An adaptive rehabilitation system and method based on magnetic micro-robot and reinforcement learning framework

Abstract

This invention belongs to, but is not limited to, the field of adaptive rehabilitation technology, and discloses an adaptive rehabilitation system and method based on a magnetic microrobot and a reinforcement learning framework. The system includes: a physical interaction end: serving as the physical carrier for the patient to directly perform rehabilitation tasks; a visual perception end: employing a high-resolution real-time imaging system, vertically fixed above the operation panel via an external bracket; the lens optical axis is vertically downward, and the field of view completely covers the entire operation panel, ensuring real-time acquisition of video stream data including the patient's affected limb and the microrobot without blind spots; a central control end: employing a data processing device equipped with a high-performance processor; its data input interface is connected to the visual perception end for receiving the video stream in real time; its communication output interface, such as a serial port or local area network, is connected to the driver of the magnetic drive unit for issuing control commands. This invention employs a magnetically coupled non-contact driving method, physically decoupling the microrobot from the drive device, supporting flexible expansion of diverse planar drive devices, and eliminating the risk of mechanical strain on the patient.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of adaptive rehabilitation technology, and particularly relates to an adaptive rehabilitation system and method based on a magnetic microrobot and a reinforcement learning framework. Background Technology

[0002] With the increasing aging of my country's population, upper limb motor dysfunction caused by neurological diseases, such as stroke, is becoming increasingly common. Hand-eye coordination is fundamental for daily activities such as eating, dressing, writing, and retrieving objects. Impairment of this ability means that patients will directly lose their ability to live independently, causing not only a huge psychological blow but also a significant increase in the long-term care burden on families. Therefore, how to provide effective, accessible, and sustainable rehabilitation services for the large number of patients with upper limb dysfunction has become a major social and livelihood issue that urgently needs to be addressed in my country's elderly care and rehabilitation services.

[0003] Currently, hand-eye coordination rehabilitation mainly relies on one-on-one physical and occupational therapy by therapists, promoting neural remodeling through repetitive tasks. However, against the backdrop of rapidly growing rehabilitation needs, its limitations are becoming increasingly prominent: the training content is monotonous and tedious, leading to low patient compliance; professional rehabilitation resources are severely insufficient and unevenly distributed, making it difficult to guarantee sufficient training intensity and frequency; and there is a lack of personalized dynamic adjustment capabilities based on objective data. Another type of rehabilitation robot, large exoskeleton robots, alleviates the pressure on human resources to some extent, but their large size, high cost, and rigidity make them unsuitable for fine hand-eye training, hindering their widespread application in community and home settings.

[0004] Against this backdrop, miniature magnetically controlled soft robots, integrating new materials, magnetic drive control, and intelligent additive manufacturing technologies, have brought new solutions to the rehabilitation field. Their small size and easy deployment allow them to easily enter everyday environments such as home desktops or community clinics, significantly lowering the barrier to entry for rehabilitation and making high-frequency home training a reality. Leveraging the omnidirectional, multi-degree-of-freedom motion capabilities provided by magnetic field actuation, microrobots can also flexibly construct rich interactive tasks, simulating everyday life scenarios and stimulating interest through gamification, providing a playable and adaptable physical medium for hand-eye coordination training. However, despite the enormous potential of magnetic microrobots at the hardware level, such systems currently face a key bottleneck in clinical translation: the level of intelligent control lags behind hardware development. Existing magnetic control systems mostly rely on open-loop control or real-time manual remote control, and the movement trajectory of microrobots is often preset and fixed. When faced with the complex pathological characteristics of stroke or Parkinson's patients, traditional control strategies cannot respond in a timely manner. For example, if a patient is unable to grasp a target due to tremors, continuing to move at the original speed will only increase the patient's frustration; conversely, if the system does not increase the difficulty as the patient's abilities improve, the training efficiency will be low. This lack of a closed-loop perception-decision-feedback mechanism hinders the effectiveness of rehabilitation.

[0005] Based on the above analysis, the urgent technical problems that need to be solved in the existing technology are:

[0006] (1) Traditional manual hand-eye coordination rehabilitation model has the defects of lack of professional rehabilitation resources, boring training content and lack of personalized dynamic adjustment ability, making it difficult to guarantee rehabilitation effect and service accessibility;

[0007] (2) Existing large exoskeleton rehabilitation robots are large in size, high in cost and have strong structural rigidity, which are not suitable for fine hand-eye coordination training and cannot be widely used in home and community scenarios;

[0008] (3) The emerging magnetically controlled Microsoft rehabilitation robot has sufficient hardware potential, but its intelligent control level is seriously lagging behind. It lacks perception-decision-feedback closed-loop control capabilities and cannot adapt to the complex pathology and real-time rehabilitation status of patients, which seriously restricts its rehabilitation efficacy and clinical translation. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention provides an adaptive rehabilitation system and method based on magnetic microrobots and reinforcement learning frameworks.

[0010] This invention is implemented as follows: an adaptive rehabilitation system based on magnetic microrobots and reinforcement learning framework, consisting of three major modules: a physical interaction end, a visual perception end, and a central control end.

[0011] The visual sensing end includes a high-resolution real-time imaging system. The real-time imaging system is vertically fixed above the operation panel by an external bracket. The lens optical axis is vertically downward and the field of view completely covers the entire operation panel. It collects video stream data including the patient's limb and the micro-robot in real time.

[0012] The central control unit adopts a data processing device equipped with a high-performance processor. The data input interface is connected to the visual perception terminal to receive video stream data in real time, and the communication output interface is connected to the magnetic drive unit driver to issue control commands.

[0013] The physical interaction terminal serves as the direct carrier for patient rehabilitation work. Through magnetic coupling, the driving device and the interaction unit are physically decoupled. Together with the visual perception terminal and the central control terminal, it completes closed-loop patient motor function assessment and rehabilitation training.

[0014] Furthermore, the physical interaction terminal includes an operation panel, a magnetic microrobot, and a magnetic drive unit;

[0015] The control panel is placed horizontally and is made of a non-magnetic material with a certain structural support strength, which combines the function of supporting the affected limb and the function of magnetic field penetration.

[0016] Furthermore, the magnetic microrobot is placed above the control panel as a dynamic physical interaction target for patient visual tracking and grasping.

[0017] Furthermore, the magnetic drive unit is concealed below the operation panel and includes a magnetic field generating unit and a two-dimensional motion actuator;

[0018] The magnetic poles of the driving magnet attract the magnetic poles inside the microrobot, forming a non-contact magnetic coupling across the control panel. The two-dimensional motion actuator receives commands from the central control unit, driving the driving magnet to move in a two-dimensional plane, thus remotely pulling the microrobot to move synchronously.

[0019] Furthermore, the two-dimensional motion actuator adopts an XY-axis linear module.

[0020] This invention also provides an adaptive rehabilitation assessment method based on a magnetic microrobot and a reinforcement learning framework, including standardized motor function limit testing interaction steps and clinical assessment index calculation steps;

[0021] In the standardized motor function limit test interactive steps, the system controls the microrobot to move at a preset high speed. The microrobot actively avoids the patient's hand approaching, and the patient performs chasing and grasping actions to test the patient's maximum speed limit and range of motion.

[0022] In the calculation step of the clinical assessment indicators, the visual perception end converts the acquired image into physical coordinates, and the central control end extracts multi-dimensional kinematic assessment indicators based on the hand and microrobot trajectory data.

[0023] Furthermore, the multi-dimensional kinematic evaluation index includes one or more combinations of the following indices: root mean square tracking distance, average motion velocity, range of motion (ROM) area, trajectory temporal smoothness (Jerk), and trajectory spatial curvature;

[0024] The root mean square tracking distance is used to characterize the average spatial distance between the affected limb and the microrobot; the average motion speed is used to characterize the patient's upper limb muscle strength and bradykinesia; the range of motion (ROM) area is used to quantify the effective extension boundary of the affected limb; the trajectory time smoothness (Jerk) is used to measure the continuity of the affected limb's movements; and the trajectory space curvature is used to identify deviations in the affected limb's motion control.

[0025] The present invention also provides an adaptive rehabilitation training method based on a magnetic microrobot and a reinforcement learning framework, which performs training parameter initialization steps, real-time interactive control steps, and dynamic difficulty adjustment steps.

[0026] In the training parameter initialization step, the system determines the initial training parameters based on clinical evaluation indicators through parameter mapping logic.

[0027] In the real-time interactive control step, the interactive control of the microrobot is achieved through a reinforcement learning dynamic game mechanism;

[0028] In the dynamic difficulty adjustment step, the difficulty of rehabilitation training tasks is adjusted based on the patient interaction effect within a fixed period.

[0029] Furthermore, in the real-time interactive control step, the reinforcement learning strategy drives the microrobot to perform dynamic evasion actions while maintaining an effective induction zone around the affected limb, ensuring that the microrobot is always in a critical position accessible to the patient.

[0030] Furthermore, in the dynamic difficulty adjustment step, the system sets a fixed time window and counts the success rate of patient capture tasks within the window;

[0031] When the task success rate is higher than the preset upper limit, the speed limit of the micro-robot is increased and the effective radius of the grasping judgment is reduced; when the task success rate is lower than the preset lower limit, the speed limit of the micro-robot is reduced.

[0032] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0033] Addressing the characteristics of existing rehabilitation systems, this invention provides an adaptive rehabilitation system based on magnetic microrobots and a reinforcement learning framework. It constructs an intelligent magnetic micro-soft robot system for hand-eye coordination rehabilitation. By integrating intelligent additive manufacturing of magnetic composite materials, computer vision, and deep reinforcement learning, the system empowers the microrobot with autonomous perception and decision-making, thereby achieving a leap in the intelligence of rehabilitation training and providing patients with more precise, efficient, and personalized rehabilitation solutions. Specific advantages are as follows:

[0034] (1) Intrinsic safety and low cost: The magnetic coupling non-contact drive method is adopted to physically decouple the microrobot from the drive device, support flexible expansion of diverse planar drive devices, and eliminate the risk of mechanical strain on patients.

[0035] (2) Multidimensional quantitative assessment: Based on computer vision, digital biomarkers such as the patient's grasping speed, smoothness, trajectory curvature, and tracking efficiency are extracted in real time to realize a closed loop from training to assessment.

[0036] (3) Personalized rehabilitation training: Based on reinforcement learning algorithms, the difficulty and mode of rehabilitation training are adaptively adjusted according to the patient's ability to maximize the rehabilitation effect.

[0037] As further supporting evidence of the inventiveness of this invention, the following important aspects are also reflected:

[0038] (1) The expected benefits and commercial value of the technical solution of this invention after transformation are as follows:

[0039] This system achieves multi-dimensional rehabilitation assessment and personalized adaptive training at a relatively low hardware cost (compared to large exoskeleton robots). It is suitable for large-scale deployment in hospital rehabilitation departments and offers a digital medical device product with significant commercial potential for the broader community and home rehabilitation market.

[0040] (2) The technical solution of this invention fills a technical gap in the industry both domestically and internationally:

[0041] This invention overcomes the limitations of traditional rehabilitation assessments that rely on subjective judgment, achieving objective quantitative assessments without human intervention throughout the entire process. It fills the technological gap in high-precision, digital, and automated assessment of upper limb motor function in stroke / Parkinson's patients. Through a high-speed evasive pursuit task using a microrobot, it objectively measures the ultimate speed and true range of motion, providing reproducible benchmark data for rehabilitation assessments and filling the technological gap in quantitative testing of upper limb rehabilitation limits.

[0042] (3) The technical solution of the present invention solves a technical problem that people have long wanted to solve but have never been able to solve successfully:

[0043] This invention achieves autonomous intelligent decision-making to replace fixed trajectories and manual remote control, and proactive induced closed-loop control. It constructs a complete intelligent control closed loop of perception-decision-execution-feedback, enabling fully automatic dynamic adaptive adjustment of training difficulty. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the overall architecture of the adaptive rehabilitation system based on magnetic microrobots and reinforcement learning framework provided in an embodiment of the present invention;

[0045] Figure 2 This is the overall architecture of the adaptive rehabilitation system based on virtual-real migration provided in the embodiments of the present invention;

[0046] Figure 3 These are real-life photos of the physical system setup provided in this embodiment of the invention;

[0047] Figure 4 This is a patient ability assessment diagram provided in an embodiment of the present invention: (a) for a healthy subject and (b) for a simulated hand-eye coordination disorder subject;

[0048] Figure 5 This is a schematic diagram illustrating the mapping relationship between the simulation training environment and real visual perception provided in this embodiment of the invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0050] The adaptive rehabilitation system based on magnetic microrobots and reinforcement learning framework provided in this invention forms a closed-loop control system through visual perception, intelligent decision-making, and magnetic drive execution. At the hardware level, the system mainly consists of three modules: a physical interaction terminal, a visual perception terminal, and a central control terminal, including:

[0051] The physical interaction terminal serves as the physical medium through which patients directly perform rehabilitation tasks;

[0052] At the visual perception end, a high-resolution real-time imaging system is used, which is vertically fixed above the operation panel by an external bracket; the lens optical axis is vertically downward, and the field of view completely covers the entire operation panel, ensuring real-time acquisition of video stream data including the patient's affected limb and the micro-robot without blind spots;

[0053] The central control unit employs a data processing device equipped with a high-performance processor; its data input interface is connected to the visual sensing unit for receiving video streams in real time; its communication output interface, such as a serial port or local area network, is connected to the driver of the magnetic drive unit for issuing control commands.

[0054] In this embodiment of the invention, the physical interaction terminal includes:

[0055] Operating panel: horizontally placed, made of non-magnetic material with a certain structural support strength; this panel serves as a support surface for the patient's hand to slide, providing both physical support and allowing magnetic fields to pass through without damage;

[0056] Magnetic microrobot: Placed above the control panel. It internally encapsulates a powerful permanent magnet, such as neodymium iron boron, and externally wraps itself with a biocompatible flexible elastomer to protect the patient from impacts with hard objects. It serves as a dynamic, interactive target for the patient's visual tracking and grasping.

[0057] Magnetic drive unit: Concealed below the control panel; contains a drive magnet and a two-dimensional motion actuator, such as an XY axis linear module or a robotic arm; the magnetic poles of the drive magnet attract the magnetic poles inside the microrobot, forming a non-contact magnetic coupling across the control panel; the motion actuator drives the drive magnet to move in a two-dimensional plane according to the signal sent by the control terminal, thereby remotely pulling the microrobot above to move synchronously.

[0058] In this embodiment of the invention, an adaptive rehabilitation method based on magnetic microrobots and a reinforcement learning framework includes the following steps:

[0059] S1: Standardized interactive test of motor function limits;

[0060] S2: Calculation of clinical assessment indicators;

[0061] S3: Initialize training parameters;

[0062] S4: Real-time interactive control;

[0063] S5: Dynamic difficulty adjustment.

[0064] In this embodiment of the invention, S1 includes:

[0065] The system controls the microrobot to move at a preset high speed within the operating plane and is programmed to actively avoid the patient's hand approaching. The patient needs to chase and grab the microrobot as quickly and accurately as possible. Through this high-intensity chase task, the patient can exert their maximum current motor ability, thereby objectively measuring the patient's upper limit of movement speed and actual range of motion under extreme conditions.

[0066] In this embodiment of the invention, S2 includes:

[0067] During the adversarial interaction, the visual perception end captures the image in real time and converts it into physical coordinates. The control end extracts multi-dimensional kinematic indicators based on the acquired hand and micro-robot trajectory data, specifically including:

[0068] Root mean square tracking distance: the average spatial distance between the affected limb and the microrobot during the pursuit process, used to assess the patient's spatial perception and tracking accuracy;

[0069] Mean speed of movement: The average displacement rate of a patient under full-force pursuit, used to assess the patient's upper limb muscle strength and degree of bradykinesia;

[0070] Range of motion (ROM) area: The area of ​​a two-dimensional convex hull constructed based on the trajectory of the affected limb, used to quantify the effective extension boundary and movement restriction of the patient's upper limb;

[0071] Trajectory temporal smoothness Jerk: Calculates the acceleration characteristics of the affected limb movement, measures the abruptness and continuity of the movement, and is used to assess the degree of spasticity or motor fragmentation after stroke.

[0072] Trajectory space curvature: measures the degree of bending and oscillation of the movement path of the affected limb, and is used to identify whether the patient has control deviations such as tremor or ataxia.

[0073] In this embodiment of the invention, S3 includes:

[0074] Based on the above evaluation indicators, the system automatically calculates the initial training parameters suitable for the patient's current state through the built-in parameter mapping logic, such as the robot's basic moving speed limit and the effective radius of grasping judgment, thereby completing the system hot start for personalized rehabilitation.

[0075] In this embodiment of the invention, S4 includes:

[0076] Through a continuous dynamic game mechanism based on reinforcement learning, the system can continuously stimulate the patient's pursuit intention and motor potential, ensuring the training quality and concentration during each interaction.

[0077] Dynamic avoidance: Reinforcement learning strategies drive the microrobot to actively avoid approaching and grasping the affected limb, preventing the target from being easily captured, thereby ensuring that rehabilitation training has a basic level of challenge.

[0078] Maintaining an effective induction range: While avoiding being grabbed, the reinforcement learning strategy is constrained by the set induction reward to prevent the microrobot from escaping to an ineffective area that is too far away from the affected limb; the system controls the microrobot to always move within the effective induction distance around the affected limb, that is, to stay at a critical position that the patient needs to make a moderate effort to reach.

[0079] In this embodiment of the invention, S5 includes:

[0080] Based on real-time trajectory planning, the control unit is also responsible for dynamically adjusting the task difficulty based on the patient's stage performance;

[0081] The system sets a fixed time window or number of attempts, and statistically analyzes the patient's interaction effects within that period, such as the success rate of the capture task;

[0082] With a built-in difficulty adjustment algorithm, if the current success rate continues to be higher than the preset rehabilitation upper limit, it means that the task has lost its challenge for the patient, and the system will appropriately increase the robot's speed limit and reduce the judgment radius; if the success rate is lower than the preset lower limit, it means that the patient is experiencing frustration, and the system will reduce the robot's speed limit. Through this feedback adjustment mechanism, the system can dynamically adjust the error of the initialization parameters and the patient's fatigue effect to ensure the long-term benefits of rehabilitation training.

[0083] In an embodiment of the present invention, the computer device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the adaptive rehabilitation method based on magnetic microrobots and reinforcement learning framework.

[0084] In this embodiment of the invention, a computer program is stored, which, when executed by a processor, causes the processor to perform the steps of the adaptive rehabilitation method based on magnetic microrobots and reinforcement learning framework.

[0085] like Figure 1 As shown, this embodiment of the invention provides an adaptive rehabilitation system based on a magnetic microrobot and a reinforcement learning framework. Figure 1 The diagram details the closed-loop control data flow of the system at the physical level, comprising three core modules: the visual perception unit at the top (responsible for capturing the position of the affected limb and the microrobot), the physical interaction unit in the middle (including the operation panel and the flexible magnetic microrobot), and the magnetic drive unit hidden below. It also illustrates how the central control unit uses reinforcement learning algorithms to convert visual input into underlying drive commands, forming a complete human-machine interaction closed loop of perception-decision-execution-feedback.

[0086] The system mainly consists of three modules at the hardware level: a physical interaction terminal, a visual perception terminal, and a central control terminal, including:

[0087] Physical interaction terminal: serving as the physical medium through which patients directly perform rehabilitation tasks;

[0088] Visual perception end: A high-resolution real-time imaging system is used, which is vertically fixed above the operation panel by an external bracket; the lens optical axis is vertically downward, and the field of view completely covers the entire operation panel, ensuring real-time acquisition of video stream data including the patient's affected limb and the micro-robot without blind spots;

[0089] Central control unit: It adopts a data processing device equipped with a high-performance processor; its data input interface is connected to the vision sensing unit for receiving video streams in real time; its communication output interface, such as serial port or local area network, is connected to the driver of the magnetic drive unit for issuing control commands.

[0090] Furthermore, the physical interaction terminal includes:

[0091] Operating panel: horizontally placed, preferably made of non-magnetic material with a certain structural support strength; this panel serves as a support surface for the patient's hand to slide, providing both physical support and allowing magnetic fields to pass through without damage;

[0092] Magnetic microrobot: Placed above the control panel. It internally encapsulates a powerful permanent magnet, such as neodymium iron boron, and externally wraps itself with a biocompatible flexible elastomer to protect the patient from impacts with hard objects. It serves as a dynamic, interactive target for the patient's visual tracking and grasping.

[0093] Magnetic drive unit: Concealed below the control panel; contains a drive magnet and a two-dimensional motion actuator, such as an XY axis linear module or a robotic arm; the magnetic poles of the drive magnet attract the magnetic poles inside the microrobot, forming a non-contact magnetic coupling across the control panel; the motion actuator drives the drive magnet to move in a two-dimensional plane according to the signal sent by the control terminal, thereby remotely pulling the microrobot above to move synchronously.

[0094] This system relies on magnetic non-contact driving technology and reinforcement learning intelligent decision-making framework to realize real-time perception, intelligent control and adaptive adaptation of patients' limb rehabilitation training. The whole process revolves around the closed-loop collaboration of three core modules: physical interaction, visual perception and central control. It takes into account the safety, interactivity and personalization of rehabilitation training. The core working principle can be divided into four coherent links: real-time perception and acquisition, intelligent decision analysis, magnetic coupling drive execution and adaptive iterative optimization.

[0095] After the system is started, the visual perception end is the first to enter the working state. The high-resolution real-time imaging system uses a vertically fixed bracket to fully cover the operation panel with a vertically downward optical axis, collecting high-definition video stream data including the patient's affected limb, the position and movement trajectory of the magnetic microrobot in real time without blind spots. This data is then synchronously transmitted to the central control end, completing the full-dimensional visual perception of the rehabilitation scene. This provides raw data support for subsequent decision-making and avoids control deviations caused by perception blind spots.

[0096] As the core of the system, the central control unit receives video streams transmitted from the visual perception unit and uses built-in image processing algorithms to accurately analyze and extract data on the patient's limb movement posture, positional deviation, and real-time coordinates of the microrobot. On the other hand, relying on the built-in reinforcement learning framework, it combines multi-dimensional data such as preset rehabilitation goals, real-time limb responses of the patient, and training completion rate to quickly complete intelligent decision analysis and generate microrobot movement trajectories, speeds, and stop point commands adapted to the patient's current rehabilitation status, abandoning the fixed training mode and realizing dynamic adjustment of commands.

[0097] The physical interaction terminal is responsible for executing rehabilitation interactions. The control panel is made of a non-magnetic material with a certain structural support strength. It is placed horizontally to provide a smooth support surface for the patient's affected limb, while ensuring that the magnetic field does not penetrate without damage. The magnetic drive unit is concealed under the panel. The drive magnet and the neodymium iron boron permanent magnet inside the microrobot form a non-contact magnetic coupling across the panel. After the central control terminal sends commands through the serial port or local area network, the two-dimensional motion actuator accurately drives the drive magnet to move. The microrobot is pulled in the air to complete the planar motion in sync. As a dynamic interactive target, it guides the patient's affected limb to complete rehabilitation actions such as tracking and grasping. The microrobot wrapped in a flexible elastic body can effectively avoid collisions with hard objects and ensure training safety.

[0098] The reinforcement learning framework continuously records feedback data such as the patient's limb reaction speed, movement accuracy, and completion efficiency during the training process. By constructing a closed-loop control mechanism based on state-action-feedback, it automatically adjusts the difficulty, speed, and trajectory complexity of the microrobot's movements, gradually adapting to the patient's rehabilitation progress. This achieves an adaptive transition from passive training to active rehabilitation, eliminating the need for frequent manual parameter adjustments and efficiently improving the targeting and effectiveness of rehabilitation training.

[0099] Figure 2 The simulation training phase demonstrates the process of simulating the training of a reinforcement learning model in a virtual environment; the real physical deployment phase demonstrates the process of transferring the trained model to a real physical environment and performing interactive control.

[0100] Figure 3 The device is marked with a main-view visual acquisition device at the top, an interactive panel for the patient's arm to rest and slide on, a magnetic microrobot above the panel that serves as an interactive target, and a six-degree-of-freedom collaborative robotic arm and permanent magnet actuator that are completely physically isolated and hidden below the panel.

[0101] Figure 4 The study compared the differences in kinematic scores among different subjects across multiple quantitative indicators.

[0102] Figure 5The coordinate system transformation and geometric dimensionality reduction modeling logic of this invention are explained. It demonstrates how to accurately map high-dimensional pixel-level image data captured by a real camera into an environment state vector that a reinforcement learning agent can understand, through camera calibration and matrix transformation.

[0103] Another objective of this invention is to provide a method based on the aforementioned magnetic microrobot and reinforcement learning framework.

[0104] An adaptive rehabilitation system based on magnetic microrobots and a reinforcement learning framework is proposed, which specifically includes:

[0105] S1: Standardized interactive test of motor function limits;

[0106] S2: Calculation of clinical assessment indicators;

[0107] S3: Initialize training parameters;

[0108] S4: Real-time interactive control;

[0109] S5: Dynamic difficulty adjustment.

[0110] Furthermore, S1 specifically includes:

[0111] The system controls the microrobot to move at a preset high speed within the operating plane and is programmed to actively avoid the patient's hand approaching. The patient needs to chase and grab the microrobot as quickly and accurately as possible. Through this high-intensity chase task, the patient can exert their maximum current motor ability, thereby objectively measuring the patient's upper limit of movement speed and actual range of motion under extreme conditions.

[0112] Furthermore, S2 specifically includes:

[0113] During the adversarial interaction, the visual perception end captures the image in real time and converts it into physical coordinates. The control end extracts multi-dimensional kinematic indicators based on the acquired hand and micro-robot trajectory data, specifically including:

[0114] Root mean square tracking distance: the average spatial distance between the affected limb and the microrobot during the pursuit process, used to assess the patient's spatial perception and tracking accuracy;

[0115] Mean speed of movement: The average displacement rate of a patient under full-force pursuit, used to assess the patient's upper limb muscle strength and degree of bradykinesia;

[0116] Range of motion (ROM) area: The area of ​​a two-dimensional convex hull constructed based on the trajectory of the affected limb, used to quantify the effective extension boundary and movement restriction of the patient's upper limb;

[0117] Trajectory temporal smoothness Jerk: Calculates the acceleration characteristics of the affected limb movement, measures the abruptness and continuity of the movement, and is used to assess the degree of spasticity or motor fragmentation after stroke.

[0118] Trajectory space curvature: measures the degree of bending and oscillation of the movement path of the affected limb, and is used to identify whether the patient has control deviations such as tremor or ataxia.

[0119] Furthermore, S3 specifically includes:

[0120] Based on the above evaluation indicators, the system automatically calculates the initial training parameters suitable for the patient's current state through the built-in parameter mapping logic, such as the robot's basic moving speed limit and the effective radius of grasping judgment, thereby completing the system hot start for personalized rehabilitation.

[0121] Furthermore, S4 specifically includes:

[0122] Through a continuous dynamic game mechanism based on reinforcement learning, the system can continuously stimulate the patient's pursuit intention and motor potential, ensuring the training quality and concentration during each interaction.

[0123] Dynamic avoidance: Reinforcement learning strategies drive the microrobot to actively avoid approaching and grasping the affected limb, preventing the target from being easily captured, thereby ensuring that rehabilitation training has a basic level of challenge.

[0124] Maintaining an effective induction range: While avoiding being grabbed, the reinforcement learning strategy is constrained by the set induction reward to prevent the microrobot from escaping to an ineffective area that is too far away from the affected limb; the system controls the microrobot to always move within the effective induction distance around the affected limb, that is, to stay at a critical position that the patient needs to make a moderate effort to reach.

[0125] Furthermore, S5 specifically includes:

[0126] Based on real-time trajectory planning, the control unit is also responsible for dynamically adjusting the task difficulty based on the patient's stage performance;

[0127] The system sets a fixed time window or number of attempts, and statistically analyzes the patient's interaction effects within that period, such as the success rate of the capture task;

[0128] With a built-in difficulty adjustment algorithm, if the current success rate continues to be higher than the preset rehabilitation upper limit, it means that the task has lost its challenge for the patient, and the system will appropriately increase the robot's speed limit and reduce the judgment radius; if the success rate is lower than the preset lower limit, it means that the patient is experiencing frustration, and the system will reduce the robot's speed limit. Through this feedback adjustment mechanism, the system can dynamically adjust the error of the initialization parameters and the patient's fatigue effect to ensure the long-term benefits of rehabilitation training.

[0129] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0130] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An adaptive rehabilitation system based on magnetic microrobots and a reinforcement learning framework, characterized in that, It consists of three main modules: physical interaction terminal, visual perception terminal, and central control terminal; The visual sensing end includes a high-resolution real-time imaging system. The real-time imaging system is vertically fixed above the operation panel by an external bracket. The lens optical axis is vertically downward and the field of view completely covers the entire operation panel. It collects video stream data including the patient's limb and the micro-robot in real time. The central control unit adopts a data processing device equipped with a high-performance processor. The data input interface is connected to the visual perception terminal to receive video stream data in real time, and the communication output interface is connected to the magnetic drive unit driver to issue control commands. The physical interaction terminal serves as the direct carrier for patient rehabilitation work. Through magnetic coupling, the driving device and the interaction unit are physically decoupled. Together with the visual perception terminal and the central control terminal, it completes closed-loop patient motor function assessment and rehabilitation training.

2. The adaptive rehabilitation system according to claim 1, characterized in that, The physical interaction terminal includes an operation panel, a magnetic microrobot, and a magnetic drive unit; The control panel is placed horizontally and is made of a non-magnetic material with a certain structural support strength, which combines the function of supporting the affected limb and the function of magnetic field penetration.

3. The adaptive rehabilitation system according to claim 2, characterized in that, The magnetic microrobot is placed above the control panel and serves as a dynamic physical interaction target for patient visual tracking and grasping.

4. The adaptive rehabilitation system according to claim 2, characterized in that, The magnetic drive unit is concealed below the operation panel and includes a magnetic field generating unit and a two-dimensional motion actuator. The magnetic poles of the driving magnet attract the magnetic poles inside the microrobot, forming a non-contact magnetic coupling across the control panel. The two-dimensional motion actuator receives commands from the central control unit, driving the driving magnet to move in a two-dimensional plane, thus remotely pulling the microrobot to move synchronously.

5. The adaptive rehabilitation system according to claim 4, characterized in that, The two-dimensional motion actuator adopts an XY-axis linear module.

6. An adaptive rehabilitation assessment method based on magnetic microrobots and a reinforcement learning framework, applied to the adaptive rehabilitation system described in any one of claims 1-5, characterized in that, This includes standardized interactive steps for testing the limits of motor function and steps for calculating clinical assessment indicators; In the standardized motor function limit test interactive steps, the system controls the microrobot to move at a preset high speed. The microrobot actively avoids the patient's hand approaching, and the patient performs chasing and grasping actions to test the patient's maximum speed limit and range of motion. In the calculation steps of the clinical assessment indicators, the visual perception end converts the acquired image into physical coordinates, and the central control end extracts multi-dimensional kinematic assessment indicators based on the hand and microrobot trajectory data.

7. The adaptive rehabilitation assessment method according to claim 6, characterized in that, The multi-dimensional kinematic evaluation metrics include one or more combinations of the following metrics: root mean square tracking distance, average motion velocity, range of motion (ROM) area, trajectory temporal smoothness (Jerk), and trajectory spatial curvature. The root mean square tracking distance is used to characterize the average spatial distance between the affected limb and the microrobot; the average motion speed is used to characterize the patient's upper limb muscle strength and bradykinesia; the range of motion (ROM) area is used to quantify the effective extension boundary of the affected limb; the trajectory time smoothness (Jerk) is used to measure the continuity of the affected limb's movements; and the trajectory space curvature is used to identify deviations in the affected limb's motion control.

8. An adaptive rehabilitation training method based on a magnetic microrobot and a reinforcement learning framework, applied to the adaptive rehabilitation system described in any one of claims 1-5, characterized in that, After the evaluation method described in claims 6-7 is used to complete the index calculation, the training parameter initialization step, the real-time interactive control step, and the dynamic difficulty adjustment step are executed sequentially. In the training parameter initialization step, the system determines the initial training parameters based on clinical evaluation indicators through parameter mapping logic. In the real-time interactive control step, the interactive control of the microrobot is achieved through a reinforcement learning dynamic game mechanism; In the dynamic difficulty adjustment step, the difficulty of rehabilitation training tasks is adjusted based on the patient interaction effect within a fixed period.

9. The adaptive rehabilitation training method according to claim 8, characterized in that, In the real-time interactive control step, the reinforcement learning strategy drives the microrobot to perform dynamic evasion actions while maintaining an effective induction zone around the affected limb, and the microrobot is always in a critical position that the patient can reach.

10. The adaptive rehabilitation training method according to claim 8, characterized in that, In the dynamic difficulty adjustment step, the system sets a fixed time window and counts the success rate of patients grasping tasks within the window. When the task success rate is higher than the preset upper limit, the speed limit of the micro-robot is increased and the effective radius of the grasping judgment is reduced; when the task success rate is lower than the preset lower limit, the speed limit of the micro-robot is reduced.

Images