Franka-panda robot arm fault-tolerant control method and device based on data driving and data packet loss compensation

Abstract

The application discloses a data-driven Franka-Panda robot arm fault-tolerant tracking control method and device, and the method comprises the following steps: 1) collecting system historical input and output information, and constructing a dynamic data model containing fault information; 2) establishing a fault compensation mechanism based on a radial basis function neural network; 3) constructing a sensor data packet loss compensation algorithm based on the dynamic data model; and 4) deducing a data-driven fault-tolerant controller. The method based on the data packet loss compensation and the data-driven fault compensation ensures good control effect in the case that the system model is unknown.

Description

Technical Field

[0001] This invention belongs to the field of control technology, specifically providing a fault-tolerant control method for the Franka-Panda robotic arm based on data-driven operation and data packet loss compensation, which ensures good intelligent control performance. Background Technology

[0002] Flexible robotic arms possess high flexibility and adaptability, and high-precision control methods enable them to adapt to different work scenarios and task requirements. By employing technologies such as adaptive control, model predictive control, and reinforcement learning, robotic arms can acquire adaptive capabilities in the face of uncertainty and environmental changes, achieving more flexible and intelligent operation. These research findings contribute to improving the performance and reliability of flexible robotic arms, further promoting their application in manufacturing.

[0003] The dynamics of flexible robotic arms are complex, and dynamic parameters such as the inertia matrix on the link side are difficult to obtain accurately, which greatly limits the practical application of related model-based robotic arm control algorithms. Data-driven control utilizes sensor data to provide real-time feedback on the position, velocity, and force information of the robotic arm. By adjusting the control commands of the joint motors through control algorithms, high-precision motion control can be effectively achieved. This real-time feedback and adjustment capability can improve the motion accuracy and stability of the robotic arm, enabling it to perform high-precision tasks in complex working environments.

[0004] Furthermore, the complex operating environment of robotic arms inevitably leads to network congestion, sensor malfunctions, and data packet loss, posing a significant research challenge for robotic arm control systems. Designing a data-driven, fault-tolerant tracking control method can leverage historical system data to design fault compensation, enhancing system fault tolerance and improving the robotic arm's trajectory tracking accuracy. By designing a compensation method based on a dynamic data model, the system can maintain high tracking performance even in the event of sensor data loss. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides a fault-tolerant control method for the Franka-Panda robotic arm based on data-driven operation and packet loss compensation, ensuring better control performance.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] The Franka-Panda robotic arm fault-tolerant control method based on data-driven operation and packet loss compensation includes the following steps:

[0008] 1) Collect historical input and output information from the system and establish a dynamic data model that includes fault information:

[0009] Consider the following robot system:

[0010] y(k+1)=g(y(k),...,y(kn a ),u(k),...,u(kn b ))+f s (k) (1)

[0011] Where y(k) represents the angle in the joint space of the robotic arm, u(k) represents the control input torque, g(.) is an unknown function, and f s (k) represents the system's fault information, n a and n b These are unknown parameters;

[0012] To design a joint space flexible robotic arm controller, equation (1) is transformed into the following form through derivation:

[0013]

[0014] Where Φ1(k)=[φ1(k),φ2(k),...,φ L (k)] T φ(k) is the pseudo-partial derivative parameter of the system, f s (k) represents the fault function. The vector is unknown;

[0015] 2) Design a fault compensation algorithm based on radial basis function neural network, the process is as follows:

[0016] It was found from the dynamic model (2) that the control performance of the robot controller is affected by the fault function. The influence of radial basis functions. Due to the nonlinear nature of radial basis functions, the training output of a radial basis function neural network can compensate for nonlinear fault functions.

[0017] Define network input: R(k) = [y(k)ε(k)] T , where ε(k)=y * (k)-y(k), the compensation values ​​of the following neural network are obtained:

[0018] ξ(k)=ω *T ψ(k)+σ(k) (3)

[0019] Where ω * Let ψ(k) be the expected weight, ψ(k) be the Gaussian vector, and σ(k) be the approximation error.

[0020] The following online weight estimation method is designed:

[0021]

[0022] in These are parameters that can be freely chosen.

[0023] Therefore, the following network output equation is determined:

[0024]

[0025] 3) Design a sensor data loss compensation algorithm based on a dynamic data model. The process is as follows:

[0026] Substituting (5) into (2), we obtain a new system dynamic data model:

[0027]

[0028] Define variables α(k) and

[0029]

[0030]

[0031] in, It is the predicted value of the system output angle y(k).

[0032] Therefore, when sensor data y(k) is lost, it can be effectively estimated using (7)-(8):

[0033]

[0034] 4) The data-driven fault-tolerant control algorithm is derived as follows:

[0035] At each time step, key parameters are estimated to minimize a predefined cost function, leading to the design of a data-driven controller. This includes the following steps:

[0036] 4.1) Construct a parameter estimation algorithm based on input and output data:

[0037] Define the following criterion function that includes tracking error:

[0038]

[0039] Substitute (2) into (10) and proceed. The following optimal estimation algorithm can be obtained:

[0040]

[0041] Where η∈(0,1] can increase the flexibility of the controller.

[0042] 4.2) Solving for the fault-tolerant controller:

[0043] Introduce the following input criterion function:

[0044]

[0045] Where λ>0 is a parameter that is manually selected.

[0046] Minimizing the criterion function yields the following control input:

[0047]

[0048] A second aspect of the present invention relates to a data-driven fault-tolerant tracking control device for a Franka-Panda robotic arm, comprising a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the data-driven fault-tolerant tracking control method for a Franka-Panda robotic arm of the present invention.

[0049] A third aspect of the invention relates to a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the data-driven Franka-Panda robotic arm fault-tolerant tracking control method of the present invention.

[0050] The technical concept of this invention is as follows: when the system model is completely unknown, a dynamic data model is established using the historical input and output data of the robotic arm; a radial basis function neural network is designed to dynamically compensate for the system's fault function; in addition, a sensor data loss compensation algorithm is designed to reasonably estimate the lost data based on the constructed system dynamic equation, and finally realizes intelligent fault-tolerant tracking control of the flexible robotic arm.

[0051] The robot used in the experiment was the Franka Emika Panda flexible collaborative arm. Each joint of the Franka robot is equipped with a high-precision torque sensor. Its control frequency is 1000Hz, and it achieves high-speed real-time communication based on the User Datagram Protocol (UDP). It has two real-time control interfaces: the first is to send joint or Cartesian position and velocity signals to the robot, and the robot's internal controller follows the corresponding motion signals to achieve robot motion control; the second is a torque controller, which directly controls the robotic arm by sending torque signals. This experiment primarily uses the torque control mode. The host computer for the experiment is an Intel Core i7-8700K processor running Ubuntu 16.04 LTS. To achieve the 1000Hz control frequency for real-time robot control, the Ubuntu system uses the real-time kernel Linux 4.14.12-rt10. The software platform is based on the Robot Operating System (ROS), version Kinetic Kame, used to coordinate and manage the communication and interaction between the robot's hardware and software components.

[0052] Compared with general control systems, flexible robotic arm systems face the following challenges in practical applications:

[0053] First, a flexible robotic arm system is an open system that needs to interact with its external environment. Compared to a closed, independent system, a flexible robotic arm needs to sense and respond to changes in the external environment to adapt to constantly changing task requirements and working conditions.

[0054] Second: Robotic arm control systems typically require data transmission and communication via networks. In complex network environments, network congestion can lead to data transmission delays, packet loss, and communication interruptions. Lost data packets can cause delays or errors in control commands.

[0055] The above issues significantly increase the complexity of robotic arm control. Therefore, only by solving these problems can the precise control effect of the robotic arm system be guaranteed.

[0056] This invention provides a fault-tolerant control method for the Franka-Panda robotic arm based on data-driven operation and sensor data packet loss compensation. It derives an incremental dynamic data time-varying model using historical input and output information of the system; dynamically compensates for the system's fault function by designing a radial basis function neural network; and designs a sensor data loss compensation algorithm to estimate the lost sensor data based on the constructed system dynamic equations. The control effect of this method can meet the requirements of practical applications and does not require internal system knowledge.

[0057] The beneficial effects of this invention are: when the system model is unknown, the method based on packet loss compensation and data-driven fault compensation ensures a good control effect. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the Franka-Panda robotic arm control platform.

[0059] Figure 2 This is a comparison chart of the fault-tolerant control effects of data-driven robotic arms;

[0060] Figure 3 This is a comparison chart of the packet loss control effects of a robotic arm. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be further described below in conjunction with the accompanying drawings and actual experiments.

[0062] Example 1

[0063] Reference Figures 1-3 A data-driven fault-tolerant tracking control method for the Franka-Panda robotic arm is proposed. This method derives an incremental dynamic linear time-varying model using system input and output data. It also estimates lost sensor output data by designing a data loss compensation algorithm. Furthermore, an online RBF neural network is designed based on the dynamic linearized data model of the control system to compensate for system faults.

[0064] This embodiment of a data-driven, fault-tolerant tracking control method for the Franka-Panda robotic arm includes the following steps:

[0065] 1) Collect historical input and output information of the system to build a dynamic data model;

[0066] 2) Establish a fault compensation mechanism based on neural networks;

[0067] 3) Construct a compensation algorithm for sensor data packet loss;

[0068] 4) Derive a data-driven fault-tolerant controller.

[0069] In step 1), historical input / output information of the system is collected to establish a dynamic data model containing fault information.

[0070] Consider the following robot system:

[0071] y(k+1)=g(y(k),...,y(kn a ),u(k),...,u(kn b))+f s (k) (1)

[0072] Where y(k) represents the angle in the joint space of the robotic arm, u(k) represents the control input torque, g(.) is an unknown function, and f s (k) represents the system's fault information, n a and n b These are unknown parameters;

[0073] To design a joint space flexible robotic arm controller, equation (1) is transformed into the following form through derivation:

[0074]

[0075] Where Φ1(k)=[φ1(k),φ2(k),...,φ L (k)] T φ(k) is the pseudo-partial derivative parameter of the system, f s (k) represents the fault function. The vector is unknown;

[0076] In step 2), a fault compensation algorithm based on a radial basis function neural network is designed, and the process is as follows:

[0077] It was found from the dynamic model (2) that the control performance of the robot controller is affected by the fault function. The influence of radial basis functions. Due to the nonlinear nature of radial basis functions, the training output of a radial basis function neural network can compensate for nonlinear fault functions.

[0078] Define network input: R(k) = [y(k)ε(k)] T , where ε(k)=y * (k)-y(k), the compensation values ​​of the following neural network are obtained:

[0079] ξ(k)=ω *T ψ(k)+σ(k) (3)

[0080] Where ω * Let ψ(k) be the expected weight, ψ(k) be the Gaussian vector, and σ(k) be the approximation error.

[0081] The following online weight estimation method is designed:

[0082]

[0083] in These are parameters that can be freely chosen.

[0084] Therefore, the following network output equation is determined:

[0085]

[0086] In step 3), a sensor data loss compensation algorithm based on a dynamic data model is designed:

[0087] Substituting (5) into (2), we obtain a new system dynamic data model:

[0088]

[0089] Define variables α(k) and

[0090]

[0091]

[0092] in, It is the predicted value of the system output angle y(k).

[0093] Therefore, when sensor data y(k) is lost, it can be effectively estimated using (7)-(8):

[0094]

[0095] In step 4), the data-driven fault-tolerant control algorithm is derived as follows:

[0096] At each time step, key parameters are estimated to minimize a predefined cost function, leading to the design of a data-driven controller. This includes the following steps:

[0097] 1. Construct a parameter estimation algorithm based on input and output data:

[0098] Define the following criterion function that includes tracking error:

[0099]

[0100] Substitute (2) into (10) and proceed. The following optimal estimation algorithm can be obtained:

[0101]

[0102] Where η∈(0,1] can increase the flexibility of the controller.

[0103] 2. Solving for the fault-tolerant controller:

[0104] Introduce the following input criterion function:

[0105]

[0106] Where λ>0 is a parameter that is manually selected.

[0107] Minimizing the criterion function yields the following control input:

[0108]

[0109] In this embodiment, the joint position q(0) of the system is initialized to 0;

[0110] Furthermore, a fault compensation algorithm based on neural networks is constructed, as follows:

[0111] The control period T is set to 0.001s, meaning that data obtained from the system every 0.001s is used to construct the dynamic data model. Parameters λ = 0.55, ρ = 0.88, μ = 0.1 are set. For each iteration, the weights φ(k) are updated based on the measurement data along the trajectory. The neural network parameters are set as follows:

[0112] σ=[1.41,2.12,0.43,0.48,0.16,0.41] T

[0113]

[0114] Solving using nearest neighbor data points in each control cycle That is, within each time interval (T = 0.001s).

[0115] To demonstrate the superiority of this method, a comparison with the traditional data-driven strategy MFAC is added. Furthermore, to highlight the excellent fault-tolerant performance of this method, the fault-tolerant control algorithm MFAFTC is added to the comparative experiments.

[0116] From practical results ( Figure 2 It can be seen that a significant fault occurred in the system at 20 seconds. After the fault occurred, the MFAC algorithm had the largest tracking error. Furthermore, compared with the MFATTC algorithm, this method has a faster convergence speed and a relatively smaller tracking error.

[0117] To demonstrate that this method can effectively address situations where some sensor data is lost, a comparative experiment was conducted using traditional data-driven control methods, which suffer from a 40% sensor data loss rate. Figure 3 It can be seen that the error of traditional control methods changes significantly after sensor data packet loss. Furthermore, from a transient performance perspective, traditional data-driven controllers exhibit severe overshoot after sensor data packet loss.

[0118] Example 2

[0119] This embodiment relates to a data-driven Franka-Panda robotic arm fault-tolerant tracking control device, including a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the data-driven Franka-Panda robotic arm fault-tolerant tracking control method of Embodiment 1.

[0120] Example 3

[0121] This embodiment relates to a computer-readable storage medium storing a program that, when executed by a processor, implements the data-driven Franka-Panda robotic arm fault-tolerant tracking control method of Embodiment 1.

[0122] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. A data-driven, fault-tolerant tracking control method for the Franka-Panda robotic arm, comprising the following steps: 1) Collect historical input and output information of the system and construct a dynamic data model containing fault information; 2) Establish a fault compensation mechanism based on radial basis function neural networks; 3) Construct a compensation algorithm for sensor data packet loss based on a dynamic data model; 4) Derive a data-driven fault-tolerant controller; Step 1) specifically includes: The following robot system is used: (1) in Indicates the angle in the joint space of the robotic arm. Indicates the control input torque. For unknown functions, Represents the fault function. and These are unknown parameters; To design a joint space flexible robotic arm controller, equation (1) is transformed into the following form through derivation: （2） in , These are the pseudo-partial derivative parameters of the system. The vector is unknown; Step 2) specifically includes: It can be seen from formula (2) that the control performance of the robot controller is affected by the fault function. The impact; due to the nonlinear nature of radial basis functions, the training output of a radial basis function neural network can compensate for nonlinear fault functions. ; Define network input: ,in The following compensation values ​​for the neural network were obtained: （3） in For expected weights, It is a Gaussian vector. To approximate the error; The following online weight estimation method is designed: （4） in These are parameters that can be freely chosen; Therefore, the following network output equation is determined: （5） Step 3 specifically includes: Substituting equation (5) into equation (2), we obtain the new system dynamic data model: （6） Define variables and : （7） （8） in, It is the system output angle The predicted value; Therefore, in sensor data When lost, use equations (8)-(9) for effective estimation: （9） Step 4) specifically includes: at each time step, estimating key parameters to minimize a predefined cost function, and then designing a data-driven controller, including the following steps:

41. Construct a parameter estimation algorithm based on input and output data: Define the following criterion function that includes tracking error: （10） Substitute (2) into (10) and proceed. The following optimal estimation algorithm is obtained: （11） in It can increase the flexibility of the controller; 42. Solve for the fault-tolerant controller: Introduce the following input criterion function: （12） in Parameters selected by humans; Minimizing the criterion function yields the following control input: （13）。 2. The data-driven, fault-tolerant tracking control method for the Franka-Panda robotic arm as described in claim 1, characterized in that: Construct a sensor data loss compensation algorithm based on a dynamic data model, including: The control period T is set as follows: That is, using each Build a dynamic data model from data obtained from the system; set parameters. For each iteration, the weights are updated by considering the measurement data along the trajectory. The neural network parameters are set as follows: Solving using nearest neighbor data points in each control cycle That is, within each time interval .

3. A data-driven, fault-tolerant tracking control device for the Franka-Panda robotic arm, characterized in that, The system includes a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the data-driven Franka-Panda robotic arm fault-tolerant tracking control method according to any one of claims 1-2.

4. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements the data-driven Franka-Panda robotic arm fault-tolerant tracking control method as described in any one of claims 1-2.

Images