors can be arranged layer by layer along the needle's axis to sense the mechanical response and tissue impedance changes experienced by the needle as it passes through different tissue layers. The multi-parameter acupoint sensor is placed on the local skin surface at the acupoint to collect local temperature, tissue pressure, and ion concentration signals. The brain-computer interface signal acquisition device is used to collect the patient's electroencephalogram (EEG) signals during acupuncture; this device can take the form of a scalp electrode cap or implanted electrodes.

[0028] In this embodiment, the synchronous acquisition of multimodal physical signals is achieved through a hardware-triggered synchronization mechanism. Specifically, all sensors and acquisition devices share the same clock source, and the control unit sends synchronization acquisition commands to each acquisition channel at a fixed sampling frequency to ensure that all signals are strictly aligned in time. The sampling frequency is preferably 1000 Hz to meet the high-fidelity acquisition requirements of transient physiological signals.

[0029] In the step of acquiring multimodal acupoint state parameters in real time, the control unit synchronously acquires multimodal physical signals reflecting the real-time interaction between the needle and the acupoint microenvironment. These multimodal physical signals include at least two of the following: acupuncture technique information, local acupoint temperature signal, tissue pressure signal, ion concentration signal, and EEG characteristic signals at the time of obtaining qi. The acupuncture technique information is acquired by force and position sensors mounted on the end effector, including parameters such as needle insertion speed, needle insertion force, twisting angle, twisting frequency, lifting and thrusting amplitude, and lifting and thrusting frequency. The local acupoint temperature signal is acquired by a temperature sensor, reflecting the temperature change of the local tissue at the acupoint during acupuncture. The tissue pressure signal is acquired by a pressure sensor, reflecting the mechanical pressure on the tissue surrounding the needle. The ion concentration signal is acquired by an ion-selective electrode, reflecting the concentration changes of key ions such as calcium, potassium, and sodium ions in the local tissue at the acupoint. These ion concentration changes are closely related to the physiological response of the local tissue microenvironment induced by acupuncture. The EEG characteristic signals are acquired by a brain-computer interface signal acquisition device, reflecting the brain's neural response to acupuncture stimulation, especially the changes in specific frequency band EEG activity related to the sensation of "deqi".

[0030] After acquiring the aforementioned multimodal physical signals, the control unit preprocesses each signal. Preprocessing includes filtering and noise reduction, baseline correction, and feature extraction. For temperature and pressure signals, a low-pass filter is used to remove high-frequency noise; for ion concentration signals, a moving average filter is used to smooth the signal; for EEG feature signals, a band-pass filter is used to extract EEG activity in specific frequency bands, and independent component analysis is used to remove artifacts such as electrooculography (EOG) and electromyography (EMG). After preprocessing, the control unit extracts feature vectors from each signal for gas delivery assessment.

[0031] In this embodiment, the step of multimodal spatiotemporal coupling weighting to generate the confidence score for "deqi" involves performing real-time spatiotemporal coupling analysis on the multimodal physical signal to generate a quantified confidence score for "deqi". The confidence score for "deqi" is a continuous value between 0 and 1, used to quantitatively characterize the probability that the current acupuncture state reaches the traditional Chinese medicine standard for "deqi". This step specifically includes the following sub-steps.

[0032] The first sub-step involves performing time-domain analysis on various multimodal physical signals to extract characteristic events related to the arrival of qi and their occurrence times. For local temperature signals at acupoints, the characteristic event is the abrupt change in the rate of temperature change, corresponding to local metabolic changes caused by the needle penetrating a specific tissue layer. For tissue pressure signals, the characteristic event is the step change in pressure value, corresponding to the mechanical release when the needle breaks through fascia and other tissue structures. For ion concentration signals, the characteristic event is the extreme point of the rate of change in ion concentration, corresponding to changes in local cell membrane permeability caused by acupuncture. For electroencephalogram (EEG) characteristic signals, the characteristic event is the moment of significant enhancement in the power spectral density of a specific frequency band, corresponding to the neural response of the brain to acupuncture stimulation. For acupuncture technique information, the characteristic event is the change in technique parameters, such as abrupt changes in needle insertion speed or changes in the direction of rotation.

[0033] The second sub-step involves calculating the time difference between the abrupt change times of at least two peripheral physical signals and the neural response times of the EEG characteristic signals. Peripheral physical signals include at least two of the following: local temperature signals at acupoints, tissue pressure signals, ion concentration signals, and acupuncture technique information. Specifically, the control unit extracts the occurrence times of characteristic events for each peripheral physical signal and calculates the time difference between these times and the neural response times of the EEG characteristic signals.

[0034] The third sub-step involves determining whether all the aforementioned time differences fall within a preset physiological delay window. The physiological delay window is a pre-defined time interval based on the physiological characteristics of neural conduction, for example, it can be set to 50 to 300 milliseconds. This delay window reflects the physiological conduction time required from the peripheral receptor receiving the acupuncture stimulus to the brain generating a corresponding neural response. Only when all calculated time differences fall within this preset physiological delay window are the peripheral physical signals and EEG signals considered to be temporally consistent, meaning the EEG signal is indeed triggered by the current acupuncture stimulus and not by other factors. In this case, the control unit uses the EEG characteristic signal to correct the confidence level of the desired effect. If the time difference between any abrupt change in the peripheral physical signal and the EEG response exceeds the preset physiological delay window, the causal relationship between the EEG signal and the current acupuncture stimulus is considered unclear, and the EEG signal does not participate in the correction of the confidence level of the desired effect.

[0035] The fourth sub-step involves calculating the comprehensive confidence level of Qi attainment based on the eigenvalues ​​of each modality signal and their spatiotemporal consistency test results. Specifically, the control unit first calculates a preliminary estimate of the probability of Qi attainment for each peripheral physical signal based on its eigenvalues. For temperature signals, a higher rate of temperature change and a longer duration of change result in a higher probability of Qi attainment. For pressure signals, a greater amplitude of pressure release and a smoother release process result in a higher probability of Qi attainment. For ion concentration signals, a higher probability of Qi attainment occurs when the amplitude and direction of ion concentration changes closely match the preset Qi attainment indicators. For acupuncture manipulation information, a higher probability of Qi attainment occurs when the frequency, amplitude, and intensity of the manipulation are closer to the preset Qi attainment manipulation parameters. Then, the control unit performs a weighted fusion of the preliminary estimates of the probability of Qi attainment for each modality, with the weighting coefficients preset based on the signal-to-noise ratio and reliability of each modality signal. Finally, if the EEG feature signal passes the spatiotemporal consistency test, it is used as a correction factor to adjust the confidence level of Qi attainment after weighted fusion. For example, when the EEG feature signal shows a significant Qi attainment-related neural response, the confidence level of Qi attainment is adjusted upwards.

[0036] In this embodiment, the step of generating physical drive commands based on air arrival confidence involves comparing the air arrival confidence with the desired physical interaction state to dynamically generate physical control commands for directly driving the motor of the end effector. Specifically, this step employs a speed-force hybrid control strategy based on multimodal state feedback.

[0037] The core of the speed-force hybrid control strategy lies in dynamically adjusting the ratio of needle insertion speed to needle insertion force based on the confidence level of the "deqi" (the arrival of Qi). Specifically, the control unit presets a first threshold and a second threshold, where the first threshold is lower than the second threshold. The first threshold can be set, for example, to 0.3, indicating the initial detection of "deqi" signs; the second threshold can be set, for example, to 0.7, indicating that the "deqi" state has been definitively achieved. When the confidence level of "deqi" is lower than the first threshold, it indicates that the needle has not yet entered the effective stimulation area. At this time, the control strategy prioritizes speed control and supplements it with force control. That is, the control unit prioritizes ensuring the needle insertion speed to quickly advance the needle to the vicinity of the target depth, while monitoring the needle insertion force to ensure it does not exceed the safety limit. When the confidence level of "deqi" reaches the first threshold but is lower than the second threshold, it indicates that the needle has entered the vicinity of the effective stimulation area. At this time, the control strategy prioritizes force control and supplements it with speed control. That is, the control unit prioritizes controlling the needle insertion force to finely adjust the interaction intensity between the needle and the tissue, while adjusting the needle insertion speed accordingly based on changes in force. When the confidence level of "deqi" reaches the second threshold, it indicates that "deqi" has been successfully achieved. The control unit then performs smooth deceleration or stops the needle insertion to avoid overstimulation.

[0038] The specific implementation of speed-force hybrid control is as follows: The control unit calculates the speed weighting coefficient and the force weighting coefficient based on the current air supply confidence level, and the sum of the two is 1. The speed weighting coefficient monotonically decreases as the air supply confidence level increases, while the force weighting coefficient monotonically increases as the air supply confidence level increases. The control unit calculates the target speed command based on the speed weighting coefficient and the preset maximum needle insertion speed, and simultaneously calculates the target force command based on the force weighting coefficient and the preset target needle insertion force. Then, the control unit converts the target speed command and the target force command into the current command or voltage command required to drive the motor, and outputs them to the motor of the end effector through the servo driver.

[0039] In this embodiment, the step of generating physical drive commands based on the air delivery confidence level also incorporates predictive inference commands from the lightweight digital twin subsystem as feedforward references, thereby constituting a feedforward-feedback composite control.

[0040] The lightweight digital twin subsystem is a streamlined physical perception and intention coupling model specifically designed for real-time simulation of the physical state of needle-tissue interaction. This model is built upon the dynamic equations of the end effector and a tissue mechanics model. Its inputs are current drive commands and sensor feedback signals, and its output is a prediction of the needle-tissue interaction state over a future period. The lightweight digital twin subsystem's real-time simulation cycle is less than 10 milliseconds, meaning it completes a full state prediction and command generation within 10 milliseconds, thus meeting the requirements of high-frequency physical control.

[0041] Unlike existing digital twin solutions, the lightweight digital twin subsystem of this application does not include a meridian and visceral state layer for macroscopic pathogenesis analysis. Specifically, the model only models and extrapolates the kinematics and dynamics of the needle in the tissue, as well as the force interaction between the needle and the tissue, without involving the dialectical analysis and extrapolation of the patient's overall meridian and visceral state. This design choice is intentional, aiming to concentrate computational resources on the physical layer control, which has the highest real-time requirements, and avoids computational delays caused by mixing complex macroscopic extrapolations with real-time physical layer control.

[0042] The lightweight digital twin subsystem is constructed as follows: First, a dynamic model of the end effector is established based on its mechanical structure and transmission parameters, including the inertia matrix, Coriolis force and centrifugal force matrices of each joint, and gravity terms. Second, a tissue mechanics model is established based on the mechanical properties of biological tissues, including the elasticity model of skin, the viscoelasticity model of subcutaneous tissue, and the rupture model of fascia. Then, the dynamic model of the actuator is coupled with the tissue mechanics model to establish an overall physical model of the needle-tissue interaction. Finally, the model is lightweighted, including model order reduction, parameter aggregation, and computational optimization, enabling the model to complete a full forward derivation in less than 10 milliseconds.

[0043] In feedforward-feedback composite control, the lightweight digital twin subsystem extrapolates and predicts the needle-tissue interaction state over several future control cycles based on current sensor feedback signals and drive commands, generating predictive feedforward drive commands accordingly. These feedforward commands are superimposed or weighted and fused with feedback control commands based on air delivery confidence to form the final physical control command output to the motor driver. Feedforward control can compensate for system inertia and delay in advance, improving control response speed and tracking accuracy; feedback control can correct model errors and external disturbances in real time, ensuring control stability and robustness. The combination of these two technologies gives the system both rapid response capability and good anti-interference performance.

[0044] In this embodiment, the TCM prior knowledge required for multimodal spatiotemporal coupling weight determination and physical driving instruction generation is provided by an externally callable knowledge base independent of the lightweight digital twin subsystem. This externally callable knowledge base is a structured TCM acupuncture knowledge system that communicates with the control unit through a standardized data interface.

[0045] The externally accessible knowledge base includes the following four knowledge modules. The first is the expert experience and techniques module, which stores clinical manipulation experience data from renowned veteran TCM doctors. This data includes personalized manipulation parameters of these doctors under different acupoints and syndrome types, such as rotation angle, rotation frequency, lifting and thrusting amplitude, lifting and thrusting frequency, and needle insertion force, as well as corresponding operational essentials and experiential characteristics for judging the arrival of Qi. This experience data can be obtained through the applicant's prior method for collecting acupuncture manipulation information based on multimodal brain-computer interface technology, or it can be extracted and organized from publicly available literature reports.

[0046] The second module is the Classic Techniques module, which stores the kinematic and dynamic characteristic parameters of standard acupuncture techniques. These standard techniques include, but are not limited to, twisting tonification and sedation techniques, lifting and thrusting tonification and sedation techniques, slow and fast tonification and sedation techniques, and breathing tonification and sedation techniques. For each standard technique, this module stores its typical kinematic parameters, such as twisting amplitude, twisting frequency, lifting and thrusting amplitude, lifting and thrusting frequency, and needle insertion speed, as well as corresponding dynamic parameters, such as needle insertion force and twisting torque. These parameters constitute the standardized technique motion space that the robot can execute, providing a basic technique template for the control unit.

[0047] The third module is the standardized operation method module, which stores operational guidelines such as acupoint location, needling depth, needling angle, needle retention time, and contraindications. These guidelines are established based on the national standard GB / T 21709 "Standardized Operation Methods for Acupuncture Techniques," industry guidelines, and clinical pathways. For each acupoint, this module stores its standard location method, recommended needling depth range, recommended needling angle, standard needle retention time, and related contraindications and precautions. These guidelines constitute the safety constraints for the robot during acupuncture operations.

[0048] The fourth module is the Ancient Literature Information Source module, which stores structured text data related to acupoint indications, tonification and sedation principles, signs of Qi attainment, and appropriate and inappropriate needling techniques from classic medical texts. These classic medical texts include *Ling Shu*, *Su Wen*, *Zhen Jiu Da Cheng*, and *Zhen Jiu Jia Yi Jing*. This module digitizes and structures the original texts, transforming discussions related to acupuncture operations into machine-readable and searchable data, providing classical theoretical basis for the control unit's decision-making.

[0049] The modules of the externally accessible knowledge base are independent of each other and can be updated and expanded independently. The data of each module is stored in a standardized format, and the control unit can query and call the modules through a unified application programming interface.

[0050] In this embodiment, before generating physical drive instructions, the control unit invokes the Qi-de-qi physiological marker library and various knowledge modules in an externally callable knowledge base for comprehensive reasoning. Specifically, the control unit first matches the real-time feature vectors of the acquired multimodal physical signals with the probability mapping models of each Qi-de-qi state in the knowledge base. The Qi-de-qi physiological marker library stores multimodal physiological signal feature templates corresponding to different types of Qi-de-qi states. Each template includes typical feature values ​​and their variation patterns for each modality, such as temperature, pressure, ion concentration, and electroencephalogram (EEG) signals. The control unit calculates the similarity between the real-time feature vectors and each template to obtain a preliminary probability distribution of the Qi-de-qi state.

[0051] At the same time, the control unit retrieves the experiential characteristics of renowned traditional Chinese medicine practitioners in judging the arrival of Qi under similar syndrome conditions from the expert experience and techniques module. Specifically, based on the current patient's individual information, such as constitution type and syndrome classification, the control unit retrieves matching experiential data from the expert experience and techniques module and extracts the experiential characteristics and techniques preferences related to the judgment of arrival of Qi.

[0052] Then, the control unit comprehensively compares and infers the probability distribution of the Qi-attaining state based on real-time signals with the Qi-attaining judgment characteristics based on expert experience. This inference process is not a simple threshold judgment, but a multi-dimensional comprehensive decision that integrates classical theoretical basis, operational standard constraints, expert experience reference, and real-time signal verification. For example, when real-time temperature and pressure signals indicate that the needle has reached the predicted depth, but the confidence level of Qi attainment has not yet reached the standard, the control unit will query the classical discussions on the signs of Qi attainment at that acupoint in the ancient literature information source module, refer to the safe operating range in the standard operating method module, and search for the handling methods of famous traditional Chinese medicine doctors in similar situations in the expert experience and manipulation module. After integrating this information, it will decide whether to continue needle insertion, adjust the manipulation technique, or stop the operation.

[0053] Through the aforementioned four-dimensional knowledge fusion reasoning mechanism, the control unit can make treatment decisions that conform to classical TCM theories, follow modern operating procedures, and draw on the experience of renowned doctors, thus enabling the acupuncture robot to possess genuine TCM knowledge reasoning capabilities.

[0054] In this embodiment, the feedforward-feedback composite control step further includes a personalized active learning mechanism. When the needle reaches the predicted depth but the confidence level for obtaining qi does not meet the standard, it indicates a deviation between the feedforward prediction based on the general model and the actual situation of the patient. At this time, the control unit will not simply report an error or stop operation, but will switch to a feedback-driven control mode, controlling the needle to autonomously perform micro-movement exploration within the safety constraints obtained from the standard operation method module.

[0055] Specifically, the control unit queries the standard operating procedure module for information such as the safe depth range, safe angle range, and contraindications of the current acupoint, using this information as the hard constraint boundary for micro-motion exploration. Then, within the safe constraint range, the control unit makes small adjustments to the position, angle, and manipulation parameters of the needle, such as performing small-amplitude lifting and thrusting probes in the depth direction and small-amplitude twisting probes in the horizontal direction. During the micro-motion exploration, the control unit continuously monitors changes in multimodal physical signals and calculates the confidence level of Qi attainment in real time. When a certain location or combination of manipulation parameters is found to significantly increase the confidence level of Qi attainment and reach a preset standard, the control unit records the characteristic values ​​of the multimodal state parameters at that moment, including the needle position, manipulation parameters, and characteristic values ​​of each modality of physiological signal.

[0056] The control unit associates and stores this set of parameters that successfully induced Deqi with the patient's individual information, constructing a personalized Deqi model for that patient. This personalized model records the patient's unique Deqi sensitive locations, sensitive manipulation parameters, and corresponding multimodal physiological response characteristics. In subsequent treatments, the control unit can directly call upon this personalized model to quickly guide the needle to the optimal stimulation location and employ the most suitable manipulation parameters for the patient, thereby significantly improving the accuracy and efficiency of treatment.

[0057] In this embodiment, the control unit is further configured to implement an iterative evolution mechanism for data closure. After completing a preset treatment task, such as after needle removal and retention, the control unit encapsulates the complete data of this operation into an incremental learning sample. This incremental learning sample contains a three-dimensional data sequence: twin prediction value, sensor measurement value, and knowledge base reference value. The twin prediction value is the feedforward prediction data generated by the lightweight digital twin subsystem at each moment throughout the treatment process. The sensor measurement value is the multimodal physical signal data actually collected by each sensor throughout the treatment process. The knowledge base reference value is the reference data that the control unit queries and retrieves from an externally accessible knowledge base during the decision-making process.

[0058] After encapsulation, the control unit analyzes the incremental learning sample and optimizes it on two levels. The first level is optimizing the externally callable knowledge base. The control unit compares the actual parameters of successfully inducing Deqi in this treatment with the existing Deqi physiological markers in the knowledge base, updating the probability mapping relationship in the Deqi physiological marker library so that the probability mapping model more accurately reflects the statistical regularities in real clinical data. At the same time, the successful cases of this treatment are added to the expert experience and techniques module as new experience data for subsequent treatment reference.

[0059] The second layer involves calibrating the lightweight digital twin subsystem. The control unit compares the actual tissue-sensor interaction data recorded during the treatment with the prediction data of the twin model under the same conditions, calculating the prediction error. Then, the control unit uses system identification or parameter estimation methods to fine-tune and calibrate the model parameters of the lightweight digital twin subsystem using this prediction error, so that the prediction accuracy of the twin model continuously improves with the increase of treatment sessions.

[0060] Through the aforementioned iterative evolution mechanism of data closure, the knowledge and skills of the acupuncture robot continuously grow with the increase in the number of treatments. The more patients treated, the more successful cases and data the system accumulates, the higher the accuracy of Qi detection, and the stronger the personalized adaptation capability of the manipulation parameters. This realizes the transformation from an automated execution tool to a cognitive intelligent agent with continuous evolution capabilities.

[0061] This application also provides a tactile-EEG fusion-based safety emergency stop mechanism. During acupuncture, the control unit simultaneously monitors tactile signals from a force sensor and EEG signals from a brain-computer interface. When the force sensor detects an abnormal resistance surge, such as when the needle encounters unforeseen hard tissue causing a sharp increase in insertion force, or when the brain-computer interface detects specific pain-related EEG patterns, such as the appearance of P300 event-related potentials or an abnormal increase in power at a specific frequency band, the control unit immediately triggers an emergency stop command. This emergency stop command causes all motors of the end effector to immediately stop operating and lock in their current position, while simultaneously issuing an audible and visual alarm signal to notify the operator. The emergency stop can only be manually deactivated once both the tactile and EEG signals have returned to normal. This dual-channel redundant safety emergency stop mechanism effectively ensures the safety of the acupuncture process.

[0062] The technical solution of the embodiments of this application will be further described below through a specific application example.

[0063] In this example, the acupuncture robot performed tonifying acupuncture on the Zusanli acupoint for a patient with spleen and stomach deficiency syndrome. Before the treatment began, the control unit first loaded the patient's basic information, including constitution type, syndrome classification, and past medical history.

[0064] During the needle insertion phase, the control unit performs initialization and knowledge loading operations. The control unit receives predictive needle insertion instructions from the lightweight digital twin subsystem based on the current needle position and tissue model. Simultaneously, the control unit initiates a multi-dimensional query to an externally accessible knowledge base. From the standardized operation method module, the control unit finds that the safe depth range for the Zusanli acupoint is 1.0 to 1.5 cun, the recommended insertion angle is perpendicular, and contraindications include caution in pregnant women. From the classic manipulation module, the control unit obtains the standard kinematic parameters for the lifting and thrusting tonification technique, including the operation sequence of superficial insertion followed by deeper insertion, the key points of heavy insertion and light lifting, the lifting and thrusting amplitude of 3 to 5 mm, and the frequency of 40-60 times / minute. From the ancient literature information source module, the control unit retrieves the original text about the Zusanli acupoint from the *Compendium of Acupuncture and Moxibustion*, which clearly states that this acupoint treats symptoms such as cold in the stomach, abdominal distension, borborygmus, and weakness of the internal organs, consistent with the current patient's spleen and stomach deficiency syndrome.

[0065] During needle insertion, the control unit continuously acquires force and impedance signals from the layered integrated quantitative sensing needle, temperature and pressure signals from the multi-parameter sensors at the acupoint, and electroencephalogram (EEG) signals from the brain-computer interface. The control unit performs real-time spatiotemporal coupling analysis on these multimodal signals. When the needle depth reaches approximately 1.2 inches, the temperature signal exhibits a significant abrupt change in its rate of change, and the pressure signal shows a step-like release. The time difference between the abrupt changes in these two peripheral signals and the moment of power enhancement in a specific frequency band of the EEG signal is approximately 120 milliseconds, falling within the preset physiological delay window. Based on these signal characteristics, the control unit calculates the current confidence level of "Deqi" (the arrival of Qi) to be 0.75, exceeding the preset second threshold of 0.7. Therefore, the control unit determines that "Deqi" has been successfully achieved, performs smooth deceleration, and stops needle insertion.

[0066] During the needling stage, the control unit enters the needling decision-making mode. When generating needling technique parameters, the control unit again calls upon the externally accessible knowledge base. It retrieves the standard kinematic parameter space for the lifting and thrusting tonification method from the classic techniques module, searches the expert experience techniques module for information that several renowned traditional Chinese medicine practitioners prefer a combination of small-amplitude, high-frequency twisting combined with gentle lifting and thrusting when needling the Zusanli acupoint in cases of spleen and stomach deficiency, and confirms the tonification principles for the Zusanli acupoint from the ancient literature information source module. The control unit integrates this multi-dimensional knowledge, within the safety constraints set by the standardized operation method module, using classic techniques as a baseline, adjusting parameters with reference to expert experience preferences, and drawing on ancient texts as theoretical basis, ultimately generating needling technique instructions suitable for the patient's current condition.

[0067] During acupuncture, if the preset feedforward depth and standard technique fail to induce Deqi (the confidence level for Deqi remains below the first threshold), the control unit will activate a personalized micro-motion exploration mode. Within the safety constraints provided by the standardized operating procedure module, the control unit slightly adjusts the depth and angle of the needle and attempts different combinations of manipulation parameters. When a specific depth and manipulation combination is found to significantly increase the confidence level for Deqi and reach the preset standard, the control unit records this successful set of parameters and constructs a personalized Deqi model for that patient. In subsequent treatments, the control unit can directly call upon this model to quickly and accurately achieve Deqi.

[0068] After treatment, the control unit enters the experience encapsulation phase. The system automatically encapsulates the three-dimensional data sequence of this treatment—the twin prediction value, the sensor measurement value, and the knowledge base reference value—into incremental learning samples. These samples are analyzed and used to update the probability mapping relationships in the Qi-de (arrival) physiological marker library, contributing new successful cases to the expert experience manipulation module. Furthermore, the predictive model parameters of the lightweight digital twin subsystem are fine-tuned using actual tissue-sensor interaction data. Through this iteration, the system's accuracy in Qi-dejudgment and its ability to adapt to different manipulation techniques have been further improved.

[0069] This application also provides a closed-loop control system for an acupuncture robot based on multimodal perception fusion and Qi attainment state assessment. The system includes a multimodal signal acquisition module, a spatiotemporal coupling weighting module, an instruction generation module, a drive module, a lightweight digital twin subsystem, and an externally callable knowledge base.

[0070] The multimodal signal acquisition module is used to acquire multimodal physical signals that reflect the real-time interaction between the needle body and the acupoint microenvironment. This module includes hardware devices such as force sensors, position sensors, temperature sensors, pressure sensors, ion-selective electrodes, and brain-computer interface signal acquisition devices, as well as corresponding signal conditioning and acquisition circuits.

[0071] The spatiotemporal coupling weighting module is used to perform real-time spatiotemporal coupling analysis on multimodal physical signals and generate quantified confidence scores. This module has been described in detail in the above method embodiments and will not be repeated here.

[0072] The instruction generation module compares the air arrival confidence level with the desired physical interaction state to dynamically generate physical control instructions. This module has been described in detail in the above method embodiments and will not be repeated here.

[0073] The drive module is used to directly drive the motor of the end effector according to physical control commands. This module includes a servo driver and a motor; the servo driver converts the low-level control signal output from the control unit into a high-power current signal required to drive the motor.

[0074] The lightweight digital twin subsystem is used to perform real-time simulations of the physical state of needle-tissue interaction, with a simulation cycle of less than 10 milliseconds. This subsystem does not include a meridian and viscera state layer for macroscopic pathogenesis analysis. An externally accessible knowledge base exists independently of the lightweight digital twin subsystem and includes at least one of the following modules: expert experience and techniques module, classic techniques module, standardized operating methods module, and ancient literature information source module. The specific content of each module has been described in detail in the above method embodiments.

[0075] This application also provides an embodied intelligent acupuncture robot, including a multi-axis collaborative robotic arm, an end effector mounted at the end of the robotic arm, and a control unit. When the control unit executes the computer program stored therein, it implements the aforementioned closed-loop control method for the acupuncture robot based on multimodal perception fusion.

[0076] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method.

[0077] In summary, this application's embodiments, by constructing a closed-loop motion control system completely independent of human physical domain, achieve a technological leap from technology-dependent humans to machines that understand Traditional Chinese Medicine (TCM), can think, learn, and iteratively improve. By acquiring multimodal physical signals in real time, including acupuncture technique information, acupoint temperature, tissue pressure, ion concentration, and EEG characteristic signals, and performing spatiotemporal coupling analysis, it is the first to incorporate multiple parameters of the acupoint microenvironment into the robot's closed-loop control, achieving precise physical interaction aimed at achieving "deqi" (the arrival of Qi). By employing a layered architecture of a lightweight digital twin and an externally accessible knowledge base, a simplified physical deduction model is decoupled from a four-dimensional TCM knowledge system integrating expert experience, classic techniques, operational norms, and ancient literature. This ensures a real-time deduction cycle of less than 10 milliseconds to meet high-frequency physical control requirements while endowing the robot with a solid foundation in TCM theory and clinical experience for knowledge reasoning capabilities. Through autonomous micro-motion exploration and personalized model construction, the robot can actively learn and accurately adapt to individual patient differences during treatment. Through data closure and incremental learning mechanisms, the system can continuously optimize the knowledge base and twin model parameters as the number of treatments increases, thereby achieving continuous skill evolution.

[0078] This invention has illustrated its principles and implementation methods using specific examples. The descriptions of these embodiments are merely illustrative of the method and its core ideas; furthermore, those skilled in the art will recognize that modifications may be made to the specific implementation methods and application scope based on the principles of this invention. Therefore, the content of this specification should not be construed as limiting the invention.

Claims

1. A closed-loop control method for an acupuncture robot based on multimodal perception fusion and Qi attainment state assessment, characterized in that, Includes the following steps: Real-time acquisition of multimodal acupoint status parameters: synchronously acquire multimodal physical signals reflecting the real-time interaction status of the needle body-acupoint microenvironment, the multimodal physical signals including acupuncture technique information and at least two of the following: acupoint local temperature signal, tissue pressure signal, ion concentration signal and EEG characteristic signal at the time of obtaining Qi; Multimodal spatiotemporal coupling weighting to generate gas yield confidence: Real-time spatiotemporal coupling analysis is performed on the multimodal physical signals to generate quantified gas yield confidence. Physical drive commands are generated based on the air delivery confidence: The air delivery confidence is compared with the desired physical interaction state to dynamically generate physical control commands for directly driving the motor of the end effector. Online iterative optimization until the physical interaction goal is achieved: continuously monitor and iteratively optimize the physical control commands in real time to form a motion control closed loop in the physical domain until the preset air intake state is confirmed.

2. The closed-loop control method for acupuncture robots based on multimodal perception fusion according to claim 1, characterized in that, The multi-mode spatiotemporal coupling weighting step further includes: The time difference between the abrupt change times of at least two peripheral physical signals and the neural response times of the EEG characteristic signals is calculated. The confidence level of the air intake is corrected only when the time difference falls within the preset physiological delay window.

3. The closed-loop control method for acupuncture robots based on multimodal perception fusion according to claim 1, characterized in that, The step of generating physical drive commands based on the confidence level of Qi arrival specifically involves a "speed-force" hybrid control based on multimodal state feedback, including: dynamically adjusting the ratio of needle insertion speed to needle insertion force according to the confidence level of Qi arrival; when the confidence level of Qi arrival is lower than a first threshold, speed control is the primary method and force control is the secondary method; when the confidence level of Qi arrival reaches the first threshold but is lower than a second threshold, force control is the primary method and speed control is the secondary method; when the confidence level of Qi arrival reaches the second threshold, smooth deceleration or needle insertion is executed.

4. The closed-loop control method for acupuncture robots based on multimodal perception fusion according to claim 1, characterized in that, The step of generating physical driving commands based on the confidence level of Qi is to integrate predictive inference commands from the lightweight digital twin subsystem as feedforward references to form a "feedforward-feedback" composite control. The lightweight digital twin subsystem is a simplified physical perception and intention coupling model, which is specifically used to perform real-time inference of the physical state of needle-tissue interaction, with a real-time inference cycle of less than 10ms. The model does not include the meridian and viscera state layer for macroscopic pathogenesis analysis.

5. The closed-loop control method for acupuncture robots based on multimodal perception fusion according to claim 4, characterized in that, The TCM prior knowledge required for the multimodal spatiotemporal coupling weight determination and the generation of the physical driving instructions is provided by an external callable knowledge base independent of the lightweight digital twin subsystem. The externally callable knowledge base includes at least one of the following modules: an expert experience and techniques module, used to store clinical techniques and experience data of renowned veteran Chinese medicine practitioners; The Classic Techniques module stores the kinematic and dynamic parameters of standard acupuncture techniques; the Standardized Operation Methods module stores the operation specifications for acupoint location, acupuncture depth, angle, needle retention time, and contraindications; and the Ancient Literature Information Source module stores structured text data related to acupoint indications, tonification and sedation principles, Qi attainment signs, and appropriate and inappropriate acupuncture techniques from classic medical texts.

6. The closed-loop control method for acupuncture robots based on multimodal perception fusion according to claim 5, characterized in that, Before generating the physical drive command, the control unit calls the Qi-de-physiological marker library and various knowledge modules in the external callable knowledge base, matches the real-time feature vector of the multimodal physical signal with the probability mapping model of each Qi-de-state in the knowledge base, and combines the expert experience features retrieved from the expert experience and techniques module to comprehensively assist in generating the physical drive command.

7. The closed-loop control method for acupuncture robots based on multimodal perception fusion according to claim 6, characterized in that, The "feedforward-feedback" composite control steps also include: When the needle reaches the predicted depth but the confidence level of obtaining Qi does not meet the standard, the needle is controlled in a feedback mode to autonomously perform micro-movement exploration within the safety constraints obtained from the standard operation method module, and the characteristic values ​​of the multimodal state parameters obtained when obtaining Qi are recorded in real time to construct a personalized obtaining Qi model for the patient.

8. The closed-loop control method for acupuncture robots based on multimodal perception fusion according to claim 1, characterized in that, The control unit is also configured to, after completing the preset treatment task, encapsulate the three-element data sequence of the twin prediction value, sensor measurement value and knowledge base reference value of this operation into an incremental learning sample, which is used to optimize the probability mapping relationship and the content of each knowledge module in the externally callable knowledge base, and to calibrate the model parameters of the lightweight digital twin subsystem.

9. A closed-loop control system for an acupuncture robot based on multimodal perception fusion and Qi attainment state assessment, characterized in that, include: The multimodal signal acquisition module is used to acquire multimodal physical signals that reflect the real-time interaction status of the needle body-acupoint microenvironment. The spatiotemporal coupling weighting module is used to perform real-time spatiotemporal coupling analysis on the multimodal physical signals and generate quantified air confidence scores. The instruction generation module is used to compare the air confidence level with the expected physical interaction state and dynamically generate physical control instructions. The drive module is used to directly drive the motor of the end effector according to the physical control command; The system also includes a lightweight digital twin subsystem and an externally callable knowledge base; the lightweight digital twin subsystem is used to perform real-time simulation of the physical state of needle-tissue interaction, with a real-time simulation cycle of less than 10ms; the externally callable knowledge base is independent of the lightweight digital twin subsystem and includes at least one of the following modules: expert experience and techniques module, classic techniques module, standardized operation method module, and ancient literature information source module.

10. A hygienically intelligent acupuncture robot, comprising a multi-axis collaborative robotic arm, characterized in that, The end effector of the multi-axis collaborative robotic arm is equipped with an end effector, which is arranged sequentially from the proximal end to the distal end along its own axis as follows: The push module includes a guide rail extending along the axial direction and a pusher slidably disposed on the guide rail, the pusher having a first working surface facing the end of the needle handle; A clamping module, located at the distal end of the pushing module, includes a clamping body and a first driving component for driving the clamping body to move. The clamping body has a coupling surface that matches a preset structure at the end of the needle handle. The end effector has a first working state and a second working state. In the first working state, the pusher slides along the guide rail to the distal end and applies an axial thrust to the end of the needle handle through the first action surface. In the second working state, the pusher retracts to the proximal end and disengages from the needle handle, and the clamping body engages with the end of the needle handle through the coupling surface. It also includes a control unit, which, when executing a computer program stored therein, implements the closed-loop control method for acupuncture robots based on multimodal perception fusion as described in any one of claims 1 to 8.

Images