A method and system for upper limb rehabilitation robot training based on doctor-patient interaction behavior learning and LSTM generalization

Abstract

The application discloses a kind of upper limb rehabilitation robot training method and system based on doctor-patient interactive behavior learning and LSTM generalization, method includes: obtaining the dynamic sensation and force sensation data of physician, robot and patient in hand-to-hand movement rehabilitation skill demonstration process;Based on the dynamic sensation and force sensation data of physician, robot and patient in hand-to-hand movement rehabilitation skill demonstration process, make robot perceive the dynamic sensation and force sensation behavior of doctor-patient interaction, form doctor-patient interaction behavior observation sequence;Based on dynamic motion primitive model, learn the doctor-patient interaction behavior in hand-to-hand movement rehabilitation skill demonstration process, obtain doctor-patient compliant interaction primitive sequence;According to the doctor-patient compliant interaction primitive sequence learned in hand-to-hand movement rehabilitation skill demonstration process, construct movement rehabilitation skill training data set, train LSTM network using the movement rehabilitation skill training data set, realize the generalization of the movement rehabilitation skill of upper limb rehabilitation robot under the condition of patient condition change.

Description

Technical Field

[0001] This invention belongs to the field of upper limb rehabilitation robot technology, and relates to an upper limb rehabilitation robot training method and system based on doctor-patient interaction behavior learning and LSTM generalization. Background Technology

[0002] Patients suffering from upper limb motor dysfunction due to neurological diseases such as stroke not only severely impact their daily lives but also impose a heavy burden on society. Traditional hands-on training methods based on rehabilitation physicians are not only inefficient and insufficient in intensity, but their therapeutic effects and rehabilitation evaluations are also easily constrained by the physician's subjective clinical experience. To address the numerous problems existing in traditional motor rehabilitation therapy, relevant research institutions at home and abroad have been competing to conduct research on rehabilitation robot technology in recent years, achieving some representative results. In robot-assisted rehabilitation training systems, designing compliant, intelligent, and inclusive human-machine collaborative rehabilitation strategies is the fundamental guarantee for achieving natural and harmonious human-machine interaction.

[0003] Most existing robot-assisted rehabilitation training strategies only establish human-machine collaboration within the patient's environment. Users need to use prior knowledge of human-machine dynamics and rehabilitation medicine to perform traditional robot control programming. When faced with unstructured interactive environments such as patients with different rehabilitation conditions or different anthropometry, the ability to adapt to new environments is obviously insufficient. Summary of the Invention

[0004] Objective: In view of at least one of the above technical problems, the present invention provides a training method and system for upper limb rehabilitation robots based on doctor-patient interaction behavior learning and LSTM generalization, which improves the compliance of rehabilitation robot motor assistive behavior through robot skill learning.

[0005] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present invention provides an upper limb rehabilitation robot training method, comprising:

[0007] Step S1: Construct a physician-robot-patient hands-on motor rehabilitation training demonstration experimental system for robot-assisted upper limb rehabilitation training, and acquire kinesiological and force sensory data of physician, robot and patient during hands-on motor rehabilitation skill demonstration; wherein the kinesiological and force sensory data of physician and patient include limb movement posture and surface electromyography signals; the kinesiological and force sensory data of robot include robot end effector movement posture and end effector torque;

[0008] Step S2: Based on the kinesthetic and force sensory data of the doctor, robot and patient during the hands-on motor rehabilitation skills demonstration, enable the robot to perceive the kinesthetic and force sensory behaviors of doctor-patient interaction and form a doctor-patient interaction behavior observation sequence.

[0009] Step S3: Based on the dynamic motion primitive model, learn the doctor-patient interaction behavior during the hands-on exercise rehabilitation skill demonstration to obtain the doctor-patient compliant interaction primitive sequence;

[0010] Step S4: Based on the patient-doctor compliant interaction primitive sequence learned during the hands-on demonstration of motor rehabilitation skills, construct a motor rehabilitation skills training dataset, and use the motor rehabilitation skills training dataset to train an LSTM network to achieve the generalization of the motor rehabilitation skills of the upper limb rehabilitation robot under changes in the patient's condition.

[0011] In some embodiments, step S1 involves constructing a physician-robot-patient hands-on motor rehabilitation training demonstration experimental system for robot-assisted upper limb rehabilitation training, and acquiring kinesthetic and force sensory data from the physician, robot, and patient during the hands-on motor rehabilitation skill demonstration, including:

[0012] Step S11: The physician-robot-patient hands-on exercise rehabilitation training demonstration experimental system includes: a virtual rehabilitation training environment, a multi-degree-of-freedom compliant robotic arm, a six-dimensional force / torque sensor, an inertial measurement unit, and a surface electromyography sensor; wherein, the six-dimensional force / torque sensor is installed at the end of the robotic arm, and the inertial measurement unit and surface electromyography sensor are worn on the upper arm and forearm of the physician and patient, respectively.

[0013] Step S12: Adjust the robot's gravity compensation coefficient to put the robot in a fully gravity-compensated state. After full gravity compensation, the robot, dragged by the physician, pulls the patient attached to the end of the robot to complete different types of motor rehabilitation training skill demonstrations in the virtual rehabilitation training environment.

[0014] Step S13: Acquire the motion posture collected by the inertial measurement unit, the surface electromyography signals of the patient's and doctor's limbs collected by the surface electromyography unit, and the end effector torque collected by the robot's end effector and the six-dimensional torque sensor during the hands-on motor rehabilitation skill demonstration.

[0015] In some embodiments, step S2, based on the kinesthetic and force sensory data of the physician, robot, and patient during the hands-on motor rehabilitation skill demonstration, enables the robot to perceive the kinesthetic and force sensory behaviors of doctor-patient interaction, forming a doctor-patient interaction behavior observation sequence, including:

[0016] The kinesthetic behaviors of doctor-patient interaction include the rotation angles and angular velocities of the shoulder and elbow joints, which are obtained through pose calculations collected by the inertial measurement unit. Specifically, the poses include the inertial poses of the upper arm and the forearm. The rotation angles and angular velocities of the shoulder joints are calculated based on the inertial poses of the upper arm and the elbow joints are calculated based on the inertial poses of the upper arm and the forearm.

[0017] Force-sensory behaviors in doctor-patient interactions include limb biostiffness and damping, which are estimated using surface electromyography (EMG) signals collected by surface EMG units. Specifically, this includes:

[0018] The biomechanical stiffness and damping of the patient's and doctor's limbs are estimated based on the electromyographic signals on the limb surface.

[0019] Furthermore, the biomechanical stiffness and damping of the patient's and doctor's limbs are estimated based on the electromyographic signals on the limb surface, including:

[0020] Step S201: Offline identification process of biomechanical stiffness and damping of the patient's limbs: While keeping the subject's limbs in a constant posture, a disturbance torque is applied to the limbs and electromyographic signals on the limb surface are collected simultaneously; according to the quasi-linear dynamic equation shown in equation (1), the biomechanical stiffness and damping of the patient's limbs are estimated offline using the least squares method with a forgetting factor:

[0021]

[0022] In the formula, ξ(t) represents the surface electromyography, angle, and angular velocity vectors of the limb; B(ξ(t)) and K(ξ(t)) represent the limb biodamping and stiffness, respectively. and T(t) represents the angular displacement and angular velocity of the limb joint; T(t) represents the joint torque, which is calculated from the robot's end effector torque.

[0023] Step S202: Online identification process of biomechanical stiffness and damping of the patient's limbs:

[0024] Stiffness radial basis network model and damping radial basis network model are constructed respectively for online estimation of the biostiffness and damping of the limbs of medical staff and patients;

[0025] The stiffness radial basis function network model is trained offline by using the limb surface electromyography signals generated offline as input and the corresponding limb bio-stiffness as output; the damping radial basis function network model is trained offline by using the limb surface electromyography signals generated offline as input and the corresponding limb bio-damping as output.

[0026] Based on the collected electromyographic signals from the limb surface, the bio-stiffness of the doctor and patient's limbs is estimated online using a trained stiffness radial basis function network model, and the bio-damping of the doctor and patient's limbs is estimated online using a trained damping radial basis function network model.

[0027] In some embodiments, a sequence of doctor-patient interaction behavior observations is formed, including:

[0028] The doctor-patient interaction behavior observation sequence is K is the number of sequences, where the j-th sequence is:

[0029]

[0030] in and These represent the patient's condition constraints during the j-th rehabilitation skill demonstration: patient's movement posture, patient's limb stiffness, and damping, respectively.

[0031] and These represent the robot end effector pose, physician limb biomechanical stiffness, and damping, respectively, during the demonstration of physician-assisted rehabilitation using a traction robot, under the constraints of the current patient's condition. T is the length of the observation sequence.

[0032] In some embodiments, the doctor-patient interaction behavior during a hands-on demonstration of motor rehabilitation skills is learned based on a dynamic motion primitive model, resulting in a doctor-patient compliant interaction primitive sequence, including:

[0033] Step S32: Based on the dynamic motion primitive model, learn the doctor-patient kinesthetic interaction behavior during the hands-on demonstration of motor rehabilitation skills, and construct doctor-patient kinesthetic interaction primitives;

[0034] The dynamic motion primitive model is represented as:

[0035]

[0036]

[0037]

[0038] The linear component determines the basic shape of the rehabilitation training trajectory and ensures that the trajectory y converges to the expected value g; the nonlinear term f(s) is defined as the radial basis function. The weighted average is used to adjust the motion trajectory while ensuring shape similarity; y0 is the initial position of the motion trajectory, τ is the system time constant, s is the system phase, and α is the system phase. z and β z represents a constant; L represents the number of radial basis functions. c represents the weight of the radial basis function of motion in the doctor-patient kinesthetic interaction primitive. m with h m Let represent the mean and variance of the radial basis functions, respectively.

[0039] Equation (5) is further expressed as a second-order system, and let Calculate primitive parameters for doctor-patient kinesthetic interaction The learning target is nonlinear terms

[0040]

[0041] After aligning the observation sequence of doctor-patient interaction behavior using dynamic time warping and learning the parameters of doctor-patient kinesthetic interaction primitives using local weighted regression, and incorporating the pose constraints of the affected limb, the doctor-patient kinesthetic interaction primitives are obtained, represented as follows: It is used to illustrate the motion interaction process between "patient's motion posture constraints and physician's corresponding motion assistance";

[0042] Step S33: Based on the method of constructing the target nonlinear term f(s) of the doctor-patient kinesthetic interaction primitive, and based on the radial basis function... The physician's limb biomechanical stiffness and damping are encoded in the following manner to construct the doctor-patient force perception interaction primitive;

[0043]

[0044]

[0045] in These represent the weights of the stiffness and damping radial basis functions in the doctor-patient force-sensory interaction primitive, respectively.

[0046] Define the system phase final value s t Primarily to ensure human-computer interaction safety, during subsequent behavior reproduction, when the robot's end-effector trajectory exceeds the expected value g, the robot's end-effector impedance is maintained at K_therapist(s). t ) and B_therapist(a t )constant;

[0047] Learn doctor-patient force-sensory interaction primitive parameters according to the doctor-patient kinesthetic interaction primitive parameter learning process. Then, by incorporating muscle strength constraints of the affected limb, a doctor-patient force-sensory interaction primitive is obtained, represented as... It is used to illustrate the compliant interaction between “muscle weakness of the affected limb and corresponding force assistance from the physician”, expressed in terms of the stiffness and damping of the patient’s and physician’s limbs.

[0048] Step S34: Integrate the doctor-patient kinesthetic interaction primitives and doctor-patient force-sensory interaction primitives to obtain a sequence of doctor-patient compliant interaction primitives generated based on doctor-patient interaction behavior learning, wherein the doctor-patient compliant interaction primitive z CIP Represented as:

[0049]

[0050] Step S35: Using the physician's motor assistance and force assistance parameters learned in step S34, construct a human-machine compliant interaction interface based on equations (5), (7) and (8) and the admittance control method to realize the upper limb rehabilitation robot's motor rehabilitation assistance training for patients.

[0051] In some embodiments, step S4, constructing a motor rehabilitation skill training dataset based on the patient-doctor compliant interaction primitive sequence learned during the hands-on motor rehabilitation skill demonstration, and training an LSTM network using the motor rehabilitation skill training dataset to achieve generalization of the upper limb rehabilitation robot's motor rehabilitation skills under changing patient conditions, includes:

[0052] Step S41: Extract the patient's condition parameter ε from the doctor-patient compliant interaction primitive sequence. patient ={P_patient,K_patient,B_patient} as input, physician exercise rehabilitation auxiliary parameters As output, construct a training dataset η={ε patient ,θ assist};

[0053] Step S42: Using the training dataset η={ε patient ,θ assist Train the LSTM network to obtain the trained LSTM network model;

[0054] Step S43: When the patient's condition changes within a certain period of time, the parameters of the human-computer compliant interaction interface are adjusted through the trained LSTM network model, thereby providing motor rehabilitation assistance adapted to the patient's new condition and realizing the generalization of the upper limb rehabilitation robot's motor rehabilitation skills.

[0055] In a second aspect, the present invention provides an upper limb rehabilitation robot training system, including a processor and a storage medium;

[0056] The storage medium is used to store instructions;

[0057] The processor is configured to operate according to the instructions to perform the steps of the method according to the first aspect.

[0058] Thirdly, the present invention provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0059] Fourthly, the present invention provides a computing device, comprising:

[0060] One or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described in the first aspect.

[0061] Beneficial Effects: The upper limb rehabilitation robot training method and system based on doctor-patient interaction behavior learning and LSTM generalization provided by this invention have the following advantages: The upper limb rehabilitation robot directly learns the compliant interaction behavior between doctors and patients from the hands-on demonstration of motor rehabilitation skills. On the one hand, the learning of doctor-patient interaction behavior overcomes the fact that most existing rehabilitation robot learning methods only learn the doctor's auxiliary behavior and ignore the patient's corresponding limb behavior learning during the training process. On the other hand, the learning of the compliantness of doctor-patient interaction not only contains the limb movement information of doctor-patient interaction in the traditional rehabilitation training process, but also contains the limb interaction force information of doctor-patient interaction. This makes up for the fact that most existing rehabilitation robot learning methods only learn movement information and lack the learning of force information, resulting in insufficient compliantness of the rehabilitation robot's motor auxiliary behavior.

[0062] After learning compliant interaction primitive sequences that characterize the interaction between doctors and patients during motor rehabilitation, the upper limb rehabilitation robot further extracts the parameters of the compliant interaction primitive sequences to train an LSTM network, thereby achieving the generalization of motor rehabilitation skills. On the one hand, the introduction of the LSTM network with memory function can adjust the motor rehabilitation assistance according to the changes in the patient's condition within a certain time range, overcoming the shortcomings of existing robot motor rehabilitation assistance behaviors that are mostly adjusted based on the patient's instantaneous condition. On the other hand, extracting primitive parameters containing doctor-patient interaction behaviors from the learned compliant interaction primitive sequences to train the LSTM network is closer to the essence of motor rehabilitation skills than directly extracting the dataset from the original doctor-patient interaction behavior observation sequences. Attached Figure Description

[0063] Figure 1 This is a schematic diagram of the training method for an upper limb rehabilitation robot based on doctor-patient interaction behavior learning and LSTM generalization according to an embodiment of the present invention.

[0064] Figure 2 This is a schematic diagram of the hands-on exercise rehabilitation skills demonstration system in the embodiment;

[0065] Figure 3 This is a schematic diagram of the motion and force perception behavior sensing scheme for doctor-patient interaction in the embodiment.

[0066] Figure 4 This is a schematic diagram of the LSTM network training structure based on the primitive parameter set in the embodiment. Detailed Implementation

[0067] The present invention will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be used to limit the scope of protection of the present invention.

[0068] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0069] In the description of this invention, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0070] Example 1

[0071] A training method for an upper limb rehabilitation robot based on doctor-patient interaction behavior learning and LSTM generalization includes:

[0072] Step S1: Construct a physician-robot-patient hands-on motor rehabilitation training demonstration experimental system for robot-assisted upper limb rehabilitation training, and acquire kinesiological and force sensory data of physician, robot and patient during hands-on motor rehabilitation skill demonstration; wherein the kinesiological and force sensory data of physician and patient include limb movement posture and surface electromyography signals; the kinesiological and force sensory data of robot include robot end effector movement posture and end effector torque;

[0073] Step S2: Based on the kinesthetic and force sensory data of the doctor, robot and patient during the hands-on motor rehabilitation skills demonstration, enable the robot to perceive the kinesthetic and force sensory behaviors of doctor-patient interaction and form a doctor-patient interaction behavior observation sequence.

[0074] Step S3: Based on the dynamic motion primitive model, learn the doctor-patient interaction behavior during the hands-on exercise rehabilitation skill demonstration to obtain the doctor-patient compliant interaction primitive sequence;

[0075] Step S4: Based on the patient-doctor compliant interaction primitive sequence learned during the hands-on demonstration of motor rehabilitation skills, construct a motor rehabilitation skills training dataset, and use the motor rehabilitation skills training dataset to train an LSTM network to achieve the generalization of the motor rehabilitation skills of the upper limb rehabilitation robot under changes in the patient's condition.

[0076] In some embodiments, step S1 involves constructing a physician-robot-patient hands-on motor rehabilitation training demonstration experimental system for robot-assisted upper limb rehabilitation training, and acquiring kinesthetic and force sensory data from the physician, robot, and patient during the hands-on motor rehabilitation skill demonstration, including:

[0077] Step S11: The physician-robot-patient hands-on exercise rehabilitation training demonstration experimental system includes: a virtual rehabilitation training environment, a multi-degree-of-freedom compliant robotic arm, a six-dimensional force / torque sensor, an inertial measurement unit, and a surface electromyography sensor; wherein, the six-dimensional force / torque sensor is installed at the end of the robotic arm, and the inertial measurement unit and surface electromyography sensor are worn on the upper arm and forearm of the physician and patient, respectively.

[0078] Step S12: Adjust the robot's gravity compensation coefficient to put the robot in a fully gravity-compensated state. After full gravity compensation, the robot, dragged by the physician, pulls the patient attached to the end of the robot to complete different types of motor rehabilitation training skill demonstrations in the virtual rehabilitation training environment.

[0079] Step S13: Acquire the motion posture collected by the inertial measurement unit, the surface electromyography signals of the patient's and doctor's limbs collected by the surface electromyography unit, and the end effector torque collected by the robot's end effector and the six-dimensional torque sensor during the hands-on motor rehabilitation skill demonstration.

[0080] In some embodiments, step S2, based on the kinesthetic and force sensory data of the physician, robot, and patient during the hands-on motor rehabilitation skill demonstration, enables the robot to perceive the kinesthetic and force sensory behaviors of doctor-patient interaction, forming a doctor-patient interaction behavior observation sequence, including:

[0081] Step S21: The kinesthetic behavior of doctor-patient interaction includes the rotation angle and angular velocity of the shoulder and elbow joints, which are obtained by the pose calculation collected by the inertial measurement unit. Specifically, the pose includes the inertial pose of the upper arm and the inertial pose of the forearm. The rotation angle and angular velocity of the shoulder joint are calculated based on the inertial pose of the upper arm, and the rotation angle and angular velocity of the elbow joint are calculated based on the inertial poses of the upper arm and forearm.

[0082] Force-sensory behaviors in doctor-patient interactions include limb biostiffness and damping, which are estimated using surface electromyography (EMG) signals collected by surface EMG units. Specifically, this includes:

[0083] The biomechanical stiffness and damping of the patient's and doctor's limbs are estimated based on the electromyographic signals on the limb surface.

[0084] Furthermore, the biomechanical stiffness and damping of the patient's and doctor's limbs are estimated based on the electromyographic signals on the limb surface, including:

[0085] Step S22: Offline identification process of biomechanical stiffness and damping of the patient's limbs: While keeping the subject's limbs in a constant posture, a disturbance torque is applied to the limbs and electromyographic signals on the limb surface are collected simultaneously; according to the quasi-linear dynamic equation shown in equation (1), the biomechanical stiffness and damping of the patient's limbs are estimated offline using the least squares method with a forgetting factor:

[0086]

[0087] In the formula, ξ(t) represents the surface electromyography, angle, and angular velocity vectors of the limb; B(ξ(t)) and K(ξ(t)) represent the limb biodamping and stiffness, respectively. and T(t) represents the angular displacement and angular velocity of the limb joint; T(t) represents the joint torque, which is calculated from the robot's end effector torque.

[0088] Step S23: Online identification process of biomechanical stiffness and damping of the patient's limbs:

[0089] Stiffness radial basis network model and damping radial basis network model are constructed respectively for online estimation of the biostiffness and damping of the limbs of medical staff and patients;

[0090] The stiffness radial basis function network model is trained offline by using the limb surface electromyography signals generated offline as input and the corresponding limb bio-stiffness as output; the damping radial basis function network model is trained offline by using the limb surface electromyography signals generated offline as input and the corresponding limb bio-damping as output.

[0091] Based on the collected electromyographic signals from the limb surface, the bio-stiffness of the doctor and patient's limbs is estimated online using a trained stiffness radial basis function network model, and the bio-damping of the doctor and patient's limbs is estimated online using a trained damping radial basis function network model.

[0092] In some embodiments, a sequence of doctor-patient interaction behavior observations is formed, including:

[0093] The doctor-patient interaction behavior observation sequence is K is the number of sequences, where the j-th sequence is:

[0094]

[0095] in and These represent the patient's condition constraints during the j-th rehabilitation skill demonstration: patient's movement posture, patient's limb stiffness, and damping, respectively. and These represent the robot end effector pose, physician limb biomechanical stiffness, and damping, respectively, during the demonstration of physician-assisted rehabilitation using a traction robot, under the constraints of the current patient's condition. T is the length of the observation sequence.

[0096] In some embodiments, the doctor-patient interaction behavior during a hands-on demonstration of motor rehabilitation skills is learned based on a dynamic motion primitive model, resulting in a doctor-patient compliant interaction primitive sequence, including:

[0097] Step S32: Based on the dynamic motion primitive model, learn the doctor-patient kinesthetic interaction behavior during the hands-on demonstration of motor rehabilitation skills, and construct doctor-patient kinesthetic interaction primitives;

[0098] The dynamic motion primitive model is represented as:

[0099]

[0100]

[0101]

[0102] The linear component determines the basic shape of the rehabilitation training trajectory and ensures that the trajectory y converges to the expected value g; the nonlinear term f(s) is defined as the radial basis function. The weighted average is used to adjust the motion trajectory while ensuring shape similarity; y0 is the initial position of the motion trajectory, τ is the system time constant, s is the system phase, and α is the system phase. z and β z represents a constant; L represents the number of radial basis functions. c represents the weight of the radial basis function of motion in the doctor-patient kinesthetic interaction primitive. m with h m Let represent the mean and variance of the radial basis functions, respectively.

[0103] Equation (5) is further expressed as a second-order system, and let Calculate primitive parameters for doctor-patient kinesthetic interaction The learning target is nonlinear terms

[0104]

[0105] After aligning the observation sequence of doctor-patient interaction behavior using dynamic time warping and learning the parameters of doctor-patient kinesthetic interaction primitives using local weighted regression, and incorporating the pose constraints of the affected limb, the doctor-patient kinesthetic interaction primitives are obtained, represented as follows: It is used to illustrate the motion interaction process between "patient's motion posture constraints and physician's corresponding motion assistance";

[0106] Step S33: Based on the method of constructing the target nonlinear term f(s) of the doctor-patient kinesthetic interaction primitive, and based on the radial basis function... The physician's limb biomechanical stiffness and damping are encoded in the following manner to construct the doctor-patient force perception interaction primitive;

[0107]

[0108]

[0109] in These represent the weights of the stiffness and damping radial basis functions in the doctor-patient force-sensory interaction primitive, respectively.

[0110] Define the system phase final value s t Primarily to ensure human-computer interaction safety, during subsequent behavior reproduction, when the robot's end-effector trajectory exceeds the expected value g, the robot's end-effector impedance is maintained at K_therapist(s). t ) and B_therapist(s t )constant;

[0111] Learn doctor-patient force-sensory interaction primitive parameters according to the doctor-patient kinesthetic interaction primitive parameter learning process. Then, by incorporating muscle strength constraints of the affected limb, a doctor-patient force-sensory interaction primitive is obtained, represented as... It is used to illustrate the compliant interaction between “muscle weakness of the affected limb and corresponding force assistance from the physician”, expressed in terms of the stiffness and damping of the patient’s and physician’s limbs.

[0112] Step S34: Integrate the doctor-patient kinesthetic interaction primitives and doctor-patient force-sensory interaction primitives to obtain a sequence of doctor-patient compliant interaction primitives generated based on doctor-patient interaction behavior learning, wherein the doctor-patient compliant interaction primitive z CIP Represented as:

[0113]

[0114] Step S35: Using the physician's motor assistance and force assistance parameters learned in step S34, construct a human-machine compliant interaction interface based on equations (5), (7) and (8) and the admittance control method to realize the upper limb rehabilitation robot's motor rehabilitation assistance training for patients.

[0115] In some embodiments, step S4, constructing a motor rehabilitation skill training dataset based on the patient-doctor compliant interaction primitive sequence learned during the hands-on motor rehabilitation skill demonstration, and training an LSTM network using the motor rehabilitation skill training dataset to achieve generalization of the upper limb rehabilitation robot's motor rehabilitation skills under changing patient conditions, includes:

[0116] Step S41: Extract the patient's condition parameter ε from the doctor-patient compliant interaction primitive sequence. patient ={P_patient,K_patient,B_patient} as input, physician exercise rehabilitation auxiliary parameters As output, construct a training dataset η={ε patient ,θassist};

[0117] Step S42: Using the training dataset η={ε patient ,θ assist Train the LSTM network to obtain the trained LSTM network model;

[0118] Step S43: When the patient's condition changes within a certain period of time, the parameters of the human-computer compliant interaction interface are adjusted through the trained LSTM network model, thereby providing motor rehabilitation assistance adapted to the patient's new condition and realizing the generalization of the upper limb rehabilitation robot's motor rehabilitation skills.

[0119] Example 2

[0120] Secondly, based on Embodiment 1, this embodiment provides an upper limb rehabilitation robot training system, including a processor and a storage medium;

[0121] The storage medium is used to store instructions;

[0122] The processor is configured to operate according to the instructions to perform the steps of the method according to Embodiment 1.

[0123] In some embodiments, an upper limb rehabilitation robot training system based on doctor-patient interaction behavior learning and LSTM generalization includes:

[0124] Example 3

[0125] Thirdly, based on Embodiment 1, this embodiment provides a storage medium on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the method described in Embodiment 1.

[0126] Example 4

[0127] Fourthly, based on Embodiment 1, the present invention provides a computing device, comprising:

[0128] One or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described in Embodiment 1.

[0129] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0130] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0133] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An upper limb rehabilitation robot training system, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the upper limb rehabilitation robot training method, the steps including: Step S1: Construct a physician-robot-patient hands-on motor rehabilitation training demonstration experimental system for robot-assisted upper limb rehabilitation training, and acquire kinesthetic and force data of physician, robot and patient during the hands-on motor rehabilitation skill demonstration process; wherein the kinesthetic and force data of physician and patient include limb movement posture and surface electromyography signals; the kinesthetic and force data of robot include robot end-effector movement posture and end-effector torque; the hands-on motor rehabilitation skill demonstration process is to adjust the robot's gravity compensation coefficient to make the robot in a fully gravity-compensated state, and after full gravity compensation, the robot, dragged by the physician, pulls the patient attached to the robot's end to complete different types of motor rehabilitation training skill demonstrations in a virtual rehabilitation training environment; Step S2: Based on the kinesthetic and force sensory data of the doctor, robot and patient during the hands-on motor rehabilitation skills demonstration, enable the robot to perceive the kinesthetic and force sensory behaviors of doctor-patient interaction and form a doctor-patient interaction behavior observation sequence. Step S3: Based on the dynamic motion primitive model, learn the observation sequence of doctor-patient interaction behavior during the hands-on demonstration of motor rehabilitation skills to obtain the doctor-patient compliant interaction primitive sequence; Step S4: Based on the patient-doctor compliant interaction primitive sequence learned during the hands-on demonstration of motor rehabilitation skills, construct a motor rehabilitation skills training dataset, and use the motor rehabilitation skills training dataset to train an LSTM network to achieve the generalization of the motor rehabilitation skills of the upper limb rehabilitation robot under changes in the patient's condition.

2. The upper limb rehabilitation robot training system according to claim 1, characterized in that, Step S1: Construct a physician-robot-patient hands-on motor rehabilitation training demonstration experimental system for robot-assisted upper limb rehabilitation training, and acquire kinesthetic and force sensory data of the physician, robot, and patient during the hands-on motor rehabilitation skill demonstration, including: Step S11: The physician-robot-patient hands-on exercise rehabilitation training demonstration experimental system includes: a virtual rehabilitation training environment, a multi-degree-of-freedom compliant robotic arm, a six-dimensional force / torque sensor, an inertial measurement unit, and a surface electromyography sensor; wherein, the six-dimensional force / torque sensor is installed at the end of the robotic arm, and the inertial measurement unit and surface electromyography sensor are worn on the upper arm and forearm of the physician and patient, respectively. Step S12: Demonstrate hands-on exercise rehabilitation skills; Step S13: Acquire the motion posture collected by the inertial measurement unit, the surface electromyography signals of the patient's and doctor's limbs collected by the surface electromyography unit, and the end effector torque collected by the robot's end effector and the six-dimensional torque sensor during the hands-on motor rehabilitation skill demonstration.

3. The upper limb rehabilitation robot training system according to claim 2, characterized in that, Step S2: Based on the kinesthetic and force sensory data of the physician, robot, and patient during the hands-on demonstration of motor rehabilitation skills, enable the robot to perceive the kinesthetic and force sensory behaviors of doctor-patient interaction, forming an observation sequence of doctor-patient interaction behaviors, including: The kinesthetic behaviors of doctor-patient interaction include the rotation angles and angular velocities of the shoulder and elbow joints, which are obtained through pose calculations collected by the inertial measurement unit. Specifically, the poses include the inertial poses of the upper arm and the forearm. The rotation angles and angular velocities of the shoulder joints are calculated based on the inertial poses of the upper arm and the elbow joints are calculated based on the inertial poses of the upper arm and the forearm. The force perception behavior of doctor-patient interaction includes limb biostiffness and damping, which are estimated by the limb surface electromyography signals collected by the surface electromyography unit. Specifically, it includes estimating the biostiffness and damping of the doctor-patient limbs based on the limb surface electromyography signals.

4. The upper limb rehabilitation robot training system according to claim 3, characterized in that, The biomechanical stiffness and damping of the patient's and doctor's limbs are estimated based on the electromyographic signals on the limb surface, including: Step S201: Offline identification process of biomechanical stiffness and damping of the patient's limbs: While keeping the subject's limbs in a constant posture, a disturbance torque is applied to the limbs and electromyographic signals on the limb surface are collected simultaneously; according to the quasi-linear dynamic equation shown in equation (1), the biomechanical stiffness and damping of the patient's limbs are estimated offline using the least squares method with a forgetting factor: (1)； In the formula For surface electromyography, angle, and angular velocity vectors of the limb; and These are limb biodamping and stiffness, respectively; and For angular displacement and angular velocity of limb joints; The joint torque is calculated from the robot's end effector torque. Step S202: Online identification process of biomechanical stiffness and damping of the patient's limbs: Stiffness radial basis network model and damping radial basis network model are constructed respectively for online estimation of the biostiffness and damping of the limbs of medical staff and patients; The stiffness radial basis function network model is trained offline by using the limb surface electromyography signals generated offline as input and the corresponding limb bio-stiffness as output; the damping radial basis function network model is trained offline by using the limb surface electromyography signals generated offline as input and the corresponding limb bio-damping as output. Based on the collected electromyographic signals from the limb surface, the bio-stiffness of the doctor and patient's limbs is estimated online using a trained stiffness radial basis function network model, and the bio-damping of the doctor and patient's limbs is estimated online using a trained damping radial basis function network model.

5. The upper limb rehabilitation robot training system according to claim 1, characterized in that, A sequence of observations of doctor-patient interactions was formed, including: The doctor-patient interaction behavior observation sequence is , Let be the number of sequences, where the th is the th . The sequences are: (2)； in They represent the first During a rehabilitation skills demonstration, under the constraints of the patient's condition, the patient's movement posture, limb stiffness, and damping were observed during the doctor-assisted rehabilitation demonstration using a traction robot. These represent the robot's end effector posture, the physician's limb biomechanical stiffness, and damping, respectively, during a demonstration of a physician-assisted rehabilitation robot under the constraints of the patient's current condition. The length of the observation sequence.

6. The upper limb rehabilitation robot training system according to claim 5, characterized in that, Based on the dynamic motion primitive model, the observed sequence of doctor-patient interaction behaviors during a hands-on demonstration of motor rehabilitation skills was learned, resulting in a doctor-patient compliance interaction primitive sequence, including: Step S32: Based on the dynamic motion primitive model, learn the observation sequence of doctor-patient kinesthetic interaction behavior during the hands-on demonstration of motor rehabilitation skills, and construct doctor-patient kinesthetic interaction primitives; The dynamic motion primitive model is represented as: ， (3)； (4)； (5)； The linear component determines the basic shape of the rehabilitation training movement trajectory and ensures the trajectory. It can converge to the expected value Nonlinear terms Defined as radial basis function The weighted average is used to adjust the motion trajectory while ensuring shape similarity. The initial position of the motion trajectory, The system time constant, For the system phase, and Represents a constant; This indicates the number of radial basis functions. This represents the weights of the radial basis functions of motion in the doctor-patient kinesthetic interaction primitives. and Let represent the mean and variance of the radial basis functions, respectively. Equation (5) is further expressed as a second-order system, and let , Calculate the primitive parameters used for doctor-patient kinesthetic interaction The learning target is nonlinear terms , of which elements The expression is as follows: (6)； After aligning the observation sequence of doctor-patient interaction behavior using dynamic time warping and learning the parameters of doctor-patient kinesthetic interaction primitives using local weighted regression, and incorporating the pose constraints of the affected limb, the doctor-patient kinesthetic interaction primitives are obtained, represented as follows: This is used to illustrate the motion interaction process between "patient's motion posture constraints and physician's corresponding motion assistance"; Step S33: Based on the target nonlinear term of the doctor-patient kinesthetic interaction primitive The construction method is based on radial basis functions. The physician's limb biomechanical stiffness and damping are encoded in the following manner to construct the doctor-patient force perception interaction primitive; (7)； (8)； in These represent the weights of the stiffness and damping radial basis functions in the doctor-patient force-sensory interaction primitive, respectively. Define the system phase final value Primarily to ensure safety in human-computer interaction, if the robot's end effector trajectory exceeds the expected value during subsequent behavior reproduction... At that time, the robot end effector impedance remains and constant; Learn doctor-patient force-sensory interaction primitive parameters according to the doctor-patient kinesthetic interaction primitive parameter learning process. Then, by incorporating muscle strength constraints of the affected limb, a doctor-patient force-sensory interaction primitive is obtained, represented as... It is used to illustrate the compliant interaction process between the muscle weakness of the affected limb, expressed in terms of the stiffness and damping of the patient's and doctor's limbs, and the corresponding force assistance of the physician. Step S34: Integrate the doctor-patient kinesthetic interaction primitives and doctor-patient force-sensory interaction primitives to obtain a sequence of doctor-patient compliant interaction primitives generated based on doctor-patient interaction behavior learning, wherein the doctor-patient compliant interaction primitives... Represented as: (9)； Step S35: Using the physician's motor assistance and force assistance parameters learned in step S34, construct a human-machine compliant interaction interface based on equations (5), (7) and (8) and the admittance control method to realize the upper limb rehabilitation robot's motor rehabilitation assistance training for patients.

7. The upper limb rehabilitation robot training system according to claim 6, characterized in that, Step S4: Based on the patient-doctor compliant interaction primitive sequences learned during the hands-on demonstration of motor rehabilitation skills, construct a motor rehabilitation skills training dataset. Use this dataset to train an LSTM network to generalize the motor rehabilitation skills of the upper limb rehabilitation robot under changing patient conditions, including: Step S41: Extract patient condition parameters from the doctor-patient compliant interaction primitive sequence. As input, physician exercise rehabilitation auxiliary parameters As output, a training dataset for generalizing motor rehabilitation skills of upper limb rehabilitation robots is constructed. ; Step S42: Using the training dataset Train the LSTM network to obtain the trained LSTM network model; Step S43: When the patient's condition changes within a certain period of time, the parameters of the human-computer compliant interaction interface are adjusted through the trained LSTM network model, thereby providing motor rehabilitation assistance adapted to the patient's new condition and realizing the generalization of the upper limb rehabilitation robot's motor rehabilitation skills.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the upper limb rehabilitation robot training method according to any one of claims 1 to 7.

9. A computing device, characterized in that, include: One or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, the one or more programs including instructions for performing the steps of the upper limb rehabilitation robot training method according to any one of claims 1 to 7.

Images