Double-arm robot simulation rehabilitation therapist traction training method for stroke upper limb rehabilitation

Abstract

The invention belongs to the technical field of double-arm / humanoid robot control, and discloses a double-arm robot simulation rehabilitation therapist traction training method for stroke upper limb rehabilitation. From the perspective of training safety, an 8-degree-of-freedom kinematics model of the upper limb complex is established and used for evaluating the reachable working space of the tail ends and the front arms of the arms in the interaction process with the double-arm robot. Based on the characteristics of double-arm rehabilitation, a non-redundant inverse kinematics method is provided to constrain joint angles, and a safety mechanism under double constraints is established. From the perspective of training scientificity and flexibility, a potential field control strategy is introduced, so that the robot can learn traction characteristics of a rehabilitation therapist from single demonstration. In combination with master-slave control, the two-arm robot can reproduce assistance of a rehabilitation therapist and allows compliance interaction. According to the invention, the double-arm / humanoid robot can perform therapist type traction training on to-be-recovered limbs, and safe, scientific and smooth rehabilitation training like therapists is provided in clinical and family environments.

Description

Technical Field

[0001] This invention relates to the field of dual-arm / humanoid robot control technology, specifically to a dual-arm robot-based traction training method for upper limb rehabilitation after stroke, mimicking a rehabilitation therapist. Background Technology

[0002] Upper limb rehabilitation robots are divided into two types: end effector-type and exoskeleton-type. The former uses an end effector to traction the patient's hand for training. In upper limb rehabilitation training, end effector robots exhibit advantages such as strong adaptability, ease of wear, safety, comfort, and a wide range of motion. However, end effector robots cannot precisely control the movement of each upper limb joint, especially in spatial training. Therefore, most current end effector robots are low-degree-of-freedom planar robots to ensure training safety. Exoskeletons, on the other hand, are worn directly on the upper limb, directly controlling the movement of each joint and providing precise rehabilitation. However, the high degree of freedom and flexibility of the upper limb leads to complex structure, control, and wearability of exoskeleton robots, weaker adaptability to different patients, and strong coupling that can generate additional interaction forces and torques. Therefore, how to combine the advantages of end effectors and exoskeleton rehabilitation robots while improving their disadvantages to provide highly adaptable, easy-to-wear, and precise spatial training for the limbs to be rehabilitated is a problem worthy of exploration.

[0003] In traditional upper limb rehabilitation, therapists use their arms to traction the limbs being rehabilitated for training. Inspired by this, the hypothesis has emerged that dual-armed or humanoid robots could replace the therapist's arms to provide patients with highly adaptive, safe, compliant, and precise training. Current research on dual-armed and humanoid robots mimicking human manipulation supports this hypothesis. However, current research focuses more on collaborative interactions between arms and objects or between arms, objects, and humans, while upper limb rehabilitation involves close robot-human physical interaction. Therefore, it is possible to explore the safety mechanisms, scientific principles, and compliance aspects of upper limb rehabilitation to achieve rehabilitation training similar to that of a robotic therapist. Summary of the Invention

[0004] To address the problems of existing end effector upper limb rehabilitation robot systems, such as the inability to precisely control individual joints, unsuitability for spatial training, and poor training adaptability and complex wearing of exoskeleton upper limb rehabilitation robots, this invention proposes a dual-arm robot-inspired rehabilitation therapist traction training method for upper limb rehabilitation after stroke.

[0005] The technical solution of this invention is: a dual-arm robot-based traction training method for upper limb rehabilitation after stroke, inspired by traditional rehabilitation therapist training methods. It uses a dual-arm robot to learn the traction movements and interaction characteristics of a rehabilitation therapist, respectively tractioning the hand and forearm posterior ends of the limb to be rehabilitated, to conduct scientific, compliant, and safe rehabilitation training in the style of a rehabilitation therapist. The dual-arm robot includes robotic arm A and robotic arm B. An 8-DOF high-precision upper limb kinematic model is constructed. Based on the characteristics of dual-arm robot traction rehabilitation, a dual safety constraint mechanism of task space plus joint space is established. A non-parametric potential field is designed to learn the traction movements and interaction characteristics of a rehabilitation therapist in a single traction demonstration. Based on the contribution of rehabilitation movements, the dual-arm robot is divided into master arms and slave arms for master-slave control, implementing master-slave compliant interaction. A dual-arm robot upper limb rehabilitation system is built to conduct dual-arm rehabilitation robot-based traction training tasks in the style of a rehabilitation therapist.

[0006] The aforementioned dual-arm robotic rehabilitation therapist-inspired traction training method for upper limb rehabilitation after stroke specifically includes the following steps:

[0007] Step 1: Measure the length of each segment of the limb to be rehabilitated, construct an 8-DOF high-precision upper limb kinematic model, and establish a dual safety constraint mechanism of task space plus joint space, taking into account the characteristics of dual-arm robot traction rehabilitation.

[0008] Step 2: Calculate the assistance required for rehabilitation training by having a rehabilitation therapist demonstrate traction to the dual-arm robot and the limb to be rehabilitated, and record the traction force, training trajectory, and interaction stiffness during the traction demonstration; design a nonparametric potential field function, and have the dual-arm robot demonstrate and learn from the traction demonstration; when the rehabilitation therapist is not present, the dual-arm robot provides the limb to be rehabilitated with the same assistance as during the traction demonstration, including equivalent traction force, training trajectory, and interaction stiffness, and provides compliant human-machine interaction;

[0009] Step 3: Based on the traction contribution of the rehabilitation therapist's arms to the limb to be rehabilitated, robotic arms A and B are classified as master arms and slave arms, respectively. Using the functional characteristics of master arms and slave arms, the task space motion is divided into absolute motion components and relative motion components. Absolute motion is used to perform large-range movements of the limb to be rehabilitated. Relative motion is used to receive the movement of the master arm to assist the rehabilitation training.

[0010] Step 1 specifically includes:

[0011] Step 1.1: Measure the acromiohumeral intercostal space, upper arm length, forearm length, and the distance from the wrist to the center of the palm of the limb to be rehabilitated. Divide the joints of the limb to be rehabilitated into a 1-DOF sternoclavicular joint, a 3-DOF shoulder joint, a 2-DOF elbow joint, and a 2-DOF wrist joint, and establish an 8-DOF high-precision upper limb kinematic model. Use the Monte Carlo method to calculate the reachable task space point set of the corresponding upper limb specific points of the dual-arm robot, and reconstruct the three-dimensional space of the reachable task space point set. Monitor the spatial position of specific points during the rehabilitation training process in real time to achieve task space constraints on the upper limb.

[0012] Step 1.2: Utilizing the characteristics of dual-arm robot traction rehabilitation, based on the specific points of the upper limb corresponding to the dual-arm robot, the 8-DOF upper limb kinematic model is divided into a 5-DOF subsystem and a 3-DOF subsystem;

[0013] The 5-DOF subsystem includes a 1-DOF sternoclavicular joint, a 3-DOF shoulder joint, and a 1-DOF elbow joint that performs flexion and extension movements; the 3-DOF subsystem includes a 1-DOF elbow joint and a 2-DOF wrist joint that perform forearm pronation and supination movements.

[0014] Numerical optimization methods for solving 5-DOF and 3-DOF subsystems were constructed to calculate the joint angles of the limb to be rehabilitated; changes in joint angles during rehabilitation training were monitored in real time to achieve spatial constraints on the joints of the upper limb.

[0015] In the 8-DOF upper limb kinematic model, the sternoclavicular joint with 1 DDOF allows for pronation and supination of the scapula; the ball joint of the shoulder with 3 DDOF allows for adduction / abduction, flexion / extension, and internal / external rotation of the arm; the elbow joint with 2 DDOF allows for flexion / extension and pronation / supination of the forearm; and the wrist joint with 2 DDOF allows for flexion / extension and adduction / abduction of the wrist.

[0016] Among them, the basic coordinate system Located at the sternoclavicular joint; local coordinate system Fixed at joint i, excluding the sternoclavicular joint; end-effector coordinate system Located at the center of the palm; an 8-DOF high-precision upper limb kinematic model is established according to the standard DH method; the transformation matrix from joint i to joint i-1. for:

[0017]

[0018] in, It is along direction and The angle between them; It is along direction and The distance between them; It is along direction and The distance between them; It is along direction and The included angle between them; symbol and These represent the sin and cos operations, respectively. It is a vector of joint variables; for the model established above, the coordinate point in the j-th coordinate system The expression relative to the base coordinate system is as follows:

[0019]

[0020] For the position corresponding to the end effector of robotic arm A The position corresponding to the end effector of robotic arm B The representation in the basic coordinate system is as follows:

[0021]

[0022] in, Indicates the distance from the elbow joint to p a2 The distance between them.

[0023] The task space constraints restrict the movement of specific segments of the upper limb to prevent non-physiological postures; based on an 8-DOF high-precision upper limb kinematic model, the Monte Carlo method is used to calculate p. a1 and p a2 Accessible task space:

[0024]

[0025] Here, E1 and E2 represent the solution point sets corresponding to the end effectors of robotic arm A and robotic arm B under the Monte Carlo method, respectively. Delaunay triangulation is performed on the solution point sets E1 and E2. A parameter X is defined, and triangles with side lengths less than 1 / X are filtered out. These triangles are then connected to generate the corresponding Alpha shape. The ray casting method is used to determine in real time whether the end effectors of robotic arm A and robotic arm B are within the created Alpha shape. If the result shows that both end effectors of robotic arm A and robotic arm B are within the created Alpha shape, it indicates that the specific segment points of the upper limb have not exceeded the reachable task space, and training continues. Otherwise, it indicates that the specific segment points of the upper limb have exceeded the reachable task space, and the rehabilitation robot stops training, thus achieving upper limb task space constraints during training.

[0026] The joint space constraint is achieved by dividing the 8-DOF kinematic model of the upper limb into a 5-DOF subsystem and a 3-DOF subsystem; the 5-DOF subsystem, point p a2 In the basic coordinate system H0, it is represented as:

[0027]

[0028] in, This indicates that the end effector is located at p a2 The transformation matrix from the base coordinate system where the robotic arm B is located to H0;

[0029] For a 3-degree-of-freedom subsystem, point p a1 In the local coordinate system H5, it is represented as:

[0030]

[0031] in, This indicates that the end effector is located at p a1 The transformation matrix from the base coordinate system where the robotic arm A is located to H0;

[0032] Based on the relationship between the end effectors of robotic arms A and B and H0 and H5, non-redundant kinematic solutions are performed for the 5-DOF subsystem and the 3-DOF subsystem, respectively.

[0033] Step 2 specifically includes:

[0034] Step 2.1: Wear end effectors on the hand and forearm, fix the upper limb to the end effector of the upper limb rehabilitation robot equipped with a six-dimensional force sensor, and drive the upper limb and the end effector of the rehabilitation robot to perform spatial movements through traction demonstration, while recording traction force, training trajectory and interaction stiffness;

[0035] Step 2.2: Design a nonparametric potential field and extract interactive features from the traction demonstration; learn the interactive features of the traction demonstration through a dual-arm robot, and provide equivalent assistance for upper limb movement by replacing the traction force, training trajectory and interactive stiffness of the rehabilitation therapist with the robot's traction force, training trajectory and interactive stiffness, and provide compliant human-computer interaction.

[0036] The traction demonstration uses robotic arms A and B instead of the rehabilitation therapist's arms; the nonlinear dynamics of the m-degree-of-freedom robotic arm A or B, whose end caps are connected to the limb to be rehabilitated, are represented as follows:

[0037]

[0038] in, Represents the joint angle vector; It is the robot's inertia matrix; It is centripetal force and Coriolis force. It is the gravitational torque vector; It is the joint input torque; The Jacobian matrix represents the associated task space and joint space; This represents the total resultant force applied in the task space; The interaction forces acting on the actuator can be broken down according to their mode of action. and control force in the control system During the traction demonstration phase, the rehabilitation therapist collaborates with the limb to be rehabilitated to complete the task, enabling... Equal to 0; the interaction force acting on the end effector Depend on , and The components represent the traction force applied by the rehabilitation therapist, the force applied to the limb to be rehabilitated, and the installation stress generated by the sensor; the dynamic relationship between force and posture during the traction demonstration is as follows:

[0039]

[0040] in, and These represent damping and stiffness properties, respectively; actual attitude. With the expected posture The difference between them represents the attitude deviation, and its derivative is... Indicates speed deviation; by... regarded as the desired posture It also accumulates incremental posture updates, with the end effector tracking the rehabilitation therapist's traction in both position and orientation:

[0041]

[0042] in, The desired pose for the next control cycle is represented by: The current position and its deviation are represented by: and The current direction and its deviation are respectively represented as and Velocity and pose deviation and The calculation is as follows:

[0043]

[0044]

[0045] in, To control the cycle; during the demonstration, the attitude and interaction forces by Frequency recording in Hz; downsampling the collected dataset to... A uniformly distributed set of samples, called attraction points, is used to construct the dataset. :

[0046]

[0047]

[0048] definition For the first A tangent-normal coordinate system for each attraction point; the tangent-normal coordinate system is based on the attraction point. Centered on the point of attraction, the tangent direction points towards the next point of attraction. The normal direction points to the current position of the end effector. Expected Attraction Points The calculation method is to minimize the current position. with point set The Euclidean distance between them.

[0049] The demonstration learning involves extracting and learning interactive features from traction demonstrations, combining them with nonparametric potential fields to provide the same assisted rehabilitation for the limb to be rehabilitated as in the traction demonstrations; at this stage, control force... Replace demonstration power Provides equivalent assistance for upper limb movement:

[0050]

[0051]

[0052] and They represent the points of attraction, respectively. The attraction and the direction to the next attraction The driving force; Provide compliant constraints during training, while To provide assistance, in order to jointly generate the tangential and normal forces applied to the current position, a nonparametric potential field is constructed, which uses a stiffness of... The virtual spring connects the position of the end effector to the corresponding attraction point. To construct; the One attraction exist Potential energy at the point for:

[0053]

[0054] Kernel regression function is used, based on each energy element. Construct the total energy function; at the current position The contribution of energy elements is determined by the Gaussian kernel:

[0055]

[0056] Through parameters Define the smoothness of the Gaussian kernel function; select... front and back The total potential energy is calculated using a point of attraction; The total potential energy at the point is:

[0057]

[0058] Attracting points Attraction normalization; each point Compared to The attraction coefficient is defined as:

[0059]

[0060] Therefore, the total potential energy simplifies to:

[0061]

[0062] By definition and This allows the limb to receive the same assistance as in the traction demonstration, including traction force and interactive stiffness:

[0063]

[0064]

[0065] in, and These represent the demonstration process. The minimum and maximum traction forces in the set; and It is the limit of the interaction stiffness; through a constant Adjust the auxiliary width to meet training requirements.

[0066] The master-slave control is used during the demonstration and learning process, where the current master arm pose is... The corresponding desired pose is From the desired pose corresponding to the arm According to the sampling set Obtain; under the action of interaction forces, from the current pose of the boom. Represented as:

[0067]

[0068]

[0069] The limbs to be rehabilitated are guided to perform coordinated joint movements by using the primary and secondary arms in traction.

[0070] Compared with existing end effector-type and exoskeleton-type upper limb rehabilitation robots, the present invention has the following advantages:

[0071] 1) Balancing adaptability and precision: By simultaneously pulling the patient's hand and forearm with a dual-arm robot, safe, scientific and compliant spatial training can be achieved, just like that of a rehabilitation therapist. It maintains the advantages of end effector robots, such as strong adaptability and simple wearability, while also taking into account the advantages of exoskeleton robots, such as precise control of individual joints.

[0072] 2) Improved safety and compliance: This invention constructs a dual safety constraint mechanism of task space and joint space, and introduces nonparametric potential field learning of the traction movements and interaction characteristics of rehabilitation therapists, effectively avoiding secondary injuries caused by non-standard training movements and strong human-machine interaction, thus improving the safety and compliance during training.

[0073] 3) Intelligent and personalized rehabilitation: By learning from the traction demonstrations of rehabilitation therapists and combining master-slave control strategies, the dual-arm robot can be assigned roles according to the rehabilitation therapist's contribution to the movement of the rehabilitated limbs. It can be used in different rehabilitation stages and various training tasks, adapting to the rehabilitation needs of different patients and achieving personalized and scientific rehabilitation training. Attached Figure Description

[0074] Figure 1 This is a schematic diagram of a dual-arm robot traction training system that mimics a rehabilitation therapist.

[0075] Figure 2 This is a diagram of a rehabilitation training exercise program. (a)-(c) are the initial movement, movement one, and movement two, respectively.

[0076] Figure 3 This is a diagram of the inertial and surface electromyography (EMG) sensor wearing scheme;

[0077] Figure 4 This is a schematic diagram of a traction demonstration;

[0078] Figure 5 This is a schematic diagram of 8-DOF upper limb kinematic modeling; (a) shows the correspondence between each joint of the upper limb and the kinematic model coordinate system, and (b) shows the visualization effect of the reachable task space of a specific segment of the upper limb.

[0079] Figure 6 These are diagrams showing the effect of non-redundant inverse kinematics joint constraints, with (a)-(h) corresponding to different joints;

[0080] Figure 7This is a diagram illustrating the traction demonstration and demonstration learning process;

[0081] Figure 8 These are the effect diagrams of different parameters on the potential field, with (a)-(d) corresponding to different parameters;

[0082] Figure 9 These are the results of the double-arm traction training; (a-1)-(a-6) show the joint angle changes corresponding to the first training movement, and (b-1)-(b-6) show the joint angle changes corresponding to the second training movement.

[0083] Figure 10 The images show the results of multi-mode compliance training; (a) shows the potential field visualization effect; (b) shows the compliance test effect under different perturbation forces; and (c) shows the training effect under four different modes.

[0084] Figure 11 This is a schematic diagram of a dual-arm robot mimicking a rehabilitation therapist's traction training technique.

[0085] In the diagram: a-robotic arm; b-connector; c-force sensor; d-end effector. Detailed Implementation

[0086] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0087] Dual-arm robotic rehabilitation therapist traction training system, such as Figure 1 As shown, it consists of a robotic arm (a), a connector (b), a force sensor (c), and an end effector (d). Specifically, two Doosan M1509 collaborative robots are used to simulate the arms of a rehabilitation therapist. Each robot's end effector is equipped with a six-axis force sensor (ATI Axia80) for traction demonstrations and demonstration learning. The limb to be rehabilitated is secured to the end effector using training gloves and Velcro to ensure comfortable and convenient wearing.

[0088] Two common rehabilitation exercises are performed in this example. For example... Figure 2 (b) and Figure 2 As shown in (c), these are arm elevation (robotic arm A as the master arm and robotic arm B as the slave arm) and chest extension (robotic arm A as the slave arm and robotic arm B as the master arm), respectively. The multi-joint coordinated movements involved include shoulder abduction, shoulder flexion, elbow flexion, forearm supination, wrist flexion, and wrist abduction. Inertial and surface electromyography sensors are used... Figure 3 It is fitted onto the subject's body in a specific way to collect upper limb movement angles and muscle activity. For example... Figure 4 As shown, a rehabilitation training demonstration was conducted by applying traction force to the subject's upper limbs. A dual-arm robot was used for the demonstration and learning, and five reciprocating rehabilitation training sessions were performed.

[0089] A dual-arm robotic traction training method, modeled after a rehabilitation therapist, for upper limb rehabilitation after stroke, is implemented as follows:

[0090] Step 1: Measure the lengths of each segment of the limb to be rehabilitated, including the acromiohumeral intercostal space, upper arm length, forearm length, and the distance from the wrist to the center of the palm, and establish an 8-DOF high-precision upper limb kinematic model. Based on the 8-DOF high-precision upper limb kinematic model, establish safety constraint mechanisms in task space and joint space;

[0091] Step 1.1: Upper limb kinematic modeling and task space constraints;

[0092] use , , and This indicates the acromiohumeral space, upper arm length, forearm length, and the distance from the wrist joint to the center of the palm. Use , They represent the elbow joint to The distance of (end effector A), and arrive The distance of (end-effector B). As shown in Table 1, the kinematic parameters of the limb to be rehabilitated are calculated according to the standard DH convention. Non-physiological postures are prevented by restricting movement at specific segmental points of the upper limb. Based on the above 8-DOF kinematic model, the Monte Carlo method is used to calculate... and The reachable training space is as follows:

[0093]

[0094]

[0095] Where E1 and E2 represent the point sets solved by end effector A and end effector B respectively under the Monte Carlo method. Figure 5 As shown, Delaunay triangulation is performed on point sets E1 and E2. A parameter X is defined, and triangles with side lengths less than 1 / X are filtered out. These triangles are then connected to generate the corresponding Alpha shape. The ray casting method is used to determine in real-time whether the current end effectors A and B are within the created Alpha shape, thus achieving the detection and constraint of the range of motion of the limb to be rehabilitated.

[0096] Table 1. DH kinematic parameters of the 8-DOF upper limb

[0097]

[0098] 1.2 Non-redundant joint space constraints

[0099] The 8-DOF kinematic model of the upper limb was decomposed into 5-DOF and 3-DOF subsystems, and non-redundant kinematic solutions were performed for each. The calculated DH parameters for the 5-DOF and 3-DOF subsystems are shown in Tables 2 and 3.

[0100] Table 2. Kinematic parameters of the upper limb of the 5-DOF subsystem (DH).

[0101]

[0102] Table 3. Kinematic parameters of the upper limb of the 3-DOF subsystem (DH)

[0103]

[0104] For a 5-DOF subsystem, the following relationship is established:

[0105]

[0106] For a 3-DOF subsystem, the following relationship is established:

[0107]

[0108] Numerical optimization models for 5-DOF and 3-DOF non-redundant subsystems were constructed. The solution results are as follows: Figure 6 As shown.

[0109] Step 2: Calculate the required assistance by having a rehabilitation therapist demonstrate the procedure to the robot. In the absence of a therapist, the robot provides the limb to be rehabilitated with the same assistance received during the traction demonstration, including equivalent traction force, training trajectory, and interaction stiffness. Personalized, compliant, and safe training of the rehabilitation robot is achieved through traction demonstrations and demonstration learning.

[0110] Specific implementation: such as Figure 7 As shown, impedance control parameters are set to adjust the traction demonstration characteristics. A traction demonstration is performed by a rehabilitation therapist, and the traction trajectory, interaction force, and interaction stiffness are recorded for subsequent demonstration learning. The demonstration features are pre-processed by downsampling, and a tangent-normal coordinate system is established. A nonparametric potential field is used to process the demonstration trajectory, calculating the traction force, resultant force, damping, and attractive force at the current position. Based on this, the traction force and interaction stiffness are further learned to achieve demonstration learning. Figure 8 As shown, through parameters Controlling field strength and utilizing parameters Control the smoothness of the potential field. This is achieved by setting... Control traction. Increase The potential field becomes smoother, local details are reduced, resulting in a more blurred field. Reducing... The potential field strength weakens, leading to increased normal compliance. (Increase) The change directs the traction force towards the target point. This is achieved through adjustment. Based on the size or direction, design targeted rehabilitation models.

[0111] Step 3: Based on the traction contribution of the therapist's arms to the limb being rehabilitated, the robot's arms are divided into master arms and slave arms. Utilizing the functional characteristics of the master and slave arms, the task space motion is divided into absolute motion components and relative motion components. Absolute motion is used to perform large-range movements of the limb being rehabilitated. Relative motion is used to receive the master arm's motion, thus providing movement assistance for rehabilitation training.

[0112] Specific implementation: The current main arm position is... The corresponding desired pose is Through sampling sets Calculate the desired pose of the arm. The current pose of the driven arm is determined using the following formula. Solve the following:

[0113]

[0114]

[0115] The limb to be rehabilitated is guided to perform coordinated joint movements through coordinated traction using the primary and secondary arms. This mimics the effect of a rehabilitation therapist's bi-arm traction. Figure 9 As shown. Using robotic arm A for spatial training (rehabilitation movement 1) ensures the accuracy of the upper limb endpoint position, but it cannot precisely control individual joints, resulting in significant deviations in the overall training posture. Using robotic arm B for spatial training (rehabilitation movement 2) improves the tracking accuracy of the shoulder and elbow, but it lacks sufficient control over the distal fine joints (i.e., the wrist joint), resulting in poor tracking performance and low wrist joint motion correlation. Increasing arm assistance through a dual-arm rehabilitation system improves joint motion accuracy. The effects of multi-mode compliant training are as follows... Figure 10 As shown, seven circular cross-sections were sampled to represent the compliance characteristics and anisotropic attraction distribution of the potential field at different demonstration stages. After applying a perturbation force, the upper limb offset distances were 4.9-5.8 mm, 10.2-11.5 mm, 15.5-17.3 mm, and 21.0-23.2 mm, reflecting the compliance training characteristics of the rehabilitation system. After the perturbation force was removed, the potential field guided the upper limb smoothly back to the desired trajectory, demonstrating the compliance and safety characteristics of the rehabilitation system.

Claims

1. A dual-arm robotic traction training method for upper limb rehabilitation after stroke, characterized in that, A dual-arm robot is used to learn the traction movements and interaction characteristics of a rehabilitation therapist, and to perform traction movements on the hand and forearm posterior ends of the limb to be rehabilitated, respectively, to conduct rehabilitation therapy-style scientific, compliant, and safe rehabilitation training. The dual-arm robot includes robotic arm A and robotic arm B. An 8-DOF high-precision upper limb kinematic model is constructed, and a dual safety constraint mechanism of task space and joint space is established based on the characteristics of dual-arm robot traction rehabilitation. A non-parametric potential field is designed to learn the traction movements and interaction characteristics of a rehabilitation therapist in a single traction demonstration. Based on the contribution of movement, the dual-arm robot is divided into master arm and slave arm, and master-slave control is applied to implement master-slave compliant interaction. A dual-arm robot upper limb rehabilitation system is built to conduct dual-arm rehabilitation robot traction training tasks that mimic rehabilitation therapists.

2. The dual-arm robotic rehabilitation therapist-inspired traction training method for upper limb rehabilitation after stroke, as described in claim 1, is characterized in that... Specifically, the steps include the following: Step 1: Measure the length of each segment of the limb to be rehabilitated, construct an 8-DOF high-precision upper limb kinematic model, and establish a dual safety constraint mechanism of task space plus joint space, taking into account the characteristics of dual-arm robot traction rehabilitation. Step 2: Calculate the assistance required for rehabilitation training by having a rehabilitation therapist demonstrate traction to the dual-arm robot and the limb to be rehabilitated, and record the traction force, training trajectory, and interaction stiffness during the traction demonstration; design a nonparametric potential field function, and have the dual-arm robot demonstrate and learn from the traction demonstration; when the rehabilitation therapist is not present, the dual-arm robot provides the limb to be rehabilitated with the same assistance as during the traction demonstration, including equivalent traction force, training trajectory, and interaction stiffness, and provides compliant human-machine interaction; Step 3: Based on the traction contribution of the rehabilitation therapist's arms to the limb to be rehabilitated, robotic arms A and B are classified as master arms and slave arms, respectively. Using the functional characteristics of master arms and slave arms, the task space motion is divided into absolute motion components and relative motion components. Absolute motion is used to perform large-range movements of the limb to be rehabilitated. Relative motion is used to receive the movement of the master arm to assist the rehabilitation training.

3. The dual-arm robotic rehabilitation therapist-inspired traction training method for upper limb rehabilitation after stroke, as described in claim 2, is characterized in that... Step 1 specifically includes: Step 1.1: Measure the acromiohumeral intercostal space, upper arm length, forearm length, and the distance from the wrist to the center of the palm of the limb to be rehabilitated. Divide the joints of the limb to be rehabilitated into a 1-DOF sternoclavicular joint, a 3-DOF shoulder joint, a 2-DOF elbow joint, and a 2-DOF wrist joint, and establish an 8-DOF high-precision upper limb kinematic model. Use the Monte Carlo method to calculate the reachable task space point set of the corresponding upper limb specific points of the dual-arm robot, and reconstruct the three-dimensional space of the reachable task space point set. Monitor the spatial position of specific points during the rehabilitation training process in real time to achieve task space constraints on the upper limb. Step 1.2: Utilizing the characteristics of dual-arm robot traction rehabilitation, based on the specific points of the upper limb corresponding to the dual-arm robot, the 8-DOF upper limb kinematic model is divided into a 5-DOF subsystem and a 3-DOF subsystem; The 5-DOF subsystem includes a 1-DOF sternoclavicular joint, a 3-DOF shoulder joint, and a 1-DOF elbow joint that performs flexion and extension movements; the 3-DOF subsystem includes a 1-DOF elbow joint and a 2-DOF wrist joint that perform forearm pronation and supination movements. Numerical optimization methods for solving 5-DOF and 3-DOF subsystems were constructed to calculate the joint angles of the limb to be rehabilitated; changes in joint angles during rehabilitation training were monitored in real time to achieve spatial constraints on the joints of the upper limb.

4. The dual-arm robotic traction training method for upper limb rehabilitation after stroke, as described in claim 3, is characterized in that... In the 8-DOF upper limb kinematic model, the sternoclavicular joint with 1 DDOF allows for pronation and supination of the scapula; the ball joint of the shoulder with 3 DDOF allows for adduction / abduction, flexion / extension, and internal / external rotation of the arm; the elbow joint with 2 DDOF allows for flexion / extension and pronation / supination of the forearm; and the wrist joint with 2 DDOF allows for flexion / extension and adduction / abduction of the wrist. Among them, the basic coordinate system Located at the sternoclavicular joint; local coordinate system Fixed at joint i, excluding the sternoclavicular joint; end-effector coordinate system Located at the center of the palm; an 8-DOF high-precision upper limb kinematic model is established according to the standard DH method; the transformation matrix from joint i to joint i-1. for: ； in, It is along direction and The angle between them; It is along direction and The distance between them; It is along direction and The distance between them; It is along direction and The angle between them; symbol and These represent the sin and cos operations, respectively. It is a vector of joint variables; for the model established above, the coordinate point in the j-th coordinate system The expression relative to the base coordinate system is as follows: ； For the position corresponding to the end effector of robotic arm A The position corresponding to the end effector of robotic arm B The representation in the basic coordinate system is as follows: ； in, Indicates the distance from the elbow joint to p a2 The distance between them.

5. A dual-arm robotic rehabilitation therapist-like traction training method for upper limb rehabilitation after stroke, as described in claim 3, is characterized in that... The task space constraints restrict the movement of specific segments of the upper limb to prevent non-physiological postures; based on an 8-DOF upper limb kinematic model, the Monte Carlo method is used to calculate p. a1 and p a2 Accessible task space: ； Here, E1 and E2 represent the solution point sets corresponding to the end effectors of robotic arm A and robotic arm B under the Monte Carlo method, respectively. Delaunay triangulation is performed on the solution point sets E1 and E2. A parameter X is defined, and triangles with side lengths less than 1 / X are filtered out. These triangles are then connected to generate the corresponding Alpha shape. The ray casting method is used to determine in real time whether the end effectors of robotic arm A and robotic arm B are within the created Alpha shape. If the result shows that both end effectors of robotic arm A and robotic arm B are within the created Alpha shape, it indicates that the specific segment points of the upper limb have not exceeded the reachable task space, and training continues. Otherwise, it indicates that the specific segment points of the upper limb have exceeded the reachable task space, and the rehabilitation robot stops training, thus achieving upper limb task space constraints during training.

6. A dual-arm robotic rehabilitation therapist-inspired traction training method for upper limb rehabilitation after stroke, as described in claim 3, is characterized in that... The joint space constraint is achieved by dividing the 8-DOF kinematic model of the upper limb into a 5-DOF subsystem and a 3-DOF subsystem; the 5-DOF subsystem, point p a2 In the basic coordinate system H0, it is represented as: ； in, This indicates that the end effector is located at p a2 The transformation matrix from the base coordinate system where the robotic arm B is located to H0; For a 3-degree-of-freedom subsystem, point p a1 In the local coordinate system H5, it is represented as: ； in, This indicates that the end effector is located at p a1 The transformation matrix from the base coordinate system where the robotic arm A is located to H0; Based on the relationship between the end effectors of robotic arms A and B and H0 and H5, non-redundant kinematic solutions are performed for the 5-DOF subsystem and the 3-DOF subsystem, respectively.

7. The dual-arm robotic rehabilitation therapist-inspired traction training method for upper limb rehabilitation after stroke, as described in claim 2, is characterized in that... Step 2 specifically includes: Step 2.1: Wear end effectors on the hand and forearm, fix the upper limb to the end effector of the upper limb rehabilitation robot equipped with a six-dimensional force sensor, and drive the upper limb and the end effector of the rehabilitation robot to perform spatial movements through traction demonstration, while recording traction force, training trajectory and interaction stiffness; Step 2.2: Design a nonparametric potential field to extract interactive features from the traction demonstration; learn the interactive features of the traction demonstration through a dual-arm robot, and provide equivalent motion assistance for upper limb movement by replacing the traction force, training trajectory and interactive stiffness of the rehabilitation therapist with the robot's traction force, training trajectory and interactive stiffness, and provide compliant human-computer interaction.

8. A dual-arm robotic traction training method for upper limb rehabilitation after stroke, as described in claim 1, is characterized in that... The traction demonstration uses robotic arms A and B instead of the rehabilitation therapist's arms; the nonlinear dynamics of the m-degree-of-freedom robotic arms A and / or B, whose ends are connected to the limb to be rehabilitated, are expressed as follows: ； in, Represents the joint angle vector; It is the robot's inertia matrix; It is centripetal force and Coriolis force. It is the gravitational torque vector; It is the joint input torque; The Jacobian matrix represents the associated task space and joint space; This represents the total resultant force applied in the task space; The interaction forces acting on the actuator can be broken down according to their mode of action. and control force in the control system During the traction demonstration phase, the rehabilitation therapist collaborates with the limb to be rehabilitated to complete the task, enabling... Equal to 0; the interaction force acting on the end effector Depend on , and The components represent the traction force applied by the rehabilitation therapist, the force applied to the limb to be rehabilitated, and the installation stress generated by the sensor; the dynamic relationship between force and posture during the traction demonstration is as follows: ； in, and These represent damping and stiffness properties, respectively; actual attitude. With expected posture The difference between them represents the attitude deviation, and its derivative is... Indicates speed deviation; by... regarded as the desired posture It also accumulates incremental posture updates, with the end effector tracking the rehabilitation therapist's traction in both position and orientation: ； in, The desired pose for the next control cycle is represented by: The current position and its deviation are represented by: and The current direction and its deviation are respectively represented as and Velocity and pose deviation and The calculation is as follows: ； ； in, To control the cycle; during the demonstration, the attitude and interaction forces by Frequency recording in Hz; downsampling the collected dataset to... A uniformly distributed set of samples, called attraction points, is used to construct the dataset. : ； ； definition For the first A tangent-normal coordinate system for each attraction point; the tangent-normal coordinate system is based on the attraction point. Centered on the point of attraction, the tangent direction points towards the next point of attraction. The normal direction points to the current position of the end effector. Expected Attraction Points The calculation method is to minimize the current position. with point set The Euclidean distance between them.

9. A dual-arm robotic traction training method for upper limb rehabilitation after stroke, as described in claim 2, is characterized in that... The demonstration learning involves extracting and learning interactive features from traction demonstrations, combining them with nonparametric potential fields to provide the same assisted rehabilitation for the limb to be rehabilitated as in the traction demonstrations; at this stage, control force... Replace demonstration power Provides equivalent assistance for upper limb movement: ； ； and They represent the points of attraction, respectively. The attraction and the direction to the next attraction The driving force; Provide compliant constraints during training, while To provide assistance, in order to jointly generate the tangential and normal forces applied to the current position, a nonparametric potential field is constructed, which uses a stiffness of... The virtual spring connects the position of the end effector to the corresponding attraction point. To construct; the One attraction exist Potential energy at the point for: ； Kernel regression function is used, based on each energy element. Construct the total energy function; at the current position The contribution of energy elements is determined by the Gaussian kernel: ； Through parameters Define the smoothness of the Gaussian kernel function; select... front and back The total potential energy is calculated using a point of attraction; The total potential energy at the point is: ； Attracting points Attraction normalization; each point Compared to The attraction coefficient is defined as: ； Therefore, the total potential energy simplifies to: ； By definition and This allows the limb to receive the same assistance as in the traction demonstration, including traction force and interactive stiffness: ； ； in, and These represent the demonstration process. The minimum and maximum traction forces in the set; and It is the limit of the interaction stiffness; through a constant Adjust the auxiliary width to meet training requirements.

10. A dual-arm robotic rehabilitation therapist-inspired traction training method for upper limb rehabilitation after stroke, as described in claim 1, is characterized in that... The master-slave control is used during the demonstration and learning process, where the current master arm pose is... The corresponding desired pose is From the desired pose corresponding to the arm According to the sampling set Obtain; under the action of interaction forces, from the current pose of the boom. Represented as: ； ； The limbs to be rehabilitated are guided to perform coordinated joint movements by using the primary and secondary arms in traction.

Images