A robotic arm control method and system based on deterministic learning and knowledge fusion

Abstract

The present invention proposes a robot arm control method and system based on deterministic learning and knowledge fusion, and a new neural network control strategy based on knowledge fusion. While ensuring system stability and tracking performance, the strategy achieves accurate approximation of closed-loop unknown dynamics in a larger learning domain. The new neural network control strategy based on knowledge fusion proposed in the present invention enhances the generalization ability and robustness of radial basis function neural networks in closed-loop control, as well as their portability in multiple control task scenarios.

Description

Technical Field

[0001] The present invention belongs to the technical field related to robot control, and in particular relates to a robot arm control method and system based on deterministic learning and knowledge fusion. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the development and progress of industrial automation, the application of industrial robots has been valued and promoted by the industry. Industrial robot systems are widely used in semiconductor manufacturing, automotive industry assembly, aerospace automation and other fields. At the same time, the problem of automatic control of robot systems has also become a hot topic in academia and industry. Existing research mainly focuses on improving the intelligence level of robot control systems, enhancing their environmental adaptability and realizing multi-task collaborative operations. By continuously optimizing control algorithms and introducing advanced learning mechanisms, researchers are committed to solving the problems of adaptive control and precise operation of robots in complex dynamic environments, thereby promoting the further development and application of industrial robot technology.

[0004] Robot control systems play a key role in modern industry and service fields. Due to their high precision, high speed and high programmability, they have become the core components of automated manufacturing and play a vital role in industrial production. The control of the robot involves multidisciplinary knowledge, and its design and implementation involve multiple complex technical links, including mechanical engineering, electronic engineering, computer science and control theory. An efficient robot control system requires not only precise motion control, but also high intelligence and adaptability. To achieve these goals, modern robot control systems usually integrate advanced control algorithms, sensor technology and artificial intelligence. In recent years, many studies have been devoted to applying advanced control theory and intelligent algorithms to robot systems. In terms of control algorithms, commonly used methods include proportional integral differential (PID) control, adaptive control and predictive control, and data-driven intelligent control methods (such as neural network control and fuzzy control) have been widely used and developed. These methods can provide stable and accurate control effects. Sensor technology is used to obtain information such as the position, speed, acceleration, etc. of the robot, and feed it back to the control system in real time to ensure that the movement of the robot meets the expected trajectory. The combination of these methods not only improves the control performance of the robot, but also promotes its application in more complex and dynamic industrial environments. In short, with the advancement of intelligent manufacturing, the control technology of robot arm systems will continue to innovate and develop. The future robot arms will be more intelligent, autonomous and flexible, and will be widely used in various fields such as manufacturing, services, medical care, agriculture, etc., promoting the overall improvement of social productivity. Through interdisciplinary cooperation and technological integration, the potential of robot arms will be fully utilized, bringing more convenience and efficiency to modern industry and daily life.

[0005] However, the robotic arm systems in actual industrial applications are often highly nonlinear and have unknown and uncertain information. The mathematical model of the controlled object is often difficult to obtain accurately, which leads to the inability of existing advanced control algorithms to obtain accurate and high-performance control effects in actual applications. Therefore, how to control such complex robotic systems with unknown nonlinearities is a problem worth studying. In the past few decades, many scholars have adopted a series of intelligent control algorithms for such complex systems, such as the neural network control algorithm and fuzzy logic controller mentioned above. However, the essence of true intelligent control is to learn dynamic knowledge in closed-loop control tasks and to use this knowledge in subsequent control tasks to further improve control performance and effects. The proposal of deterministic learning theory starts from the essence of intelligent control and solves the problem of adaptive neural network control learning from closed-loop control tasks. Therefore, based on the deterministic learning theory, some scholars have studied the deterministic learning neural network control of robotic arm systems and proved the effectiveness of this method in simulation experiments. However, it is worth noting that the existing deterministic learning control methods for robotic arm systems are all learning and controlling along a single reference trajectory (or control task), that is, a single task and a fixed environment. However, the actual environment is often a complex task and a changing environment. Therefore, it is necessary to further expand the knowledge representation domain of neural networks in closed-loop control, explore closed-loop learning and control problems in a larger trajectory range, and further improve the knowledge generalization and system robustness of the robotic arm system in the face of complex scenarios and multiple tasks.

[0006] In summary, the current problems with robot control are: Most of the current robot deterministic learning control methods are for single tasks and fixed environments, and have weak generalization and robustness in adapting to complex tasks and changing environments. Summary of the invention

[0007] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a robotic arm control method and system based on deterministic learning and knowledge fusion, which realizes intelligent and high-performance control of complex unknown robotic arm nonlinear systems in complex multi-task control scenarios.

[0008] To achieve the above object, a first aspect of the present invention provides a robot arm control method based on deterministic learning and knowledge fusion, comprising:

[0009] Construct reference trajectory model and continuous time model of the robotic arm system;

[0010] Based on the reference trajectory model and continuous time model of the robotic arm system, the closed-loop tracking error and tracking filter error of the robotic arm system are defined;

[0011] The closed-loop tracking error and weight estimation error of each robot system are used to construct a robot linear time-varying system;

[0012] The radial basis function neural network is used to identify the closed-loop unknown dynamics of the robotic arm system. Based on the deterministic learning theory and the exponential stability theorem of the robotic arm linear time-varying system, the radial basis function neural network weights corresponding to each robotic arm system converge to the common optimal weights, and the converged common optimal weights are saved as constant neural networks.

[0013] Constructing a knowledge fusion learning controller based on the tracking filter error and the learned constant neural network;

[0014] The constructed knowledge fusion learning controller is used to control the robotic arm system.

[0015] A second aspect of the present invention provides a robotic arm control system based on deterministic learning and knowledge fusion, comprising:

[0016] A first building module is used to build a reference trajectory model and a continuous time model of the robotic arm system;

[0017] A definition module is used to define a closed-loop tracking error and a tracking filter error of the robotic arm system based on a reference trajectory model and a continuous time model of the robotic arm system;

[0018] The second building module is used to build a linear time-varying system of the robot arm by using the closed-loop tracking error and the weight estimation error of each robot arm system;

[0019] The convergence module is used to identify the closed-loop unknown dynamics of the manipulator system using a radial basis function neural network. Based on the deterministic learning theory and the exponential stability theorem of the manipulator linear time-varying system, the weights of the radial basis function neural network corresponding to each manipulator system converge to the common optimal weights, and the converged common optimal weights are saved as constant value neural networks.

[0020] A third building module is used to build a knowledge fusion learning controller based on the tracking filter error and the learned constant neural network;

[0021] The control module is used to control the robotic arm system using the constructed knowledge fusion learning controller.

[0022] The third aspect of the present invention provides a computer device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, a robot arm control method based on deterministic learning and knowledge fusion is performed.

[0023] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, executes a robot arm control method based on deterministic learning and knowledge fusion.

[0024] One or more of the above technical solutions have the following beneficial effects:

[0025] The present invention proposes a new neural network control strategy based on knowledge fusion, which achieves accurate approximation of closed-loop unknown dynamics in a larger learning domain while ensuring system stability and tracking performance.

[0026] The new knowledge fusion neural network control strategy proposed in the present invention enhances the generalization ability and robustness of radial basis function neural networks in closed-loop control, as well as their portability in multiple control task scenarios.

[0027] When faced with different control tasks, the method of the present invention can achieve high-performance control based on the experience learning controller constructed by integrating closed-loop dynamic knowledge without the need for online neural network weight updates, embodying the concept of human-like learning and intelligent control in multi-task control scenarios.

[0028] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0030] Figure 1(a)-Figure 1(c) is a schematic diagram of different reference trajectories generated by the reference model in the first embodiment of the present invention;

[0031] Figure 2(a)-Figure 2(b) All of them are norm convergence diagrams of online knowledge fusion neural network in Embodiment 1 of the present invention;

[0032] Figure 3(a)-Figure 3(b) All of them are online knowledge fusion control effect diagrams in Example 1 of the present invention;

[0033] Figure 4(a)-Figure 4(b) All of them are offline knowledge fusion control effect diagrams in the first embodiment of the present invention;

[0034] Figure 5(a)-Figure 5(b) All of them are the fusion neural network fitting effect diagrams in Example 1 of the present invention;

[0035] Figure 6 This is a knowledge fusion learning control framework diagram in the first embodiment of the present invention. DETAILED DESCRIPTION

[0036] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0037] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0038] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0039] Embodiment 1

[0040] This embodiment discloses a robot arm control method based on deterministic learning and knowledge fusion, including:

[0041] Construct reference trajectory model and continuous time model of the robotic arm system;

[0042] Based on the reference trajectory model and continuous time model of the robotic arm system, the closed-loop tracking error and tracking filter error of the robotic arm system are defined;

[0043] The closed-loop tracking error and weight estimation error of each robot system are used to construct a robot linear time-varying system;

[0044] The radial basis function neural network is used to identify the closed-loop unknown dynamics of the manipulator system. Based on the deterministic learning theory and the exponential stability theorem of the manipulator linear time-varying system, the manipulator linear time-varying system converges to a small neighborhood near zero, that is, the radial basis function neural network weights corresponding to each manipulator system converge to the common optimal weights, and the converged common optimal weights are saved as constant neural networks.

[0045] Constructing a knowledge fusion learning controller based on the tracking filter error and the learned constant neural network;

[0046] The constructed knowledge fusion learning controller is used to control the robotic arm system.

[0047] In order to address the shortcomings and deficiencies of traditional deterministic learning neural network controllers in the intelligent control of robot systems in complex multi-task scenarios, this embodiment proposes a robot system knowledge fusion deterministic learning control method based on deterministic learning, aiming to achieve intelligent learning and high-performance control in complex multi-task control scenarios.

[0048] Below is combined with Figure 1- Figure 6 The control method proposed in this embodiment is described in detail:

[0049] The first part is the design of online knowledge fusion learning controller, which mainly includes the following steps:

[0050] Step (1): Establish the continuous time model of the n-joint robot and the reference trajectory model:

[0051]

[0052] in, They are respectively represented as the joint angle, joint velocity and joint acceleration of the i-th robot arm; τ i represents the input torque vector of the i-th controller; M(q i ) represents the unknown inertia matrix of the robot; represents the unknown Coriolis force matrix; G(q i ) represents the gravity vector, represents the unknown system friction.

[0053] Reference trajectory model:

[0054]

[0055] in, Represents the reference model system state; Y di ∈R n is the reference output of the i-th robotic arm system; F d (X di )∈R n is a smooth nonlinear function.

[0056] Step (2): Convert the above n-joint robot model into the form of state space expression:

[0057]

[0058] Among them, X i，1 =q i ∈R n , It is represented as the system state after conversion, corresponding to the joint angle information and joint velocity information in the original system respectively.

[0059]

[0060] Among them, A(X i )=-M -1 [F+G+BX i，2 ].

[0061] Step (3): Online knowledge fusion learning controller design, defining the tracking error of the i-th robotic arm system:

[0062]

[0063] Among them, e il represents the tracking error of the lth joint of the i-th robotic arm system, represents the output of the i-th reference system.

[0064] Define system preset performance functions:

[0065]

[0066] in, are two constants greater than 0, β l (t) is a monotonically decreasing performance constraint function.

[0067]

[0068] Among them, β l0 , β l∞ Represent the function β l (t) initial and final values.

[0069] The defined tracking error should satisfy the following constraints:

[0070] e il (t) = β l (t)T l (z il (t)) (8)

[0071] Where T l (z il ) represents the state transformation function, satisfying

[0072] Furthermore, we can get the following about the transformed unconstrained variable z il The closed-loop dynamic equation is:

[0073]

[0074] in,

[0075] Further, we can get:

[0076]

[0077] in,

[0078] Z i =[z i1 ,…,z in ] T ∈R n ,

[0079] Υ i =diag{γ i1 ,γ i2 ,…,γ in},

[0080]

[0081] Define the tracking filter error:

[0082] Among them, the unknown nonlinear dynamic A(X i ) can be fitted using the following neural network:

[0083] A(X i )=W *T S(X i )+∈ i (11)

[0084] Among them, ∈ i ∈R n represents the fitting error, represents the ideal weight matrix, S(X i ) represents the regression vector of the system.

[0085] Design the following neural network controller:

[0086]

[0087] Among them, K i represents the control gain;

[0088]

[0089] Design the following neural network weight update law:

[0090]

[0091] Among them, Γ, σ i , η is the weight update law parameter, a ij >0 indicates the communication coefficient, and represent the neural network weights of the i-th and j-th robotic arm systems respectively.

[0092] Furthermore, Lyapunov stability is used to prove that all closed-loop signals of the above-mentioned multi-manipulator closed-loop system are ultimately uniformly stable, and the tracking error of each system can converge exponentially to a small neighborhood near zero.

[0093] Step (4): Multi-robot collaborative learning and online knowledge fusion

[0094] In summary, we can further obtain the following linear time-varying (LTV) system that combines the preset performance closed-loop errors and weight update closed-loop errors of N robotic arm systems, where: The transformed system tracking error is used to eliminate the affine term:

[0095]

[0096] in,

[0097] in, is a symmetric matrix satisfying represents the estimated error of each neural network weight in each robot arm along the trajectory, represents the neural network approximation error of each robot along the trajectory, Γ ζ is the corresponding learning gain.

[0098] According to deterministic learning theory, when the input vector of the radial basis function neural network (RBF NN) satisfies the cyclic trajectory, the neurons along the trajectory can always meet the continuous excitation (PE) condition. Therefore, the regression matrix B of each subsystem of the closed-loop multi-manipulator system is i Meet PE conditions.

[0099] It can be further shown that the closed-loop LTV system (14) satisfies the collaborative PE condition. Through the exponential stability theorem of the LTV system and the perturbation theorem of the nonlinear system, it can be proved that the weight estimation error of each subsystem of the LTV system converges to a common optimal weight, that is, the unknown dynamics of different manipulators of each joint can be accurately identified by a common convergent neural network, and the converged common optimal weight is saved as a constant neural network, and a knowledge fusion learning controller is designed based on the trained constant radial basis function neural network.

[0100] Furthermore, the converged public weights are saved as the following constant neural network form:

[0101]

[0102] Among them, [t a , t b ] represents the steady-state stage after the neural network weights converge.

[0103] Then, based on the fusion constant neural network obtained through online learning, the following knowledge fusion learning controller is designed:

[0104]

[0105] The present embodiment below describes in detail the design of the offline knowledge fusion learning controller of the second part, specifically including:

[0106] Step (1): Determine the learning closed-loop training for a single control task;

[0107] Among them, the design steps of the deterministic learning preset performance controller for a single control task in step (1) are roughly the same as the design of the online knowledge fusion learning controller in the first part. The specific design steps can refer to (5)-(12) in the first part. Different from the first part, the weight update rate involved does not contain the interaction terms between other robot arm systems. The weight update rate of the neural network in this part is designed as follows:

[0108]

[0109] Introducing state transitions We can further get:

[0110]

[0111] in,

[0112] Among them, S ζ is the regressor vector near the neuron along the trajectory.

[0113] Furthermore, through the exponential stability theorem and perturbation theorem of the LTV system mentioned above, combined with the lemma of deterministic learning about PE conditions, it can be obtained that the above LTV system (19) satisfies exponential stability, that is, the neural network weight estimation error converges to a small neighborhood around zero.

[0114] Furthermore, the converged constant neural network under each control task can be obtained:

[0115]

[0116] Furthermore, the following linear equations are established for the neural network of each joint of the robotic arm system under different training tasks:

[0117]

[0118] Among them, W fi is the weight after fusion; W i1 is the constant neural network corresponding to the i-th control task.

[0119] Furthermore, the above offline knowledge fusion problem can be transformed into a least squares minimum norm solution problem for a linear equation system, and the offline fused neural network is obtained by the following calculation method:

[0120]

[0121] Furthermore, based on the offline fusion neural network obtained above, an offline knowledge fusion learning controller is constructed:

[0122]

[0123] Through online or offline fusion, intelligent high-performance control of complex unknown robotic arm nonlinear systems in complex multi-task control scenarios is achieved. The generalization and knowledge representation domain of neural networks in closed-loop control tasks are further expanded, thereby greatly improving the task portability and robustness of the robotic arm system.

[0124] This embodiment proposes a new neural network control strategy based on knowledge fusion. While ensuring system stability and tracking performance, the strategy achieves accurate approximation of closed-loop unknown dynamics in a larger learning domain.

[0125] The knowledge fusion neural network control strategy proposed in this embodiment enhances the generalization ability and robustness of radial basis function neural networks in closed-loop control, as well as their portability in multiple control task scenarios.

[0126] The method proposed in this embodiment can achieve high-performance control based on the experience learning controller constructed by integrating closed-loop dynamic knowledge without updating the weights of the online neural network when facing different control tasks. This reflects the concept of human-like learning and intelligent control in multi-task control scenarios.

[0127] Embodiment 2

[0128] The purpose of this embodiment is to provide a robotic arm control system based on deterministic learning and knowledge fusion, including:

[0129] A first building module is used to build a reference trajectory model and a continuous time model of the robotic arm system;

[0130] A definition module is used to define a closed-loop tracking error and a tracking filter error of the robotic arm system based on a reference trajectory model and a continuous time model of the robotic arm system;

[0131] The second building module is used to build a linear time-varying system of the robot arm by using the closed-loop tracking error and the weight estimation error of each robot arm system;

[0132] The convergence module is used to identify the closed-loop unknown dynamics of the manipulator system using a radial basis function neural network. Based on the deterministic learning theory and the exponential stability theorem of the manipulator linear time-varying system, the weights of the radial basis function neural network corresponding to each manipulator system converge to the common optimal weights, and the converged common optimal weights are saved as constant value neural networks.

[0133] A third building module is used to build a knowledge fusion learning controller based on the tracking filter error and the learned constant neural network;

[0134] Control module, used to control the robot system using the constructed knowledge fusion learning controller Example 3

[0135] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the program.

[0136] Embodiment 4

[0137] The purpose of this embodiment is to provide a computer-readable storage medium.

[0138] A computer-readable storage medium stores a computer program, which executes the steps of the above method when executed by a processor.

[0139] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0140] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0141] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A robot arm control method based on deterministic learning and knowledge fusion, characterized in that: include: Construct reference trajectory model and continuous time model of the robotic arm system; Based on the reference trajectory model and continuous time model of the robotic arm system, the closed-loop tracking error and tracking filter error of the robotic arm system are defined; The reference trajectory model and continuous time model of the robotic arm system are used to define the tracking filter error of the robotic arm system; specifically: The continuous time model of the robot system is converted into a state space expression, and the tracking error of the robot system is defined based on the state space expression of the robot system; Based on the constraints satisfied by the tracking error corresponding to each joint of the manipulator system, a closed-loop dynamic equation with no constrained variables is obtained after conversion; According to the closed-loop dynamic equation corresponding to each joint of the robotic arm system, the closed-loop dynamic equation of the robotic arm system is obtained; According to the closed-loop dynamic equation of the robot system, the tracking filter error of the robot system is defined; The closed-loop tracking error and weight estimation error of each robot system are used to construct a robot linear time-varying system; The radial basis function neural network is used to identify the closed-loop unknown dynamics of the robotic arm system. Based on the deterministic learning theory and the exponential stability theorem of the robotic arm linear time-varying system, the radial basis function neural network weights corresponding to each robotic arm system converge to the common optimal weights, and the converged common optimal weights are saved as constant neural networks. Constructing a knowledge fusion learning controller based on the tracking filter error and the learned constant neural network; The constructed knowledge fusion learning controller is used to control the robotic arm system.

2. A robot arm control method based on deterministic learning and knowledge fusion as claimed in claim 1, characterized in that: A neural network controller is constructed based on the tracking filter error and the neural network weights corresponding to the robotic arm system; and a neural network weight update law is designed by exchanging weight update information between the robotic arm systems; The radial basis function neural network is trained using the reference trajectory model of the robotic arm system. The converged common weights are saved in the form of a constant value neural network, and the constant value neural network is fused through online learning to obtain a knowledge fusion learning controller.

3. A robot arm control method based on deterministic learning and knowledge fusion as claimed in claim 2, characterized in that: The neural network weight update law is specifically: Among them, Γ, σ i , η is the weight update law parameter, a ij >0 indicates the communication coefficient, and denote the neural network weights of the i-th and j-th robotic arm systems, respectively, S(X i ) represents the regression vector of the system, Z si represents the tracking filter error, and the superscript T represents the transpose.

4. A method for controlling a robotic arm based on deterministic learning and knowledge fusion as claimed in claim 1, characterized in that: Constructing a neural network controller and a neural network weight update law based on the tracking filter error and the neural network weight corresponding to the robotic arm system; The radial basis function neural network is trained using the reference trajectory model of the robotic arm system; a linear equation group is constructed for the neural network of each joint of the robotic arm system under different training tasks; an offline fused neural network is obtained based on the constructed linear equation group; and a knowledge fusion learning controller is constructed based on the offline fused neural network.

5. A robot arm control method based on deterministic learning and knowledge fusion as claimed in claim 1, characterized in that: A state transformation is introduced based on the tracking filter error, and the state transformation is used to eliminate the affine term of the linear time-varying system of the robotic arm.

6. A method for controlling a robotic arm based on deterministic learning and knowledge fusion as claimed in claim 4, characterized in that: According to the linear equations, the knowledge fusion problem is converted into a least squares minimum norm solution problem of the linear equations, and the neural network after offline fusion is calculated.

7. A robotic arm control system based on deterministic learning and knowledge fusion, characterized in that: include: A first building module is used to build a reference trajectory model and a continuous time model of the robotic arm system; A definition module is used to define a closed-loop tracking error and a tracking filter error of the robotic arm system based on a reference trajectory model and a continuous time model of the robotic arm system; The reference trajectory model and continuous time model of the robotic arm system are used to define the tracking filter error of the robotic arm system; specifically: The continuous time model of the robot system is converted into a state space expression, and the tracking error of the robot system is defined based on the state space expression of the robot system; Based on the constraints satisfied by the tracking error corresponding to each joint of the manipulator system, a closed-loop dynamic equation with no constrained variables is obtained after conversion; According to the closed-loop dynamic equation corresponding to each joint of the robotic arm system, the closed-loop dynamic equation of the robotic arm system is obtained; According to the closed-loop dynamic equation of the robot system, the tracking filter error of the robot system is defined; The second building module is used to build a linear time-varying system of the robot arm by using the closed-loop tracking error and the weight estimation error of each robot arm system; The convergence module is used to identify the closed-loop unknown dynamics of the manipulator system using a radial basis function neural network. Based on the deterministic learning theory and the exponential stability theorem of the manipulator linear time-varying system, the weights of the radial basis function neural network corresponding to each manipulator system converge to the common optimal weights, and the converged common optimal weights are saved as constant value neural networks. A third building module is used to build a knowledge fusion learning controller based on the tracking filter error and the learned constant neural network; The control module is used to control the robotic arm system using the constructed knowledge fusion learning controller.

8. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, a robot arm control method based on deterministic learning and knowledge fusion as described in any one of claims 1 to 6 is performed.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the robot arm control method based on deterministic learning and knowledge fusion as described in any one of claims 1 to 6 is executed.

Images