Physiotherapy robot intelligent control system and method

Abstract

The invention discloses an intelligent control system and method for a physiotherapy robot, and belongs to the technical field of robot control. The intelligent control system comprises an initial planning module, a real-time sensing module and a correction updating module; the initial planning module is used for acquiring initial data of a target physiotherapy area, constructing a target body surface feature set, acquiring a preparation posture, performing initial planning in combination with a sampling probability gradient, generating an initial movement track and performing pre-starting verification; the real-time sensing module is used for binding an execution timestamp, constructing a sensing reference library, screening high-weight feature points, collecting a physiotherapy image, calculating a single-frame displacement die length, and performing displacement risk assessment and adjustment once the single-frame displacement die length exceeds a single-frame displacement early warning line; the correction updating module is used for performing adjustment verification, calculating a safety compliance rate, displacement correction precision and energy uniformity, and generating a target physiotherapy summary; precise adaptation of the physiotherapy track and the parameters in a dynamic scene is achieved, and physiotherapy area offset caused by micro-motion of a user is effectively avoided.

Description

Technical Field

[0001] This invention relates to an intelligent control system and method for a physiotherapy robot, belonging to the field of robot control technology. Background Technology

[0002] With the continued growth in healthcare needs, physiotherapy service robots are gradually becoming an important tool in rehabilitation treatment. These robots not only need to possess high precision and reliability, but also need to provide personalized services to meet the needs of different patients.

[0003] Chinese patent application CN119589705A discloses a control system and method for a physiotherapy service robot, comprising: a machine vision device, a robot, a central control system, an interactive device, and a physiotherapy instrument; the interactive device is connected to the machine vision device and sends a start signal to the machine vision device, causing the machine vision device to start acquiring human body images and selecting a physiotherapy area and planning a physiotherapy trajectory; the machine vision device is connected to the central control system to send physiotherapy area and physiotherapy trajectory data; the central control system is connected to the robot and the physiotherapy instrument to send the physiotherapy area and physiotherapy trajectory data to the robot and the physiotherapy instrument, enabling the robot and the physiotherapy instrument to perform physiotherapy services respectively, and receiving the running trajectory data and status information sent by the robot and the physiotherapy instrument, as well as sending them to the interactive device.

[0004] Although existing technologies can acquire patient data in real time in remote environments and adjust treatment parameters based on the data to provide more flexible and effective treatment services, they do not consider the real-time adaptation capability of motion trajectories in dynamic physiotherapy scenarios. In particular, when the user makes slight displacements during physiotherapy, such as unconsciously turning over or adjusting their sitting posture, the robot still performs actions according to the initial preset coordinates, which can easily lead to the physiotherapy area shifting. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent control system and method for physiotherapy robots. By scanning and denoising, and model registration to establish a unique feature set, combined with optimization algorithms and pre-start verification, a precise initial trajectory is generated to reduce initial errors. By binding sensor timing and path points, tracking feature points to capture user micro-movements in real time, and dynamically adapting according to risk level, combined with trimodal cross-verification, the robot is prevented from executing according to preset parameters, thus solving the problem of physiotherapy deviation caused by user micro-movements.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A physiotherapy robot intelligent control system includes: an initial planning module, a real-time sensing module, and a correction and update module;

[0008] The initial planning module is used to obtain initial data of the user's target physiotherapy area, calculate the physiotherapy diameter, configure initial physiotherapy parameters, construct a user-specific target body surface feature set, obtain the ready posture of the physiotherapy head, combine the sampling probability gradient, use the RRT* algorithm and multi-target ant colony algorithm for initial planning, generate the initial motion trajectory, drive the physiotherapy head to the ready posture, and perform pre-start verification based on coordinate deviation.

[0009] The real-time sensing module is used to bind the sampling time sequence with the execution timestamp of the path point, drive the physiotherapy head to move to the target coordinates to stabilize the contact force, build a sensing benchmark library, screen high-weight feature points, collect physiotherapy images, calculate the displacement modulus of a single frame, and immediately conduct a displacement risk assessment and make adjustments once the single frame displacement warning line is exceeded.

[0010] The correction and update module is used to perform adjustment verification, calculate safety compliance rate, displacement correction accuracy, energy uniformity, and generate a summary of target physiotherapy.

[0011] Specifically, the initial planning steps include:

[0012] The target physiotherapy area is scanned to generate raw three-dimensional point cloud data. The physiotherapy point cloud is obtained by combining a bilateral filtering algorithm, and the physiotherapy diameter of the target physiotherapy area is calculated using the maximum diameter distance method.

[0013] The standard model of the corresponding part of the target physiotherapy area is retrieved from the body surface feature library and registered with the physiotherapy point cloud to construct the target body surface feature set, and the initial contact reference value and initial body surface temperature are obtained simultaneously.

[0014] Obtain the user's physiotherapy type, call the default parameter set, set the correction coefficient, and adjust the default parameter set in conjunction with the physiotherapy diameter to obtain the initial physiotherapy parameters;

[0015] A three-layer constraint system is constructed, and trajectory smoothing index, parameter adaptation index, and safety redundancy index are defined to construct a multi-objective optimization function.

[0016] Specifically, the initial planning steps also include:

[0017] Based on the target body surface feature set, the corresponding weights of various feature points are retrieved, and the target coordinates of the target physiotherapy area are calculated by weighted calculation method. The normal vector of the physiotherapy point is then combined with the cloud computing normal vector.

[0018] Set position safety limits, calculate end-effector position parameters, and generate ready attitude;

[0019] A secondary distance segmentation threshold is set to divide the target physiotherapy area into a target neighborhood, a transition region, and a far-field region, and corresponding sampling probabilities are configured to segment the sampling resources.

[0020] Combining the aforementioned three-layer constraint system, the RRT* algorithm is used to generate several candidate planning paths;

[0021] Using the ant colony algorithm, in each iteration, the candidate path corresponding to each ant is selected by probability, and the fitness value is calculated using a multi-objective optimization function;

[0022] Once all ants have completed their path selection, update the pheromone concentration of all planned candidate paths and repeat the iterative calculation until the preset iteration coefficient is reached.

[0023] The candidate path with the highest pheromone concentration is used as the initial motion trajectory, and combined with the robot's inverse kinematics, an initial physiotherapy path is generated.

[0024] Specifically, the pre-startup verification steps include:

[0025] Based on the initial physiotherapy path, the physiotherapy head is driven to the ready posture, and a real-time image of the target physiotherapy area is acquired, and a real-time body surface feature set is extracted.

[0026] Real-time values ​​and target values ​​are obtained from the real-time body surface feature set and the target body surface feature set, respectively. The coordinate deviations between the real-time values ​​and target values ​​of various feature points are calculated, and deviation verification values ​​are selected. Set the verification threshold To determine if there is a positioning deviation;

[0027] like If the pre-startup verification is deemed successful; This triggers automatic fine-tuning, converting the constructed deviation vector into joint angle fine-tuning amounts, and then re-evaluating for compliance.

[0028] Specifically, the steps for displacement risk assessment include:

[0029] Extract the execution timestamp corresponding to each path point in the initial physiotherapy path and bind it with the sampling timing of the sensing sensor. For the execution operation of each path point, synchronously trigger the directional sampling of the sensing sensor.

[0030] The physiotherapy head is driven to move to the target coordinates, and a dynamic reference zero point is configured with the initial contact reference value. The image of the target physiotherapy area is acquired, the average coordinates of each type of feature point are calculated, and a perception reference library is constructed.

[0031] Based on the weighted association table, high-weight feature points and backup feature points are divided;

[0032] According to the initial physiotherapy path, activate the energy output device of the physiotherapy head;

[0033] The physiotherapy robot is driven to move along the initial physiotherapy path. A sensing synchronization signal is triggered every time the robot moves a preset distance, and multi-dimensional data acquisition is initiated.

[0034] Specifically, the steps of displacement risk assessment also include:

[0035] For the target physiotherapy area, a set of physiotherapy images is acquired, and the high-weight feature points are tracked first to extract their real-time coordinates.

[0036] If the high-weight feature point is not detected, the backup feature point will be automatically called for supplementary tracking;

[0037] The effective contact force is obtained, and the real-time displacement vector and displacement magnitude of each high-weight feature point in the physiotherapy image set are calculated in combination with the aforementioned perception benchmark library.

[0038] The displacement magnitude of a single frame is calculated by combining the displacement weights of each high-weight feature point.

[0039] A single-frame displacement warning line is set based on the product of a preset displacement limit coefficient and a position safety limit.

[0040] If it exists If the single-frame displacement magnitude of a frame image is greater than the single-frame displacement warning line, it is determined that there is continuous displacement. Displacement risk assessment is performed, and displacement risk, stage risk, and safety redundancy risk are calculated separately and weighted to obtain comprehensive displacement risk. A secondary risk threshold is set, and the risk level of the comprehensive displacement risk is divided into low, medium, and high risks, and the corresponding adjustment mechanism is triggered.

[0041] Otherwise, if there is no effective displacement, continue to collect physiotherapy image sets according to the initial physiotherapy path.

[0042] Specifically, the risk level adjustment mechanism includes:

[0043] If the risk is low, the parameter adaptation mechanism is triggered, the corresponding compensation mechanism is called, and the parameter fine-tuning amount is calculated in combination with real-time displacement data.

[0044] Verify the parameter fine-tuning and monitor the continuous performance after adjustment. The displacement of high-weight feature points in the frame is calculated, and the displacement decrease rate is calculated. If the displacement decrease rate is greater than the lower limit of parameter fine-tuning, the adjustment is determined to be effective; otherwise, the local correction mechanism is switched.

[0045] If the risk level is medium, a local correction mechanism is triggered to calculate the global displacement vector and decompose it into translational and rotational deviations. A pose deviation matrix is ​​constructed to correct local path points and generate a local correction trajectory.

[0046] The local correction trajectory is verified by calculating the displacement deviation between the coordinates of high-weight feature points and the path points in the local correction trajectory. If the maximum displacement deviation is less than the upper limit of trajectory adjustment, the trajectory fine-tuning is deemed effective. Otherwise, the local trajectory fine-tuning is repeated. If the trajectory fine-tuning is still ineffective after reaching the upper limit of adjustment times, the path incremental planning mechanism is immediately switched.

[0047] Specifically, the risk level adjustment mechanism also includes:

[0048] If it is high risk, the path incremental planning mechanism is triggered. Based on the real-time coordinates of high-weight feature points, the new target coordinates are calculated by weighting the mind, and the new normal vector is fitted by combining the physiotherapy point cloud to determine the new preparation posture. The bidirectional RRT algorithm is used to generate candidate incremental paths, the multi-objective optimization function is reused, the multi-objective optimization value of each candidate incremental path is calculated, and the path corresponding to the minimum value is selected as the incremental correction trajectory.

[0049] Before executing the incremental correction trajectory, a pre-start verification is triggered. After the verification is successful, the physiotherapy is restarted, the pause duration is automatically calculated, and the total physiotherapy duration is extended.

[0050] Specifically, the steps for adjusting and verifying include:

[0051] The driving end-effector moves along a path point sequence and performs trimodal cross-validation, including visual validation, mechanical validation, and temperature validation.

[0052] If a single modality check is abnormal, only the adjustment items associated with the abnormal modality are corrected; if two or more modalities are abnormal, the physical therapy is immediately suspended and the three-modal cross-check is performed again to confirm whether the abnormality is a transient interference. If it is a single modality check abnormality, it is determined to be a transient interference; otherwise, it is determined to be a persistent abnormality, triggering emergency braking.

[0053] Obtain the duration of compliance with security indicators, combine it with the total execution time for adjustments, calculate the security compliance rate, set the security compliance threshold, and classify the security compliance level;

[0054] Obtain the actual maximum displacement deviation after adjustment, calculate the displacement correction accuracy, and set the displacement compliance threshold to classify the displacement compliance level;

[0055] Acquire full-process temperature field data, calculate energy uniformity, set energy uniformity thresholds to classify energy uniformity levels, and thus generate a summary of target physiotherapy.

[0056] A method for intelligent control of a physiotherapy robot, comprising:

[0057] Step S1: Obtain initial data of the user's target physiotherapy area, calculate the physiotherapy diameter, configure initial physiotherapy parameters, construct a user-specific target body surface feature set, and obtain the ready posture of the physiotherapy head;

[0058] Step S2: Allocate sampling resources using the sampling probability gradient, and perform initial planning using the RRT* algorithm to generate the initial motion trajectory;

[0059] Step S3: Drive the therapy head to the prepared posture, perform pre-start verification based on coordinate deviation, and drive the therapy head to the target coordinates to stabilize the contact force, build a perception benchmark library, filter high-weight feature points, and acquire therapy images;

[0060] Step S4: Calculate the displacement modulus of a single frame. Once the displacement warning line of a single frame is exceeded, immediately conduct a displacement risk assessment and adjustment, verify the adjustment, and generate a summary of target physical therapy.

[0061] The beneficial effects of this invention are:

[0062] By constructing a user-specific feature set through scanning denoising and standard model registration, and combining multi-objective optimization with an improved RRT* algorithm to generate an initial trajectory, pre-launch verification further fine-tunes the initial positioning deviation, reducing initial positioning error from the source and avoiding offset caused by the superposition of initial deviation and user micro-movements. In the displacement risk assessment stage, sensor sampling time sequence and path points are bound together, directional collection of regional data and tracking of high-weight feature points are performed to capture in real time the user's unconscious micro-movements such as turning over and adjusting sitting posture. Single-frame displacement modulus calculation is used to promptly identify offset trends, avoiding the lag of responding only after offset accumulation. A graded adjustment mechanism is combined to address different risks. The matching strategy involves fine-tuning parameters for low-risk cases, locally correcting trajectories for medium-risk cases, and incrementally replanning for high-risk cases. This ensures that the trajectory and parameters dynamically adapt when micro-movements occur, rather than executing according to preset coordinates. This fundamentally solves the problem of insufficient trajectory adaptation. The adjustment and verification process uses three-modal verification to correct anomalies in real time. Combined with compliance rate, accuracy, and other indicators and user feedback, the system is optimized to ensure the accuracy of current physiotherapy and accumulate data for future adaptation capabilities. This achieves accurate adaptation of physiotherapy trajectories and parameters in dynamic scenarios, effectively avoiding physiotherapy area deviations caused by user micro-movements. At the same time, it ensures the continuity and safety of physiotherapy and improves the consistency of physiotherapy effects for users of different body types. Attached Figure Description

[0063] Figure 1 This is a structural diagram of an intelligent control system for a physiotherapy robot.

[0064] Figure 2 This is a flowchart of the preliminary planning in this invention;

[0065] Figure 3 This is a flowchart of the displacement risk assessment in this invention;

[0066] Figure 4 This is a flowchart of an intelligent control method for a physiotherapy robot. Detailed Implementation

[0067] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0068] Example 1

[0069] refer to Figures 1 to 3 As shown in the figure, this embodiment introduces an intelligent control system for a physiotherapy robot, including: an initial planning module, a real-time sensing module, a dynamic correction module, and a linkage adjustment module;

[0070] The initial planning module uses a series of devices configured on the physiotherapy robot, such as cameras, force sensors, and infrared thermal imagers, to collect initial data on the user's target physiotherapy area, including 3D point clouds, initial contact reference values, and physiotherapy type. It sets initial physiotherapy parameters based on the physiotherapy diameter of the target area, constructs a user-specific target surface feature set based on the physiotherapy type, and obtains the ready posture of the physiotherapy head. It allocates sampling resources using sampling probability gradients, and combines the collected and calculated data with the RRT* algorithm and multi-target ant colony algorithm for initial planning, generating the initial motion trajectory and driving the physiotherapy head to the ready posture. Pre-start verification is performed based on coordinate deviation to ensure accurate initial positioning. This effectively reduces the physiotherapy area offset caused by initial positioning deviations combined with the user's unconscious turning over or adjusting sitting posture, improving the accuracy and safety of the physiotherapy.

[0071] Specifically, the preliminary planning steps include:

[0072] When the user is in a preset physiotherapy posture, such as a preset knee joint physiotherapy posture, a scanning device, such as a 3D structured light camera, is used to scan the target physiotherapy area to cover the full range of action of the physiotherapy head at the end of the physiotherapy robot, generating original three-dimensional point cloud data. A bilateral filtering algorithm is used to remove environmental noise, such as false points caused by clothing wrinkles, to avoid positioning deviations caused by data noise in subsequent planning, resulting in a physiotherapy point cloud. All point pairs in the physiotherapy point cloud are traversed, and the Euclidean distance between each point pair is calculated. The maximum value of the Euclidean distance is selected as the physiotherapy diameter of the target physiotherapy area.

[0073] The standard model of the corresponding part of the target physiotherapy area is retrieved from the body surface feature library, such as the standard 3D model of the adult knee joint. It is converted into a reference point cloud with the same coordinate system as the physiotherapy point cloud to provide standard position anchor points for feature points. This avoids misjudgment of feature points due to individual differences when directly extracting features from the user's physiotherapy point cloud. The iterative nearest point algorithm is used to register the physiotherapy point cloud with the reference point cloud to extract the physiotherapy target feature points corresponding to the feature points marked in the standard model in the physiotherapy point cloud. These points serve as the positioning benchmark, thereby forming the user's target body surface feature set.

[0074] The robotic arm is driven to bring the physiotherapy head close to the target physiotherapy area at a preset start speed. Sensors are used to monitor whether the target physiotherapy area is in contact with the physiotherapy head. Once the sensor detects a non-zero contact force for the first time and it remains stable for a preset period of time, it indicates that the physiotherapy head has just made contact with the body surface without pressure. This is recorded as the initial contact reference value. At the same time, temperature monitoring equipment, such as an infrared thermal imager, is used to collect the initial body surface temperature of the target physiotherapy area.

[0075] The user's treatment type, such as knee phototherapy, is obtained through the user's preset treatment posture or the interactive screen of the physiotherapy robot. The default parameter set corresponding to the treatment type is retrieved from the preset physiotherapy parameter library, and a correction coefficient is set to represent the diameter of the standard physiotherapy area. Based on the ratio of the correction coefficient to the physiotherapy diameter, the adaptation ratio is calculated, and the default parameter set is adjusted. The initial physiotherapy parameters are obtained by multiplying the adaptation ratio with the parameter values ​​in the default parameter set. This allows the physiotherapy parameters to dynamically adapt to the size of the area, ensuring that the physiotherapy effect is consistent and safe for users of different body types.

[0076] Based on the DH parameter table of the physiotherapy robot, the length and torsion angle of each joint link are recorded, the pose constraints, joint angle constraints, and motion speed constraints of the robotic arm end are defined, and motion execution constraints are constructed. The parameter boundaries are defined with the initial physiotherapy parameters as the benchmark to construct physiotherapy parameter constraints. The initial contact benchmark value and the initial body surface temperature are used as benchmarks to define the contact force fluctuation range and the upper limit of the physiotherapy temperature to construct safety boundary constraints, and finally a three-layer constraint system is generated.

[0077] The sum of squares of the rate of change of the joint angles of the robotic arm is defined as the trajectory smoothness index; the deviation between the initial physiotherapy parameters and the constraint boundary is defined as the parameter adaptation index; and the ratio of the difference between the actual contact force and the upper limit of the contact force is defined as the safety redundancy index. Through weighted calculation, a multi-objective optimization function is constructed to avoid planning defects caused by a single objective, such as the shortest path but large jitter, or parameter adaptation but insufficient safety redundancy. The weights are obtained through training with a large amount of sample data.

[0078] Because different types of feature points exhibit varying stability—for example, feature points in skin folds are prone to shifting while those in bony prominences are more stable—the feature point types in this application are categorized based on their stability levels. Different types of feature points have different weights. For instance, skin folds have low stability. Based on the target body surface feature set, the corresponding weights for each type of feature point are retrieved from a pre-defined feature stability mapping table, and the center coordinates of the target therapeutic area are calculated using a weighted algorithm. Using the target coordinates as the basis, and combining the normal vector of the target physiotherapy area with cloud computing data from the physiotherapy points, a location safety boundary is also set. To characterize the distance limit from the end of the physiotherapy head to the center coordinate, the end-effector position parameters of the physiotherapy robot are calculated. Using the normal vector as the attitude parameter, the ready pose of the end effector of the physiotherapy robot is generated; where, , , The target therapeutic area is centered in three-dimensional space. axis, axis, The coordinate components of the axes;

[0079] Using the ready posture as the sampling center, a sampling probability gradient is set to divide the sampling resources, including: setting a secondary distance division threshold based on the physiotherapy diameter of the target physiotherapy area and the characteristics of the robot's workspace, such as the first-level threshold. 1.5 times the diameter of the physiotherapy device, second-level threshold. Using the sampling distance between any point in the target physiotherapy area and the sampling center as the basis, and taking 0.6 times the robot's working radius as the dividing line, the target physiotherapy area is divided into a target neighborhood, a transition region, and a far-field region. The target neighborhood is defined as a region with a sampling distance less than [missing information]. The transition region is the area where the sampling distance is within , The region between, and the far-field region is the sampling distance greater than In areas where frequent sampling is unnecessary, corresponding sampling probabilities are configured, and the sum of the sampling probabilities is 1. The sampling probability represents the probability that an area will be selected to generate sampling points, which determines the number density of sampling points in different areas and the dynamic allocation of the total number of samplings per unit time. The sampling resources of the physiotherapy robot are obtained, including sampling points and sampling frequency. Sampling resources are configured for each area based on the sampling probability, and the sampling resources are divided into different areas according to the sampling probability ratio. At the same time, when sampling in the target neighborhood, the distance from the sampling point to the sampling center is inversely proportional to the corresponding sampling probability. A distance decay function is constructed to adjust the distribution of sampling points, avoid uniform sampling in the target neighborhood, and form local sampling resources in each area.

[0080] Combining a three-layer constraint system, with the target location as the starting point of the physical therapy, the RRT* algorithm is used to generate... The proposed candidate paths include a sequence of path points at the end of the robotic arm and the corresponding physiotherapy parameter adaptation values ​​for each path point.

[0081] Based on the multi-objective ant colony algorithm, each ant corresponds to a planning candidate path. The initial pheromone concentration of all planning candidate paths is initialized. The multi-objective optimization function is used as the heuristic function for planning candidate paths. At the same time, ant colony parameters are configured, such as the number of ants being the number of planning candidate paths, the iteration coefficient being 50, the pheromone emission coefficient being 0.2, and the pheromone enhancement coefficient being 100. Using the probability formula of the ant colony algorithm, the probability of an ant choosing a certain planning candidate path is calculated in each iteration, and the corresponding path is selected based on the maximum probability.

[0082] After each ant selects a corresponding candidate path, the fitness value is recalculated based on the multi-objective optimization function. After all ants have completed path selection, the pheromone concentration of all candidate paths is updated according to the pheromone update formula.

[0083] Repeat the selection of the next planning candidate path and the update of pheromone concentration until the iteration coefficient is reached, and take the planning candidate path with the highest pheromone concentration as the initial motion trajectory.

[0084] By solving the robot's inverse kinematics, the initial motion trajectory is converted into a sequence of joint angles based on the end pose of each path point in the initial motion trajectory, ensuring that each joint angle is within the limit range. Combined with the adaptation values ​​of physiotherapy parameters, the initial physiotherapy path is generated, and the end pose planning is converted into joint execution commands so that the physiotherapy robot can accurately reproduce the initial motion trajectory.

[0085] Specifically, the pre-startup verification steps include:

[0086] Based on the initial physiotherapy path, the physiotherapy head at the end of the physiotherapy robot is controlled to move along the initial motion trajectory. When the physiotherapy head reaches the ready posture, a scanning device is used to acquire a real-time image of the target physiotherapy area, and a real-time body surface feature set is extracted. Real-time and target values ​​are obtained from the real-time and target body surface feature sets, respectively. The coordinate deviations between the real-time and target values ​​of various feature points are calculated, and the maximum coordinate deviation is selected as the deviation verification value. Set the verification threshold To determine if there is a positioning deviation;

[0087] like If the pre-startup verification is successful, the formal execution instruction is generated; if The system triggers automatic fine-tuning, constructs a deviation vector based on the coordinate deviations of various feature points, and converts the deviation vector into joint angle fine-tuning amounts using the inverse kinematics Jacobian matrix according to the robot's DH parameter table. The single fine-tuning amount is limited according to preset rules to prevent the robotic arm from shaking significantly and causing other feature point deviations to worsen. After adjustment, real-time feature points are re-acquired, the coordinate deviations of all feature points are calculated, and the deviation verification value is used to determine whether it is qualified. This process continues until the verification is qualified, with a maximum of 3 fine-tuning attempts. If it is still unqualified, the system prompts for equipment calibration.

[0088] The real-time perception module binds the sensor sampling time sequence with the path point execution timestamp, only collecting data in the corresponding physiotherapy area to avoid data invalidation due to misalignment between perception and execution positions. It drives the physiotherapy head to move to the target coordinates, establishes a perception benchmark library after stabilizing the contact force, and divides high-weight feature points and backup feature points into a weighted association table to adapt to posture changes and reduce feature point misjudgments caused by posture changes. At the same time, it pre-activates physiotherapy parameters to avoid parameter switching delays during formal execution. During formal execution, the physiotherapy head triggers a perception synchronization signal every time it moves a preset distance to collect physiotherapy images, prioritizing the tracking of high-weight feature points. If no high-weight feature point is detected, backup points are activated. The module calculates the single-frame displacement modulus, and if it exceeds the single-frame displacement warning line, it immediately performs a displacement risk assessment and makes adjustments to compensate for the initial planning's insufficient adaptation to dynamic scenes.

[0089] Specifically, the steps for displacement risk assessment include:

[0090] The joint angle sequence in the initial physiotherapy path is analyzed, and the execution timestamp corresponding to each path point is extracted. The sampling time sequence collected by the sensing sensor is bound to each execution timestamp. For the execution operation of each path point, the directional sampling of the sensing sensor is triggered synchronously. Only the physiotherapy area corresponding to the path point is sampled, rather than the full range sampling, to avoid invalid sensing data caused by misalignment between the sampling area and the execution position.

[0091] Drive the physiotherapy head to the target coordinates of the initial physiotherapy path and ensure that the contact force is stable at the initial contact reference value. Set the torque sensor reading to the dynamic reference zero point and call the target body surface feature set. At the current position, continuously acquire consecutive frames of target physiotherapy area images, calculate the average coordinates of each type of feature point, build a perception reference library, and avoid reference deviation caused by fluctuations in single frame data.

[0092] To adapt to dynamic posture changes during physiotherapy path execution, such as when the physiotherapy head tilts, where outer feature points are more stable than inner feature points, a weighted association table containing the correspondence between path points and feature points is established. This reduces feature point misjudgments caused by posture changes, ensures the perception benchmark library matches the current path posture, and improves the stability of feature point tracking. For each path point, feature points are sorted from highest to lowest weight, and the top-ranked features are grouped... One feature point is defined as a high-weight feature point, and the remaining feature points are defined as spare feature points to avoid misjudgment of feature points due to pose changes;

[0093] Based on the therapeutic parameter adaptation values ​​and duration of action in the initial therapeutic path, the energy output device of the therapeutic head is activated in advance to put it in a ready-to-output state. At this time, the energy is not released, but the parameters are loaded, avoiding parameter switching delays during formal execution.

[0094] After completing the pre-treatment preparations, the physiotherapy robot drives the treatment head to move according to the joint angle sequence of the initial physiotherapy path. Each time it moves a preset distance, it triggers a sensing synchronization signal, initiating multi-dimensional data acquisition to capture the user's subtle movements and potential risks in real time. For the target physiotherapy area, it continuously collects data. The set of physiotherapy images of frames uses the KLT optimized optical flow method to prioritize the tracking of high-weight feature points in the perception benchmark library. By calculating the optical flow vector of the feature points in consecutive frames, the real-time coordinates of the high-weight feature points are extracted. If a high-weight feature point is not detected due to occlusion, such as the user's clothing shifting, a backup feature point is automatically called for supplementary tracking to avoid visual perception interruption.

[0095] Based on a torque sensor, the contact force between the physiotherapy head and the body surface is synchronously collected. After removing mechanical vibration noise through a first-order low-pass filter, the effective contact force is obtained. Combined with the feature point coordinates in the perception benchmark library, the real-time displacement vector and displacement magnitude of each high-weight feature point in the physiotherapy image set are calculated. Combined with the weights corresponding to each high-weight feature point, normalization processing is performed to obtain the displacement weight of each high-weight feature point. Combined with the corresponding displacement magnitude, a weighted operation is performed to calculate the displacement magnitude of a single frame.

[0096] The displacement limit coefficient is determined by multiple sets of user micro-motion test data, and a single-frame displacement warning line is set based on the product of the displacement limit coefficient and the position safety limit.

[0097] If physiotherapy images are concentrated If the single-frame displacement magnitude of a frame image exceeds the single-frame displacement warning line, continuous displacement is determined, and a displacement risk assessment is performed to adjust the initial physiotherapy path. The displacement risk is calculated based on the ratio of the single-frame displacement magnitude to the positional safety limit, reflecting the severity of displacement deviation from the target area. Corresponding weights are applied according to the physiotherapy progress to define stage risks. The safety redundancy risk is obtained by calculating the ratio of the difference between the upper limit of contact force and the effective contact force to the upper limit of contact force, reflecting the safety margin of contact force. Each risk is weighted to obtain the comprehensive displacement risk. A secondary risk threshold is set to classify the comprehensive displacement risk into low, medium, and high risk levels, triggering corresponding adjustment mechanisms. The weights are determined through training with multiple sets of samples, prioritizing displacement accuracy.

[0098] Otherwise, if there is no effective displacement, continue acquiring physiotherapy image sets according to the initial physiotherapy path; among which, .

[0099] Specifically, the risk level adjustment mechanism is as follows:

[0100] If the risk is low, the parameter adaptation mechanism is triggered. The trajectory is not adjusted; instead, minor displacement effects are offset by fine-tuning the physiotherapy parameters. This adapts to the boundary range constrained by the physiotherapy parameters. Combined with a preset displacement compensation mapping table, the corresponding compensation mechanism is invoked. The parameter fine-tuning amount is calculated based on real-time displacement data, synchronously sent to the physiotherapy head, and the parameter fine-tuning is verified and monitored for continuity after adjustment. The displacement of high-weight feature points in the frame, based on The displacement magnitude of a single frame of the image is used to calculate the average displacement magnitude after adjustment. Simultaneously, combined with the average displacement magnitude before adjustment, the displacement decrease rate after adjustment is calculated. If the displacement decrease rate exceeds the lower limit of parameter fine-tuning, the parameter adaptation is deemed effective, and execution continues according to the current parameters and initial trajectory. Otherwise, if the parameter fine-tuning cannot offset the displacement effect, the local correction mechanism is immediately switched to prevent a continuous decline in the therapeutic effect. Specifically, through fitting multi-scenario experimental data, typical therapeutic areas and types are selected. Different minute displacement amplitudes are simulated under a controllable environment. Combining the weight levels of high-weight feature points, the amount of therapeutic parameter adjustment required to offset the displacement effect under each combination of displacement amplitude and feature point weights is recorded, thereby generating a displacement compensation mapping table.

[0101] If the risk level is medium, a local correction mechanism is triggered to fine-tune the local trajectory. Combining displacement weights, the real-time displacement vector is normalized to eliminate the influence of weight differences between high-weight feature points, resulting in a global displacement vector. This global displacement vector is decomposed into translational and rotational deviations, constructing a pose deviation matrix to characterize the impact of user micro-movements on the therapeutic head pose. Based on the DH parameter table, the inverse kinematics Jacobi matrix corresponding to the current therapeutic head pose is calculated to reflect the mapping relationship between joint angle changes and end-effector pose changes. The joint angle fine-tuning is solved using the pseudo-inverse of the Jacobi matrix, correcting only local path points in the initial therapeutic path. The original pose of each path point is multiplied by the deviation matrix to obtain a new pose, generating a locally corrected trajectory. The locally corrected trajectory is then validated, controlling the therapeutic head to move to the first corrected path point, collecting the coordinates of high-weight feature points, and calculating... The displacement deviation of path points in the local correction trajectory is calculated. If the maximum displacement deviation is less than the upper limit of trajectory adjustment, the trajectory fine-tuning is deemed effective, the physiotherapy is resumed, and the local correction trajectory is executed. Otherwise, the local trajectory fine-tuning is repeated. If the trajectory fine-tuning is still ineffective after reaching the upper limit of adjustment times, the path incremental planning mechanism is immediately switched. The acquisition of local path points includes: extracting the next continuous segment of the currently executed path point from the initial physiotherapy path as the initial screening segment; calculating the ratio of the sum of the products of the real-time displacement vector, the displacement change rate, and the single segment execution time to the distance between path points to obtain the number of path points; extracting continuous points with coverage and safety redundancy from the initial screening segment; pre-calculating the theoretical joint angle fine-tuning amount of each path point using the DH parameter table and the inverse kinematics Jacobian matrix; and eliminating points whose fine-tuning amount exceeds the three-layer constraint system to generate local path points.

[0102] If the risk is high, the path incremental planning mechanism is triggered, the physiotherapy is immediately paused, and path replanning is initiated. Using the real-time coordinates of high-weight feature points as a reference, the new target coordinates are calculated using a weighted method. Combined with the physiotherapy point cloud, a new normal vector is fitted to determine the new preparation posture. A bidirectional RRT algorithm is used to generate candidate paths. Starting from the current physiotherapy head position and ending at the new target pose, the path tree is expanded bidirectionally to filter out illegal sampling points. Finally, multiple candidate incremental paths that conform to the three-layer constraint system are generated. The multi-objective optimization function is reused to calculate the multi-objective optimization value of each candidate incremental path. The path corresponding to the minimum value is selected as the incremental correction trajectory to replace the remaining unexecuted segments in the initial physiotherapy path. Before executing the incremental correction trajectory, a pre-start verification is triggered. After the verification is passed, the physiotherapy is restarted, the pause duration is automatically calculated, and the total physiotherapy duration is extended to ensure that the user receives complete physiotherapy services. At the same time, the replanning process data is recorded.

[0103] The correction and update module is used to perform adjustment and verification after trajectory and parameter adjustments. It drives the physiotherapy head to move along the path and loads the adapted parameters. After the first new path point is successfully located, energy output is started, effectively avoiding misalignment between energy output and trajectory execution. Simultaneously, monitoring equipment is activated to perform cross-verification of visual, mechanical, and temperature modes. For single-modal anomalies, only related items are corrected. For multi-modal anomalies, transient interference is investigated first before processing, significantly reducing the probability of falsely triggering emergency braking. The module also calculates and classifies safety compliance rate, displacement correction accuracy, and energy uniformity to reflect the safety stability of the adjustment process, trajectory correction accuracy, and parameter adaptation effectiveness. User feedback questionnaires are pushed through the interactive screen. The verification results and physiotherapy data are archived by timestamp to generate a target physiotherapy summary, laying the foundation for ensuring physiotherapy effects and subsequent optimization.

[0104] Specifically, the steps for adjusting and verifying include:

[0105] After the trajectory and parameters are adjusted, the end-effector physiotherapy head moves along the path point sequence. The dynamically adapted physiotherapy parameters are loaded into the energy output module of the physiotherapy head. Once the physiotherapy head reaches the first new path point and the positioning error meets the requirements, the physiotherapy energy output is started to ensure that the parameters take effect as soon as the trajectory is in place, thus avoiding misalignment between energy output and trajectory execution.

[0106] The hardware devices used for safety monitoring are activated simultaneously to perform trimodal cross-verification, including visual verification, mechanical verification, and temperature verification. Visual verification calculates and judges the real-time deviation of high-weight feature points when a new path point is executed. If the real-time deviation of multiple consecutive frames is less than the verification deviation threshold, the visual verification is considered normal. Mechanical verification collects contact force data and calculates the fluctuation value of multiple consecutive sampling points. If the fluctuation is greater than the verification mechanical threshold, it is determined that the trajectory is unstable in contact with the body surface, triggering a fine adjustment of the path point posture. Temperature verification acquires and judges the temperature field data. If the temperature difference of the body surface at different locations within the same physiotherapy area exceeds the verification temperature threshold, it is determined that the energy distribution is uneven due to the physiotherapy parameters, and the physiotherapy parameters are adjusted in reverse to correct the temperature gradient.

[0107] If a single modality verification fails, only the adjustment items associated with the failed modality are corrected; if two or more modalities fail, physiotherapy is immediately suspended, and a three-modal cross-verification is performed again to confirm whether the failure is a transient interference, in order to avoid triggering emergency braking. If a single modality verification fails, it is determined to be a transient interference; otherwise, it is determined to be a persistent failure, triggering emergency braking. The adjustment items are adjustable data or parameters directly associated with the failed modality.

[0108] The time taken for all three modal cross-validations to pass during the entire adjustment process is recorded as the safety indicator compliance time. The ratio of this time to the total adjustment execution time is recorded as the safety compliance rate. A safety compliance threshold is set to classify safety compliance levels, including high compliance rate, safety qualified, and safety unqualified, to reflect the safety stability of the adjustment process. The total adjustment execution time is the total time from triggering the adjustment mechanism to fully executing all adjustment paths.

[0109] Based on visual verification, the maximum actual displacement deviation after adjustment is calculated by the three-dimensional Euclidean distance between the real-time coordinates of the adjusted high-weight feature points and the adjusted path points. The ratio of this deviation to the upper limit of trajectory adjustment is recorded as the displacement correction accuracy. A displacement compliance threshold is set to classify the displacement compliance level, including high accuracy, qualified displacement, and unqualified displacement, to reflect the accuracy of trajectory correction.

[0110] Acquire full-process temperature field data, calculate the standard deviation of the temperature field at each time point, and calculate the average temperature difference based on all standard deviations, which is recorded as energy uniformity. Set an energy uniformity threshold to classify energy uniformity levels, including high uniformity, qualified energy, and unqualified energy, to reflect the effectiveness of parameter adaptation.

[0111] The user feedback questionnaire is pushed to the interactive screen of the physiotherapy robot. It includes comfort rating and subjective feelings input. The verification results and data related to the user's physiotherapy process are organized and archived according to timestamps to generate a summary of the target physiotherapy.

[0112] Example 2

[0113] Please see Figure 4 Another embodiment of the present invention provides: an intelligent control method for a physiotherapy robot, comprising the following steps:

[0114] Step S1: Using a series of devices configured on the physiotherapy robot, obtain the initial data of the user's target physiotherapy area, calculate the physiotherapy diameter, configure the initial physiotherapy parameters, construct a user-specific target body surface feature set in combination with the physiotherapy type, and obtain the ready posture of the physiotherapy head;

[0115] Step S2: Allocate sampling resources using the sampling probability gradient, and perform initial planning using the RRT* algorithm to generate the initial motion trajectory;

[0116] Step S3: Drive the therapy head to the ready position, perform pre-start verification based on coordinate deviation, and bind the sensor sampling time sequence with the path point execution timestamp after the verification is passed. Only collect data in the corresponding therapy area to avoid data invalidation due to misalignment between the sensing and execution positions. Drive the therapy head to the target coordinates, establish a sensing benchmark library after stabilizing the contact force, divide high-weight feature points and spare feature points based on the weight association table, and collect therapy images.

[0117] Step S4: Calculate the displacement modulus of a single frame. Once the displacement warning line of a single frame is exceeded, an immediate displacement risk assessment and adjustment are performed, and a three-modal cross-validation is conducted. For single-modal anomalies, only the associated items are corrected. For multi-modal anomalies, transient interference is investigated first before processing, which significantly reduces the probability of falsely triggering emergency braking. The safety compliance rate, displacement correction accuracy, and energy uniformity are calculated and classified to reflect the safety stability, trajectory correction accuracy, and parameter adaptation effectiveness of the adjustment process. User feedback questionnaires are pushed through the interactive screen, and the verification results and physiotherapy data are archived by timestamp to generate a target physiotherapy summary.

[0118] Working principle and effects:

[0119] When the user is in a preset therapy posture, the 3D point cloud of the target therapy area is acquired. After bilateral filtering and noise reduction, the therapy diameter is calculated. A standard model is retrieved for registration to construct a user-specific target body surface feature set. Combining the therapy type and a multi-objective optimization function, an improved RRT* algorithm is used to generate the initial motion trajectory, reducing initial positioning errors and avoiding therapy deviations caused by subsequent superposition with the user's micro-movements. The therapy head is then driven to the ready posture. Pre-start verification and fine-tuning of coordinate deviations are performed to ensure accurate initial positioning and further reduce the risk of initial deviation. During the actual therapy, the sensor sampling time sequence and path point execution time are bound together for directional data acquisition. For corresponding regional data, high-weight feature points are prioritized for tracking. Contact force is simultaneously obtained using torque sensors, and the displacement modulus of a single frame is calculated. If the warning line is exceeded, the overall risk is calculated to capture subtle movements such as unconscious turning over or adjusting sitting posture in real time, avoiding the lag in response of traditional solutions. Subsequently, adjustments are made according to risk levels to ensure dynamic adaptation of trajectory and parameters, effectively preventing deviation of the physiotherapy area. Anomalies are corrected through three-modal verification, and indicators such as compliance rate are calculated. Combined with user feedback and archived optimization, this ensures both the accuracy of current physiotherapy and the ability to dynamically adapt to subsequent treatments. At the same time, it ensures the continuity and safety of physiotherapy, achieving consistent treatment effects for users of different body types.

[0120] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent control system for a physiotherapy robot, characterized in that, include: Initial planning module, real-time sensing module, and correction and update module; The initial planning module is used to obtain initial data of the user's target physiotherapy area, calculate the physiotherapy diameter, configure initial physiotherapy parameters, construct a user-specific target body surface feature set, obtain the ready posture of the physiotherapy head, combine the sampling probability gradient, use the RRT* algorithm and multi-target ant colony algorithm for initial planning, generate the initial motion trajectory, drive the physiotherapy head to the ready posture, and perform pre-start verification based on coordinate deviation. The real-time sensing module is used to bind the sampling time sequence with the execution timestamp of the path point, drive the physiotherapy head to move to the target coordinates to stabilize the contact force, build a sensing benchmark library, screen high-weight feature points, collect physiotherapy images, calculate the displacement modulus of a single frame, and immediately conduct a displacement risk assessment and make adjustments once the single frame displacement warning line is exceeded. The correction and update module is used to perform adjustment verification, calculate safety compliance rate, displacement correction accuracy, energy uniformity, and generate a summary of target physiotherapy.

2. The intelligent control system for a physiotherapy robot according to claim 1, characterized in that, The initial planning steps include: The target physiotherapy area is scanned to generate raw three-dimensional point cloud data. The physiotherapy point cloud is obtained by combining a bilateral filtering algorithm, and the physiotherapy diameter of the target physiotherapy area is calculated using the maximum diameter distance method. The standard model of the corresponding part of the target physiotherapy area is retrieved from the body surface feature library and registered with the physiotherapy point cloud to construct the target body surface feature set, and the initial contact reference value and initial body surface temperature are obtained simultaneously. Obtain the user's physiotherapy type, call the default parameter set, set the correction coefficient, and adjust the default parameter set in conjunction with the physiotherapy diameter to obtain the initial physiotherapy parameters; A three-layer constraint system is constructed, and trajectory smoothing index, parameter adaptation index, and safety redundancy index are defined to construct a multi-objective optimization function.

3. The intelligent control system for a physiotherapy robot according to claim 2, characterized in that, The initial planning steps also include: Based on the target body surface feature set, the corresponding weights of various feature points are retrieved, and the target coordinates of the target physiotherapy area are calculated by weighted calculation method. The normal vector of the physiotherapy point is then combined with the cloud computing normal vector. Set position safety limits, calculate end-effector position parameters, and generate ready attitude; A secondary distance segmentation threshold is set to divide the target physiotherapy area into a target neighborhood, a transition region, and a far-field region, and corresponding sampling probabilities are configured to segment the sampling resources. Combining the aforementioned three-layer constraint system, the RRT* algorithm is used to generate several candidate planning paths; Using the ant colony algorithm, in each iteration, the candidate path corresponding to each ant is selected by probability, and the fitness value is calculated using a multi-objective optimization function; Once all ants have completed their path selection, update the pheromone concentration of all planned candidate paths and repeat the iterative calculation until the preset iteration coefficient is reached. The candidate path with the highest pheromone concentration is used as the initial motion trajectory, and combined with the robot's inverse kinematics, an initial physiotherapy path is generated.

4. The intelligent control system for a physiotherapy robot according to claim 3, characterized in that, The pre-boot verification steps include: Based on the initial physiotherapy path, the physiotherapy head is driven to the ready posture, and a real-time image of the target physiotherapy area is acquired, and a real-time body surface feature set is extracted. Real-time values ​​and target values ​​are obtained from the real-time body surface feature set and the target body surface feature set, respectively. The coordinate deviations between the real-time values ​​and target values ​​of various feature points are calculated, and deviation verification values ​​are selected. Set the verification threshold To determine if there is a positioning deviation; like If the pre-startup verification is deemed successful; This triggers automatic fine-tuning, converting the constructed deviation vector into joint angle fine-tuning amounts, and then re-evaluating for compliance.

5. The intelligent control system for a physiotherapy robot according to claim 4, characterized in that, The steps involved in displacement risk assessment include: Extract the execution timestamp corresponding to each path point in the initial physiotherapy path and bind it with the sampling timing of the sensing sensor. For the execution operation of each path point, synchronously trigger the directional sampling of the sensing sensor. The physiotherapy head is driven to move to the target coordinates, and a dynamic reference zero point is configured with the initial contact reference value. The image of the target physiotherapy area is acquired, the average coordinates of each type of feature point are calculated, and a perception reference library is constructed. Based on the weighted association table, high-weight feature points and backup feature points are divided; According to the initial physiotherapy path, activate the energy output device of the physiotherapy head; The physiotherapy robot is driven to move along the initial physiotherapy path. A sensing synchronization signal is triggered every time the robot moves a preset distance, and multi-dimensional data acquisition is initiated.

6. The intelligent control system for a physiotherapy robot according to claim 5, characterized in that, The steps of displacement risk assessment also include: For the target physiotherapy area, a set of physiotherapy images is acquired, and the high-weight feature points are tracked first to extract their real-time coordinates. If the high-weight feature point is not detected, the backup feature point will be automatically called for supplementary tracking; The effective contact force is obtained, and the real-time displacement vector and displacement magnitude of each high-weight feature point in the physiotherapy image set are calculated in combination with the aforementioned perception benchmark library. The displacement magnitude of a single frame is calculated by combining the displacement weights of each high-weight feature point. A single-frame displacement warning line is set based on the product of a preset displacement limit coefficient and a position safety limit. If it exists If the single-frame displacement magnitude of a frame image is greater than the single-frame displacement warning line, it is determined that there is continuous displacement. Displacement risk assessment is performed, and displacement risk, stage risk, and safety redundancy risk are calculated separately and weighted to obtain comprehensive displacement risk. A secondary risk threshold is set, and the risk level of the comprehensive displacement risk is divided into low, medium, and high risks, and the corresponding adjustment mechanism is triggered. Otherwise, if there is no effective displacement, continue to collect physiotherapy image sets according to the initial physiotherapy path.

7. The intelligent control system for a physiotherapy robot according to claim 6, characterized in that, The risk level adjustment mechanism includes: If the risk is low, the parameter adaptation mechanism is triggered, the corresponding compensation mechanism is called, and the parameter fine-tuning amount is calculated in combination with real-time displacement data. Verify the parameter fine-tuning and monitor the continuous performance after adjustment. The displacement of high-weight feature points in the frame is calculated, and the displacement decrease rate is calculated. If the displacement decrease rate is greater than the lower limit of parameter fine-tuning, the adjustment is determined to be effective; otherwise, the local correction mechanism is switched. If the risk level is medium, a local correction mechanism is triggered to calculate the global displacement vector and decompose it into translational and rotational deviations. A pose deviation matrix is ​​constructed to correct local path points and generate a local correction trajectory. The local correction trajectory is verified by calculating the displacement deviation between the coordinates of high-weight feature points and the path points in the local correction trajectory. If the maximum displacement deviation is less than the upper limit of trajectory adjustment, the trajectory fine-tuning is deemed effective. Otherwise, the local trajectory fine-tuning is repeated. If the trajectory fine-tuning is still ineffective after reaching the upper limit of adjustment times, the path incremental planning mechanism is immediately switched.

8. The intelligent control system for a physiotherapy robot according to claim 7, characterized in that, The risk level adjustment mechanism also includes: If it is high risk, the path incremental planning mechanism is triggered. Based on the real-time coordinates of high-weight feature points, the new target coordinates are calculated by weighting the mind, and the new normal vector is fitted by combining the physiotherapy point cloud to determine the new preparation posture. The bidirectional RRT algorithm is used to generate candidate incremental paths, the multi-objective optimization function is reused, the multi-objective optimization value of each candidate incremental path is calculated, and the path corresponding to the minimum value is selected as the incremental correction trajectory. Before executing the incremental correction trajectory, a pre-start verification is triggered. After the verification is successful, the physiotherapy is restarted, the pause duration is automatically calculated, and the total physiotherapy duration is extended.

9. The intelligent control system for a physiotherapy robot according to claim 8, characterized in that, The adjustment and verification steps include: The driving end-effector moves along a path point sequence and performs trimodal cross-validation, including visual validation, mechanical validation, and temperature validation. If a single modality check is abnormal, only the adjustment items associated with the abnormal modality are corrected; if two or more modalities are abnormal, the physical therapy is immediately suspended and the three-modal cross-check is performed again to confirm whether the abnormality is a transient interference. If it is a single modality check abnormality, it is determined to be a transient interference; otherwise, it is determined to be a persistent abnormality, triggering emergency braking. Obtain the duration of compliance with security indicators, combine it with the total execution time for adjustments, calculate the security compliance rate, set the security compliance threshold, and classify the security compliance level; Obtain the actual maximum displacement deviation after adjustment, calculate the displacement correction accuracy, and set the displacement compliance threshold to classify the displacement compliance level; Acquire full-process temperature field data, calculate energy uniformity, set energy uniformity thresholds to classify energy uniformity levels, and thus generate a summary of target physiotherapy.

10. A method for intelligent control of a physiotherapy robot, which is based on an intelligent control system for a physiotherapy robot as described in any one of claims 1-9, characterized in that, include: Step S1: Obtain initial data of the user's target physiotherapy area, calculate the physiotherapy diameter, configure initial physiotherapy parameters, construct a user-specific target body surface feature set, and obtain the ready posture of the physiotherapy head; Step S2: Allocate sampling resources using the sampling probability gradient, and perform initial planning using the RRT* algorithm to generate the initial motion trajectory; Step S3: Drive the therapy head to the prepared posture, perform pre-start verification based on coordinate deviation, and drive the therapy head to the target coordinates to stabilize the contact force, build a perception benchmark library, filter high-weight feature points, and acquire therapy images; Step S4: Calculate the displacement modulus of a single frame. Once the displacement warning line of a single frame is exceeded, immediately conduct a displacement risk assessment and adjustment, verify the adjustment, and generate a summary of target physical therapy.

Images