Hemiplegic hand rehabilitation training system based on multi-dimensional task complexity and adaptive decision

Abstract

This invention relates to a rehabilitation training system for hemiplegic hand based on multidimensional task complexity and adaptive decision-making. The system includes a first autonomous movement module, a second autonomous movement module, and an edge computing central control system. The first movement module can adjust the gripping force and present various object grasping postures. The second movement module has a multidimensional adjustable object tray on its top, whose opening size, height, and dynamic / static modes are adjustable. The central control system uses visual recognition to obtain the object posture angle and the number of targets, and based on the cumulative success rate, it adaptively and jointly adjusts multidimensional difficulty parameters such as gripping force, posture angle, number of targets, difficulty of the object tray, and distance between the two modules using a reinforcement learning algorithm. This achieves closed-loop adaptive optimization of difficulty for each dynamic change in the training space coordinates. This invention combines realistic tactile feedback with high-precision automatic acquisition of kinematic data, constructing a three-dimensional difficulty gradient system, suitable for refined rehabilitation training of hemiplegic hand function after stroke.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation robot technology, and in particular to a rehabilitation training system for hemiplegic hand based on multidimensional task complexity and adaptive decision-making. Background Technology

[0002] In current clinical practice for hand function rehabilitation of hemiplegic patients, the mainstream training models can be divided into the following two categories: Traditional desktop task-oriented training uses everyday objects such as snowflakes, wooden blocks, screws, and small sticks, requiring patients to repeatedly grasp, move, and place them. Its advantage is that the training items are derived from daily life, and the training content is closely integrated with real-life scenarios, which is conducive to the transfer of rehabilitation effects. However, its fundamental drawbacks are: the points of retrieving and placing objects are relatively fixed, the training trajectory is monotonous and repetitive, easily leading to patient boredom and fatigue; the training difficulty cannot be dynamically adjusted according to the patient's condition on a given day or the long-term recovery process; the training process heavily relies on one-on-one observation and manual recording by the therapist, resulting in high human resource consumption and low treatment efficiency.

[0003] Digital and robotic-assisted training devices, such as upper limb exoskeletons, virtual reality interactive screens, and force feedback joysticks, offer advantages such as novel interactive interfaces, built-in gamification elements to enhance short-term interest, and automatic recording of training data, allowing therapists to monitor multiple patients simultaneously. However, their significant drawback is that training often involves virtual graphical operations on a screen or passive / assisted movements of robotic arms, detached from the tactile feedback of real physical objects and the context of handling everyday objects. This results in poor transfer of functional improvements gained through training to real-life activities.

[0004] Chinese patent application publication number CN107928989A discloses a comprehensive hand function rehabilitation training platform, including lockable casters, a lifting motor, an intelligent workbench assembly, and various training components. These components are connected by threaded joints to form a four-corner frame structure, supporting height adjustment and movement. It provides 12 training components to train different upper limb joints, combined with an intelligent control system and game functionality. However, it is merely a fixed-position, single-dimensional collection of strength training components, lacking dynamic changes in spatial positioning for picking and placing, as well as adaptive closed-loop decision-making capabilities. It cannot achieve three-dimensional, automated, and precise control of difficulty within a real grasping-transfer-placement operational loop.

[0005] In summary, there is currently a lack of desktop rehabilitation training systems that can deeply integrate the advantages of physical operation training and dynamic adaptive control. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a rehabilitation training system for hemiplegic hand based on multidimensional task complexity and adaptive decision-making. This system aims to solve or partially solve the problems of fixed training positions, rigid and monotonous difficulty adjustment, and incompatibility between real tactile operation and automated data acquisition in traditional hemiplegic hand rehabilitation training, which result in poor interest, high dependence on therapists, and difficulty in objectively evaluating rehabilitation effects.

[0007] The objective of this invention can be achieved through the following technical solutions: One aspect of the present invention provides a rehabilitation training system for hemiplegic hand based on multidimensional task complexity and adaptive decision-making, comprising: The first moving module can move autonomously within the desktop plane and is equipped with a storage compartment, a catapult mechanism, and a multi-degree-of-freedom flipping fixture. The flipping fixture is used to carry training objects and can present various gripping postures and adjust the clamping force on the objects. The second mobile module can move independently within the desktop plane. It has a multi-dimensional adjustable storage tray on the top, and the size of the tray opening, the height of the tray surface / edge fence height, and the static / dynamic working mode are adjustable. An edge computing central control subsystem, communicatively connected to the first and second mobile modules, includes: The visual recognition module is used to obtain the current object posture angle of the flipping fixture and the number of target objects ejected in a single ejection. The training space management module is used to define the training space range; The data acquisition and evaluation module is used to record and calculate the success rate of object retrieval and placement, the time of a single task, and the spatial coverage of the upper limbs in real time. An adaptive decision-making module is used to jointly adjust multi-dimensional difficulty parameters based on accumulated time-series training data, with the optimization goal of maintaining the success rate of object retrieval within a preset target range. The multi-dimensional difficulty parameters include at least the clamping force, the object posture angle, the number of target objects, the difficulty parameters of the storage dish, and the spatial distance between the first moving module and the second moving module.

[0008] As a preferred technical solution, the storage compartment has a single-shot channel and a scattering channel. The ejection mechanism is configured to selectively execute a single-target mode or a multi-target stacking / scattering mode according to the set cognitive sorting difficulty. In the multi-target stacking / scattering mode, 3-8 homogeneous or heterogeneous training objects are ejected at one time.

[0009] As a preferred technical solution, the flipping clamp is configured to carry the training object in at least three gripping postures under the drive of a micro servo motor: horizontal full palm position, side-standing pinch position, and suspended upside down position, and the gripping force is continuously adjustable in the range of 0.5N to 8N.

[0010] As a preferred technical solution, the storage dish includes an annular aperture or telescopic baffle to adjust the effective diameter of the dish opening, and a miniature electric push rod to adjust the height of the dish surface and the height of the edge fence. The storage dish also has a dynamic drive mechanism, which enables it to perform low-speed reciprocating translational or rotational motion on a horizontal plane according to a preset trajectory.

[0011] As a preferred technical solution, the bottom of the first moving module and the second moving module are respectively provided with an omnidirectional wheel chassis and a drive motor.

[0012] As a preferred technical solution, the visual recognition module uses a deep learning target detection algorithm to segment the global desktop image captured by the wide-angle camera to identify the effective area boundary of the desktop and the location of obstacles, and extracts the object posture angle and the number of ejected target objects at the flipping fixture in real time.

[0013] As a preferred technical solution, the adaptive decision-making module uses a reinforcement learning algorithm or a Bayesian regression model to automatically optimize and adjust the multidimensional difficulty parameters, with the success rate of object retrieval stabilizing at 60% to 80% as the preset target range.

[0014] As a preferred technical solution, a touch display interactive screen is also included. The data acquisition and evaluation module is also used to calculate the average speed of hand movements and automatically generate an electronic report containing trend charts and functional level changes after training.

[0015] In another aspect, the present invention provides a control method for a hemiplegic hand rehabilitation training system, for controlling the aforementioned hemiplegic hand rehabilitation training system, the method comprising the following steps: Pre-fill the storage compartment of the first mobile module, and input patient information and initial rating; Drive the first and second mobile modules to perform edge scanning and build a desktop digital twin map; Calculate the initial safe distance based on the initial rating, move the first mobile module to the starting retrieval point, and move the second mobile module to the starting placement point; The training task is executed in a loop: the first moving module is controlled to pop out and fix the training object. After the patient is detected to put the object into the storage tray of the second moving module, a successful task is recorded. After each task is completed, the success rate and kinematic statistics are updated, and the adaptive decision-making module automatically determines the next set of difficulty parameters and coordinate offsets based on the cumulative success rate. The first and second moving modules are controlled to move autonomously to the new pick-up and place-up points according to the coordinate offset, and the next task is started, so that the pick-up and place-up positions are different in each time. After the training timer ends, an electronic report containing trend charts and changes in functional levels is generated.

[0016] As a preferred technical solution, the steps of the adaptive decision-making module automatically determining the next set of difficulty parameters include: The goal is to maintain a success rate of 60% to 80%. When the success rate is lower than this target range, the attitude angle of the object is reduced or the number of target objects in a single launch is increased to reduce the difficulty. When the success rate is higher than the target range, prioritize increasing the dynamic disturbance frequency of the storage dish or increasing the spatial distance to increase the difficulty.

[0017] Compared with the prior art, the present invention has at least one of the following beneficial effects: (1) Enhanced the richness of training: Traditional physical training has a fixed position, which leads to repetitive training actions and poor interest. It is difficult to effectively stimulate the patient's spatial exploration ability. The present invention sets up a first moving module and a second moving module that can move independently within the desktop plane. The first moving module and the second moving module are controlled to move autonomously to a new picking point and placement point according to the coordinate offset, so that the picking and placing positions are different each time. This fundamentally breaks the training limitations of fixed scenes and introduces infinite spatial coordinate transformation into desktop physical training, so that the trajectory and spatial coordinates of each picking and placing action are not repeated, which greatly enhances the richness, interest and the patient's desire for spatial exploration.

[0018] (2) Providing precisely matched training: Traditional training equipment can only adjust a single parameter, and the difficulty system is flat, which cannot precisely adapt to the complex needs of patients in different stages from the relaxation phase to the recovery phase. In this regard, the flipping clamp of the present invention is used to carry training objects and can present multiple gripping postures and adjust the clamping force on the objects. The size of the plate opening, the plate height / edge fence height, and the static / dynamic working mode are adjustable. The ejection mechanism is configured to selectively execute a single target mode or a multi-target stacking / scattering mode. The adaptive decision module is used to jointly adjust multi-dimensional difficulty parameters. The multi-dimensional difficulty parameters include at least the clamping force, the object posture angle, the number of target objects, the difficulty parameters of the plate, and the spatial distance between the first moving module and the second moving module. Thus, a three-dimensional gradient system covering multiple dimensions such as object retrieval difficulty, target reception difficulty, spatial exploration difficulty, and cognitive sorting difficulty is constructed. It can precisely combine and adjust various parameters to provide precisely matched training programs for patients in different recovery stages such as Brunnstrom II to V.

[0019] (3) The high ecological validity of physical training is preserved: The system of the present invention integrates force and visual perception. The flip clamp is used to adjust the clamping of the object, and the clamping force is continuously adjustable in the range of 0.5N to 8N. The data acquisition and evaluation module is used to record and calculate the success rate of picking up and placing objects, the time of a single task and the spatial coverage of the upper limb in real time. Thus, on the same physical platform, the synchronous automatic acquisition of real tactile force feedback from gentle touch to tight fit and high-precision kinematic data is realized. This not only preserves the high ecological validity of physical training, but also gives it the ability to make objective quantitative evaluation.

[0020] (4) Ensuring training efficiency: In this invention, the adaptive decision module is used to adjust multi-dimensional difficulty parameters based on accumulated time-series training data, with the optimization goal of maintaining the success rate of object retrieval to a preset target range. Specifically, when the success rate is lower than the target range, the system prioritizes reducing the object's attitude angle or increasing the number of target objects in a single ejection to reduce the difficulty. When the success rate is higher than the target range, the system prioritizes increasing the dynamic disturbance frequency of the storage dish or increasing the spatial distance to increase the difficulty. This achieves real-time closed-loop adaptive adjustment of training difficulty. The system can automatically optimize based on the patient's continuous performance, just like an experienced therapist, and always maintain the success rate in the optimal challenge range. This avoids excessive frustration, ensures training efficiency, and significantly reduces the cost of manual supervision.

[0021] (5) Improve the ecological validity of rehabilitation training: In this invention, the flipping fixture is configured to carry the training object in at least three gripping postures under the drive of a micro servo motor: horizontal full palm posture, side-standing pinch posture, and suspended inverted posture. By physically reconstructing the spatial posture of the object, it can simulate the gripping mode of key life actions such as picking up a bank card, taking out a card, and unscrewing a bottle cap with high fidelity. This allows the fingertip coordination and three-dimensional positioning ability obtained from training to be directly transferred to real life scenarios, greatly improving the ecological validity of rehabilitation training.

[0022] (6) Synergistic effect of motor and cognitive rehabilitation is achieved: In this invention, the ejection mechanism is configured to selectively execute a single target mode or a multi-target stacking / scattering mode. The visual recognition module is used to obtain the current object posture angle of the flipping fixture and the number of target objects ejected in a single ejection. The system can intelligently switch between high-precision isolated target and multi-object mixed sorting mode according to the patient's cognitive impairment type. It can force the patient to focus their attention on a single target on the affected side to counteract neglect, and can also train their visual search and sorting ability in a cluttered background, thus achieving synergistic effect of motor and cognitive rehabilitation. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of a hemiplegic hand rehabilitation training system based on multidimensional task complexity and adaptive decision-making in the embodiment. Figure 2 This is a flowchart of the control method for the hemiplegic hand rehabilitation training system in the embodiment. Figure 3 This is a schematic diagram of the electronic device in the embodiment. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0026] Example 1 To address the problems of fixed training positions, rigid difficulty, poor engagement, and high dependence on therapists in traditional training methods, this embodiment provides a hemiplegic hand rehabilitation training system based on multidimensional task complexity and adaptive decision-making. It aims to achieve real-time dynamic changes in the spatial positions of the object-grabbing and placement targets on the table, ensuring that the trajectory and spatial coordinates of each patient's training movements are unique, thus enhancing the richness and engagement of the training. Furthermore, it constructs a multi-dimensional, quantifiable, and adjustable three-dimensional difficulty gradient system, ranging from object grasping difficulty and target placement precision difficulty to the overall training spatial range difficulty, precisely adapting to the needs of patients at different recovery stages. In addition, it introduces visual perception and reinforcement learning algorithms to achieve automatic identification of training boundaries, closed-loop adaptive adjustment of training difficulty, and objective quantitative recording of training data throughout the entire process, significantly reducing the cost of manual supervision.

[0027] See Figure 1 The system includes the following components: (1) Autonomous mobile object retrieval module (i.e., the first mobile module).

[0028] It has an omnidirectional wheel chassis and a drive motor at the bottom, which can move freely in the two-dimensional plane of the desktop.

[0029] The internal casing contains multiple storage compartments for batch filling with everyday training objects such as snowflakes, small screws, and sticks. The storage compartments are divided into single-shot and scattered channels. The system can control the launch mechanism to execute two modes based on the currently set cognitive sorting difficulty: Single-target mode, also known as high-precision mode: Only one object is exposed at a fixed point on the table. Due to the lack of tactile reference and visual redundancy from surrounding objects, the patient must use extremely high fingertip control to accurately locate and grasp this isolated target. The slightest deviation will cause it to slip. This is precisely the most difficult precision control training.

[0030] Multi-object stacking / scattering mode, also known as sorting mode: 3-8 homogeneous or heterogeneous objects are ejected at once, such as mixed snowflakes, screws, and sticks, simulating a cluttered scene in real life. In this mode, patients do not need to worry too much about the precision of their grasping; instead, they can instinctively pick up an easier object to complete the task. The difficulty is relatively low, and it is mainly used to train patients' visual search ability and hand volume grasping ability.

[0031] The exit of the chamber is equipped with an electromagnetic or mechanical multi-degree-of-freedom flipping gripper. Driven by a micro servo motor, this gripper can carry the training object in three or more typical grasping postures in space: 1. Horizontal full palm position, 0°: The object is laid flat on the table. The patient can grasp it with the whole palm or pinch it with three fingers. This is the easiest position.

[0032] 2. Side-pinch position, 45°-90° tilt: When an object, such as a snowflake, is placed with its side facing up or upright, the patient must use the radial side of the thumb and index finger to pinch it to remove it. This greatly increases the precision required for the fingertip contact surface and makes it easy to slip.

[0033] 3. Suspended inverted position, 180° or a specific angle of elevation: The object is attracted to the lower end by the clamp, and the patient needs to overcome gravity and maintain wrist pronation / supination stability to pull it out. Simultaneously, the clamp at the exit of the compartment can adjust the clamping force according to the current difficulty level, such as continuously adjustable from 0.5N to 8N: at low difficulty, the object floats on the exit surface and can be easily pinched by the patient; at high difficulty, the object is tightly embedded or clamped, and the patient needs to apply greater fingertip pinching force or three-finger gripping force to pull it out.

[0034] (2) Autonomous moving placement module (i.e., the second moving module).

[0035] The bottom also features an omnidirectional wheel chassis and drive motor for independent and autonomous movement. The top features a multi-dimensional adjustable dish. This dish integrates three adjustment mechanisms: adjusting the effective diameter via a ring-shaped aperture or telescopic baffle to achieve dish opening size adjustment; adjusting the vertical height and edge baffle height via a miniature electric actuator to achieve dish surface and edge guard height adjustment; and switching between static and dynamic modes, allowing it to remain static or perform low-speed reciprocating translation or rotational movement on a horizontal plane along a preset trajectory, increasing the difficulty of placement and aiming.

[0036] (3) Edge computing central control subsystem.

[0037] Including a wireless communication module, a touch display screen, and a high-performance embedded processor, it also includes: The built-in visual recognition module uses a top-mounted wide-angle camera on either the first or second moving module to capture a global image of the desktop. Deep learning object detection algorithms, such as YOLO or UNet, are then used to segment the effective desktop area boundaries and obstacle locations. Furthermore, the module must continuously identify the current object orientation angle θ at the object retrieval port of the first moving module and the number of objects (n) ejected.

[0038] The training space management module defines an elliptical or irregular polygonal training area based on the patient's initial assessment results. The better the function, the larger the area, such as 40cm × 60cm. The worse the function, the smaller the area, such as 20cm × 30cm.

[0039] The data acquisition and evaluation module records and calculates key kinematic indicators in real time, including the success rate of retrieving and placing objects, the time taken for a single task, the coverage range of the patient's upper limb in space, and the average speed of hand movements. Among these, the coverage range of the patient's upper limb in space can be indirectly estimated through changes in position.

[0040] The adaptive decision-making module, based on accumulated time-series training data, employs reinforcement learning algorithms such as DQN or PPO, or alternatively, a Bayesian regression model. It automatically adjusts the joint difficulty parameters for the next round or task, with the success rate stabilizing between 60% and 80% as the optimal target range. The adaptive algorithm will jointly adjust the clamping force F, attitude angle θ, number of targets n, and difficulty D of the storage dish. B The algorithm considers five dimensions, including spatial distance L. Based on patient success rate feedback, it automatically optimizes the algorithm. For example, if a patient frequently slips due to insufficient finger strength, the system prioritizes lowering the posture angle, adjusting it from a 90° sideways position back to a 0° flat position, or increasing the gripping force. If a patient is unable to grasp isolated objects due to hand-eye coordination difficulties, the system automatically switches to a multi-target scattered mode to lower the threshold for retrieving objects and restore the patient's confidence.

[0041] The working process of this device is as follows: Step 1: The therapist pre-fills the first mobile module storage compartment and enters the patient ID and initial rating into the central system.

[0042] Step 2: The system drives the first and second mobile modules to perform edge scanning and establish a desktop digital twin map.

[0043] Step 3: The system calculates the initial safe distance based on the initial rating, moves the first moving module to the starting pick-up point, and moves the second moving module to the starting placement point.

[0044] Step 4: Start the training cycle. The first moving module pops out and fixes an object; the patient reaches out and grabs it; the patient transfers the object and puts it into the storage dish of the second moving module; the sensor built into the second moving module detects the object falling in and sends a success signal.

[0045] Step 5: After a single task is completed, the system immediately updates the success rate and kinematic statistics, and uses the built-in algorithm to determine the next set of coordinate offsets and difficulty increments.

[0046] Step 6: The first and second moving modules autonomously travel to the new position according to the new coordinates and start the next cycle, so that the position changes every time it is picked up and put down.

[0047] Step 7: After the training timer ends, the system will automatically generate an electronic report that includes trend charts and functional level change prompts.

[0048] This system breaks through the technical bias of fixed scenarios. In traditional physical operation training, both therapists and patients assume that training tools must be placed statically on the table. This system introduces the spatial coordinate transformation capability of mobile robots into desktop fine rehabilitation training through dual mobile chassis collaborative technology, enabling physical training to have the ability to change infinite scenarios like virtual reality. This is a fundamental change to the traditional physical training mode.

[0049] Furthermore, while existing digital devices can record data, they lose the sense of touch; traditional physical training, while providing tactile feedback, cannot objectively record it. This system integrates force perception and visual motion tracking, achieving simultaneous acquisition of realistic tactile feedback and high-precision kinematic data on the same physical platform. This resolves the contradiction between realistic tactile feedback and automated data processing.

[0050] Furthermore, existing devices can only adjust a single parameter, such as resistance or speed. This system innovatively combines and regulates multiple physical dimensions, including the difficulty of object retrieval, the difficulty of target reception, the difficulty of spatial exploration, spatial perception, and visual field, forming a three-dimensional rehabilitation gradient. This significantly expands the applicable disease stages for patients, covering Brunnstrom stages II to V.

[0051] Finally, in this system, the reinforcement learning algorithm is not limited to character control in virtual games, but is applied to control the spatial position and mechanical parameters of two physical entities in the real world, namely the first and second movement modules, to achieve physical rehabilitation training.

[0052] In a specific implementation scenario, taking an occupational therapy room as an example, the operation follows the process of initial assessment, parameter prescription setting, real-time monitoring and compensatory correction, closed-loop adaptive adjustment of difficulty, and periodic reassessment of efficacy. The following details each step: (1) Preparation and environmental setup before treatment.

[0053] Choose a sturdy wooden or metal treatment table with a length ≥1.2m, width ≥0.7m, and height 73~78cm. Keep the tabletop clean and scratch-free to avoid affecting the uniformity of friction for the omnidirectional casters. After powering on the first and second movement modules, place them side by side in the center of the tabletop, approximately 30cm apart. Place the central control tablet on the therapist's workbench to the side, ensuring the therapist can simultaneously observe patient movements and screen data.

[0054] Based on the training objectives, the therapist selects everyday training objects of appropriate size (hereinafter referred to as training objects). A recommended standard training object library is as follows: Level 1, Large Grip Items: Soft sponge blocks with a diameter ≥ 4cm, colored EVA building blocks; Level 2, Three-finger pinching type: wooden cylinders with a diameter of 2-3cm, large snowflake-shaped pieces, and a thickness of 3mm; Level 3, Fine Pinching Type: Small screws, coins, small snowflakes with a diameter of 1~1.5cm and a thickness of 1.5mm, and plastic pill simulations.

[0055] The therapist injects approximately 30 to 50 selected training objects into the vibration storage chamber at the top of the first moving module.

[0056] (2) Initial functional assessment and personalized prescription setting.

[0057] With the therapist's assistance, the patient sits in front of the treatment table with the affected upper limb placed on the table, the shoulder joint naturally relaxed, the elbow joint flexed at about 90°, and the forearm in a neutral position.

[0058] Step 1: Power on the system and create patient records.

[0059] The therapist selects a new patient on the control panel and enters the patient's basic information, diagnosis, course of illness, and previous rehabilitation assessment scale scores, such as the FMA-UE wrist and hand score, modified Ashworth spasticity grade, and Brunnstrom stage.

[0060] Step 2, initial motion threshold detection.

[0061] Clicking the initial assessment button will automatically run a standardized probing procedure to determine the patient's current functional limits, avoiding subjective biases from therapists. The first moving module stops at the center of the table. The second moving module starts 10cm away from the first module and gradually moves further away at a speed of 3cm per movement. The system provides a voice prompt to the patient to reach as far as they can to touch A and B. The system uses the built-in visual cameras in A and B to capture the actual spatial positions reached by the patient's upper limb and automatically calculates the current safe and effective range of motion, D. initial .

[0062] The first moving module sequentially releases training objects with four clamping forces: 0.5N, 1.5N, 3N, and 5N, with the default orientation being a flat 0° angle. The second moving module's object placement dish is set to its maximum diameter and static position. The patient's success rate in retrieving objects at each clamping force is recorded. If a patient fails twice consecutively at a certain clamping force, the test is terminated. A cognitive-motor interference test is then conducted. Without the patient's knowledge, the system randomly switches between two modes: single isolated target and multiple scattered targets, three times each. The patient's reaction time and success rate differences between the two modes are recorded to assess the patient's visuospatial attention and sorting abilities.

[0063] Step 3: The therapist confirms and generates the initial training prescription.

[0064] Based on the above detection results, the system automatically recommends an initial five-dimensional training prescription. Therapists can make minor adjustments based on their clinical experience and then click to confirm. 1. Clamping force F: Take 70% of the highest clamping force setting achieved by the patient in two consecutive successful clamping attempts as the starting value.

[0065] 2. Posture angle θ: If FMA wrist and hand ≤ 20 minutes, set 0°; if ≥ 30 minutes, set 45°~60°.

[0066] 3. Number of targets n: If multiple targets are significantly better than single targets in the visual cognition test, the initial setting is multiple targets scattered; otherwise, the initial setting is single isolated targets.

[0067] 4. Initial distance L between AB (i.e., the first moving module and the second moving module): Take D initial 50% of the initial spacing is used.

[0068] 5. The difficulty of the second moving module's storage tray should initially be set to static, large diameter, and low edge position.

[0069] The therapist clicks "Start Training," and the system enters autonomous operation mode.

[0070] (3) Therapist supervision and manual intervention guidelines during the training process.

[0071] While the system is operating autonomously, therapists do not need to monitor the operation continuously, but they still need to perform the following advanced clinical supervisory duties: 1. Real-time observation and voice / manual correction of compensatory posture.

[0072] Common compensatory patterns in hemiplegic patients include: trunk lateral flexion compensating for upper limb extension, shoulder shrugging compensating for elbow flexion, and excessive ulnar deviation of the wrist compensating for finger pinching. When a therapist observes these compensatory patterns...

[0073] For mild to moderate compensation, the therapist does not need to stop the system. Simply press the preset voice prompt button on the control panel, and the system will play the corresponding prompt through the built-in speaker. At the same time, the system will add a compensation mark to this action and separately count the frequency of compensation in the final report.

[0074] For severely compensatory or risky movements, such as excessive traction in cases of subluxation of the affected shoulder joint, the therapist immediately presses the pause button. The system instantly locks all motors in both the first and second movement modules, entering a zero-speed hold state. Simultaneously, the support plate switches to maximum support mode to prevent objects from falling. After the therapist assists the patient to return to the correct sitting posture, they click to continue the recovery training.

[0075] 2. Difficulty fine-tuning knob operation based on real-time system feedback.

[0076] The central control panel features a five-dimensional real-time fine-tuning wheel on the right side of the main training interface, allowing therapists to make smooth adjustments within a ±20% range based on the patient's immediate performance without stopping the training. For example: The patient was agitated, moved too quickly, and exhibited significant tremors. The therapist could increase the dynamic perturbation frequency of the dish in real time, for example, from 0.2Hz to 0.8Hz, to force the patient to slow down and improve stability. If the patient failed to retrieve an object five times in a row and showed signs of frustration, the therapist could immediately reduce the posture angle from 60° to 0° and switch from a single-object to a multi-object scattered approach, quickly reducing the difficulty to rebuild confidence.

[0077] 3. Adjust the training strategy for the affected side using the system's spatial heat map.

[0078] Every 10 tasks completed, the system automatically generates a thermal map of the upper limb's spatial coverage, projected onto a two-dimensional plane on the desktop. The therapist observes the distribution of the thermal map. If the thermal map is significantly biased towards the healthy side of the desktop, it indicates that the patient is unwilling to extend towards the affected side. In the next cycle, the therapist should use the forced affected-side offset function on the control panel to shift the coordinates of A and B as a whole towards the affected side by 10-15 cm, and provide a voice prompt asking the patient to extend their hand to the blue-marked area, forcing the patient to break the spatial neglect. If the thermal map is concentrated in a very small area in front of the patient, it indicates limited mobility. The therapist should gradually and manually expand the virtual safety boundary range automatically generated by the system to induce the patient to perform a wider range of shoulder and elbow combined movements.

[0079] (4) Interpretation of the report after training and decision-making for the next course of treatment.

[0080] After each training session, the system automatically generates a structured hand function rehabilitation training record form, which may include the following clinical decision support data that is directly available to therapists: Summary of core kinematic indicators: average object retrieval success rate (%), average time per repetition (s), maximum spatial reach (cm), and total trajectory length (m) during this training session. All data were marked with significant differences compared to the average of the previous 10 training sessions to help therapists visually assess trends in functional changes.

[0081] The system automatically tracks compensatory behavior events, recording the number of compensatory events manually marked by the therapist and the number of abnormal trunk displacements detected during training, and generates a bar chart. If the number of compensatory events increases for three consecutive days, the system will display a prompt suggesting additional trunk stability training.

[0082] The optimal timeframe analysis automatically identifies the 10-minute window of highest success rate during the day's training and presents it as a line graph. Therapists can use this information to suggest that patients advance or postpone their training time the following day, or optimize the timing of the training with other treatments such as physical therapy and acupuncture, for example, scheduling the training during the period when the patient's spasms are most relieved.

[0083] With synchronized electronic medical records, therapists can export PDF / XML reports with a single click and upload them directly to the hospital's information system or rehabilitation electronic medical record system, serving as an objective basis for the next rehabilitation assessment.

[0084] (5) Efficient management process for treating multiple patients simultaneously.

[0085] Given the staff shortage in the rehabilitation department, the system supports one therapist simultaneously managing 3-5 desktop systems via a multi-device management mode on the central control panel. After the morning shift handover, therapists sequentially import the corresponding patients' pre-stored prescriptions into the systems of each bed / desktop and then start them sequentially. The central control panel screen is divided into multiple sub-windows, which poll and update the real-time success rate and current action status of each patient every 5 seconds. Only when a patient's success rate is below 20% for 3 consecutive minutes or they remain motionless for more than 1 minute will that patient's window flash and emit a soft alarm sound, allowing the therapist to then handle the situation, thus achieving exception management.

[0086] (6) The timing coordination standard with other treatment methods in the rehabilitation department.

[0087] To ensure deep integration of this system into the existing rehabilitation treatment system, any proven temporal integration scheme can be adopted: (1) PT first, OT follow-up mode. The patient first undergoes 30 minutes of lower limb / trunk core stability training by a physical therapist. Within 15 minutes after the neuromuscular excitability is increased, the patient is immediately transferred to this system for hand function training. At this time, the central nervous system is in a facilitated state, and the system automatically selects a high-difficulty initiation strategy. The initial success rate is set at 70%, making full use of the post-activation enhancement effect brought about by stability training.

[0088] (2) Synchronous playback integration of mirror therapy. This system can be used in conjunction with the mirror therapy box next to the treatment table. During system training, the external display screen on the central control panel synchronously plays a mirror image of the patient's unaffected hand performing the same picking and placing actions. After the therapist activates this mode, the difficulty calculation is no longer based solely on the physical success rate, but rather on the mirror fixation duration as an evaluation indicator of the patient's attention participation. If the fixation duration is less than 40% of the total training time, the system automatically increases the frequency of voice prompts to enhance the activation effect of mirror neurons.

[0089] The above operational procedures, from environmental preparation, initial assessment, prescription setting, real-time monitoring, standard treatment templates, report interpretation to multi-patient management and cross-treatment collaboration, constitute a complete and replicable clinical operation system. Any licensed rehabilitation therapist, after reading this section, can independently use this invention to carry out daily clinical rehabilitation treatment work after 1-2 teaching demonstrations.

[0090] Example 2 Based on Example 1, this example provides a control method for a hemiplegic hand rehabilitation training system, characterized in that it is used to control the hemiplegic hand rehabilitation training system of Example 1, see [link to example]. Figure 2 The method includes the following steps: Step S1: Pre-fill the storage compartment of the first mobile module and input patient information and initial rating.

[0091] Step S2: Drive the first and second mobile modules to perform edge scanning and establish a desktop digital twin map.

[0092] Step S3: Calculate the initial safe distance based on the initial rating, move the first moving module to the starting retrieval point, and move the second moving module to the starting placement point.

[0093] Step S4, cyclically execute the training task: control the first moving module to pop out and fix the training object, and record a successful task after detecting that the patient puts the object into the storage tray of the second moving module.

[0094] Step S5: After each task is completed, update the success rate and kinematic statistics, and the adaptive decision module automatically decides the next set of difficulty parameters and coordinate offsets based on the cumulative success rate.

[0095] Step S6: Control the first and second moving modules to move autonomously to the new pick-up and place-up points according to the coordinate offset, and start the next task, so that the pick-up and place-up positions are different each time.

[0096] Step S7: After the training timer ends, generate an electronic report containing trend charts and changes in functional levels.

[0097] The present invention has the following beneficial effects: 1. The entire process uses real everyday objects, rather than virtual graphics. The grip control, interdigital coordination, and three-dimensional spatial positioning abilities that patients gain during training can be directly transferred to real-life scenarios such as dressing, eating, and screwing in screws, resulting in extremely high training ecological validity.

[0098] 2. Since the first and second moving modules quietly move to a new position on the desktop each time, the patient needs to search and plan the movement path again each time they pick up or put down the module. This effectively stimulates the patient's spatial attention and desire to explore, reduces the boredom of rehabilitation training, and significantly improves the patient's participation and concentration.

[0099] 3. The system has a complete cycle of assessment-training-reassessment. A therapist can monitor the training process of 3-5 desktops simultaneously through the central console. The system will only issue an alarm when an abnormality occurs, such as when the training is not completed for a long time, thereby improving the efficiency of comprehensive treatment and greatly freeing up therapists' human resources.

[0100] 4. The clamping force can be set to a minimum of gentle contact, suitable for assisted passive training with muscle strength of 0-1, and a maximum of tight clamping, suitable for resistance training with muscle strength of 4. The dynamic frequency of the storage plate can be steplessly adjusted from 0.1Hz to 1.5Hz to meet the training needs of the entire cycle from relaxation to recovery, and has the ability to finely adjust the gradient.

[0101] 5. The system output reports not only include frequency statistics, but also in-depth indicators such as motion smoothness, reaction time, and spatial coverage heatmaps, providing therapists with objective quantitative basis for adjusting medication or physical therapy plans and enabling data-driven precision rehabilitation decisions.

[0102] 6. For patients with visuospatial agnosia or unilateral neglect, the system can force a single isolated target mode, compelling the patient to concentrate their attention on an isolated object on the affected side or in the center. Combined with the random displacement of the first motion module, this effectively trains the patient's spatial search and attention maintenance abilities, enabling specialized training for patients with agnosia / apraxia. 7. By adjusting the posture angle, high-value daily life actions such as picking up a bank card from a table and taking a card out of a slot can be simulated. The training ecological validity far exceeds that of a virtual screen and seamlessly connects to the daily life action library.

[0103] Example 3 Based on Embodiment 2, this embodiment provides an electronic device, including: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing a control method for the hemiplegic hand rehabilitation training system as described in Embodiment 2.

[0104] like Figure 3As shown, at the hardware level, this electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above. Figure 2 The method described herein. Of course, in addition to software implementation, this invention does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0105] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0106] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0107] 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 person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A hemiplegic hand rehabilitation training system based on multi-dimensional task complexity and adaptive decision, characterized in that, include: The first moving module can move autonomously within the desktop plane and is equipped with a storage compartment, a catapult mechanism, and a multi-degree-of-freedom flipping fixture. The flipping fixture is used to carry training objects and can present various gripping postures and adjust the clamping force on the objects. The second mobile module can move independently within the desktop plane. It has a multi-dimensional adjustable storage tray on the top, and the size of the tray opening, the height of the tray surface / edge fence height, and the static / dynamic working mode are adjustable. An edge computing central control subsystem, communicatively connected to the first and second mobile modules, includes: The visual recognition module is used to obtain the current object posture angle of the flipping fixture and the number of target objects ejected in a single ejection. The training space management module is used to define the training space range; The data acquisition and evaluation module is used to record and calculate the success rate of object retrieval and placement, the time of a single task, and the spatial coverage of the upper limbs in real time. An adaptive decision-making module is used to jointly adjust multi-dimensional difficulty parameters based on accumulated time-series training data, with the optimization goal of maintaining the success rate of object retrieval within a preset target range. The multi-dimensional difficulty parameters include at least the clamping force, the object posture angle, the number of target objects, the difficulty parameters of the storage dish, and the spatial distance between the first moving module and the second moving module.

2. The hemiplegic hand rehabilitation training system based on multi-dimensional task complexity and adaptive decision according to claim 1, characterized in that, The storage compartment has a single-shot channel and a scattering channel. The ejection mechanism is configured to selectively execute a single-target mode or a multi-target stacking / scattering mode according to the set cognitive sorting difficulty. In the multi-target stacking / scattering mode, 3-8 homogeneous or heterogeneous training objects are ejected at once.

3. The hemiplegic hand rehabilitation training system based on multi-dimensional task complexity and adaptive decision according to claim 1, characterized in that, The flipping clamp is configured to carry the training object in at least three gripping postures under the drive of a micro servo motor: horizontal full palm position, side-standing pinch position, and suspended upside down position, and the gripping force is continuously adjustable in the range of 0.5N to 8N.

4. The hemiplegic hand rehabilitation training system based on multi-dimensional task complexity and adaptive decision according to claim 1, characterized in that, The storage dish includes an annular aperture or telescopic baffle to adjust the effective diameter of the dish opening, and a miniature electric push rod to adjust the height of the dish surface and the height of the edge guardrail. The storage dish also has a dynamic drive mechanism that enables it to perform low-speed reciprocating translational or rotational motion on a horizontal plane according to a preset trajectory.

5. The hemiplegic hand rehabilitation training system based on multidimensional task complexity and adaptive decision-making according to claim 1, characterized in that, The bottom of the first moving module and the second moving module are respectively equipped with an omnidirectional wheel chassis and a drive motor.

6. The hemiplegic hand rehabilitation training system based on multidimensional task complexity and adaptive decision-making according to claim 1, characterized in that, The visual recognition module uses a deep learning target detection algorithm to segment the global desktop image captured by the wide-angle camera to identify the effective area boundary of the desktop and the location of obstacles, and extracts the object posture angle and the number of ejected target objects at the flipping fixture in real time.

7. The hemiplegic hand rehabilitation training system based on multidimensional task complexity and adaptive decision-making according to claim 1, characterized in that, The adaptive decision-making module uses reinforcement learning algorithms or Bayesian regression models to automatically optimize and adjust the multidimensional difficulty parameters, with the success rate of object retrieval stabilizing between 60% and 80% as the preset target range.

8. The hemiplegic hand rehabilitation training system based on multidimensional task complexity and adaptive decision-making according to claim 1, characterized in that, It also includes a touch-screen interactive display. The data acquisition and evaluation module is also used to calculate the average speed of hand movements and automatically generate an electronic report containing trend charts and changes in functional levels after training.

9. A control method for a rehabilitation training system for hemiplegic hand, characterized in that, The method for controlling the hemiplegic hand rehabilitation training system as described in any one of claims 1-8 comprises the following steps: Pre-fill the storage compartment of the first mobile module, and input patient information and initial rating; Drive the first and second mobile modules to perform edge scanning and build a desktop digital twin map; Calculate the initial safe distance based on the initial rating, move the first mobile module to the starting retrieval point, and move the second mobile module to the starting placement point; The training task is executed in a loop: the first moving module is controlled to pop out and fix the training object. After the patient is detected to put the object into the storage tray of the second moving module, a successful task is recorded. After each task is completed, the success rate and kinematic statistics are updated, and the adaptive decision-making module automatically determines the next set of difficulty parameters and coordinate offsets based on the cumulative success rate. The first and second moving modules are controlled to move autonomously to the new pick-up and place-up points according to the coordinate offset, and the next task is started, so that the pick-up and place-up positions are different in each time. After the training timer ends, an electronic report containing trend charts and changes in functional levels is generated.

10. The control method for a hemiplegic hand rehabilitation training system according to claim 9, characterized in that, The steps by which the adaptive decision-making module automatically determines the next set of difficulty parameters include: The goal is to maintain a success rate of 60% to 80%. When the success rate is lower than this target range, the attitude angle of the object is reduced or the number of target objects in a single launch is increased to reduce the difficulty. When the success rate is higher than the target range, prioritize increasing the dynamic disturbance frequency of the storage dish or increasing the spatial distance to increase the difficulty.

Images