Rehabilitation robot motion planning method based on neural network

Abstract

The invention discloses a rehabilitation robot motion planning method based on a neural network, and aims to solve the problems of insufficient individualization, poor adaptability, low intelligent degree and the like of a traditional method. An individualized rehabilitation exercise plan is generated in real time, and the rehabilitation effect is improved; according to individual features such as the age, the physical condition and the rehabilitation stage of the patient, an individualized rehabilitation exercise plan meeting the requirements of the patient is generated; in the rehabilitation process, the motion plan is adjusted in real time according to the motion state and the rehabilitation effect of the patient, and smooth proceeding of the rehabilitation process is ensured; the motion state and rehabilitation effect of the patient are monitored in real time, intelligent analysis and decision making are conducted through the neural network, targeted rehabilitation suggestions and guidance are provided, and a more accurate and individualized rehabilitation scheme is provided for the rehabilitation patient.

Description

Technical Field

[0001] The present invention relates to the technical field of rehabilitation robots, and in particular to a rehabilitation robot motion planning method based on a neural network. Background Art

[0002] (1) Current status of existing technologies

[0003] In the field of rehabilitation medicine, rehabilitation robots have become an important tool for helping patients recover their limb motor function, playing an irreplaceable role in improving patients' quality of life and reducing the workload of rehabilitation therapists. With the continuous advancement of science and technology, rehabilitation robotics has made significant progress, evolving from simple auxiliary support devices to intelligent systems with multiple movement modes and complex control functions.

[0004] Currently, traditional rehabilitation robot motion planning methods are primarily based on preset motion trajectories and fixed control algorithms. For example, linear interpolation methods based on joint angles pre-set the joint angles at different time points and use a linear interpolation algorithm to calculate the angle curve of the joints throughout the entire motion process, thereby controlling the rehabilitation robot to achieve the corresponding motion. There are also trajectory tracking control methods based on dynamic models. This method establishes a mathematical model based on the dynamic characteristics of the rehabilitation robot, plans the robot's motion trajectory by solving the model equations, and uses feedback control strategies to enable the robot to accurately track the planned trajectory.

[0005] Although these traditional methods can realize the basic movement functions of rehabilitation robots to a certain extent and play a certain role in certain stages of rehabilitation treatment, their limitations are becoming increasingly prominent as the demand for individualized and precise treatment in rehabilitation medicine continues to increase.

[0006] (2) Defects of existing technology

[0007] Lack of individualization: Traditional methods use a unified exercise planning strategy and are unable to individualize exercise planning based on individual patient differences (such as age, physical condition, rehabilitation stage, etc.). Different patients have significant differences in the degree of limb motor function impairment and the speed of motor recovery. A unified planning method is difficult to meet the individual needs of patients, resulting in poor rehabilitation results. For example, for older and weaker patients, if the same exercise intensity and difficulty are used as for younger and better-healthy patients, it may increase the patient's physical burden and even cause new injuries.

[0008] Poor adaptability: During the rehabilitation process, patients' motor abilities and rehabilitation needs change over time. As rehabilitation progresses, patients' limb motor function gradually recovers, placing new demands on the intensity, difficulty, and methods of rehabilitation training. However, traditional methods struggle to adjust motion planning in real time to accommodate these changes, often requiring manual intervention and parameter adjustments. This not only increases the workload of rehabilitation therapists but also makes it difficult to meet patients' rehabilitation needs in a timely and accurate manner.

[0009] Low intelligence: Traditional methods lack the ability to monitor and evaluate a patient's motor status and rehabilitation outcomes in real time during rehabilitation training. During rehabilitation training, they lack timely information about a patient's motor performance, fatigue level, and recovery progress, making it impossible to make intelligent rehabilitation decisions based on this information. For example, if a patient exhibits abnormal movements or excessive fatigue, traditional methods are unable to detect and adjust the exercise plan in a timely manner, potentially affecting rehabilitation outcomes and even endangering the patient's safety. Summary of the Invention

[0010] The present invention aims to provide a neural network-based motion planning method for rehabilitation robots, aiming to address the challenges of traditional methods, including insufficient individualization, poor adaptability, and low intelligence. This method leverages the powerful learning and nonlinear mapping capabilities of neural networks to generate personalized rehabilitation motion plans in real time based on the patient's individual characteristics and rehabilitation needs, thereby improving rehabilitation outcomes.

[0011] To achieve the above objectives, the present invention provides the following technical solutions:

[0012] A neural network-based rehabilitation robot motion planning method comprises the following steps:

[0013] Step S1: Data acquisition

[0014] Using sensors to collect real-time motion data of the patient during rehabilitation, the motion data including at least one or more of joint angle, muscle strength, movement speed, and muscle tension;

[0015] Step S2: Neural network model construction and training

[0016] Constructing a neural network model and training the neural network model using a pre-collected historical data set containing patient movement data and corresponding rehabilitation effect evaluation results, so that the neural network model can accurately evaluate the patient's movement status and rehabilitation effect based on the input movement data;

[0017] Step S3: Real-time motion planning

[0018] During the rehabilitation process, the real-time collected patient motion data is input into the trained neural network model to obtain the patient's motion status and rehabilitation effect evaluation results. Combined with the patient's individual characteristics and rehabilitation needs, an individualized rehabilitation exercise plan is generated based on the preset motion planning algorithm;

[0019] Step S4: Execution control

[0020] The generated rehabilitation motion plan is converted into control instructions for the rehabilitation robot to control the rehabilitation robot to perform corresponding motion actions.

[0021] Furthermore, in step S1, the sensor includes an accelerometer, a gyroscope, and an electromyoelectric sensor, and after collecting data, the collected motion data is preprocessed, and the preprocessing operation includes filtering and normalization to improve the quality and availability of the data.

[0022] Furthermore, the neural network model is a combination of one or more of a convolutional neural network, a recurrent neural network, a long short-term memory network or a gated recurrent unit.

[0023] Furthermore, in step S3, the individual characteristics of the patient include age, physical condition, and rehabilitation stage, the rehabilitation needs are determined according to the patient's rehabilitation goals, and the motion planning algorithm adopts an optimization method based on reinforcement learning, genetic algorithm or particle swarm optimization algorithm to generate an optimal individualized rehabilitation motion plan.

[0024] Furthermore, it also includes a feedback adjustment step, which monitors the patient's movement status and rehabilitation effect in real time during the rehabilitation robot's execution of movement movements, and inputs the feedback information into the neural network model to dynamically adjust and optimize the neural network model, and at the same time adjust the rehabilitation movement plan in real time according to the feedback information.

[0025] Furthermore, the feedback information includes changes in physiological indicators of the patient during exercise, exercise completion, and improvement in rehabilitation effects. The changes in physiological indicators include heart rate, blood pressure, and electromyographic signal strength. The improvement in rehabilitation effects is obtained by comparing with preset rehabilitation goals.

[0026] Furthermore, the rehabilitation motion plan includes the motion trajectory, motion speed, motion amplitude and motion duration parameters of the rehabilitation robot. When generating the rehabilitation motion plan, the motion planning algorithm comprehensively considers the patient's motion ability, rehabilitation goals and the motion performance limitations of the rehabilitation robot.

[0027] Furthermore, the method also includes a multimodal information fusion step. When generating a rehabilitation exercise plan, the patient's physiological indicators, exercise status, rehabilitation effect and psychological state multimodal information are comprehensively considered. The psychological state is obtained through the patient's expression, voice or specific psychological assessment scale.

[0028] The neural network-based rehabilitation robot motion planning method of the present invention has the following beneficial effects:

[0029] 1. Individualized rehabilitation: Based on the individual characteristics of the patient and the rehabilitation stage, a tailored exercise plan is developed to precisely meet the patient's needs and enhance the targeted nature of rehabilitation.

[0030] 2. Real-time adaptation to changes: Real-time monitoring of patient movement and physiological data, dynamic adjustment of plans, response to changes in rehabilitation progress and emergencies, to ensure safe and effective rehabilitation.

[0031] 3. Intelligent assessment and decision-making: Deeply analyze data, accurately evaluate rehabilitation effects, and intelligently decide on rehabilitation strategies to provide a scientific basis for treatment and optimize treatment plans.

[0032] 4. Improve rehabilitation efficiency: Optimize the training process, reduce time waste, and reduce the workload of therapists, allowing them to focus more on individualized care and program development.

[0033] 5. Enhance patient confidence: Provide real-time feedback on rehabilitation effects, give positive encouragement, provide individualized training experience, and improve patient rehabilitation enthusiasm and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The present invention is a schematic diagram of the implementation process of a neural network-based rehabilitation robot motion planning method. DETAILED DESCRIPTION

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0036] In the description of the embodiments of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0037] Example:

[0038] A motion planning method for rehabilitation robots based on neural networks

[0039] 1. Overall Architecture

[0040] The rehabilitation robot motion planning method of this embodiment mainly consists of data acquisition, neural network processing, motion planning decision-making, execution control and feedback optimization.

[0041] 2. Architecture of the rehabilitation robot motion planning method based on neural network:

[0042] Hardware layer

[0043] Rehabilitation robot: Choose a high-precision rehabilitation robot with multiple degrees of freedom. For example, an upper limb rehabilitation robot should be able to achieve flexible movement of multiple joints, such as the shoulder, elbow, and wrist. A lower limb rehabilitation robot should be able to simulate movements such as walking and stepping. Ensure that the robot has sufficient load capacity and motion stability to meet the rehabilitation needs of different patients.

[0044] Sensor system: Various sensors are installed, including motion sensors (angle sensors, acceleration sensors) for real-time monitoring of the patient's joint angle, movement speed and acceleration; physiological sensors (electromyographic sensors, heart rate sensors) for collecting the patient's electromyographic signals and heart rate physiological data; and force sensors for measuring the interaction force between the patient and the robot.

[0045] Software Layer

[0046] Data processing module: pre-processes the raw data collected by the sensor, including filtering, denoising, and normalization operations to improve the quality and usability of the data;

[0047] Neural network model: Build a neural network model suitable for rehabilitation robot motion planning. This can use recurrent neural network (RNN), long short-term memory network (LSTM), or convolutional neural network (CNN) structures from deep learning. The model can be trained and optimized based on the characteristics of the rehabilitation task and data features.

[0048] Motion planning algorithm: Based on the output of the neural network model, combined with rehabilitation medicine principles and motion control theory, a motion planning algorithm is designed to convert the prediction results of the neural network into specific motion instructions for the rehabilitation robot.

[0049] Human-computer interaction layer

[0050] User interface: Develop an intuitive and easy-to-use user interface for rehabilitation therapists to input patients' personal information, rehabilitation goals, and training parameters; at the same time, it displays patients' movement data, rehabilitation effect evaluation results, and robot movement status information in real time;

[0051] Feedback mechanism: Establish a feedback mechanism between the patient and the robot to provide real-time training feedback to the patient through visual, auditory and tactile means, such as motion trajectory display, voice prompts and vibration feedback, to enhance patient participation and rehabilitation effect.

[0052] 3. Implementation process of the neural network-based rehabilitation robot motion planning method:

[0053] (1) Data collection

[0054] Sensor selection and placement

[0055] Select appropriate sensors based on the application scenario of the rehabilitation robot and the patient's rehabilitation area;

[0056] For upper limb rehabilitation training, high-precision accelerometers, gyroscopes, and electromyographic sensors are used. Accelerometers and gyroscopes are placed at the joints of the patient's upper limbs to collect motion data such as joint angles and movement speed; electromyographic sensors are attached to the surface of relevant muscles to collect muscle strength information.

[0057] For lower limb rehabilitation training, in addition to the above sensors, plantar pressure sensors can be added and placed on the plantar support part of the rehabilitation robot to obtain plantar pressure distribution data during the patient's walking process.

[0058] Data collection and preprocessing

[0059] The sensor collects the patient's motion data in real time and transmits it to the central processing unit of the data acquisition module via wireless or wired means. The central processing unit filters the collected data to remove noise interference and uses a low-pass filter to filter out high-frequency noise;

[0060] The filtered data is normalized to unify data of different dimensions into the same numerical range for subsequent processing by the neural network model. For example, the joint angle data is normalized to the interval [0,1].

[0061] (2) Neural Network Processing

[0062] Neural network model construction

[0063] According to the requirements of rehabilitation robot motion planning, a suitable neural network model is constructed. This embodiment adopts a hybrid neural network model that combines a long short-term memory network (LSTM) with a convolutional neural network (CNN). The LSTM network is used to process time series motion data and capture the temporal characteristics of the patient's movement process; the CNN network is used to extract spatial features from the motion data.

[0064] For example, for electromyographic signal data, due to its time series characteristics, the LSTM network is used for processing; for the spatial data of joint angle and movement speed, the CNN network is used for feature extraction.

[0065] Model training

[0066] Collect a large amount of historical data sets containing patient movement data and corresponding rehabilitation effect evaluation results. The rehabilitation effect evaluation results are assessed by professional rehabilitation physicians based on the patient's movement performance and rehabilitation progress, including exercise ability scores and rehabilitation goal achievement;

[0067] The constructed hybrid neural network model is trained using historical data sets, and the backpropagation algorithm and stochastic gradient descent optimization method are used to continuously adjust the weight parameters of the neural network, so that the model can accurately evaluate the patient's movement status and rehabilitation effect based on the input movement data.

[0068] (3) Motion planning decision

[0069] Individualized feature extraction

[0070] Before the start of rehabilitation training, individual characteristics of the patients were obtained through questionnaires and physical examinations, including age, physical condition (such as whether there were other diseases), and rehabilitation stage (such as different Brunnstrom stages);

[0071] The patient's individual characteristic information is combined with the real-time collected motion data as input for motion planning decisions.

[0072] Motion planning algorithm implementation

[0073] According to the patient's movement status and rehabilitation effect evaluation results output by the neural network model, combined with the patient's individual characteristics and rehabilitation needs, a motion planning algorithm based on the particle swarm optimization algorithm is used to generate an individualized rehabilitation movement plan;

[0074] The particle swarm optimization algorithm searches for the optimal rehabilitation motion planning parameters in the solution space by simulating the group behavior of bird flocks or fish schools. For example, when planning the upper limb rehabilitation motion trajectory, the particle swarm optimization algorithm is used to search for the optimal motion trajectory parameters with maximizing the rehabilitation effect as the objective function and the robot's motion performance limitations (such as joint range of motion and motion speed limitations) as constraints.

[0075] (4) Execution Control

[0076] Control instruction generation

[0077] The rehabilitation motion planning parameters generated by the motion planning decision module are converted into control instructions for the rehabilitation robot. For example, for joint angle and motion speed parameters, the proportional-integral-derivative (PID) control algorithm is used to generate the corresponding motor control signals.

[0078] Robot motion control

[0079] After receiving control instructions, the rehabilitation robot's controller drives the motor to move according to the specified motion trajectory, speed and amplitude, assisting the patient in completing rehabilitation training movements; during the movement process, the controller monitors the robot's motion status in real time, such as joint position and speed error, and makes dynamic adjustments based on the monitoring results to ensure the accuracy and stability of the robot's movement.

[0080] (V) Feedback Optimization

[0081] Feedback information collection

[0082] During the rehabilitation robot's exercise, it collects the patient's exercise status and rehabilitation effect feedback in real time. In addition to the exercise data collected by the data acquisition module, it also collects the patient's physiological index change information through the heart rate sensor and blood pressure sensor equipment;

[0083] At the same time, cameras and voice recognition technology are used to collect patients' facial expressions and voice information to assess their mental state.

[0084] Neural network model optimization

[0085] The collected feedback information is input into the neural network model of the neural network processing module, and the model is dynamically adjusted and optimized. For example, according to the patient's rehabilitation effect feedback information, the weight parameters of the model are adjusted to improve the accuracy of the model's assessment of the patient's movement status and rehabilitation effect.

[0086] Motion planning adjustments

[0087] The rehabilitation motion plan is adjusted in real time based on feedback information. If the patient's motor ability improves or the rehabilitation goal changes, the motion planning decision module will regenerate the individualized rehabilitation motion plan and control the rehabilitation robot to perform the new motion through the execution control module.

[0088] 4. An implementation example of a neural network-based rehabilitation robot motion planning method (taking the Brunnstrom classification of lower limb hemiplegia patients as an example)

[0089] 1. Basic patient information

[0090] Patient Zhang, a 55-year-old male, suffered right lower limb hemiplegia due to a stroke. The following description uses Brunnstrom stage III and IV as examples. The patient was in good health and had no other serious underlying medical conditions.

[0091] 2. Implementation of Brunnstrom Stage III

[0092] (1) Data collection

[0093] Sensor placement

[0094] Accelerometers and gyroscopes are placed at the hip, knee, and ankle joints of the patient's right lower limb to monitor joint angle changes, movement speed, acceleration, and other motion data in real time.

[0095] A pressure sensor array is installed on the sole of the patient's foot to collect pressure distribution data in different areas of the sole to assess the patient's gait characteristics and sole force conditions;

[0096] Electromyographic sensors are attached to the main muscle groups of the patient's right lower limb (such as the quadriceps femoris, tibialis anterior, etc.) to collect the muscle electromyographic signals, which reflect the degree of muscle activation and fatigue status.

[0097] Data collection frequency

[0098] The data acquisition frequency of the accelerometer, gyroscope, and plantar pressure sensor is set to 100 Hz to ensure that subtle changes in the patient's lower limb movements can be accurately captured;

[0099] The data acquisition frequency of the EMG sensor was set to 1000 Hz to obtain high-quality EMG signals.

[0100] Data Example

[0101] Motion data: At a certain moment, the hip joint angle is 25°, the angular velocity is 4° / s, and the angular acceleration is 1.5° / s 2 ; Knee joint angle is 50°, angular velocity is 2.5° / s, angular acceleration is 0.8° / s 2; Ankle joint angle is 10°, angular velocity is 1.5° / s, angular acceleration is 0.3° / s 2 The plantar pressure distribution shows that the pressure in the heel area is 70N, the pressure in the midfoot is 40N, and the pressure in the forefoot is 20N;

[0102] Electromyographic signal: The amplitude of the electromyographic signal of the quadriceps femoris is 1.8 mV, and the amplitude of the electromyographic signal of the tibialis anterior is 1.2 mV.

[0103] (2) Neural Network Model Training

[0104] Dataset preparation

[0105] Historical data of 300 Brunnstrom stage III lower limb rehabilitation patients were collected, including movement data, physiological data (such as heart rate and blood pressure), rehabilitation assessment results (such as Fugl-Meyer lower limb motor function score), etc. The dataset was preprocessed, including data cleaning and normalization.

[0106] Model selection and construction

[0107] A hybrid neural network model was selected, combining the strengths of convolutional neural networks (CNNs) and recurrent neural networks (RNNs). CNNs were used to extract spatial features from motion data and electromyographic signals, while RNNs were used to process time series data and capture dynamic changes in motion states. The model consisted of two convolutional layers, one pooling layer, one LSTM layer, and one fully connected layer.

[0108] Model training parameters

[0109] The learning rate is set to 0.002, the batch size is set to 24, and the training epochs are set to 80. The Adam optimizer is used to update the model parameters.

[0110] Training Results

[0111] After 80 rounds of training, the model achieved an accuracy of 80%, a recall rate of 75%, and an F1 value of 0.77 on the validation set, indicating that the model can accurately assess the patient's lower limb movement status and rehabilitation effect.

[0112] (3) Individualized exercise planning

[0113] Acquisition of individual characteristic information

[0114] Before rehabilitation training began, we collected information on the patient's age (55 years), physical condition (general), and stage of rehabilitation (stage III). Based on the patient's rehabilitation goals and needs, the training objectives were to promote lower limb muscle synergy and enhance joint mobility.

[0115] Motion plan generation

[0116] During training, real-time motion data (such as joint angles and movement speed), electromyographic signals, and individual characteristics are input into the motion planning and decision-making module. Using a particle swarm optimization algorithm, a personalized lower limb rehabilitation exercise plan is generated based on the patient's specific needs. For example, a walking speed of 0.5 m / s, a stride length of 0.4 m, and a gait cycle of 1.5 s are set, with an emphasis on coordinated movement of the lower limb joints.

[0117] Planning Adjustment

[0118] During the training process, if it is found that the patient's motor ability does not match the preset plan, such as limited joint mobility, the system will promptly adjust the motion planning parameters, such as appropriately reducing the joint motion range or increasing the joint mobility training content.

[0119] (4) Execution Control

[0120] Control instruction conversion

[0121] The generated motion planning parameters (walking speed, stride length, gait cycle, etc.) are converted into control instructions for the rehabilitation robot, which then controls the robot to assist the patient in walking training. For example, by adjusting the robot's joint angles and movement speed, the patient can walk according to the set gait pattern.

[0122] Real-time monitoring

[0123] During training, the patient's exercise status and rehabilitation results are monitored in real time. A heart rate sensor is used to monitor heart rate fluctuations, which should normally be maintained between 70 and 90 beats per minute. A sphygmomanometer is used to monitor blood pressure fluctuations, which should remain within the normal range. Furthermore, a motion analysis system is used to assess improvements in the patient's joint range of motion.

[0124] Data Example

[0125] At a certain moment, the patient's heart rate was 80 beats / minute, blood pressure was 115 / 75 mmHg, and joint mobility increased compared to before training, indicating that the patient's exercise state and rehabilitation effect were good.

[0126] (V) Feedback Optimization

[0127] Patient feedback collection

[0128] After the training, we communicated with the patient and collected his feedback. The patient said that he felt some muscle soreness during the training, but his joint mobility improved.

[0129] Model and planning adjustments

[0130] Based on the patient's feedback, we analyzed the possible causes. Taking into account the patient's muscle soreness, we appropriately adjusted the neural network model's training parameters to reduce the model's sensitivity to motion data and mitigate the risk of overtraining. We also adjusted the rehabilitation exercise plan, appropriately reducing training intensity and increasing muscle relaxation training to ensure the effectiveness and safety of rehabilitation training.

[0131] Optimization results

[0132] After optimization, the patient underwent rehabilitation training again and reported that muscle soreness was reduced and joint mobility continued to improve, verifying the effectiveness of the feedback optimization mechanism.

[0133] 3. Implementation of Brunnstrom Stage IV

[0134] (1) Data collection

[0135] Sensor placement

[0136] Similar to Phase III, accelerometers and gyroscopes are placed at the hip, knee, and ankle joints of the patient's right lower limb, a pressure sensor array is installed on the sole of the patient's foot, and electromyographic sensors are attached to the main muscle groups of the patient's right lower limb (such as the quadriceps femoris, tibialis anterior, etc.).

[0137] Data collection frequency

[0138] The data acquisition frequency of the accelerometer, gyroscope, and plantar pressure sensor was set to 120 Hz, and the data acquisition frequency of the electromyography sensor was set to 1200 Hz to obtain richer data.

[0139] Data Example

[0140] Motion data: At a certain moment, the hip joint angle is 35°, the angular velocity is 6° / s, and the angular acceleration is 2.5° / s 2 ; Knee joint angle is 65°, angular velocity is 3.5° / s, angular acceleration is 1.2° / s 2 ; Ankle joint angle is 20°, angular velocity is 2.5° / s, angular acceleration is 0.6° / s 2 The plantar pressure distribution showed that the pressure in the heel area was 90N, the pressure in the midfoot was 60N, and the pressure in the forefoot was 40N.

[0141] Myoelectric signal: The amplitude of the EMG signal of the quadriceps femoris is 2.5mV, and the amplitude of the EMG signal of the tibialis anterior is 2mV.

[0142] (2) Neural Network Model Training

[0143] Dataset preparation

[0144] Historical data of 400 Brunnstrom stage IV lower limb rehabilitation patients were collected, including movement data, physiological data (such as heart rate and blood pressure), rehabilitation assessment results (such as Fugl-Meyer lower limb motor function score), etc. The dataset was preprocessed, including data cleaning and normalization.

[0145] Model selection and construction

[0146] A hybrid neural network model was chosen, combining the strengths of convolutional neural networks (CNNs) and recurrent neural networks (RNNs). CNNs were used to extract spatial features from motion data and electromyographic signals, while RNNs were used to process time series data and capture dynamic changes in motion states. The model consisted of three convolutional layers, two pooling layers, two LSTM layers, and one fully connected layer.

[0147] Model training parameters

[0148] The learning rate is set to 0.001, the batch size is set to 32, and the number of training rounds is set to 100. The Adam optimizer is used to update the model parameters.

[0149] Training Results

[0150] After 100 rounds of training, the model achieved an accuracy of 85%, a recall rate of 80%, and an F1 value of 0.82 on the validation set, indicating that the model can accurately assess the patient's lower limb movement status and rehabilitation effect.

[0151] (3) Individualized exercise planning

[0152] Acquisition of individual characteristic information

[0153] Before rehabilitation training began, we collected information on the patient's age (55 years), physical condition (general), and rehabilitation stage (stage IV). Based on the patient's rehabilitation goals and needs, the training objectives were set to improve walking speed and gait stability.

[0154] Motion plan generation

[0155] During training, real-time motion data (such as joint angles and movement speed), electromyographic signals, and individual characteristics are input into the motion planning and decision-making module. Using a particle swarm optimization algorithm, a personalized lower limb rehabilitation exercise plan is generated based on the patient's specific needs. For example, a walking speed of 0.8 m / s, a stride length of 0.6 m, and a gait cycle of 1.2 s are set, with an emphasis on gait coordination and stability.

[0156] Planning Adjustment

[0157] During the training process, if it is found that the patient's motor ability does not match the preset plan, such as insufficient gait stability, the system will promptly adjust the motion planning parameters, such as appropriately reducing the walking speed or increasing the gait stability training content.

[0158] (4) Execution Control

[0159] Control instruction conversion

[0160] The generated motion planning parameters (walking speed, stride length, gait cycle, etc.) are converted into control instructions for the rehabilitation robot, which then controls the robot to assist the patient in walking training. For example, by adjusting the robot's joint angles and movement speed, the patient can walk according to the set gait pattern.

[0161] Real-time monitoring

[0162] During training, the patient's exercise status and rehabilitation results are monitored in real time. A heart rate sensor is used to monitor heart rate fluctuations, which should normally be maintained at 80-100 beats per minute. A sphygmomanometer is used to monitor blood pressure fluctuations, which should remain within the normal range. A motion analysis system is also used to assess the patient's gait stability, such as gait symmetry and gait periodicity.

[0163] Data Example

[0164] At a certain moment, the patient's heart rate was 90 beats / minute, blood pressure was 120 / 80 mmHg, gait symmetry index was 0.88 (normal range 0.8-1.0), and gait periodicity index was 0.92 (normal range 0.8-1.0), indicating that the patient's exercise state and rehabilitation effect were good.

[0165] (V) Feedback Optimization

[0166] Patient feedback collection

[0167] After the training, we communicated with the patient and collected his feedback. The patient said that he felt a little tired during the training, but the overall training effect was good and his walking speed improved.

[0168] Model and planning adjustments

[0169] Based on patient feedback, we analyzed possible causes. Taking into account patient fatigue, we adjusted the neural network model's training parameters appropriately to reduce the model's sensitivity to motion data and mitigate the risk of overtraining. We also adjusted the rehabilitation exercise plan, appropriately reducing training intensity and increasing rest time to ensure the effectiveness and safety of rehabilitation training.

[0170] Optimization results

[0171] After optimization, the patient underwent rehabilitation training again and reported that fatigue was reduced, while walking speed and gait stability continued to improve, verifying the effectiveness of the feedback optimization mechanism.

[0172] Through the implementation examples of the above different Brunnstrom stages, the neural network-based rehabilitation robot motion planning method can realize individualized and intelligent rehabilitation motion planning according to the individual differences, rehabilitation stages and rehabilitation needs of patients with lower limb hemiplegia, and effectively improve the patients' rehabilitation treatment effects.

[0173] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A rehabilitation robot motion planning method based on neural network, characterized in that: The following steps are involved: Step S1: Data acquisition Using sensors to collect real-time motion data of the patient during rehabilitation, the motion data including at least one or more of joint angle, muscle strength, movement speed, and muscle tension; Step S2: Neural network model construction and training Constructing a neural network model and training the neural network model using a pre-collected historical data set containing patient movement data and corresponding rehabilitation effect evaluation results, so that the neural network model can accurately evaluate the patient's movement status and rehabilitation effect based on the input movement data; Step S3: Real-time motion planning During the rehabilitation process, the real-time collected patient motion data is input into the trained neural network model to obtain the patient's motion status and rehabilitation effect evaluation results. Combined with the patient's individual characteristics and rehabilitation needs, an individualized rehabilitation exercise plan is generated based on the preset motion planning algorithm; Step S4: Execution control The generated rehabilitation motion plan is converted into control instructions for the rehabilitation robot to control the rehabilitation robot to perform corresponding motion actions.

2. The neural network-based rehabilitation robot motion planning method according to claim 1, characterized in that: In step S1, the sensors include an accelerometer, a gyroscope, and an electromyographic sensor, and after collecting data, preprocessing operations are performed on the collected motion data, and the preprocessing operations include filtering and normalization processing to improve the quality and usability of the data.

3. The neural network-based rehabilitation robot motion planning method according to claim 1, characterized in that: The neural network model is a combination of one or more of a convolutional neural network, a recurrent neural network, a long short-term memory network or a gated recurrent unit.

4. The neural network-based rehabilitation robot motion planning method according to claim 1, characterized in that: In step S3, the individual characteristics of the patient include age, physical condition, and rehabilitation stage. The rehabilitation needs are determined according to the patient's rehabilitation goals. The motion planning algorithm adopts an optimization method based on reinforcement learning, genetic algorithm, or particle swarm optimization algorithm to generate an optimal individualized rehabilitation motion plan.

5. The neural network-based rehabilitation robot motion planning method according to claim 1, characterized in that: It also includes a feedback adjustment step, which monitors the patient's movement status and rehabilitation effect in real time during the rehabilitation robot's execution of movement movements, inputs the feedback information into the neural network model, dynamically adjusts and optimizes the neural network model, and adjusts the rehabilitation movement plan in real time according to the feedback information.

6. The neural network-based rehabilitation robot motion planning method according to claim 5, characterized in that: The feedback information includes changes in physiological indicators of the patient during exercise, exercise completion, and improvement in rehabilitation effects. The changes in physiological indicators include heart rate, blood pressure, and electromyographic signal strength. The improvement in rehabilitation effects is obtained by comparing with preset rehabilitation goals.

7. The neural network-based rehabilitation robot motion planning method according to claim 1, characterized in that: The rehabilitation motion plan includes the motion trajectory, motion speed, motion amplitude and motion duration parameters of the rehabilitation robot. When generating the rehabilitation motion plan, the motion planning algorithm comprehensively considers the patient's motion ability, rehabilitation goals and the motion performance limitations of the rehabilitation robot.

8. A neural network-based rehabilitation robot motion planning method according to any one of claims 1 to 7, characterized in that: The method also includes a multimodal information fusion step, which comprehensively considers the patient's physiological indicators, movement status, rehabilitation effect and psychological state multimodal information when generating a rehabilitation exercise plan. The psychological state is obtained through the patient's expression, voice or specific psychological assessment scale.

Images