A Feedforward Control Method and System for Robot Inverse Dynamics Based on Fusion Model

Abstract

The present invention proposes a robot inverse dynamics feedforward control method and system based on a fusion model, the method comprising: constructing a robot inverse dynamics mechanism model and a neural network data model; obtaining the robot's joint angular position error, angular velocity error and angular acceleration error; respectively calculating the output of the feedback controller, the output of the neural network data model and the output of the mechanism model; adding the output of the feedback controller, the output of the neural network data model and the output of the mechanism model to obtain the robot's control torque. The present invention couples the neural network data model on the basis of the robot's inverse dynamics mechanism model, can overcome the influence of the inaccuracy of the inverse dynamics mechanism model on the control performance, and improve the robot's trajectory tracking control accuracy and anti-interference ability.

Description

Technical Field

[0001] The present invention belongs to the technical field of robots, and in particular relates to a robot inverse dynamics feedforward control method and system based on a fusion model. Background Art

[0002] Industrial robot technology is a crucial piece of equipment that drives the transformation and upgrading of the manufacturing industry and enhances the overall strength and core competitiveness of China's manufacturing sector. All high-end industrial applications of industrial robots require the robot's end effector to achieve precise positioning, rapid response, and robustness in the presence of external interference and varying loads.

[0003] Motion performance has a significant impact on improving the production quality and efficiency of robots. Achieving high-quality motion performance for industrial robots requires controllers with high control accuracy, dynamic response, and stability. However, industrial robots are complex, nonlinear, and strongly coupled time-varying systems. Especially at high speeds, the robot's inertia varies significantly, significantly increasing nonlinear effects. Traditional closed-loop feedback control alone is unable to meet the high-quality motion requirements of industrial robots.

[0004] Feedforward control algorithms based on a robot's inverse dynamics model can, to a certain extent, overcome the nonlinear, time-varying characteristics of the robot and are considered the most direct and effective approach to improving the robot's dynamic performance and motion accuracy. However, the complex dynamics of industrial robots, characterized by parametric and nonparametric uncertainties, make accurate inverse dynamics models difficult to obtain, and model inaccuracies significantly impact control performance. Patent CN111736460A discloses a robot adaptive iterative learning control method and system. This invention addresses the impact of industrial robot parameter uncertainty on control performance to a certain extent while also taking into account the algorithm's convergence speed. However, it does not consider the impact of the robot's nonparametric uncertainty on control performance. Prior art literature has explored, building on traditional inverse dynamics feedforward control methods, using online identification to modify the robot's inverse dynamics feedforward model in real time. However, this approach only modifies the inertia matrix within the inverse dynamics model, resulting in limited improvement in model accuracy. Other literature has considered the impact of flexibility on robot dynamics, using flexibility torque as a dynamic feedforward for the robot. However, further research has been conducted to address the impact of parameter and nonparametric uncertainty in the robot's inverse dynamics model on robot control performance.

[0005] It can be seen that the existing feedforward control method for robot inverse dynamics still has some shortcomings in overcoming the parameter uncertainty and non-parameter uncertainty of the robot inverse dynamics model. Summary of the Invention

[0006] In response to the defects of the existing technology, the present invention proposes a robot inverse dynamics feedforward control method and system based on a fusion model, which is used to solve the problem that the existing robot inverse dynamics feedforward control method fails to overcome the parameter uncertainty and non-parameter uncertainty of the robot inverse dynamics model.

[0007] In a first aspect, the present invention discloses a robot inverse dynamics feedforward control method based on a fusion model, the method comprising:

[0008] Construct a mechanism model of robot inverse dynamics based on the theory of robot inverse dynamics, and construct a neural network data model of robot inverse dynamics based on artificial neural network;

[0009] Calculate the angular position error, angular velocity error and angular acceleration error of each joint of the robot;

[0010] Input the angular position error, angular velocity error and angular acceleration error of each joint into the feedback controller and calculate the output of the feedback controller;

[0011] Inputting the desired joint angular position, angular velocity, angular acceleration, joint angular position error, and angular velocity error of the robot into the neural network data model, calculating the weights of the neural network data model, and calculating the output of the neural network data model;

[0012] Input the desired joint angular position, angular velocity, and angular acceleration of the robot into the mechanism model and calculate the output of the mechanism model;

[0013] The output of the feedback controller, the output of the neural network data model, and the output of the mechanism model are added together to obtain the control torque of the robot.

[0014] On the basis of the above technical solution, preferably, the neural network data model of the robot inverse dynamics adopts an RBF neural network structure.

[0015] On the basis of the above technical solution, preferably, the feedback controller adopts the following control law:

[0016]

[0017] in, is the joint angle position error, is the angular velocity error, is the angular acceleration error, as well as is the control parameter matrix.

[0018] On the basis of the above technical solution, preferably, the weights of the neural network data model are updated by online adaptive adjustment according to the weight formula derived from Lyapunov's stability theorem.

[0019] Based on the above technical solution, preferably, calculating the weights of the neural network data model specifically includes:

[0020] Obtain the driving torque of each joint calculated by the mechanism model and the output of the neural network data model to obtain the overall control law of the robot's inverse kinematics:

[0021]

[0022] Based on the overall control law of the robot inverse kinematics, considering the parameter uncertainty and nonparametric uncertainty The impact of , calculate the control deviation equation of the closed-loop system:

[0023]

[0024] Simulating control deviations through neural network data models , and update the control deviation equation of the closed-loop system;

[0025] Based on the updated control deviation equation of the closed-loop system, the weight formula of the neural network data model is derived according to the Lyapunov stability theorem.

[0026] On the basis of the above technical solution, preferably, the control deviation equation based on the updated closed-loop system, according to Lyapunov stability theorem, derives the weight formula of the neural network data model, specifically including:

[0027] The control error is obtained by measuring the sensor , obtain the control deviation according to the updated control deviation equation of the closed-loop system ;

[0028] Calculating control error Deviation from control The transfer function between

[0029] Based on control error Deviation from control The transfer function between them calculates the state space form of the error equation of the entire robot system;

[0030] The Lyapunov function is defined in the state space form of the error equation of the entire robot system. The stability of the control system is analyzed according to the first-order derivative of the Lyapunov function, and the weight formula of the neural network data model under the stability of the control system is obtained.

[0031] On the basis of the above technical solution, preferably, the weight of the neural network data model is set to , Satisfies the equation:

[0032]

[0033] in is a constant matrix, and , , is the Gaussian basis function of the radial basis neural network, is the error function, , is a constant diagonal matrix, and .

[0034] In a second aspect, the present invention discloses a robot inverse dynamics feedforward control system based on a fusion model, the system comprising:

[0035] Expected input module: used to input the robot's expected joint angle position, angular velocity, and angular acceleration;

[0036] Error calculation module: used to calculate the angular position error, angular velocity error and angular acceleration error of each joint of the robot and input them into the feedback controller and data model calculation module;

[0037] Feedback controller: used to perform proportional and differential operations based on the angular position error, angular velocity error, and angular acceleration error of each joint, and output feedback control torque;

[0038] Data model calculation module: used to build a neural network data model for the robot's inverse dynamics based on an artificial neural network; input the robot's desired joint angular position, angular velocity, angular acceleration, joint angular position error, and angular velocity error into the neural network data model, calculate the weights of the neural network data model, and calculate the output of the neural network data model;

[0039] Mechanism model calculation module: used to build the mechanism model of robot inverse dynamics based on the robot inverse dynamics theory, input the robot's desired joint angular position, angular velocity, and angular acceleration into the mechanism model, and calculate the output of the mechanism model;

[0040] Summing module: used to add the output of the neural network data model and the output of the mechanism model to obtain the feedforward control torque, or used to add the output of the feedback controller, the output of the neural network data model and the output of the mechanism model to obtain the control torque of the robot.

[0041] A third aspect of the present invention discloses an electronic device, comprising: at least one processor, at least one memory, a communication interface, and a bus;

[0042] The processor, memory, and communication interface communicate with each other via the bus.

[0043] The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to the first aspect of the present invention.

[0044] According to a fourth aspect of the present invention, a computer-readable storage medium is disclosed, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to the first aspect of the present invention.

[0045] The present invention has the following beneficial effects compared to the prior art:

[0046] 1) This invention couples a neural network data model with the robot's inverse dynamics mechanism model and proposes a robot feedforward control method based on a hybrid inverse dynamics model. This method fully utilizes the advantages of both mechanism modeling and data modeling, overcomes the impact of the inaccuracy of the inverse dynamics mechanism model on control performance, and improves the robot's trajectory tracking control accuracy and anti-interference ability.

[0047] 2) The present invention considers the effects of parameter uncertainty and non-parameter uncertainty when analyzing the overall control law of the robot's inverse kinematics. Based on the control deviation, the weight formula of the neural network data model is derived according to the Lyapunov stability theorem. The weight of the neural network data model is adaptively adjusted and updated online according to the formula derived from the Lyapunov stability theorem to ensure the stability of the control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 A control block diagram of a robot inverse dynamics feedforward control method based on a fusion model provided in an embodiment of the present invention;

[0050] Figure 2 Schematic diagram of the structure of a robot inverse dynamics feedforward control system based on a fusion model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] Robot dynamics are complex, subject to parametric and non-parametric uncertainty. It is generally difficult to construct an accurate robot inverse dynamics model based solely on theoretical analysis. Model inaccuracies significantly impact the control performance of the robot's inverse dynamics feedforward control. To address this issue, the present invention proposes a robot inverse dynamics feedforward control method and system based on a fusion model. This method, which leverages the advantages of both mechanism and data modeling, effectively improves the accuracy and anti-interference capability of the robot's trajectory tracking control.

[0053] See also Figure 1 The robot inverse dynamics feedforward control method based on the fusion model of the present invention is mainly divided into a feedforward control part and a feedback control part. The feedback control part adopts a feedback controller, and the feedforward control part includes two parts: a mechanism model of the robot inverse dynamics and a neural network data model. The present invention fuses the mechanism model with the neural network data model to form a hybrid dynamics model, wherein the mechanism model is constructed based on the robot inverse dynamics theory, and the neural network data model is constructed based on the radial basis neural network. The neural network data model is used to compensate for the parameters and non-parametric uncertainties of the mechanism model.

[0054] See also Figure 1 The present invention discloses a robot inverse dynamics feedforward control method based on a fusion model, the method comprising:

[0055] S1. Construct a mechanism model of robot inverse dynamics based on the theory of robot inverse dynamics, and construct a neural network data model of robot inverse dynamics based on artificial neural network.

[0056] Based on the conventional robot inverse dynamics theory, a robot inverse dynamics mechanism model is constructed, and based on the artificial neural network, a robot inverse dynamics neural network data model is constructed. The robot inverse dynamics neural network data model can adopt an RBF neural network structure. In this embodiment, the neural network data model includes a weight adaptive update module and an RBF neural network model, such as Figure 1 The weights of the RBF neural network model are updated online by the weight adaptive update module according to the weight formula derived from the Lyapunov stability theorem. This mechanism model and the neural network data model together constitute the inverse dynamics feedforward of the present invention.

[0057] S2. Calculate the angular position error, angular velocity error, and angular acceleration error of each joint of the robot.

[0058] Get the actual joint angle position of each joint from the robot , actual joint angular velocity , and actual joint angular acceleration , the calculated position error of each joint angle of the robot is , the angular velocity error is, and the angular acceleration error is ,in, is the desired joint angle position of the robot, is the desired joint angular velocity, is the expected joint angular acceleration.

[0059] S3, the angular position error and angular velocity error of each joint And the angular acceleration error is input into the feedback controller to calculate the output of the feedback controller.

[0060] The angle position error of each joint , angular velocity error and angular acceleration error Input the feedback controller to perform proportional and differential operations, and according to the control law of the feedback controller, the following output is obtained:

[0061] (1)

[0062] in, as well as is the control parameter matrix, and n is the total number of robot joints.

[0063] S4. Input the desired joint angular position, angular velocity, angular acceleration, joint angular position error, and angular velocity error of the robot into the neural network data model, calculate the weights of the neural network data model, and calculate the output of the neural network data model.

[0064] Step S4 specifically includes the following sub-steps:

[0065] S41, will The weight adaptive update module in the neural network data model is input, and the weight of the neural network data model is calculated by the weight adaptive update module.

[0066] The process of calculating the weights of the neural network data model in step S41 specifically includes the following steps:

[0067] S411. Assume that the driving torque of each joint calculated based on the mechanism model is , the output of the neural network data model is , then the overall control law of the robot inverse kinematics can be expressed as follows:

[0068] (2)

[0069] is the total control torque, is the feedforward control torque, Feedback control torque

[0070] S412, based on the overall control law of robot inverse kinematics, taking into account parameter uncertainty and nonparametric uncertainty The influence of is calculated and the control deviation equation of the closed-loop system is calculated.

[0071] set up is the difference between the true moment and the estimated moment due to parameter inaccuracy, i.e., parameter uncertainty; is the load disturbance and other unmodeled states, that is, non-parametric uncertainty. For the robot controlled object, the following equation holds:

[0072] (3)

[0073] Substitute equation (3) into equation (2) and let , the control deviation equation of the closed-loop system is calculated as follows:

[0074] (4)

[0075] S413, simulate control deviation through neural network data model , and update the control deviation equation of the closed-loop system.

[0076] The radial basis neural network has a universal approximation characteristic, so the data model of the embodiment of the present invention adopts the radial basis neural network structure. represents the input of the neural network, represents the node of the hidden layer of the neural network, is the Gaussian basis function of the neural network, and:

[0077] (5)

[0078] In formula (5) The center point of the basis function, is the radial basis width. Then the output of the radial basis neural network is:

[0079] (6)

[0080] in is the weight of the neural network. Let the weight of the neural network be the ideal value , the output of the neural network can fully simulate , then:

[0081] (7)

[0082] Similarly, let the weights of the neural network be the estimated values , use the output of the neural network as the output of the data-driven model , then:

[0083] (8)

[0084] Substituting equations (7) and (8) into the control deviation equation (4), we can obtain the updated control deviation equation of the closed-loop system:

[0085] (9)

[0086] in, is the difference between the ideal weight and the estimated weight of the RBF network.

[0087] S414. Based on the updated control deviation equation of the closed-loop system, the weight formula of the neural network data model is derived according to Lyapunov's stability theorem.

[0088] S414 specifically includes the following steps:

[0089] a) Get the control error by measuring the sensor , according to the updated closed-loop system control deviation equation (9) to obtain the control deviation .

[0090] b) Calculate the control error Deviation from control The transfer function between .

[0091] Define the following error equation:

[0092] (10)

[0093] in, is a constant diagonal matrix, and .

[0094] The transfer function between and is:

[0095] (11)

[0096] Where s is the control signal, and t=1,2,…,n.

[0097] c) Based on control error Deviation from control The transfer function between them is used to calculate the state space form of the error equation of the entire robot system.

[0098] By choosing appropriate parameters The transfer function is a strictly positive real function, then according to the positive real lemma, there exists a positive definite matrix and , such that:

[0099] (12)

[0100] Among them, the matrix 、 and is the minimum state space realization of the robot's joint error equation, that is:

[0101] (13)

[0102] in , The state space form of the error equation of the entire robot system can be further written as follows:

[0103] (14)

[0104] In the formula 、 and are all diagonal modular matrices, and the matrix 、 and Located in the matrix 、 and On the diagonal position, If you remember , , then the following formula holds:

[0105] (15)

[0106] d) Define the Lyapunov function based on the state-space form of the error equation of the entire robot system. Perform control system stability analysis based on the first-order derivative of the Lyapunov function, and obtain the weight formula of the neural network data model under stable control system.

[0107] The Lyapunov function is defined as follows:

[0108] (16)

[0109] in, and Taking the derivative of the Lyapunov function and combining equations (9), (14) and (15), we can obtain:

[0110] (17)

[0111] Therefore, if the , then the first derivative of the Lyapunov function is , obviously it is less than 0, so the control system is stable.

[0112] Depend on It can be seen that:

[0113] (18)

[0114] Simplified and organized, the weight adaptive update law of the neural network data model is as follows:

[0115] (19)

[0116] In summary, as long as the neural network weights in the neural network data model satisfy the change rule of formula (19), the control system can be guaranteed to be stable. Based on formula (19), the weight formula is expressed as:

[0117] (20)

[0118] in is a constant matrix, and , , is the Gaussian basis function of the radial basis neural network, is the error function, , is a constant diagonal matrix, and .

[0119] Based on the hybrid dynamics model composed of the robot inverse dynamics mechanism model and the neural network data model, the present invention performs control system stability analysis according to the Lyapunov stability theorem, and derives the weight formula of the neural network data model under the stability of the control system. According to the weight formula, the weight is adjusted and updated online adaptively to ensure the stability of the control system.

[0120] S42, input the weight output by the weight adaptive update module into the RBF neural network model, and obtain the output of the neural network data model as .

[0121] S5. Input the desired joint angular position, angular velocity, and angular acceleration of the robot into the mechanism model, and calculate the output of the mechanism model.

[0122] Will Input the robot inverse dynamics mechanism model for calculation, and the mechanism model output is .

[0123] S6. Add the output of the feedback controller, the output of the neural network data model, and the output of the mechanism model to obtain the control torque of the robot. , .

[0124] Based on the traditional closed-loop feedback control, the present invention adds inverse dynamics feedforward, which includes a mechanism model and a neural network data model. The inverse dynamics mechanism model is constructed based on the robot inverse dynamics theory, and the neural network data model is used to compensate for the mechanism model parameters and non-parametric uncertainties, which can effectively improve the accuracy and anti-interference ability of the robot's trajectory tracking control.

[0125] See also Figure 2 The present invention proposes a robot inverse dynamics feedforward control system based on a fusion model, the system comprising:

[0126] Expected input module 1: used to input the robot's expected joint angle position, angular velocity, and angular acceleration;

[0127] Error calculation module 2: used to calculate the angular position error, angular velocity error and angular acceleration error of each joint of the robot and input them into the feedback controller 5 and the data model calculation module 4;

[0128] The feedback controller 3 is used to perform proportional and differential operations based on the angular position error, angular velocity error and angular acceleration error of each joint, and output feedback control torque;

[0129] Data model calculation module 4: used to construct a robot inverse dynamics mechanism model based on the robot inverse dynamics theory, input the robot's desired joint angular position, angular velocity, angular acceleration, joint angular position error, and angular velocity error into the neural network data model, calculate the weights of the neural network data model, and calculate the output of the neural network data model;

[0130] The mechanism model calculation module 5 is used to construct a mechanism model of the robot's inverse dynamics based on the robot's inverse dynamics theory, input the robot's desired joint angular position, angular velocity, and angular acceleration into the mechanism model, and calculate the output of the mechanism model;

[0131] Summing module 6: used to add the output of the neural network data model and the output of the mechanism model to obtain the feedforward control torque, or to add the output of the feedback controller, the output of the neural network data model and the output of the mechanism model to obtain the control torque of the robot.

[0132] The above system embodiments and method embodiments correspond to each other. For a brief description of the system embodiments, please refer to the method embodiments.

[0133] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus; wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the aforementioned method of the present invention.

[0134] The present invention also discloses a computer-readable storage medium storing computer instructions that cause the computer to implement all or part of the steps of the method described in the embodiments of the present invention. The storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0135] The system embodiment described above is merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be distributed across multiple network units. A person skilled in the art may, without inventive effort, select some or all of the modules as needed to achieve the objectives of this embodiment.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A robot inverse dynamics feedforward control method based on a fusion model, characterized in that: The method comprises: Construct the mechanism model of robot inverse dynamics based on the robot inverse dynamics theory, and construct the neural network data model of robot inverse dynamics based on artificial neural network; Obtain the angular position error, angular velocity error and angular acceleration error of each joint of the robot; Inputting the angular position error, angular velocity error and angular acceleration error of each joint into the feedback controller, and calculating the output of the feedback controller; Inputting the desired joint angular position, angular velocity, angular acceleration, joint angular position error, and angular velocity error of the robot into the neural network data model, calculating the weights of the neural network data model, and calculating the output of the neural network data model; Input the desired joint angular position, angular velocity, and angular acceleration of the robot into the mechanism model, and calculate the output of the mechanism model; The output of the feedback controller, the output of the neural network data model and the output of the mechanism model are added together to obtain the control torque of the robot.

2. The robot inverse dynamics feedforward control method based on the fusion model according to claim 1 is characterized in that: The neural network data model of the robot inverse dynamics adopts the RBF neural network structure.

3. The robot inverse dynamics feedforward control method based on the fusion model according to claim 1 is characterized in that: The feedback controller adopts the following control law: Among them, u fb is the output of the feedback controller, e is the joint angle position error, is the angular velocity error, is the angular acceleration error, as well as is the control parameter matrix.

4. The robot inverse dynamics feedforward control method based on the fusion model according to claim 1 is characterized in that: The weights of the neural network data model are updated online through adaptive adjustment based on the weight formula derived from Lyapunov's stability theorem.

5. The robot inverse dynamics feedforward control method based on fusion model according to claim 3 is characterized in that: Calculating the weights of the neural network data model specifically includes: Get the driving torque of each joint calculated by the mechanism model and the output of the neural network data model The overall control law of the robot inverse kinematics is obtained: u is the total control torque; Based on the overall control law of the robot inverse kinematics, the parameter uncertainty Δτ and the non-parameter uncertainty f are considered. d The influence of f = Δτ + f d , calculate the control deviation equation of the closed-loop system: Simulating control deviations through neural network data models And update the control deviation equation of the closed-loop system; Based on the updated control deviation equation of the closed-loop system, the weight formula of the neural network data model is derived according to the Lyapunov stability theorem.

6. The robot inverse dynamics feedforward control method based on fusion model according to claim 5 is characterized in that: The control deviation equation of the updated closed-loop system, which is derived from the weight formula of the neural network data model according to the Lyapunov stability theorem, specifically includes: The control error E is obtained by sensor measurement, and the control deviation U is obtained according to the updated control deviation equation of the closed-loop system; Calculate the transfer function between the control error E and the control deviation U; Calculate the state space form of the error equation of the entire robot system based on the transfer function between the control error E and the control deviation U; The Lyapunov function is defined in the state space form based on the error equation of the entire robot system. The stability of the control system is analyzed according to the first-order derivative of the Lyapunov function, and the weight formula of the neural network data model under the stability of the control system is obtained.

7. The robot inverse dynamics feedforward control method based on fusion model according to claim 6 is characterized in that: Assume that the weights of the neural network data model are Satisfies the equation: Where Γ is a constant matrix, Γ = diag(λ1, λ2, ..., λ n ) and λ i >0, i=1, 2, ... n, h(x) is the Gaussian basis function of the radial basis neural network, E is the error function, ψ is a constant diagonal matrix, ψ=diag(ψ1,ψ2,...,ψ n ) and ψ i >0.

8. A robot inverse dynamics feedforward control system based on a fusion model, characterized in that: The system comprises: Expected input module: used to input the robot's expected joint angle position, angular velocity, and angular acceleration; Error calculation module: used to calculate the angular position error, angular velocity error and angular acceleration error of each joint of the robot and input them into the feedback controller and data model calculation module; Feedback controller: used to perform proportional and differential operations based on the angular position error, angular velocity error and angular acceleration error of each joint, and output feedback control torque; Data model calculation module: used to construct a neural network data model of robot inverse dynamics based on artificial neural network; input the robot's desired joint angle position, angular velocity, angular acceleration, joint angle position error, angular velocity error into the neural network data model, calculate the weight of the neural network data model, and calculate the output of the neural network data model; Mechanism model calculation module: used to build the mechanism model of robot inverse dynamics based on the robot inverse dynamics theory, input the robot's expected joint angle position, angular velocity, and angular acceleration into the mechanism model, and calculate the output of the mechanism model; Summation module: used to add the output of the neural network data model and the output of the mechanism model to obtain the feedforward control torque, or to add the output of the feedback controller, the output of the neural network data model and the output of the mechanism model to obtain the control torque of the robot.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; Wherein, the processor, memory, and communication interface communicate with each other via the bus; The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images