A method and system for joint torque compensation of remotely operated arms

Abstract

A method and system for joint torque compensation in a remote teleoperated arm is disclosed. For each joint of the teleoperated arm, a friction torque prediction model, nominal dynamic equation, and compensation residual prediction model are constructed using the arm's historical joint command torque, joint angular velocity, and joint temperature. During each real-time control cycle of the teleoperated arm, real-time compensation calculations are performed using the constructed models to obtain a reference feedforward compensation torque. The end effector force deviation of the teleoperated arm is calculated and mapped to a joint torque increment. A feedback fine-tuning compensation torque is generated based on the joint torque increment. The total compensation torque is calculated based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque. The final torque command for each joint servo driver is calculated based on the total compensation torque. This invention systematically solves the problems of model accuracy, dynamic response, and disturbance suppression, thereby achieving a truly high-fidelity force perception and presence.

Description

Technical Field

[0001] This invention belongs to the field of robot teleoperation and servo control technology, and particularly relates to a joint torque compensation method and system for remote teleoperated arms. Background Technology

[0002] In high-risk applications such as nuclear power plant operation and maintenance, deep-sea operations, and remote surgery, teleoperated arm systems serve as a crucial interface for interaction between operators and the remote environment. Operators use this system to control the robotic arm to complete precise tasks, relying on force feedback to perceive contact forces in the remote environment, thereby determining the operational status and achieving fine-grained control. The accuracy, real-time performance, and fidelity of the force feedback directly determine the success, efficiency, and safety of teleoperation tasks.

[0003] A teleoperated arm is a complex nonlinear system. The significant friction, inertia, and linkage coupling in its joint drive train consume some of the drive command torque, causing a deviation between the actual force output by the joints and the end effector and the expected command. To eliminate this interference and improve the accuracy of force feedback, precise torque compensation must be performed on the internal disturbances such as friction and inertia at the joints.

[0004] Existing technologies, such as the patent document with publication number CN112677156B, establish a Coulomb-viscous friction model with joint velocity as input through experimental calibration, and substitute the calculated friction torque into the dynamic equation for feedforward compensation. However, the friction model of this method is overly simplified, only considering the velocity factor and neglecting the nonlinear and time-varying influence of joint output load on friction. This leads to a sharp increase in model prediction error and a decrease in compensation effect when the load changes drastically (such as joints with balancing gravity torque). Furthermore, this method is an offline, open-loop static compensation, which cannot compensate for complex dynamic effects that are not accurately modeled (such as transmission elasticity and linkage flexibility).

[0005] Existing technologies, such as the patent document with publication number CN113059567A, measure end-effector pose error in real time using external high-precision sensors (such as optical positioning systems) and compensate online through control algorithms. This method directly uses end-effector force sensor feedback for real-time correction, which is essentially a delayed, lagging feedback compensation. For sudden contact forces or precise force control operations requiring millisecond-level response, its dynamic response speed is insufficient to meet the requirements of instantaneous force perception reproduction.

[0006] Therefore, there is an urgent need for a teleoperated arm joint torque compensation method that can deeply integrate high-precision physical models, intelligent learning networks, and real-time force feedback to achieve collaborative compensation of "model first, intelligent fine-tuning, and online closed loop" in order to systematically solve the problems of model accuracy, dynamic response, and disturbance suppression, thereby achieving a truly high-fidelity force perception and presence. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for joint torque compensation in remotely operated arms. For each joint of the remotely operated arm, a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model are constructed using the arm's historical joint command torque, joint angular velocity, and joint temperature. During each real-time control cycle of the remotely operated arm, real-time compensation calculations are performed using the constructed models to obtain a reference feedforward compensation torque. The end effector force deviation of the remotely operated arm is calculated and mapped to a joint torque increment. A feedback fine-tuning compensation torque is generated based on the joint torque increment. The total compensation torque is calculated based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque. The final torque command for each joint servo driver is calculated based on the total compensation torque. This invention systematically solves the problems of model accuracy, dynamic response, and disturbance suppression, thereby achieving a truly high-fidelity force perception and presence.

[0008] This invention proposes a method for joint torque compensation of a remotely operated arm, comprising: S1. For each joint of the teleoperated arm, a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model are constructed using the historical joint command torque, joint angular velocity, and joint temperature of the teleoperated arm. S2, in each real-time control cycle of the teleoperated arm, real-time compensation calculation is performed using the constructed friction torque prediction model, the nominal dynamic equation and the trained residual compensation torque prediction model to obtain the reference feedforward compensation torque; S3, calculate the end-effector force deviation of the teleoperated arm, map the end-effector force deviation to the joint torque increment; generate feedback fine-tuning compensation torque based on the joint torque increment; S4 calculates the total compensation torque based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque; and calculates the final torque command of each joint servo drive based on the total compensation torque.

[0009] More preferably, in S1, for each joint of the teleoperated arm, a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model are constructed using the teleoperated arm's historical joint command torque, joint angular velocity, and joint temperature. Specific steps include: S1.1 For each joint, construct a friction torque prediction model with historical joint command torque, joint angular velocity and joint temperature as input, and output the predicted friction torque value; S1.2, Establish the nominal dynamic equation of the teleoperated arm to obtain the theoretical joint torque without considering friction; S1.3, Construct a residual compensation torque prediction model based on an enhanced time-delay neural network; The residual compensation torque prediction model is trained based on joint angle, joint angular velocity, filtered joint angular acceleration, current joint command torque, and historical cycle compensation residuals, and the output is the residual compensation torque prediction value.

[0010] More preferably, in S1.3, the method for calculating the compensation residual of the historical period is as follows: The compensation residual for the historical cycle is obtained by subtracting the corresponding predicted friction torque and theoretical joint torque from the actual total joint torque required for the control cycle.

[0011] More preferably, in S2, during each real-time control cycle of the teleoperated arm, real-time compensation calculations are performed using the constructed friction torque prediction model, the nominal dynamic equation, and the trained residual compensation torque prediction model to obtain the reference feedforward compensation torque. Specific methods include: S2.1, Based on the current motion state and joint command torque of the teleoperated arm, calculate the predicted friction torque using the friction torque prediction model, and calculate the theoretical joint torque using the nominal dynamic equation; sum the predicted friction torque and the theoretical joint torque to obtain the reference compensation torque; wherein, the motion state includes joint angle, joint angular velocity, and joint angular acceleration; S2.2, input the current joint angle, joint angular velocity, filtered joint angular acceleration, joint command torque, and compensation residuals from the most recent real-time control cycles into the trained residual compensation torque prediction model, and output the predicted value of the residual compensation torque at the current moment; S2.3, add the base compensation torque to the predicted value of the residual compensation torque to obtain the base feedforward compensation torque.

[0012] More preferably, in S3, the end-effector force deviation of the teleoperated arm is calculated, and the end-effector force deviation is mapped to the joint torque increment; a feedback fine-tuning compensation torque is generated based on the joint torque increment, specifically including the following steps: S3.1, Acquire the actual torque vector at the end of the teleoperated arm; Calculate the desired end torque based on the current motion state and dynamic model; Calculate the difference between the actual torque vector and the desired end torque to obtain the end force deviation in Cartesian space; S3.2, using the inverse transformation of the force Jacobian matrix under the current configuration of the teleoperated arm, the end force deviation in Cartesian space is converted into the joint torque increment in joint space; S3.3 Multiply the joint torque increment by the set gain matrix to generate the final feedback fine-tuning compensation torque.

[0013] More preferably, in S4, the total compensation torque is calculated based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque. The specific method includes: S4.1, add the reference feedforward compensation torque and the feedback fine-tuning compensation torque to obtain the final total compensation torque.

[0014] S4.2, add the total compensation torque to the real-time joint torque command to obtain the final torque command sent to each joint servo driver.

[0015] More preferably, the method further includes: S5, determining whether to perform model parameter fine-tuning based on the motion state, real-time joint torque command and feedback fine-tuning compensation torque in the current and past control cycles; and using the fine-tuned model to perform torque compensation in subsequent control cycles.

[0016] More preferably, in S5, it is determined whether to perform model parameter fine-tuning, and the fine-tuned model is used for torque compensation in subsequent control cycles. Specific steps include: S5.1 Continuously collect operating status data and feedback fine-tuning compensation torque in the current and several past consecutive control cycles. When the operating status data accumulates to a preset amount, calculate the mean and standard deviation of the feedback fine-tuning compensation torque in several control cycles. When the absolute value of the mean is greater than the set mean threshold and the standard deviation is less than the set threshold, it is determined that there is a systematic deviation in the reference feedforward compensation torque generated in the current control cycle, and proceed to step S5.2 to perform model parameter fine-tuning. S5.2 Calculate the Pearson correlation coefficient between the feedback fine-tuning compensation torque and the joint angular acceleration in the running status data during the current control cycle, and at the same time calculate the correlation coefficient between the feedback fine-tuning compensation torque and the joint angular velocity in the running status data; When the absolute value of the Pearson correlation coefficient is greater than the set Pearson threshold and the ratio of the absolute value of the correlation coefficient to the absolute value of the set correlation coefficient threshold is greater than the set ratio threshold, it is determined that the current systematic deviation is caused by unmodeled dynamic characteristics, and the parameters of the residual compensation torque prediction model are fine-tuned first. If the above conditions are not met, the parameters of the friction torque prediction model should be fine-tuned. S5.3 After the model is fine-tuned, the feedback fine-tuning compensation torque of the subsequent control cycle is continuously calculated and it is determined whether there is a systematic deviation. If so, steps S5.1-S5.2 are repeated.

[0017] This invention also proposes a joint torque compensation system for a remotely operated arm, comprising a physical model construction module, a reference feedforward compensation torque calculation module, a feedback fine-tuning compensation torque generation module, and a compensation module: The physical model building module constructs a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model for each joint of the teleoperated arm, using the historical joint command torque, joint angular velocity, and joint temperature of the teleoperated arm. The reference feedforward compensation torque calculation module performs real-time compensation calculations in each real-time control cycle of the teleoperated arm using the constructed friction torque prediction model, the nominal dynamic equation, and the trained residual compensation torque prediction model to obtain the reference feedforward compensation torque. The feedback fine-tuning compensation torque generation module calculates the end-effector force deviation of the teleoperated arm and maps the end-effector force deviation to the joint torque increment; it then generates the feedback fine-tuning compensation torque based on the joint torque increment. The compensation module calculates the total compensation torque based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque; and calculates the final torque command of each joint servo drive based on the total compensation torque.

[0018] The present invention also proposes a terminal, including a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to perform the steps of the above method according to the instructions.

[0019] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the above-described method.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs an improved friction torque prediction model using joint command torque, angular velocity, and temperature as inputs. Compared to traditional models such as Coulomb-viscosity models that only consider velocity, this model can more accurately characterize the nonlinear and time-varying influence of load torque on friction, fundamentally improving the accuracy of the baseline model. Simultaneously, it introduces a residual compensation torque prediction model based on an enhanced time-delay neural network (ETDNN), which can specifically learn and compensate for model "blind zone" errors caused by unmodeled dynamics, transmission elasticity, and complex coupling effects. This "physical model first, data-driven network fine-tuning" architecture effectively overcomes the problems of overly simplified models and inability to compensate for complex dynamic effects in existing technologies, achieving higher-precision feedforward compensation.

[0021] 2. This invention introduces an online fine-tuning compensation loop based on feedback from the end-effector force sensor. This loop calculates the end-effector force deviation in real time and uses the inverse Jacobian matrix transformation to quickly map it into joint torque increments, generating feedback fine-tuning compensation torque. This constitutes a fast force control inner loop independent of the feedforward path, capable of providing immediate (millisecond-level) compensation and correction for sudden external disturbances such as inaccurate model predictions, abrupt changes in contact stiffness, and collisions with unknown obstacles. This mechanism effectively compensates for the shortcomings of pure model feedforward or single learning network compensation response lag, ensuring high fidelity and low latency in force perception under highly dynamic and uncertain environments.

[0022] 3. This invention, through an integrated online adaptive step for model parameters, enables the system to automatically determine the source of error (dynamic characteristics or frictional characteristics) by utilizing stable and reproducible feedback compensation torque deviations observed during long-term operation. It then uses online learning algorithms (such as gradient descent and recursive least squares) to perform small incremental updates to the parameters of the friction torque prediction model or ETDNN. This capability allows the compensation system to continuously approximate the real, time-varying system dynamics, gradually absorbing steady-state deviations into the feedforward model, thereby continuously optimizing performance and gradually reducing the real-time dependence on the end-force feedback loop. This solves the problem in existing technologies where static, offline compensation models cannot adapt to long-term system changes (such as wear and temperature drift), significantly improving the system's long-term stability and intelligence level. Attached Figure Description

[0023] Figure 1 This is a flowchart of a joint torque compensation method for a remotely operated arm according to the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0025] like Figure 1 As shown, this invention proposes a joint torque compensation method for a remotely operated arm, the specific steps of which include: S1. For each joint of the teleoperated arm, a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model are constructed using the historical joint command torque, joint angular velocity, and joint temperature of the teleoperated arm. S1.1 For each joint, construct a friction torque prediction model with historical joint command torque, joint angular velocity and joint temperature as input, and output the predicted friction torque value; S1.2, Establish the nominal dynamic equation of the teleoperated arm to obtain the theoretical joint torque without considering friction; S1.3, Construct a residual compensation torque prediction model based on an enhanced time-delay neural network; The residual compensation torque prediction model is trained based on joint angle, joint angular velocity, filtered joint angular acceleration, current joint command torque, and historical cycle compensation residuals, and the output is the residual compensation torque prediction value.

[0026] The compensation residual for the historical cycle is obtained by subtracting the corresponding predicted friction torque and theoretical joint torque from the actual total joint torque required for the control cycle.

[0027] S2, in each real-time control cycle of the teleoperated arm, real-time compensation calculation is performed using the constructed friction torque prediction model, the nominal dynamic equation and the trained residual compensation torque prediction model to obtain the reference feedforward compensation torque; S2.1, Based on the current motion state and joint command torque of the teleoperated arm, calculate the predicted friction torque using the friction torque prediction model, and calculate the theoretical joint torque using the nominal dynamic equation; sum the predicted friction torque and the theoretical joint torque to obtain the reference compensation torque; wherein, the motion state includes joint angle, joint angular velocity, and joint angular acceleration; S2.2, input the current joint angle, joint angular velocity, filtered joint angular acceleration, joint command torque, and compensation residuals from the most recent real-time control cycles into the trained residual compensation torque prediction model, and output the predicted value of the residual compensation torque at the current moment; S2.3, add the base compensation torque to the predicted value of the residual compensation torque to obtain the base feedforward compensation torque.

[0028] S3, calculate the end-effector force deviation of the teleoperated arm, map the end-effector force deviation to the joint torque increment; generate feedback fine-tuning compensation torque based on the joint torque increment; S3.1, Acquire the actual torque vector at the end of the teleoperated arm; Calculate the desired end torque based on the current motion state and dynamic model; Calculate the difference between the actual torque vector and the desired end torque to obtain the end force deviation in Cartesian space; S3.2, using the inverse transformation of the force Jacobian matrix under the current configuration of the teleoperated arm, the end force deviation in Cartesian space is converted into the joint torque increment in joint space; S3.3 Multiply the joint torque increment by the set gain matrix to generate the final feedback fine-tuning compensation torque.

[0029] S4 calculates the total compensation torque based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque; and calculates the final torque command of each joint servo drive based on the total compensation torque.

[0030] S4.1, add the reference feedforward compensation torque and the feedback fine-tuning compensation torque to obtain the final total compensation torque.

[0031] S4.2, add the total compensation torque to the real-time joint torque command to obtain the final torque command sent to each joint servo driver.

[0032] The method further includes: S5, determining whether to perform model parameter fine-tuning based on the motion state, real-time joint torque commands, and feedback fine-tuning compensation torque in the current and past control cycles; and using the fine-tuned model for torque compensation in subsequent control cycles.

[0033] S5.1 Continuously collect operating status data and feedback fine-tuning compensation torque in the current and several past consecutive control cycles. When the operating status data accumulates to a preset amount, calculate the mean and standard deviation of the feedback fine-tuning compensation torque in several control cycles. When the absolute value of the mean is greater than the set mean threshold and the standard deviation is less than the set threshold, it is determined that there is a systematic deviation in the reference feedforward compensation torque generated in the current control cycle, and proceed to step S5.2 to perform model parameter fine-tuning. S5.2 Calculate the Pearson correlation coefficient between the feedback fine-tuning compensation torque and the joint angular acceleration in the running status data during the current control cycle, and at the same time calculate the correlation coefficient between the feedback fine-tuning compensation torque and the joint angular velocity in the running status data; When the absolute value of the Pearson correlation coefficient is greater than the set Pearson threshold and the ratio of the absolute value of the correlation coefficient to the absolute value of the set correlation coefficient threshold is greater than the set ratio threshold, it is determined that the current systematic deviation is caused by unmodeled dynamic characteristics, and the parameters of the residual compensation torque prediction model are fine-tuned first. If the above conditions are not met, the parameters of the friction torque prediction model should be fine-tuned. S5.3 After the model is fine-tuned, the feedback fine-tuning compensation torque of the subsequent control cycle is continuously calculated and it is determined whether there is a systematic deviation. If so, steps S5.1-S5.2 are repeated.

[0034] This invention also proposes a joint torque compensation system for a remotely operated arm, comprising a physical model construction module, a reference feedforward compensation torque calculation module, a feedback fine-tuning compensation torque generation module, and a compensation module: The physical model building module constructs a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model for each joint of the teleoperated arm, using the historical joint command torque, joint angular velocity, and joint temperature of the teleoperated arm. The reference feedforward compensation torque calculation module performs real-time compensation calculations in each real-time control cycle of the teleoperated arm using the constructed friction torque prediction model, the nominal dynamic equation, and the trained residual compensation torque prediction model to obtain the reference feedforward compensation torque. The feedback fine-tuning compensation torque generation module calculates the end-effector force deviation of the teleoperated arm and maps the end-effector force deviation to the joint torque increment; it then generates the feedback fine-tuning compensation torque based on the joint torque increment. The compensation module calculates the total compensation torque based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque; and calculates the final torque command of each joint servo drive based on the total compensation torque.

[0035] The present invention also proposes a terminal, including a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to perform the steps of the above method according to the instructions.

[0036] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the above-described method.

[0037] Example 1 To improve the force perception and operational accuracy of teleoperated arms and address the issues of insufficient accuracy in traditional friction models, lag in response of data-driven models, and insensitivity to sudden disturbances, this invention proposes a joint torque compensation method for remote teleoperated arms, specifically including the following steps: S1. For each joint of the teleoperated arm, a friction torque prediction model is constructed using the teleoperated arm's historical joint command torque, joint angular velocity, and joint temperature to obtain the predicted friction torque value; the nominal dynamic equation of the teleoperated arm is constructed based on the current joint angle to obtain the theoretical joint torque; using the joint angle, the predicted friction torque value, and the theoretical joint torque, a residual compensation prediction model is trained to output the residual compensation torque prediction value. S1.1, For each joint, construct a system based on historical joint command torques. Joint angular velocity and joint temperature The input friction torque prediction model Output the predicted value of friction torque This model can characterize the nonlinear and time-varying influence of load torque on friction.

[0038] The friction torque prediction model can be, but is not limited to, a parametric polynomial / nonlinear function model, a high-dimensional lookup table model, or a feedforward neural network model.

[0039] S1.2, Based on the Lagrange method or the Newton-Euler method, establish the nominal dynamic equation of the teleoperated arm to obtain the theoretical joint torque without considering friction: .in, These are joint angle, joint angular velocity, and joint angular acceleration, respectively. The inertia matrix; The matrix represents the Coriolis force and the centrifugal force. This is the gravity vector.

[0040] S1.3 The basic architecture of the residual compensation torque prediction model is the enhanced time delay neural network (ETDNN).

[0041] Based on joint angle Joint angular velocity Filtered joint angular acceleration (through the process of) (Obtained by differentiation after low-pass filtering) Current joint command torque The system is trained using historical periodic compensation residuals, and the output is a predicted value of the residual compensation torque. ; Among them, the compensation residual of the historical period = ; Train the ETDNN to learn this residual mapping .

[0042] S2, in each real-time control cycle of the teleoperated arm, real-time compensation calculation is performed using the friction torque prediction model, the nominal dynamic equation and the trained residual compensation torque prediction model. S2.1, based on the current motion state of the teleoperated arm, including and joint command torque The predicted friction torque is calculated using a friction torque prediction model, and the theoretical joint torque is calculated using the nominal dynamic equation. The predicted friction torque and the theoretical joint torque are then summed to obtain the benchmark compensation torque. .

[0043] S2.2, will the current time The compensation residuals from the most recent real-time control cycles are input into the trained residual compensation torque prediction model, which outputs the predicted residual compensation torque value at the current moment. .

[0044] S2.3, add the predicted value of the baseline compensation torque to the predicted value of the residual compensation torque to obtain the baseline feedforward compensation torque: .

[0045] S3, calculate the end-effector force deviation of the teleoperated arm, map the end-effector force deviation to the joint torque increment; generate feedback fine-tuning compensation torque based on the joint torque increment.

[0046] S3.1, Acquire the actual torque vector at the end of the teleoperated arm. Based on the current motion state And the dynamic model, calculate the desired end moment The difference between the two is the end force deviation in Cartesian space: .

[0047] S3.2, using the inverse transformation of the force Jacobian matrix under the current configuration of the teleoperated arm, the end effector force deviation in Cartesian space is converted into the joint torque increment in joint space, thus obtaining the joint torque increment. The conversion formula is: .in, It is the geometric Jacobian matrix of the manipulator.

[0048] S3.3, Joint torque increment This reflects the instantaneous difference between the expected theoretical value and the actual value. This is achieved through a configurable gain matrix. (Typically a diagonal matrix, used to adjust the response intensity), the final feedback fine-tuning compensation torque is generated based on the joint torque increment: This invention constitutes a fast force control inner loop, capable of responding to sudden disturbances such as abrupt changes in contact stiffness and unknown obstacles within milliseconds.

[0049] S4 calculates the total compensation torque based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque; and calculates the final torque command of each joint servo drive based on the total compensation torque.

[0050] S4.1, add the reference feedforward compensation torque and the feedback fine-tuning compensation torque to obtain the final total compensation torque: .

[0051] S4.2, the total compensation torque With real-time joint torque commands The summation yields the final torque command sent to each joint servo driver: The driver controls the motor according to this instruction, achieving high-precision, high-fidelity force feedback and motion tracking.

[0052] S5, based on the motion state in the current and past control cycles, real-time joint torque commands, and feedback fine-tuning compensation torque. Determine whether to perform model parameter fine-tuning; use the fine-tuned model for torque compensation in subsequent control cycles.

[0053] S5.1 continuously collects operating status data and feedback fine-tuning compensation torque for the current and several past consecutive control cycles. Once the operating status data accumulates to a preset amount, the feedback fine-tuning compensation torque is calculated over several control cycles. The mean and standard deviation are calculated. When the absolute value of the mean is greater than the set mean threshold and the standard deviation is less than the set threshold, it is determined that there is a systematic deviation in the reference feedforward compensation torque generated in the current control cycle, and the process proceeds to step S5.2 to fine-tune the model parameters. S5.2, Calculate the feedback fine-tuning compensation torque in the current control cycle. Joint angular acceleration in running status data The Pearson correlation coefficient was calculated, and the feedback fine-tuning compensation torque was also calculated. Joint angular velocity in running status data The correlation coefficient; When the absolute value of the Pearson correlation coefficient is greater than the set Pearson threshold and the ratio of the absolute value of the correlation coefficient to the absolute value of the set correlation coefficient threshold is greater than the set ratio threshold, it is determined that the current systematic deviation is mainly caused by unmodeled dynamic characteristics (such as high-frequency inertial coupling and transmission elastic vibration). In this case, parameter fine-tuning of the residual compensation torque prediction model is prioritized because the structure of the residual compensation torque prediction model is specifically designed to compensate for such dynamically related residuals. If the above conditions are not met, the judgment deviation is mainly caused by unmodeled friction related to speed or load, and the parameters of the friction torque prediction model should be fine-tuned first. In a preferred embodiment of the present invention, the specific steps for fine-tuning the parameters of the residual compensation torque prediction model include: The observed systematic biases are used as supervision signals. An online gradient descent algorithm is employed with an extremely low learning rate (e.g., 0.0001) to iteratively update the weights of only the last few layers (e.g., the last two layers) of the residual compensation torque prediction model. This step aims to quickly absorb the systematic errors under the current operating conditions, while simultaneously preventing catastrophic forgetting in the network through a "partial update" strategy. Furthermore, as a preferred embodiment of the present invention, the specific steps for fine-tuning the parameters of the friction torque prediction model include: The friction torque prediction model is linearized in the current control cycle, and the model parameters are identified and updated online using the recursive least squares method, so that the model prediction value gradually approaches "predicted value + observation deviation".

[0054] S5.3 After the model is fine-tuned, the feedback fine-tuning compensation torque of the subsequent control cycle is continuously calculated and it is determined whether there is a systematic deviation. If so, steps S5.1-S5.2 are repeated.

[0055] Example 2 This invention also proposes a joint torque compensation system for a remotely operated arm, comprising a physical model construction module, a reference feedforward compensation torque calculation module, a feedback fine-tuning compensation torque generation module, and a compensation module: The physical model building module constructs a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model for each joint of the teleoperated arm, using the historical joint command torque, joint angular velocity, and joint temperature of the teleoperated arm. The reference feedforward compensation torque calculation module performs real-time compensation calculations in each real-time control cycle of the teleoperated arm using the constructed friction torque prediction model, the nominal dynamic equation, and the trained residual compensation torque prediction model to obtain the reference feedforward compensation torque. The feedback fine-tuning compensation torque generation module calculates the end-effector force deviation of the teleoperated arm and maps the end-effector force deviation to the joint torque increment; it then generates the feedback fine-tuning compensation torque based on the joint torque increment. The compensation module calculates the total compensation torque based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque; and calculates the final torque command of each joint servo drive based on the total compensation torque.

[0056] Example 3 The present invention also proposes a terminal, including a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to perform the steps of the above method according to the instructions.

[0057] Example 4 The present invention also proposes a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the above-described method.

[0058] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0059] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0060] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0061] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute 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 a remote computer, the remote computer may 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 may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for joint torque compensation of a remotely operated arm, characterized in that, include: S1. For each joint of the teleoperated arm, a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model are constructed using the historical joint command torque, joint angular velocity, and joint temperature of the teleoperated arm. S2, in each real-time control cycle of the teleoperated arm, real-time compensation calculation is performed using the constructed friction torque prediction model, the nominal dynamic equation and the trained residual compensation torque prediction model to obtain the reference feedforward compensation torque; S3, calculate the end-effector force deviation of the teleoperated arm and map the end-effector force deviation to the joint torque increment; Feedback-based fine-tuning compensation torque is generated based on joint torque increment; S4, calculate the total compensation torque based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque; The final torque command of each joint servo drive is obtained based on the total compensation torque calculation.

2. The method for joint torque compensation of a remotely operated arm according to claim 1, characterized in that: In S1, for each joint of the teleoperated arm, a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model are constructed using the teleoperated arm's historical joint command torque, joint angular velocity, and joint temperature. Specific steps include: S1.1 For each joint, construct a friction torque prediction model with historical joint command torque, joint angular velocity and joint temperature as input, and output the predicted friction torque value; S1.2, Establish the nominal dynamic equation of the teleoperated arm to obtain the theoretical joint torque without considering friction; S1.3, Construct a residual compensation torque prediction model based on an enhanced time-delay neural network; The residual compensation torque prediction model is trained based on joint angle, joint angular velocity, filtered joint angular acceleration, current joint command torque, and historical cycle compensation residuals, and the output is the residual compensation torque prediction value.

3. The method for joint torque compensation of a remotely operated arm according to claim 2, characterized in that: In S1.3, the calculation method for the compensation residual of the historical period is as follows: The compensation residual for the historical cycle is obtained by subtracting the corresponding predicted friction torque and theoretical joint torque from the actual total joint torque required for the control cycle.

4. The method for joint torque compensation of a remotely operated arm according to claim 1, characterized in that: In S2, during each real-time control cycle of the teleoperated arm, real-time compensation calculations are performed using the constructed friction torque prediction model, the nominal dynamic equation, and the trained residual compensation torque prediction model to obtain the reference feedforward compensation torque. Specific methods include: S2.1, Based on the current motion state and joint command torque of the teleoperated arm, calculate the predicted friction torque using the friction torque prediction model, and calculate the theoretical joint torque using the nominal dynamic equation; sum the predicted friction torque and the theoretical joint torque to obtain the reference compensation torque; wherein, the motion state includes joint angle, joint angular velocity, and joint angular acceleration; S2.2, input the current joint angle, joint angular velocity, filtered joint angular acceleration, joint command torque, and compensation residuals from the most recent real-time control cycles into the trained residual compensation torque prediction model, and output the predicted value of the residual compensation torque at the current moment; S2.3, add the predicted value of the benchmark compensation torque to the predicted value of the residual compensation torque to obtain the benchmark feedforward compensation torque.

5. The method for joint torque compensation of a remotely operated arm according to claim 1, characterized in that: In S3, the end-effector force deviation of the teleoperated arm is calculated and mapped to the joint torque increment. The specific steps for generating feedback fine-tuning compensation torque based on joint torque increments include: S3.1, Acquire the actual torque vector at the end of the teleoperated arm; Calculate the desired end torque based on the current motion state and dynamic model; Calculate the difference between the actual torque vector and the desired end torque to obtain the end force deviation in Cartesian space; S3.2, using the inverse transformation of the force Jacobian matrix under the current configuration of the teleoperated arm, the end force deviation in Cartesian space is converted into the joint torque increment in joint space; S3.3 Multiply the joint torque increment by the set gain matrix to generate the final feedback fine-tuning compensation torque.

6. The method for joint torque compensation of a remotely operated arm according to claim 1, characterized in that: In S4, the total compensation torque is calculated based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque. Specific methods include: S4.1, add the reference feedforward compensation torque and the feedback fine-tuning compensation torque to obtain the final total compensation torque; S4.2, add the total compensation torque to the real-time joint torque command to obtain the final torque command sent to each joint servo driver.

7. A method for joint torque compensation for a remotely operated arm according to claim 1, characterized in that: The method further includes: S5, determining whether to perform model parameter fine-tuning based on the motion state, real-time joint torque commands, and feedback fine-tuning compensation torque in the current and past control cycles; and using the fine-tuned model for torque compensation in subsequent control cycles.

8. A method for joint torque compensation for a remotely operated arm according to claim 7, characterized in that: In S5, it determines whether to perform model parameter fine-tuning and uses the fine-tuned model for torque compensation in subsequent control cycles. The specific steps include: S5.1 Continuously collect operating status data and feedback fine-tuning compensation torque in the current and several past consecutive control cycles. When the operating status data accumulates to a preset amount, calculate the mean and standard deviation of the feedback fine-tuning compensation torque in several control cycles. When the absolute value of the mean is greater than the set mean threshold and the standard deviation is less than the set threshold, it is determined that there is a systematic deviation in the reference feedforward compensation torque generated in the current control cycle, and proceed to step S5.2 to perform model parameter fine-tuning. S5.2 Calculate the Pearson correlation coefficient between the feedback fine-tuning compensation torque and the joint angular acceleration in the running status data during the current control cycle, and at the same time calculate the correlation coefficient between the feedback fine-tuning compensation torque and the joint angular velocity in the running status data; When the absolute value of the Pearson correlation coefficient is greater than the set Pearson threshold and the ratio of the absolute value of the correlation coefficient to the absolute value of the set correlation coefficient threshold is greater than the set ratio threshold, it is determined that the current systematic deviation is caused by unmodeled dynamic characteristics, and the parameters of the residual compensation torque prediction model are fine-tuned first. If the above conditions are not met, the parameters of the friction torque prediction model should be fine-tuned. S5.3 After the model is fine-tuned, the feedback fine-tuning compensation torque of the subsequent control cycle is continuously calculated and it is determined whether there is a systematic deviation. If so, steps S5.1-S5.2 are repeated.

9. A joint torque compensation system for a remotely operated arm using the method described in any one of claims 1-8, comprising a physical model construction module, a reference feedforward compensation torque calculation module, a feedback fine-tuning compensation torque generation module, and a compensation module, characterized in that: The physical model building module constructs a friction torque prediction model, a nominal dynamic equation, and a compensation residual prediction model for each joint of the teleoperated arm, using the historical joint command torque, joint angular velocity, and joint temperature of the teleoperated arm. The reference feedforward compensation torque calculation module performs real-time compensation calculations in each real-time control cycle of the teleoperated arm using the constructed friction torque prediction model, the nominal dynamic equation, and the trained residual compensation torque prediction model to obtain the reference feedforward compensation torque. The feedback fine-tuning compensation torque generation module calculates the end force deviation of the teleoperated arm and maps the end force deviation into joint torque increments. Feedback-based fine-tuning compensation torque is generated based on joint torque increment; The compensation module calculates the total compensation torque based on the reference feedforward compensation torque and the feedback fine-tuning compensation torque; The final torque command of each joint servo drive is obtained based on the total compensation torque calculation.

10. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-8.

Images