Robot control method, electronic device, storage medium and program product

Abstract

The invention discloses a robot control method, electronic equipment, a storage medium and a program product, and relates to the technical field of robot automation control, and the method comprises the steps: determining a neural network compensation amount through a radial basis function neural network according to a total slave end force, a first robot output and a first system state estimation value; determining a second system state estimation value through a preset third-order linear expansion state observer and a neural network compensation amount; determining a target control quantity according to the received control instruction and the second system state estimation value; and controlling the robot to execute actions according to the target control quantity to obtain second robot output. According to the method, the neural network compensation amount is generated by introducing the radial basis function neural network, uncertain disturbance caused by scene changes and individual differences can be dynamically and accurately compensated, and the accuracy of robot control in a real scene is improved.

Description

Technical Field

[0001] This application relates to the field of robot automation control technology, and in particular to robot control methods, electronic devices, storage media and program products. Background Technology

[0002] By introducing robotics-related theories into interventional vascular surgery and leveraging emerging technologies such as automated control, sensor technology, computer-aided technology, and electronic information technology, doctors can remotely control the movement of catheters and guidewires using interventional vascular surgery robots.

[0003] Currently, in the control system of vascular interventional surgical robots, ADRC (Active Disturbance Rejection Control) technology usually estimates and compensates for disturbances in the system through Extended State Observer (ESO). However, in the process of ESO estimating the system state, the adjustment of some parameters depends on the manual completion of expert experience, which cannot adapt to different surgical scenarios and individual patient differences, and is prone to leading to a decrease in control accuracy.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this application is to provide a robot control method, electronic device, storage medium, and program product, aiming to solve the technical problem of how to improve the accuracy of robot control.

[0006] To achieve the above objectives, this application proposes a robot control method, the method comprising: The neural network compensation amount is determined by using a radial basis function neural network based on the total slave force, the first robot output, and the first system state estimate. The second system state estimate is determined by using a preset third-order linear extended state observer and the neural network compensation value, wherein the first system state estimate is the corresponding value of the second system state estimate in the previous control cycle. The target control quantity is determined based on the received control command and the second system state estimate. The robot is controlled to perform actions according to the target control quantity to obtain the second robot output, wherein the first robot output is the corresponding value of the second robot output in the previous control cycle.

[0007] In one embodiment, the first system state estimate includes a first position estimate. Prior to the step of determining the neural network compensation amount using a radial basis function neural network based on the total slave force, the first robot output, and the first system state estimate, the method further includes: The observation error is determined based on the first position estimate and the first robot output; By combining the online gradient descent algorithm and the momentum accumulation mechanism, the network parameters of the radial basis function neural network are updated to minimize the observation error. The network parameters include output weights, node center vectors, and basis width parameters.

[0008] In one embodiment, before the step of determining the second system state estimate using a preset third-order linear extended state observer and the neural network compensation quantity, the method further includes: Based on the pre-defined second-order robot control model, construct the state-space expression; The total perturbation in the state-space expression is expanded into new state variables to construct the third-order linearly extended state observer.

[0009] In one embodiment, the control command includes an acceleration command, the second system state estimate includes a total disturbance estimate, and the step of determining the target control quantity based on the received control command and the second system state estimate includes: Based on the total slave force, determine the feedforward compensation amount, and based on the acceleration command, determine the feedforward control amount; The initial control quantity is determined based on the feedforward control quantity, the feedforward compensation quantity, and the second system state estimate. The target control quantity is generated based on the preliminary control quantity and the total disturbance estimate.

[0010] In one embodiment, the control command includes a speed command and a position command, the second system state estimate includes a second position estimate and a speed estimate, and the step of determining the preliminary control quantity based on the feedforward control quantity, the feedforward compensation quantity, and the second system state estimate includes: Based on the position command and the second position estimate, determine the position error, and based on the speed command and the speed estimate, determine the speed error; The position error and the velocity error are used to determine the adjustment amount of the controller gain through a fuzzy controller, wherein the controller gain includes differential gain and proportional gain; The preliminary control quantity is determined based on the feedforward control quantity, the feedforward compensation quantity, the position error, the speed error, and the adjustment amount of the controller gain.

[0011] In one embodiment, before the step of determining the preliminary control quantity based on the control command and the preset system state estimate, the method further includes: The received original command is smoothed according to a preset parabolic acceleration / deceleration algorithm to generate the control command.

[0012] In one embodiment, prior to the step of determining the neural network compensation amount using a radial basis function neural network based on the total slave force, the first robot output, and the first system state estimate, the method further includes: Based on the acquired disturbance data, the total slave force of the robot is determined, wherein the disturbance data includes the natural viscosity force of blood, contact force, combined friction force, sliding friction force, and elastic deformation force.

[0013] Furthermore, to achieve the above objectives, this application also proposes a robot control device, which includes: The compensation amount determination module is used to determine the neural network compensation amount based on the total slave force, the first robot output, and the first system state estimate through a radial basis function neural network. The estimation value determination module is used to determine the second system state estimate by using a preset third-order linear extended state observer and the neural network compensation amount, wherein the first system state estimate is the corresponding value of the second system state estimate in the previous control cycle; The control quantity determination module is used to determine the target control quantity based on the received control command and the second system state estimate. The robot control module is used to control the robot to perform actions according to the target control quantity to obtain a second robot output, wherein the first robot output is the corresponding value of the second robot output in the previous control cycle.

[0014] In addition, to achieve the above objectives, this application also proposes an electronic device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the robot control method as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the robot control method described above.

[0016] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the robot control method described above.

[0017] One or more technical solutions proposed in this application have at least the following technical effects: First, by using a radial basis function neural network to comprehensively process the total slave force, the previous round robot output (first robot output), and the previous round system state estimate (first system state estimate), the nonlinear mapping capability of the neural network is utilized to capture nonlinear factors and external disturbances in the dynamic characteristics of the system, generating a neural network compensation quantity, which effectively offsets the nonlinear errors and the influence of the total slave force that are difficult to handle in traditional control methods; furthermore, by using a preset third-order linear extended state observer and the neural network compensation quantity, the determination of the second system state estimate is based on the real-time characteristics of the current scene. This invention improves the adaptive compensation capability of robot control by introducing a radial basis function neural network to generate neural network compensation quantities. This allows for dynamic and precise compensation of uncertain disturbances caused by changes in the surgical scene and individual differences, without the need for manual adjustment of the parameters of the extended state observer, thus improving the accuracy of robot control in real-world scenarios. Furthermore, based on the received control commands and the second system state estimate, a target control quantity is determined, enabling real-time dynamic adjustment of the control commands. The robot is then controlled to perform actions based on this target control quantity, resulting in a second robot output. This achieves precise control of robot behavior, and the second robot output is used as the first robot output for the next round of control, feeding back to the neural network and the extended state observer. This forms a closed-loop control system that continuously optimizes itself based on the actual control situation, improving the accuracy of robot control in real-world scenarios. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an embodiment of the robot control method of this application. Figure 2 This is a schematic diagram of the radial basis function neural network provided in Embodiment 1 of this application; Figure 3 This is a comparison diagram of the traditional step command and the command planned by the parabolic acceleration and deceleration curve provided in Embodiment 2 of this application; Figure 4This is a system structure block diagram of the robot control method provided in Embodiment 2 of this application; Figure 5 The response performance analysis diagram of different control systems under pulse signals provided in Embodiment 2 of this application; Figure 6 This is a schematic diagram of the module structure of the robot control device according to an embodiment of this application; Figure 7 This is a schematic diagram of the device structure of the hardware operating environment involved in the robot control method in the embodiments of this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] It should be noted that in the description of this application and the appended claims, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0024] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0025] In the control system of vascular interventional surgery robots, the traditional PID method relies excessively on disturbances and system models for tuning, and cannot adequately address the model uncertainties and unknowns during the operation of cardiovascular interventional surgery robots. While ADRC technology can estimate and compensate for disturbances in the system through ESO, many parameters in the ESO and control law rely on manual adjustment based on expert experience, making it unable to adapt to different surgical scenarios and individual patient differences. Furthermore, its ESO has limited ability to estimate strongly nonlinear and time-varying disturbances (such as dynamic changes in blood viscosity and complex friction) present in vascular interventional surgery, which can easily lead to decreased control accuracy and unstable system performance.

[0026] This application provides a solution. First, a radial basis function neural network is used to comprehensively process the total slave force, the previous round robot output (first robot output), and the previous round system state estimate (first system state estimate). Utilizing the nonlinear mapping capability of the neural network, nonlinear factors and external disturbances in the system's dynamic characteristics are captured, generating a neural network compensation quantity. This effectively offsets the nonlinear errors and the influence of the total slave force, which are difficult to handle in traditional control methods. Then, through a pre-set third-order linear extended state observer and the neural network compensation quantity, the determination of the second system state estimate is based on the real-time characteristics of the current scenario, eliminating the need for manual adjustment of the extended state observer parameters to compensate for individual differences in the current scenario, thereby improving the adaptive compensation capability of robot control. Next, based on the received control command and the second system state estimate, a target control quantity is determined, enabling real-time dynamic adjustment of the control command. Finally, the robot is controlled to execute actions based on this target control quantity, obtaining the second robot output, achieving precise control of robot behavior. This second robot output is then used as the first robot output for the next round of control, fed back to the neural network and the extended state observer, forming a closed-loop control system that continuously optimizes itself based on the actual control situation, thus improving the accuracy of robot control. This application introduces a radial basis function neural network to generate neural network compensation quantities, which can dynamically and accurately compensate for uncertain disturbances caused by changes in surgical scenarios and individual differences, without the need for manual adjustment of relevant parameters of the expansion state observer, thereby improving the accuracy of robot control in real-world scenarios.

[0027] It should be noted that the executing entity in this embodiment can be an electronic device with data processing, network communication and program execution functions, such as a tablet computer, personal computer, mobile phone, etc.

[0028] Based on this, the embodiments of this application provide a robot control method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the robot control method of this application.

[0029] In this embodiment, the robot control method includes steps S10 to S40: Step S10: Determine the neural network compensation amount based on the total slave force, the first robot output, and the first system state estimate using a radial basis function neural network. Radial Basis Function Neural Network (RBFNN) is a type of feedforward neural network model that uses radial basis functions (such as Gaussian functions) as activation functions and has a three-layer structure (input layer, hidden layer, and output layer). The hidden layer consists of Gaussian radial basis function neurons.

[0030] Total end force refers to the force naturally generated during robot operation. It mainly comes from the reaction force generated when the surgical instruments interact with human tissues and organs at the end (closer to the patient). It is an unavoidable internal dynamic force during surgical operation. For example, in the process of controlling the interventional robot to insert the guidewire, the disturbance it is subjected to includes, but is not limited to: the natural viscosity of blood, contact force, combined friction force, sliding friction force, and elastic deformation force.

[0031] Robot output refers to the actual position of the robot after receiving a control command to move; while system state estimate refers to the estimate of the internal state variables of the system through a state observer (such as an extended state observer), which usually includes position, velocity, total system disturbance, etc.

[0032] To distinguish between the robot output and system state estimate in different control cycles, the robot output and system state estimate in the previous control cycle are uniformly referred to as the first robot output and the first system state estimate, respectively, and the robot output and system state estimate in the current control cycle are referred to as the second robot output and the second system state estimate, respectively.

[0033] The neural network compensation value refers to the output value calculated by the RBFNN, which is used to adjust the control input of the robot system to counteract any disturbances that the robot may experience.

[0034] Optionally, data such as total slave force, first robot output, and first system state estimate are collected in real time and preliminarily processed, such as calculating the difference between the position estimate and the first robot output in the first system state estimate as the observation error; then, the input vector containing the total slave force, controller output, first robot output, and observation error is passed to the input layer of the RBFNN, and nonlinearly transformed by the radial basis function of the hidden layer. The hidden layer nodes generate responses based on the distance between the input vector and the center point; then, the output layer generates the neural network compensation quantity by weighted summation of the responses of each node.

[0035] For example, please refer to Figure 2 , Figure 2 A schematic diagram of a radial basis function neural network is provided. The input layer of the RBFNN can include four nodes, each used to receive the total slave force F. t The input vector is obtained from the controller output u(k-1) of the previous control cycle, the first robot output y(k-1) of the previous control cycle, and the difference e between the first position estimate and the first robot output in the first system state estimate. The corresponding hidden layer can be composed of 8 Gaussian radial basis function neurons, and the output layer can obtain the neural network compensation amount by combining the output of the hidden layer with weights in a linear combination.

[0036] For example, the output expression of each neuron node in the hidden layer is as follows:

[0037] Where j represents a hidden layer node, and m is the number of hidden layer nodes; c j It is the center vector of the j-th node in the hidden layer; This represents the L2 norm (i.e., Euclidean distance). Let X represent the input vector and the center vector c. j The Euclidean distance between them; σ j It is the base width parameter (σ) of the j-th node in the hidden layer. j >0).

[0038] For example, when the output layer performs a linear combination of the outputs of each neuron node in the hidden layer, it can combine the weights of each neuron node, as shown in the following expression:

[0039] Among them, u nn w represents the compensation amount in the neural network. j Let be the weight of the j-th node in the hidden layer.

[0040] For example, the neural network compensation amount u obtained from the RBFNN output nn Next, further restrictions can be imposed on it. The formula for restricting the compensation amount of the neural network is as follows:

[0041] Here, 'sat' refers to the saturate function, used to limit the range of neuron output and control the neural network compensation amount in the final output within u. min to u max Between; u min u represents the minimum value of the preset neural network compensation amount. max This represents the maximum preset neural network compensation value; by limiting the output amplitude of RBFNN, overcompensation or system instability caused by excessive network output can be prevented.

[0042] Understandably, the nonlinear mapping capability of RBFNN can accurately capture and compensate for disturbances, reducing steady-state errors. At the same time, it can automatically learn scene changes based on the system control and output quantities of the previous cycle, automatically adapt to different patients and surgical conditions, improve the accuracy of disturbance compensation, and thus help improve the accuracy of subsequent robot control.

[0043] In one feasible implementation, the first system state estimate includes: a first position estimate, and prior to step S10, it further includes: Step S01: Determine the observation error based on the first position estimate and the first robot output; Position estimate refers to the prediction of the robot's position by the state observer in the robot control system, which is generated based on the robot's output and total slave force in the previous control cycle.

[0044] Observation error refers to the deviation between the position estimate of the previous control cycle and the robot's actual position. It is used to measure the gap between the predictive ability and the actual execution ability of the robot's control system.

[0045] Optionally, the absolute value, square, or half of the difference between the first position estimate and the first robot output can be used as the learning target of the radial basis function neural network. This embodiment does not impose specific limitations on this.

[0046] Step S02: Combining the online gradient descent algorithm and momentum accumulation mechanism, update the network parameters of the radial basis function neural network to minimize the observation error. The network parameters include output weights, node center vectors, and basis width parameters.

[0047] Online gradient descent is an iterative optimization algorithm that calculates the gradient of the objective function (observation error in this embodiment) with respect to the network parameters, and then gradually adjusts the parameter values ​​in the opposite direction of the gradient to minimize the objective function during the iteration process.

[0048] The momentum accumulation mechanism is an optimization algorithm that introduces a momentum term during gradient descent. By accumulating historical gradient information, it accelerates parameter updates in dimensions where the gradient direction is consistent and suppresses parameter oscillations in dimensions where the gradient direction changes, thereby achieving a faster and more stable convergence process.

[0049] Network parameters refer to the internal variables that a radial basis function neural network learns from data during training, including output weights, node center vectors, and basis width parameters.

[0050] The output weights refer to the weight coefficients connecting the hidden layer nodes and the output nodes of the RBFNN, which determine the contribution of each hidden layer node to the final output (compensation amount).

[0051] The node center vector refers to the "center point" position of each radial basis function in the input space of the hidden layer of RBFNN. The closer the input data is to the center of a node, the higher the degree of activation of that node. For example, the center of a hidden layer node may correspond to the state of "guidewire traveling at low speed in a moderately curved blood vessel". Updating the center vector is to adjust these states to better represent the actual surgical situation encountered.

[0052] The basis width parameter refers to the parameter used in RBFNN to control the width of the radial basis functions, which determines the response range width of each radial basis function. The larger the basis width, the more responsive the radial basis functions are to inputs that are far from the center; the smaller the basis width, the more sensitive the radial basis functions may be to inputs that are near the center.

[0053] For example, based on the gradient descent algorithm, the network parameters of a radial basis function neural network are updated with the goal of minimizing the observation error. First, based on the observation error, a loss function J is constructed, as follows:

[0054] Where e represents the observation error, z1(k-1) represents the first position estimate, and y(k-1) represents the first robot output; Furthermore, by calculating the gradient of the loss function with respect to the network parameters, the network parameters can be gradually adjusted to minimize J. The gradient update amounts for each network parameter are as follows:

[0055]

[0056]

[0057] Where, Δw j Δc j , Δσ j These represent the gradient update amounts corresponding to the output weights, node center vectors, and basis width parameters, respectively.

[0058] Simultaneously, the momentum accumulation of each network parameter in the current control cycle can be calculated based on the historical momentum accumulation corresponding to each network parameter, as shown in the following formula:

[0059] Where k represents the control period; These are the momentum accumulation amounts corresponding to the output weights, node center vectors, and base width parameters in the previous control cycle, respectively, with an initial value of 0; η represents the momentum factor corresponding to the output weight, node center vector, and basis width parameter, respectively, with a value range of [0.01, 1]. w η is the learning rate for output weights.c The learning rate is the learning rate for the node center vector. The learning rate is the base width parameter, and the value of each learning rate ranges from [0.01, 0.1]. Furthermore, the network parameters can be corrected based on the gradient update amount, as shown in the following expression:

[0060] Where k represents the control period; 、 、 These represent the values ​​of the data weights, node center vectors, and base width parameters in the current control cycle, respectively. Optionally, the learning rate and momentum factor of each network parameter can be preset values ​​or adjusted according to a preset adaptive strategy; this embodiment does not specifically limit this.

[0061] In one feasible implementation, the learning rate and momentum factor of each network parameter can be adjusted according to a preset adaptive strategy to further optimize the learning performance of the radial basis function neural network.

[0062] Among them, the adaptive strategy can dynamically adjust the learning rate of each parameter, so that the parameter update can be personalized according to its historical gradient information. The strategy appropriately suppresses the update step size in the dimension with large gradient and enhances the update magnitude in the dimension with small gradient, thereby improving the algorithm's adaptability to sparse data and non-stationary targets.

[0063] For example, the adaptive adjustment formula for the learning rate of each network parameter is as follows:

[0064] in, This represents the learning rate for each network parameter in the current control cycle. These are the base learning rates corresponding to each network parameter. Indicates the adjustment range of the learning rate ( ), represents the sensitivity parameter, all of which are preset values; tanh is the hyperbolic tangent function, a smooth S-shaped activation function with an output range between (-1, 1), used to smoothly adjust the increase or decrease of the learning rate according to the magnitude of the observation error; e represents the observation error, which is the difference between the position estimate of the ESO output in the previous control cycle and the actual output position of the robot, defined as... This error reflects the estimation accuracy of the state observer and is an important basis for adjusting the learning rate.

[0065] For example, the adaptive adjustment formula for the momentum factor of each network parameter is as follows:

[0066] in, This represents the momentum factor corresponding to each network parameter in the current control cycle. , This indicates the preset momentum factor range. The exp function represents the attenuation coefficient; it is used to achieve the effect of the momentum factor decaying exponentially as the observation error increases; e represents the observation error.

[0067] When the observation error |e| is large The learning rate approaches 1. Increasing the value accelerates the parameter convergence speed; simultaneously Approaching 0, momentum factor Reduce to To reduce the influence of historical gradients and avoid acceleration in the wrong direction, the learning rate of each network parameter is reduced when the observation error |e| decreases to prevent excessive parameter updates from causing oscillations. Simultaneously, the momentum factor of each network parameter is increased to enhance the accumulation effect of historical gradients and improve the stability of parameter updates. Thus, by introducing adaptive learning rate and momentum factor adjustment strategies, the radial basis function neural network can dynamically optimize the learning process according to the system state: emphasizing rapid convergence when the observation error is large and stable optimization when the observation error is small, thereby significantly improving the stability of parameter updates while ensuring learning efficiency. This adaptive mechanism further enhances the neural network's adaptability to dynamic changes in the surgical scenario and individual patient differences, ensuring the consistency and robustness of control performance under various unknown and time-varying environments.

[0068] Understandably, by introducing an adaptive strategy, the learning rate of each parameter is dynamically adjusted based on its historical gradient information, enabling the network parameters to have a more refined adjustment capability when facing different features and data distributions. Furthermore, by introducing a momentum factor, learning can be accelerated when the gradient direction is consistent and oscillations can be suppressed when the gradient direction changes. The synergistic effect of the two effectively improves the accuracy and response speed of the neural network to system disturbances, while ensuring the stability of the parameter update process and avoiding divergence or slow convergence caused by improper learning rate settings.

[0069] For example, before applying this radial basis function neural network, it is necessary to initialize each network parameter, the corresponding momentum accumulation, and the adaptive factor. For instance, historical input data samples can be clustered using the K-means clustering algorithm, and the cluster centers can be determined as the node center vectors. The initial value; based on the clustering results, calculate the average distance between the sample and the center point in each cluster, and then apply the formula... Determine the initial value of the base width parameter, where The maximum average distance among all clusters is denoted by , and M is the number of neurons in the hidden layer. Output weights are initialized with small random numbers, for example, randomly selected within the range [-0.1, 0.1], to avoid initial saturation. To implement the adaptive strategy, adjustable parameters (such as output weights w) need to be assigned to each class. j Node center c j and base width σ j Initialize the corresponding gradient squared cumulant, usually set as a zero vector or zero matrix. Simultaneously, initialize the momentum cumulant for each network parameter. , and It is also initialized to zero to ensure that the parameter update process starts smoothly from the initial state. In the subsequent online learning process, the adaptive strategy will operate in practice: at each iteration, after calculating the current gradient, the algorithm will first update the cumulative squared gradients (e.g., using an exponential moving average), and then use the cumulative gradients to calculate the adaptive learning rate for each parameter. For parameters with larger historical gradients, the adaptive learning rate will be reduced accordingly, thus achieving a robust update step size; for parameters with smaller historical gradients, a relatively larger learning rate will be obtained to accelerate their convergence. This dynamic adjustment mechanism, combined with momentum accumulation, allows parameter updates to utilize both the directional information (momentum) of historical gradients and adapt to historical changes in their amplitude (adaptive), thereby achieving faster and more stable convergence in complex, non-stationary surgical environments.

[0070] Understandably, by updating the RBFNN network parameters online and combining momentum accumulation with an adaptive strategy to generate accurate compensation values, not only is the reliance on precise initial parameter settings and the tedious steps of manual adjustment avoided, but the introduction of momentum accumulation and adaptive learning rate effectively smooths the parameter update process, significantly reducing parameter oscillations while maintaining rapid learning. Furthermore, since the learning process is conducted online, the RBFNN, combined with the momentum accumulation mechanism and adaptive strategy, can more stably track and adapt to the dynamic changes in the surgical scenario and individual differences among patients in real time. The adaptive strategy enhances the model's adaptability to non-stationary data by dynamically adjusting the learning step size of each parameter, while momentum accumulation further improves the coherence and stability of parameter updates by utilizing the directional characteristics of historical gradient information. This ensures the consistency and robustness of control performance under various unknown and time-varying environments, ultimately significantly improving the accuracy of robot control and the overall adaptive capability of the system.

[0071] Step S20: Determine the estimated state value of the second system by using a preset third-order linear extended state observer and neural network compensation quantity; The first system state estimate is the corresponding value of the second system state estimate in the previous control cycle. Therefore, the second system state estimate represents the newly generated system state estimate in the current control cycle, used to determine the robot's final control quantity (target control quantity).

[0072] A third-order linear extended state observer is an observer used to estimate system states and disturbances. The observer involves three state variables, including the robot's own partial states (such as position and velocity) and extended disturbance states. It treats the uncertainties of the robot control system, unmodeled dynamics, and external disturbances as a "total disturbance" and observes this total disturbance as a new state (i.e., the extended disturbance state) in real time.

[0073] For example, the pre-defined equations for a third-order linear ESO used to observe the robot's state position, velocity, and total disturbance in real time are as follows:

[0074] Where k represents the control period; This represents the position estimate in the second system state estimate of the current control output. This represents the velocity estimate in the second system state estimate. This represents the total disturbance estimate in the second system state estimation; This represents the derivative of the estimated state of the second system. y(k-1) represents the position, velocity, and total disturbance estimates in the first system state estimate of the previous control cycle; y(k-1) represents the robot output of the previous control cycle (i.e., the first robot output), and therefore e represents the estimation error (i.e., the observation error) of the robot position state. All of these are gain parameters of the third-order linear ESO, and all are fixed values; is the neural network compensation quantity output by RBFNN, used to compensate for the observation error of the observer; b is the control gain, and u(k-1) is the control quantity received by the robot in the previous control cycle; It indicates the total force from the end.

[0075] Understandably, by introducing neural network compensation, the originally "preset fixed" linear observer is made capable of coping with nonlinear and time-varying disturbances. It can adapt to different surgical scenarios and patient differences without manual online parameter adjustment, and output a second system state estimate with improved accuracy, thereby improving the precision of robot control.

[0076] In one possible implementation, prior to step S20, the method further includes: Step S201: Construct a state-space expression based on a preset second-order robot control model; A second-order robot control model is a mathematical model that describes the dynamic behavior of a robot. It is usually a second-order differential equation and includes state variables such as position, velocity, and acceleration.

[0077] State-space representation is a mathematical representation used to describe the dynamic behavior of a robot. It consists of first-order differential equations, including state equations and output equations, and is used to relate the changes of state variables to the robot's inputs and outputs.

[0078] For example, the formula for a second-order robot control model is as follows:

[0079] Where y is the robot output (corresponding to the position). Let y be the first derivative (which represents velocity). y is the second derivative (corresponding to acceleration); w(t) is the external disturbance, and f(·) is the total disturbance including the unmodeled dynamics of the system and the external disturbance; Let be the total slave force, b be the control gain, and u be the actual control quantity received by the robot.

[0080] Therefore, y can be used as the state variable. ,by State variables Construct the state-space expression for the robot control system, as follows:

[0081] Essentially, it is an assignment behavior used to transform the second-order robot equation into a state-space form, so that position, velocity, etc. can be used as state variables to intuitively observe the dynamic behavior of the robot. In this state-space form, the total disturbance received by the robot can be conveniently regarded as an extended state variable and included in the observation range of the observer to construct a third-order linear ESO, so as to realize the simultaneous estimation of system state and disturbance.

[0082] Step S202: Expand the total perturbation in the state-space expression into new state variables to construct a third-order linear extended state observer.

[0083] For example, the external disturbances and disturbance forces in the robot control process are used as new unknown state variables, that is, the total disturbance f(·) containing the unmodeled dynamics of the system and external disturbances is expanded into a new state variable (i.e., the expanded disturbance state). The state-space expression is transformed into a third-order equation:

[0084] Furthermore, adopt Representing states respectively The estimated value is adopted. This represents the estimation error in the previous control cycle, and the robot's position is re-estimated based on this error. That is, error correction is added during the position estimation process; similarly, the robot's velocity and total disturbance can be estimated based on this error, resulting in the above-mentioned third-order linear ESO equation, where the total disturbance f(·) is determined by the radial basis function neural network, corresponding to the neural network compensation amount. .

[0085] In this embodiment, a third-order linear extended state observer is constructed to achieve real-time estimation of the system state and total disturbance. Based on this information, the robot's control strategy is adjusted in a timely manner to effectively resist external interference, achieve active disturbance rejection control, and thus improve the accuracy of robot control.

[0086] Step S30: Determine the target control quantity based on the received control command and the second system state estimate; Control commands are externally input signals that instruct the robot's slave actuators on how to move. They typically include information such as target position and target velocity. For example, a control command can be represented as a pulse sequence containing information about intermediate points. This implementation does not impose specific limitations on the specific form of the control command. Generally, the position, velocity, and acceleration information in the control command can be converted to each other; obtaining one allows you to solve for the other two.

[0087] The target control quantity refers to the control signal that needs to be output in order to make the robot reach the desired state.

[0088] For example, the received control command (such as the target position r) can be compared with the position estimate in the second system state estimate to calculate the tracking error. Furthermore, based on this tracking error, a preliminary control quantity is calculated using a PD (Proportional-Differential) controller. Its form is usually as ,in and It is the preset control gain. It is the derivative of the tracking error (which can also be obtained from the target velocity and the velocity estimate). (Calculation); then, based on the principle of self-disturbance reaction, the disturbance is canceled by feedforward, and the initial control quantity is combined with the total disturbance estimate in the second system estimate to determine the target control quantity.

[0089] In one feasible implementation, the control command includes an acceleration command, the second system state estimate includes a total disturbance estimate, and step S30 includes: Step S31: Determine the feedforward compensation amount based on the total slave force, and determine the feedforward control amount based on the acceleration command; Feedforward compensation refers to the control quantity component calculated based on the total slave force. Its purpose is to offset the dynamic influence of the total slave force on robot control in a feedforward manner outside the force feedback closed loop.

[0090] The acceleration command in the control command refers to the acceleration value that the robot's execution end is expected to achieve; while the feedforward control quantity refers to the control quantity component calculated based on the acceleration command, which is used to directly provide the power required to generate the target acceleration.

[0091] For example, the calculation formulas for the feedforward compensation and feedforward control are as follows:

[0092]

[0093] in, This is the feedforward compensation amount. For total end force; This is the feedforward control quantity. 'b' represents the acceleration command; 'b' represents the control gain.

[0094] Optionally, if the control command does not include an acceleration command, the position command and / or velocity command in the control command can be used to determine the feedforward control quantity; for example, the position command or velocity command can be integrated to convert it into an acceleration command, and then the acceleration command can determine the feedforward control quantity.

[0095] Step S32: Determine the preliminary control quantity based on the feedforward control quantity, the feedforward compensation quantity, and the second system state estimate. The preliminary control quantity refers to the intermediate control quantity synthesized from the feedforward quantity and the state feedback quantity before incorporating the final disturbance compensation. The feedforward quantity includes the feedforward compensation quantity and the feedforward control quantity mentioned above. The state feedback quantity can be calculated by the PD controller. The calculation process is the same as the calculation method of the preliminary control quantity in the specific implementation of step S30 in this embodiment, so it will not be described again.

[0096] It is understandable that by providing feedforward compensation and feedforward control, i.e. providing power and compensating for resistance, we can anticipate known disturbances and plan motions in advance, reduce error accumulation, and improve the accuracy of robot control.

[0097] In one feasible implementation, the control command includes a speed command and a position command, the second system state estimate includes a second position estimate and a speed estimate, and step S32 includes: Step S321: Determine the position error based on the position command and the second position estimate, and determine the speed error based on the speed command and the speed estimate; Position commands refer to the desired spatial coordinates in control commands, while velocity commands refer to the desired speed of motion in control commands.

[0098] The second position estimate and velocity estimate in the second system state estimate are high-precision state estimates that have been compensated and calculated by a third-order linear ESO.

[0099] For example, the formulas for calculating position error and velocity error are as follows:

[0100]

[0101] in, These represent position error and velocity error, respectively. These represent position commands and speed commands, respectively. ; represent the position estimate and velocity estimate, respectively. Furthermore, it is also possible to... Differentiate to obtain ,Right now Manifestation The rate of change.

[0102] Step S322: Using a fuzzy controller, the position error and velocity error are used to determine the adjustment amount of the controller gain, wherein the controller gain includes derivative gain and proportional gain. A fuzzy controller is a controller that uses fuzzy logic to handle uncertainty and complexity. It can simulate human thinking and experience, process input error signals using fuzzy rules, and output corresponding control quantities, thereby achieving self-adaptation to different scenarios.

[0103] Controller gain refers to the proportional relationship between the controller output and the input error, which determines the controller's response to the error. Proportional gain reflects the controller's response to the magnitude of the error, while derivative gain reflects the controller's response to the rate of change of the error.

[0104] For example, the position error and velocity error are converted into fuzzy speech values ​​such as "positive large (PB)", "zero (ZO)" and "negative large (NB)" through the membership function in the fuzzy controller. Then, the fuzzy speech values ​​corresponding to the position error and velocity error are input into a preset fuzzy rule base to determine the adjustment amount of the differential gain and the proportional gain. The fuzzy rules define the mapping relationship between the fuzzy speech values ​​and the gain adjustment amount. For example, when the position error is PB and the velocity error is NB, the proportional gain is kept unchanged (adjustment amount is 0), and the differential gain is increased by 0.5 (adjustment amount is +0.5).

[0105] Step S323: Determine the preliminary control quantity based on the feedforward control quantity, feedforward compensation quantity, position error, speed error, and controller gain adjustment quantity.

[0106] Optionally, the updated proportional gain and the updated derivative gain are determined based on the initial values ​​of the proportional gain and the derivative gain, as well as the adjustment amount of the controller gain. Then, the product of the position error and the updated proportional gain, and the product of the velocity error and the updated derivative gain are added to obtain the state feedback quantity. Then, the state feedback quantity is added to the feedforward control quantity and the feedforward compensation quantity to obtain the preliminary control quantity.

[0107] For example, the formula for calculating the initial control quantity is as follows:

[0108] in, This is the feedforward control quantity. This is the feedforward compensation amount; This represents the updated proportional gain. This represents the initial value of the proportional gain. Indicates the adjustment amount of the proportional gain; This represents the updated differential gain. This represents the initial value of the differential gain. This indicates the adjustment amount for the differential gain; These represent position error and velocity error, respectively.

[0109] Step S33: Generate the target control quantity based on the preliminary control quantity and the total disturbance estimate.

[0110] For example, the formula for calculating the target control quantity is as follows:

[0111] Where u(k) represents the target control quantity for the current control period k. Indicates the initial control quantity. denoted by , represents the total disturbance estimate for the current control period k, and b is the control gain, which characterizes the amplification factor of the control quantity u on the system's dynamic response and directly affects the calculation accuracy of the target control quantity and the system stability.

[0112] It should be noted that the control gain b, as a key parameter connecting the control command and the actual motion, may drift slowly due to various factors such as mechanical wear, component aging, temperature changes, and external disturbances. Optionally, the control gain b can be calibrated using methods such as system identification calibration, frequency domain analysis calibration, and online adaptive calibration.

[0113] In one feasible implementation, the control gain is adjusted based on gradient descent and observation error.

[0114] For example, the control gain b is updated using an adaptive law based on gradient descent, and the formula for online adaptive update is as follows:

[0115] Where k represents the control period. denoted by , where represents the learning rate corresponding to the control gain; e represents the observation error (i.e., the difference between the position estimate of the ESO output in the previous control cycle and the actual output position of the robot); u(k-1) is the control quantity of the previous control cycle.

[0116] The initial value of the control gain b can be determined using a high-precision offline system identification method, specifically including: applying a pseudo-random binary sequence (PRBS) or step signal with gradually increasing amplitude to the robot control system as the control input u, and synchronously acquiring the system's acceleration response. Furthermore, based on the robot control model The parameter estimation is performed using the recursive least squares (RLS) method to obtain a high-precision initial control gain.

[0117] Optionally, after obtaining the initial control gain described above, it can be achieved through... (R-Squared, coefficient of determination) The input data during the identification period is fitted and calibrated, and the initial control gain is output when preset requirements are met; for example, preset requirements can be set: A value of ≥0.95 ensures that the identified model can explain more than 95% of the system response dynamics, thereby guaranteeing the accuracy of the initial control gain.

[0118] Optionally, the control gain can be dynamically calibrated using the aforementioned high-precision offline system identification method based on preset triggering conditions to generate a new initial control gain and reset the gradient descent-based online adaptive update process. For example, triggering conditions may include: the root mean square of the observation error continuously exceeding a preset threshold for multiple consecutive control cycles; the deviation between the control gain obtained from the online adaptive update and the initial control gain exceeding a preset deviation value, etc. The aforementioned high-precision offline system identification process can also be triggered periodically (e.g., every 6 months). This embodiment does not specifically limit the setting of triggering conditions.

[0119] In this embodiment, by adaptively adjusting the controller gain, the feedback intensity can be intelligently adjusted according to the different situations of error and its change, thereby maintaining high tracking accuracy and stability when facing internal and external disturbances, and thus improving the precision of robot control.

[0120] Step S40: Control the robot to perform actions according to the target control quantity to obtain the second robot output.

[0121] The first robot output is the corresponding value of the second robot output in the previous control cycle. Therefore, the second robot output represents the robot output in the current control cycle, that is, the actual physical state directly measured by sensors after the robot performs its action in the current control cycle, typically including information such as position and velocity.

[0122] A robot is a machine device that has the ability to perceive the environment, make decisions and plan, and perform actions.

[0123] For example, after receiving the output of the second robot, it is fed back to the third-order linear ESO and RBFNN to form a complete closed-loop control, enabling the robot control system to continuously observe the state, compensate for disturbances, adjust the control quantity, and ultimately accurately track the control commands from the upper layer.

[0124] This embodiment provides a robot control method. By introducing a radial basis function neural network and an online gradient descent algorithm, it learns and compensates for uncertain disturbances caused by changes in the surgical scenario and individual patient differences. This allows the third-order linear extended state observer to output high-precision state estimates without changing its fixed parameters. Simultaneously, a fuzzy controller dynamically adjusts the controller gain based on real-time errors, ensuring that the parameters of the feedback control law are always at their optimal values ​​under the current operating conditions, achieving fast, overshoot-free tracking. Furthermore, an initial control quantity is determined based on the feedforward control quantity, the feedforward compensation quantity, and the adaptive feedback control quantity (calculated based on position error, velocity error, and controller gain). The total disturbance estimated by the third-order linear extended state observer in the initial control quantity is then removed, resulting in the final target control quantity used to control the robot. In this embodiment, the process of determining the target control quantity, from disturbance compensation to state observation and control decision-making, is adaptive, eliminating the need for cumbersome manual parameter tuning. This achieves consistent high performance of the robot control system across different scenarios, improving the accuracy of robot control.

[0125] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description and will not be repeated hereafter. In addition, before step S30, the following is also included: Step S301: Based on the preset parabolic acceleration / deceleration algorithm, the received original command is smoothed to generate control command.

[0126] Raw commands are unprocessed signals generated directly by input devices, which may contain instantaneous large changes or steps, reflecting the user's immediate operational intentions.

[0127] Parabolic acceleration / deceleration algorithm refers to an acceleration / deceleration control algorithm designed based on the characteristics of parabolic functions. It uses the rising and falling phases of the parabola to simulate the acceleration and deceleration process of robot control. By reasonably setting the parameters of the parabola, the rate of change of acceleration and deceleration can be precisely controlled, so that the system can smoothly transition during the acceleration and deceleration phases, avoid sudden changes in speed and position, and thus reduce impact and vibration.

[0128] For example, during the smoothing process of the original command based on the parabolic acceleration / deceleration algorithm, the relationship between the acceleration command, velocity command, position command, and pulse number is as follows:

[0129]

[0130]

[0131] Where p represents the current cumulative number of pulses, These represent the preset pulse thresholds; These represent the acceleration command, velocity command, and position command in the smoothed control command, respectively. This indicates the preset maximum speed. This indicates the preset minimum speed. These represent the pulse equivalent, and are all constants.

[0132] It should be noted that the above They are interconnected, and their respective formulas can be derived from each other.

[0133] For example, please refer to Figure 3 , Figure 3 A comparison chart is provided between a traditional step command and a command planned using a parabolic acceleration / deceleration curve. Figure 3 The comparison of position, velocity, and acceleration commands reveals that traditional skip-order commands exhibit abrupt changes in position, velocity, and acceleration, while the parabolic curve planning command shows a smooth movement in position, and the parabolic curve velocity and acceleration remain relatively stable. Furthermore, magnification of the corresponding parabolic position, velocity, and acceleration curves shows that the parabolic position shift does not exceed 5mm for every 5 pulse changes, the parabolic velocity change does not exceed 0.1mm / pulse, and the parabolic acceleration change shows a gradual downward trend. Therefore, the control commands smoothed by the aforementioned parabolic acceleration / deceleration algorithm effectively reduce shocks and vibrations.

[0134] In this embodiment, a parabolic acceleration / deceleration algorithm is used to transform abrupt inputs into smooth inputs, thereby eliminating motor start-stop shocks, overshoots, and vibrations caused by command steps or abrupt changes at the source, and providing a smooth and physically achievable ideal tracking trajectory for subsequent control circuits.

[0135] In one possible implementation, prior to step S10, the method further includes: A01. Based on the acquired disturbance data, determine the total slave force of the robot. The disturbance data includes the natural viscosity of blood, contact force, combined friction force, sliding friction force, and elastic deformation force.

[0136] Perturbation data refers to a set of physical quantities that describe the complex mechanical interaction between the robot's end effector and the vascular environment. These data are usually obtained from direct sensor measurements and include, but are not limited to, the natural viscosity of blood, contact forces, combined friction forces, sliding friction forces, and elastic deformation forces.

[0137] For example, the dynamic model of the total end force (disturbance force) is as follows:

[0138] in, For the total force from the end, It is the natural viscosity of blood. This refers to the contact force between the catheter / guidewire and the blood vessel wall. For sliding friction, F f To combine frictional forces, This refers to the elastic deformation force when the guidewire or catheter contacts the blood vessel. This represents the number of points where the guidewire / catheter makes significant frictional contact with the vessel wall as it travels through the blood vessel; The number of vascular segments or contact points where the guidewire / catheter has stored significant elastic potential energy due to bending and compression (i.e., bending deformation) within the blood vessel.

[0139] In this embodiment, by accurately calculating and feeding back the total slave force, the force situation of disturbance in robot control can be intuitively perceived, thereby achieving an accurate estimation of the disturbance situation and improving the accuracy of robot control.

[0140] For example, to help understand the implementation flow of the robot control method obtained by combining this embodiment with the above embodiment one, please refer to... Figure 4 , Figure 4 A system architecture block diagram of a robot control method is provided. First, control command generation is performed. After receiving the original command, the control system transmits it to the parabolic acceleration / deceleration trajectory planning module for smoothing and generating control commands, specifically including position, velocity, and acceleration commands. Simultaneously, known disturbance processing is performed. Disturbance data is acquired through known disturbance sensing, and a known disturbance (total slave force) is generated accordingly. Then, through feedforward compensation, the known disturbance is converted into a corresponding feedforward compensation quantity. Simultaneously, the radial basis function neural network compensation module can perform signal compensation for unmodeled disturbances based on the received known disturbance, the first system state estimate from the previous control cycle, and the robot output from the previous control cycle, generating a neural network compensation quantity. Finally, a third-order linear expansion state... The observer can generate a second system state estimate for the current control cycle based on the neural network compensation value and pass it to the fuzzy adaptive feedforward feedback control law module. The fuzzy adaptive feedforward feedback control law module then generates a preliminary control quantity based on the smoothed control command and the received second system state estimate. The module then synthesizes the control quantity using this preliminary control quantity and the total disturbance estimate in the second system state estimate for the current control cycle to generate a target control quantity, which is used to control the robot to perform actions, resulting in the robot output for the current control cycle. This robot output can then be fed back to the third-order linear extended state observer and the radial basis function neural network compensation module for disturbance compensation and state estimation in the next control cycle, achieving scene and individual difference adaptation and improving the accuracy of robot control.

[0141] To further illustrate the superiority of the robot control method of this application, this embodiment provides a method for comparing and testing the pulse signal response performance of three strategies—PID control, ADRC control, and the intelligent ADRC control based on radial basis function neural network of this application—in a unified simulation environment: First, a high-density time series from 0 seconds to 10 seconds is constructed to simulate a continuous test time axis, and a standard square wave pulse signal with two periodic variations is generated as a reference signal (original command). This command undergoes high and low level transitions at specific times (such as 1 second, 3.5 seconds, 6 seconds, and 8.5 seconds) to assess the control system's ability to track and recover from sudden commands.

[0142] Furthermore, after receiving the original command, the PID control system, the ADRC control system, and the RBFNN-ADRC control system corresponding to this application generate the corresponding pulse tracking response waveform according to the original command. Please refer to... Figure 5 , Figure 5 A response performance analysis diagram of different control systems under pulse signals is provided. PID control and ADRC control both undergo abrupt changes at the specific transition moments mentioned above, while this application performs a smooth transition, which is closer to the change trend of the reference signal. It can be seen that the RBFNN-ADRC control system of this application can more accurately track the input original command and achieve more precise robot control.

[0143] In addition, to quantify the performance differences among the three control systems, as shown in Table 1, which presents the performance index results of the three control systems after command tracking and robot control under the same reference signal, Table 1 provides the corresponding performance index results of the three control systems.

[0144] Table 1. Example of performance metrics display

[0145] Where a% represents the maximum position deviation percentage, and Ta represents the settling time. Simulation results show that both traditional PID controllers and ADRC exhibit significant overshoot problems; however, the active disturbance rejection controller incorporating RBFNN and fuzzy control algorithms shows no significant overshoot during operation and achieves the shortest settling time to reach steady state. Therefore, the robot control method proposed in this application can improve the accuracy of robot control, reduce response time, and increase control efficiency.

[0146] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the robot control method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0147] This application also provides a robot control device, please refer to... Figure 6 The robot control device includes: The compensation amount determination module 10 is used to determine the neural network compensation amount based on the total slave force, the first robot output, and the first system state estimate through a radial basis function neural network. The estimation value determination module 20 is used to determine the second system state estimate by using a preset third-order linear extended state observer and neural network compensation quantity, wherein the first system state estimate is the corresponding value of the second system state estimate in the previous control cycle; The control quantity determination module 30 is used to determine the target control quantity based on the received control command and the second system state estimate. The robot control module 40 is used to control the robot to perform actions according to the target control quantity and obtain the second robot output, wherein the first robot output is the corresponding value of the second robot output in the previous control cycle.

[0148] The robot control device provided in this application, employing the robot control method described in the above embodiments, can solve the technical problem of how to improve the accuracy of robot control. Compared with the prior art, the beneficial effects of the robot control device provided in this application are the same as those of the robot control method provided in the above embodiments, and other technical features in the robot control device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0149] This application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the robot control method in the first embodiment described above.

[0150] The following is for reference. Figure 7 The diagram illustrates a structural schematic of an electronic device suitable for implementing embodiments of this application. The electronic devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 7 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0151] like Figure 7 As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. The communication device 1009 allows the electronic device to communicate wirelessly or wiredly with other devices to exchange data. Although the diagrams show electronic devices with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented alternatively.

[0152] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0153] The electronic device provided in this application, employing the robot control method described in the above embodiments, can solve the technical problem of how to improve the accuracy of robot control. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the robot control method provided in the above embodiments, and other technical features of the electronic device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0154] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0155] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0156] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the robot control method in the above embodiments.

[0157] The computer-readable storage medium provided in this application embodiment may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0158] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.

[0159] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by an electronic device, the electronic device causes the following: a radial basis function neural network is used to determine a neural network compensation amount based on the total slave force, the first robot output, and the first system state estimate; a preset third-order linear extended state observer and the neural network compensation amount are used to determine a second system state estimate, wherein the first system state estimate is the corresponding value of the second system state estimate in the previous control cycle; a target control amount is determined based on the received control command and the second system state estimate; and the robot is controlled to perform an action based on the target control amount to obtain a second robot output, wherein the first robot output is the corresponding value of the second robot output in the previous control cycle.

[0160] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0161] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0162] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0163] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-described robot control method, and can solve the technical problem of how to improve the accuracy of robot control. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the robot control method provided in the above embodiments, and will not be repeated here.

[0164] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the robot control method described above.

[0165] The computer program product provided in this application can solve the technical problem of how to improve the accuracy of robot control. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the robot control method provided in the above embodiments, and will not be repeated here.

[0166] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A robot control method, characterized in that, The robot control method includes: The neural network compensation amount is determined by using a radial basis function neural network based on the total slave force, the first robot output, and the first system state estimate. The second system state estimate is determined by using a preset third-order linear extended state observer and the neural network compensation value, wherein the first system state estimate is the corresponding value of the second system state estimate in the previous control cycle. The target control quantity is determined based on the received control command and the second system state estimate. The robot is controlled to perform actions according to the target control quantity to obtain the second robot output, wherein the first robot output is the corresponding value of the second robot output in the previous control cycle.

2. The robot control method as described in claim 1, characterized in that, The first system state estimate includes: a first position estimate. Prior to the step of determining the neural network compensation amount using a radial basis function neural network based on the total slave force, the first robot output, and the first system state estimate, the system state estimate further includes: The observation error is determined based on the first position estimate and the first robot output; By combining the online gradient descent algorithm and the momentum accumulation mechanism, the network parameters of the radial basis function neural network are updated to minimize the observation error. The network parameters include output weights, node center vectors, and basis width parameters.

3. The robot control method as described in claim 1, characterized in that, Before the step of determining the second system state estimate using a preset third-order linear extended state observer and the neural network compensation quantity, the method further includes: Based on the pre-defined second-order robot control model, construct the state-space expression; The total perturbation in the state-space expression is expanded into new state variables to construct the third-order linearly extended state observer.

4. The robot control method as described in claim 1, characterized in that, The control command includes an acceleration command, the second system state estimate includes a total disturbance estimate, and the step of determining the target control quantity based on the received control command and the second system state estimate includes: Based on the total slave force, determine the feedforward compensation amount, and based on the acceleration command, determine the feedforward control amount; The initial control quantity is determined based on the feedforward control quantity, the feedforward compensation quantity, and the second system state estimate. The target control quantity is generated based on the preliminary control quantity and the total disturbance estimate.

5. The robot control method as described in claim 4, characterized in that, The control commands include speed commands and position commands, the second system state estimate includes a second position estimate and a speed estimate, and the step of determining the preliminary control quantity based on the feedforward control quantity, the feedforward compensation quantity, and the second system state estimate includes: Based on the position command and the second position estimate, determine the position error, and based on the speed command and the speed estimate, determine the speed error. The position error and the velocity error are used to determine the adjustment amount of the controller gain through a fuzzy controller, wherein the controller gain includes differential gain and proportional gain; The preliminary control quantity is determined based on the feedforward control quantity, the feedforward compensation quantity, the position error, the speed error, and the adjustment amount of the controller gain.

6. The robot control method as described in claim 1, characterized in that, Before the step of determining the preliminary control quantity based on the control command and the preset system state estimate, the method further includes: The received original command is smoothed according to the preset parabolic acceleration and deceleration algorithm to generate the control command.

7. The robot control method as described in claim 1, characterized in that, Before the step of determining the neural network compensation amount using a radial basis function neural network based on the total slave force, the first robot output, and the first system state estimate, the method further includes: Based on the acquired disturbance data, the total slave force of the robot is determined, wherein the disturbance data includes the natural viscosity force of blood, contact force, combined friction force, sliding friction force, and elastic deformation force.

8. An electronic device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the robot control method as described in any one of claims 1 to 7.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the robot control method as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the robot control method as described in any one of claims 1 to 7.

Images