Robot skill transfer system and method based on truncated core dynamic motion primitives

Abstract

The present invention discloses a robot skill transfer system and method based on truncated kernel dynamic motion primitives. The system includes a robot skill teaching module, a robot skill learning module, and a robot skill reproduction module. A teaching person wearing an upper limb exoskeleton performs remote operation teaching on an industrial robot through the upper limb exoskeleton. The robot skill learning module uses the acquired teaching data to learn trajectory shape parameters through a skill learning model based on the truncated kernel dynamic motion primitive. Finally, the learned trajectory shape parameters are used to control the industrial robot to perform skill reproduction and trajectory reproduction accuracy verification. The present invention applies the Gaussian kernel truncation mechanism to robot skill transfer based on dynamic motion primitives, improves the robot skill transfer reproduction accuracy, and can achieve efficient learning and high-precision reproduction of robot skills. It is suitable for application fields such as intelligent manufacturing and intelligent robots.

Description

Technical Field

[0001] The present invention relates to a robot skill transfer system and method in the field of robot technology, and in particular to a robot skill transfer system and method based on truncated core dynamic motion primitives. Background Art

[0002] Robot skill learning is a key step in achieving intelligent manufacturing. It involves transferring human operational skills to robots, enabling them to perform tasks similar to those of humans. The main steps in robot skill learning include skill demonstration, skill learning, and skill reproduction. Enabling robots to accurately reproduce learned skills is crucial for completing skill transfer and expanding their intelligent applications. The foundation of robot skill learning lies in the robot skill learning model. Common skill learning models include the dynamic motion primitive model, Gaussian mixture model, and hidden Markov model. As a nonlinear second-order dynamic system, the dynamic motion primitive model has attracted considerable attention in robot skill learning due to its simplicity, high learning efficiency, and strong generalization capabilities, resulting in numerous application attempts.

[0003] To adapt to various application scenarios such as industrial production and assembly manufacturing, optimizing the original dynamic motion primitive model is the main measure to improve its application adaptability. In application scenarios involving high-precision robot skill learning, such as enabling robots to achieve high-precision assembly, traditional dynamic motion primitive models are difficult to effectively meet the application requirements of high-precision robot skill reproduction. To this end, a large number of scholars have carried out innovative research on the application and optimization of dynamic motion primitive models, which can be roughly divided into two categories:

[0004] The first type of research focuses on improving the accuracy of dynamic motion primitive modeling by obtaining more motion information, thereby achieving the goal of improving trajectory reproduction accuracy. However, the amount of information obtained will affect the trajectory reproduction accuracy.

[0005] The other category focuses on improving the modeling capabilities of dynamic motion primitives by coupling other algorithms to the dynamic motion primitive model. The coupled algorithms destroy the efficient learning characteristics of the original dynamic motion primitives. Summary of the Invention

[0006] In response to the common problem of low skill trajectory reproduction accuracy in robot skill transfer, the purpose of the present invention is to provide a dynamic motion primitive skill transfer method based on truncated kernel function, which improves the robot skill trajectory reproduction accuracy by truncation processing of the Gaussian kernel function of the dynamic motion primitive model and optimization of shape parameters.

[0007] The present invention applies Gaussian kernel truncation mechanism and shape parameter optimization to robot skill transfer based on dynamic motion primitives, improves the reproduction accuracy of skill transfer, and is suitable for application fields such as intelligent manufacturing and intelligent robots.

[0008] In the present invention, the performance of the original dynamic motion primitive model learning is maintained through the optimization method of the kernel function, which can provide an effective solution for the robot skill transfer to achieve efficient learning and high-precision skill reproduction.

[0009] The technical solution adopted by the present invention to solve its technical problem is:

[0010] 1. A robot skill transfer system based on truncated core dynamic motion primitives, characterized by:

[0011] The system includes a robot skill teaching module, a robot skill learning module and a robot skill reproduction module; the robot skill teaching module, the robot skill learning module and the robot skill reproduction module are connected in sequence.

[0012] The robot skill teaching module includes an upper limb exoskeleton, a teleoperation module and an industrial robot; the upper limb exoskeleton and the teleoperation module are connected, and the teleoperation module and the industrial robot are connected.

[0013] The robot skill learning module includes a teaching data acquisition module and a truncated core dynamic motion primitive learning module. The teaching data acquisition module is connected to each joint of the industrial robot, and the output end of the teaching data acquisition module is connected to the input end of the truncated core dynamic motion primitive learning module.

[0014] The robot skill reproduction module includes a trajectory reproduction module and a trajectory reproduction accuracy evaluation module. The input end of the trajectory reproduction module is connected to the output end of the truncated kernel dynamic motion primitive learning module, the output end of the trajectory reproduction module is connected to the industrial robot, and the input end of the trajectory reproduction accuracy evaluation module is respectively connected to the output end of the trajectory reproduction module and the output end of the teaching data acquisition module.

[0015] The upper limb exoskeleton is a serial mechanical exoskeleton.

[0016] The industrial robot is a serial industrial robotic arm.

[0017] 2. A robot skill transfer method based on truncated core dynamic motion primitives, the method mainly includes the following steps:

[0018] Step 1): The instructor wears an upper limb exoskeleton and controls the industrial robot through the teleoperation module to move according to the skill trajectory taught by the upper limb exoskeleton to complete the skill trajectory teaching.

[0019] Step 2): During the skill teaching process, the teaching data acquisition module in the robot skill learning module collects, processes and records the data of each joint of the industrial robot in real time through the angle sensor as the teaching trajectory data;

[0020] The truncated kernel dynamic motion primitive learning module uses the acquired teaching trajectory data of the industrial robot to learn the teaching trajectory and obtains the shape parameters of the teaching trajectory as the skill trajectory shape parameters.

[0021] Step 3): In the trajectory reproduction module, the shape parameters of the teaching trajectory learned by the truncated kernel dynamic motion primitive learning module are used to generate reproduction trajectory data, and then the industrial robot is controlled to move and reproduce the teaching trajectory.

[0022] In the specific implementation, the trajectory reproduction accuracy evaluation module is used to calculate the trajectory reproduction root mean square error and maximum position error using the reproduced trajectory data and the taught trajectory data to evaluate the trajectory reproduction accuracy, which can reflect and further optimize the skill transfer effect of the skill transfer system.

[0023] In step 2), the data collected by the teaching data collection module for skill trajectory learning is the data of the six joints in the joint space of the industrial robot. The data of the six joints of the industrial robot is collected in real time during the dynamic movement of the industrial robot.

[0024] In step 2), the shape parameters of the teaching trajectory are obtained, specifically:

[0025] By collecting the motion information of each joint, a one-dimensional model of dynamic motion primitives is established according to the following formula to obtain the target forced term f corresponding to each joint: target :

[0026]

[0027] Where g is the target position of the joint, α z , β z They are proportional gain and differential control parameters, specifically positive constants, τ is the time proportional factor, y d 、 are the position, velocity and acceleration of the joint respectively;

[0028] In order to obtain the target forcing term f for all joints target Afterwards, the regularization system is used to force the target terms f of all six joints target The coupling process obtains the corresponding phase variable x to ensure the synchronization of the joint movements;

[0029] The Gaussian kernel function calculation module uses the Gaussian kernel function to combine the target force term f of all joints targetThe shape parameters representing the teaching trajectory are obtained by fitting the corresponding phase variable x according to the following formula. The fitting formula is as follows:

[0030]

[0031] Where N is the number of Gaussian kernel functions, ω i ,η i and ρ i are the first, second, and third shape parameters of the trajectory, x is the phase variable, It is the Gaussian kernel function after truncation, set by the Gaussian kernel function truncation module;

[0032] The specific method of truncation is as follows: the Gaussian kernel function after truncation is Calculated as:

[0033]

[0034] Where, t i is the Gaussian kernel function The truncation time, i = 1, ..., N, N represents the number of Gaussian kernel functions, t0 is the time when the movement starts, t N The moment when the movement ends.

[0035] In the Gaussian kernel function The cut-off time t i Before and after, the relationship between two adjacent kernel functions is expressed as:

[0036]

[0037] The method of the present invention can eliminate the exponential decay at the tail of the traditional Gaussian kernel function through the above-mentioned kernel function truncation processing, improve the degree of fitting to the target forcing term, and achieve higher-precision trajectory reproduction.

[0038] The fitting is performed by using a local weighted regression algorithm to calculate the shape parameters of the teaching trajectory.

[0039] In the step 3), specifically: using the shape parameters of the teaching trajectory, the acceleration of the reproduced trajectory is calculated by integrating the Gaussian kernel function truncation processing mechanism and the dynamic motion primitive learning model optimized by multiple shape parameters, and then the speed and position information of the reproduced trajectory are calculated by Euler integral. Finally, the calculated position, speed and acceleration information of the reproduced trajectory is used to control the industrial robot to complete the reproduction of the skill trajectory according to the reproduced trajectory.

[0040] In the specific implementation of the present invention, in order to verify the effect of the robot skill system and method based on truncated kernel dynamic motion primitives on improving the trajectory reproduction accuracy, the motion trajectory of the teaching stage and the motion trajectory of the reproduction stage are compared, and the trajectory reproduction accuracy is evaluated by the root mean square error and maximum error.

[0041] The present invention has the following beneficial effects:

[0042] (1) The present invention uses an upper limb exoskeleton to remotely teach and transfer skills to an industrial robot, and can intuitively and efficiently transfer the skills demonstrated by the instructor to the industrial robot.

[0043] (2) The present invention realizes the learning of robot skills based on the dynamic motion primitive model. Thanks to the strong single-time teaching learning ability of the model, efficient robot skill transfer can be achieved.

[0044] (3) The robot skill transfer method based on the truncated kernel dynamic motion primitive is proposed in this paper. The Gaussian kernel function of the traditional DMP model is truncated to eliminate the influence of its tail exponential decay on the fitting target forcing term. By optimizing the number of shape parameters, the model can better approximate the local gradient of the target forcing term, thereby achieving the purpose of improving the accuracy of robot skill trajectory reproduction.

[0045] (4) The robot skill transfer system and method based on the truncated core dynamic motion primitive of the present invention can achieve efficient robot skill transfer and high-precision skill trajectory reproduction, and is suitable for application fields such as rapid deployment of robots in intelligent manufacturing and intelligent robot skill learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a block diagram of the skill transfer system based on truncated core dynamic motion primitives.

[0047] Figure 2 This is a block diagram of the skill transfer method based on truncated core dynamic motion primitives.

[0048] Figure 3 It is a comparison curve of Gaussian kernel function.

[0049] Figure: 1. Upper limb exoskeleton, 2. Robot skill teaching module, 3. Industrial robot, 4. Teleoperation module, 5. Robot skill learning module, 6. Teaching data acquisition module, 7. Truncated kernel dynamic motion primitive learning module, 8. Robot skill reproduction module, 9. Trajectory reproduction module, 10. Trajectory reproduction accuracy evaluation module. 11. One-dimensional model of dynamic motion primitives, 12. Canonical system, 13. Gaussian kernel function calculation module, 14. Gaussian kernel function truncation module, 15. Shape parameter improvement module, 16. Trajectory shape parameter calculation module, 17. Euler integration module, 18. Trajectory error calculation module. DETAILED DESCRIPTION

[0050] The present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0051] like Figure 1 As shown, the system includes a robot skill teaching module 2, a robot skill learning module 5 and a robot skill reproduction module 8; the robot skill teaching module 2, the robot skill learning module 5 and the robot skill reproduction module 8 are connected in sequence.

[0052] The robot skill teaching module 2 includes an upper limb exoskeleton 1, a teleoperation module 4 and an industrial robot 3; the upper limb exoskeleton 1 is connected to the teleoperation module 4, and the teleoperation module 4 is connected to the industrial robot 3.

[0053] The teaching personnel wears the upper limb exoskeleton 1 and performs movements along the teaching trajectory to complete the skill action teaching. The teleoperation module 4 maps the movements of the upper limb exoskeleton 1 to the industrial robot 3, and controls the industrial robot 3 to move synchronously according to the teaching trajectory of the upper limb exoskeleton 1, thereby completing the teleoperation teaching process, that is, the skill teaching process.

[0054] The robot skill learning module 5 includes a teaching data acquisition module 6 and a truncated core dynamic motion primitive learning module 7. The input end of the teaching data acquisition module 6 is connected to each joint of the industrial robot 3, and the output end of the teaching data acquisition module 6 is connected to the input end of the truncated core dynamic motion primitive learning module 7.

[0055] The teaching data acquisition module 6 is used to collect and record the joint data of the industrial robot 3 during the teleoperation teaching process of the upper limb exoskeleton 1 on the industrial robot 3 as teaching trajectory data.

[0056] The truncated core dynamic motion primitive learning module 7 uses the joint data of the industrial robot 3 obtained by the teaching data acquisition module 6 to learn the teaching trajectory, and obtains the shape parameters of the teaching trajectory as the skill trajectory shape parameters.

[0057] The robot skill reproduction module 8 includes a trajectory reproduction module 9 and a trajectory reproduction accuracy evaluation module 10. The input end of the trajectory reproduction module 9 is connected to the output end of the truncated kernel dynamic motion primitive learning module 7, the output end of the trajectory reproduction module 9 is connected to the industrial robot 3, and the input end of the trajectory reproduction accuracy evaluation module 10 is respectively connected to the output end of the trajectory reproduction module 9 and the output end of the teaching data acquisition module 6.

[0058] The trajectory reproduction module 9 generates reproduction trajectory data using the shape parameters of the teaching trajectory learned by the truncated kernel dynamic motion primitive learning module 7, and then controls the industrial robot 3 to move and reproduce the teaching trajectory.

[0059] The trajectory reproduction accuracy evaluation module 10 uses the reproduced trajectory data and the teaching trajectory data to calculate the maximum position error and root mean square error of the teaching trajectory reproduction, and evaluates the teaching trajectory reproduction accuracy to reflect and further optimize the skill transfer effect of the skill transfer system.

[0060] In this implementation, upper-limb exoskeleton 1 is a serial-type mechanical exoskeleton, with each joint integrated with a LAT216T-360 angle sensor. Industrial robot 3 is an EFORT ER3B-C30 serial-type industrial manipulator arm with an effective working range of 672.8 mm, a payload of 4 kg, and a repeatability of 0.03 mm.

[0061] The upper limb exoskeleton 1 and the industrial robot 3 are in a heterogeneous mapping relationship. The upper limb exoskeleton 1 establishes a mapping relationship with the industrial robot 3 through link posture constraints to achieve synchronous motion control.

[0062] The block diagram of the skill transfer process based on truncated core dynamic motion primitives is as follows Figure 2 As shown, it includes a one-dimensional model of dynamic motion primitives 11, a regular system 12, a Gaussian kernel function calculation module 13, a Gaussian kernel function truncation module 14, a shape parameter improvement module 15, a trajectory shape parameter calculation module 16, an Euler integration module 17 and a trajectory error calculation module 18.

[0063] The specific implementation process of the present invention is:

[0064] 1) The instructor wears the upper limb exoskeleton 1 and controls the industrial robot 3 through the teleoperation module 4 to move according to the skill trajectory taught by the upper limb exoskeleton 1 to complete the skill teaching.

[0065] 2) During the skill teaching process, the teaching data acquisition module 6 in the robot skill learning module 5 uses angle sensors to collect, process, and record the six joint motion data [y1, y2, y3, y4, y5, y6] of the industrial robot 3 in real time during the teaching process. The truncated core dynamic motion primitive learning module 7 uses the acquired joint motion data to learn the trajectory. The six joints are modeled using six dynamic motion primitive one-dimensional models 11, and their system parameters are set to the same: α z =25,β z =6.25, τ=1, and the target forcing term f to be fitted is obtained by calculation. target The same regular system 12 is used to couple the one-dimensional models of dynamic motion primitives corresponding to the six joints, where α x =25 / 3.

[0066] Gaussian kernel function calculation module 13 calculates the Gaussian kernel function used to fit the target forcing term The traditional Gaussian kernel function is truncated and optimized by the Gaussian kernel function truncation module 14. exist Figure 3 The distribution of 10 unweighted kernel functions is compared in Figure 3 (a) It can be seen that the tail of the traditional Gaussian kernel function is a curve that decays exponentially. After truncation, the exponential decay of the tail of the original Gaussian kernel function is eliminated, as shown in Figure 3 (b) is shown. Figure 3 At the black dotted line of the traditional Gaussian kernel function curve shown in (a), the forcing term at a certain moment on the coordinate axis is affected by the superposition of multiple kernel functions. Therefore, in order to avoid the influence of coupling between multiple kernel functions on the forcing term, the improved Gaussian kernel function curve of the present invention is as follows: Figure 3 As shown in (b), the Gaussian kernel function is truncated below the dashed line parallel to the x-axis, and the kernel function value is set to 0. Only the part above the dashed line is retained as the kernel function for fitting the forcing term. At a certain moment, it is determined by only the value of one kernel function, thereby eliminating the influence of the exponential decay coupling at the tail of the original Gaussian kernel function on the forcing term.

[0067] The shape parameter improvement module 15 is used to optimize the forcing term in the original dynamic motion primitive model to further improve the trajectory reproduction accuracy of the dynamic motion primitive model. The multi-shape parameter forcing term integrated with the Gaussian kernel function truncation processing mechanism can be expressed as:

[0068]

[0069] The shape parameters ω of the six joint motion trajectories are calculated by the trajectory shape parameter calculation module 16 i ,η i and ρ i .

[0070] 3) In the robot skill reproduction stage, the trajectory reproduction module 9 uses the skill trajectory shape parameter ω learned by the truncated kernel dynamic motion primitive learning module 7 i ,η i and ρ i The Euler integration module 17 calculates the motion information of the reproduced trajectory and controls the industrial robot 3 to reproduce the taught trajectory. The trajectory reproduction accuracy evaluation module 10 uses the trajectory error calculation module 18 to calculate the root mean square error and maximum position error between the reproduced trajectory and the taught trajectory to evaluate the trajectory reproduction accuracy and analyze and verify the skill transfer effect of the skill transfer system.

[0071] To verify the feasibility and effectiveness of the proposed industrial robot skill transfer system and method based on truncated kernel dynamic motion primitives, a comparative experimental study was conducted among different skill transfer methods. Specifically, the skill transfer methods compared included a traditional method, a standard deviation truncation method, and a skill transfer method based on truncated kernel dynamic motion primitives.

[0072] The experimental results of joint space trajectory reproduction for three different skill transfer methods are shown in Tables 1 and 2. Table 1 shows the root mean square error (RMS) of joint trajectory reproduction for different skill learning methods. The experimental results show that the maximum RMS error of joint trajectory reproduction for the traditional method is 2.838°, the maximum RMS error of joint trajectory reproduction for the standard deviation truncation method is 0.990°, and the maximum RMS error of joint trajectory reproduction for the method of the present invention is 0.550°. The experimental results show that the joint space trajectory reproduction accuracy of the method of the present invention is significantly improved compared to the traditional method and is superior to the standard deviation truncation method.

[0073] Table 1 Root mean square error of joint trajectory reproduction for different skill learning methods (unit: °)

[0074]

[0075] Table 2 shows the root mean square error (RMSE) of joint trajectory reproduction for different skill learning methods. Experimental results show that the maximum joint trajectory reproduction error for the traditional method is 8.462°, the maximum joint trajectory reproduction error for the standard deviation truncation method is 3.684°, and the maximum joint trajectory reproduction error for the proposed method is 1.754°. The experimental results demonstrate that the proposed method significantly improves joint space trajectory reproduction accuracy compared to traditional methods and outperforms the standard deviation truncation method.

[0076] Table 2 Maximum error of joint trajectory reproduction for different skill learning methods (unit: °)

[0077]

[0078] In Cartesian space, the joint angle data reproduced by the three skill learning methods mentioned above are used to control the industrial robot 3 to reproduce the skill trajectory. At the same time, the trajectory data of the end of the industrial robot 3 is collected and the trajectory reproduction accuracy is evaluated using the root mean square position error and maximum position error. The experimentally obtained end trajectory reproduction errors of different skill learning methods are shown in Table 3. The experimental results show that the root mean square position error of the Cartesian space trajectory reproduction of the traditional method is 24.103mm, and the maximum position error is 44.741mm; the root mean square position error of the Cartesian space trajectory reproduction of the standard deviation truncation method is 5.898mm, and the maximum position error is 18.784mm; the root mean square position error of the Cartesian space trajectory reproduction of the method of the present invention is 3.874mm, and the maximum position error is 15.278mm. The experimental results show that the Cartesian space trajectory reproduction accuracy of the method of the present invention is greatly improved compared with the traditional method, and is better than the standard deviation truncation method.

[0079] Table 3 End-point trajectory reproduction error of different skill learning methods (unit: mm)

[0080]

[0081] The experimental results above demonstrate that the skill transfer method based on truncated kernel dynamic motion primitives significantly improves the accuracy of skill trajectory reproduction compared to traditional methods and outperforms the standard deviation truncation method. This demonstrates the feasibility and effectiveness of the proposed robot skill transfer method based on truncated kernel dynamic motion primitives in improving the accuracy of trajectory reproduction during skill transfer.

[0082] The above specific embodiments are used to illustrate the present invention rather than to limit the present invention. Any modifications and changes made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A robot skill transfer system based on truncated core dynamic motion primitives, characterized by: The system includes a robot skill teaching module (2), a robot skill learning module (5) and a robot skill reproduction module (8); the robot skill teaching module (2), the robot skill learning module (5) and the robot skill reproduction module (8) are connected in sequence; The robot skill teaching module (2) comprises an upper limb exoskeleton (1), a teleoperation module (4) and an industrial robot (3); the upper limb exoskeleton (1) and the teleoperation module (4) are connected, and the teleoperation module (4) and the industrial robot (3) are connected; The robot skill learning module (5) includes a teaching data acquisition module (6) and a truncated core dynamic motion primitive learning module (7), the teaching data acquisition module (6) is connected to each joint of the industrial robot (3), and the output end of the teaching data acquisition module (6) is connected to the input end of the truncated core dynamic motion primitive learning module (7); The robot skill reproduction module (8) includes a trajectory reproduction module (9) and a trajectory reproduction accuracy evaluation module (10), wherein the input end of the trajectory reproduction module (9) is connected to the output end of the truncated core dynamic motion primitive learning module (7), the output end of the trajectory reproduction module (9) is connected to the industrial robot (3), and the input end of the trajectory reproduction accuracy evaluation module (10) is respectively connected to the output end of the trajectory reproduction module (9) and the output end of the teaching data acquisition module (6); In the truncated core dynamic motion primitive learning module (7), the shape parameters of the teaching trajectory are obtained in the following manner, specifically: By collecting the motion information of each joint, the dynamic motion primitive one-dimensional model (11) is established according to the following formula to obtain the target forced term f corresponding to each joint target : Where g is the target position of the joint, α z , β z are proportional gain and differential control parameters respectively, τ is the time proportional factor, are the position, velocity and acceleration of the joint respectively; In order to obtain the target forcing term f for all joints target Then, the regularization system (12) is used to force the target of all joints to target The corresponding phase variables x are obtained by coupling processing; The Gaussian kernel function is used to combine the target force terms f of all joints target The shape parameters representing the teaching trajectory are obtained by fitting the corresponding phase variable x according to the following formula. The fitting formula is as follows: Where N is the number of Gaussian kernel functions, ω i ,η i and ρ i are the first, second, and third shape parameters of the trajectory, x is the phase variable, is the Gaussian kernel function after truncation; The Gaussian kernel function after truncation Calculated as: Where, t i is the Gaussian kernel function The truncation moment of , i = 1,…,N, N represents the number of Gaussian kernel functions.

2. The robot skill transfer system based on truncated core dynamic motion primitives according to claim 1, characterized in that: The upper limb exoskeleton (1) is a serial mechanical exoskeleton.

3. The robot skill transfer system based on truncated core dynamic motion primitives according to claim 1, characterized in that: The industrial robot (3) is a serial-type industrial robot arm.

4. A robot skill transfer method based on truncated core dynamic motion primitives applied to the robot skill transfer system according to any one of claims 1 to 3, characterized in that: The method mainly includes the following steps: Step 1): The instructor wears an upper limb exoskeleton and controls the industrial robot through the teleoperation module to move according to the skill trajectory taught by the upper limb exoskeleton to complete the skill trajectory teaching; Step 2): During the skill teaching process, the teaching data acquisition module (6) in the robot skill learning module collects, processes and records the data of each joint of the industrial robot (3) in real time through the angle sensor as the teaching trajectory data; The truncated core dynamic motion primitive learning module (7) uses the acquired teaching trajectory data of the industrial robot (3) to learn the teaching trajectory and obtain the shape parameters of the teaching trajectory; Step 3): The trajectory reproduction module (9) generates reproduction trajectory data using the shape parameters of the teaching trajectory learned by the truncated kernel dynamic motion primitive learning module (7), and then controls the industrial robot (3) to move and reproduce the teaching trajectory.

5. The robot skill transfer method based on truncated core dynamic motion primitives according to claim 4, characterized in that: In the step 2), the data collected by the teaching data acquisition module for skill trajectory learning is the data of the six joints in the joint space of the industrial robot; the data of the six joints of the industrial robot are collected in real time during the dynamic movement of the industrial robot.

6. The robot skill transfer method based on truncated core dynamic motion primitives according to claim 1, characterized in that: The fitting is performed by using a local weighted regression algorithm to calculate the shape parameters of the teaching trajectory.

7. The robot skill transfer method based on truncated core dynamic motion primitives according to claim 4, characterized in that: In the step 3), specifically, the shape parameters of the teaching trajectory are used to calculate the acceleration of the reproduction trajectory by integrating the Gaussian kernel function truncation processing mechanism and the dynamic motion primitive learning model optimized by multiple shape parameters, and then the speed and position information of the reproduction trajectory are calculated by Euler integral. Finally, the calculated position, speed and acceleration information of the reproduction trajectory are used to control the industrial robot (3) to complete the reproduction of the skill trajectory according to the reproduction trajectory.

Images