Torque compensation calibration method and system based on joint mechanical arm

Abstract

The invention discloses a torque compensation calibration method and system based on a joint mechanical arm, and relates to the technical field of robot intelligent force control, and the torque compensation calibration method comprises the steps: configuring a joint angular position encoder and a current measurement module on the joint mechanical arm, setting operation parameters, generating an original joint angular position based on the joint angular position encoder, calculating a joint angular velocity and a joint angular acceleration based on the original joint angular position, and estimating a motor torque; calculating a joint external torque residual error by using joint angular velocity and acceleration and motor torque estimation, and generating a real-time residual error sequence based on the external torque residual error; contact is judged based on the real-time residual sequence, a joint angle is recorded as a contact balance point after contact, data are collected based on the contact balance point, and an annotation data set is generated; based on the labeled data set, rigidity and damping parameters are obtained through a least square algorithm, and fusion parameters are obtained based on the rigidity and damping parameters. The precision and safety of torque compensation of the mechanical arm in the time-varying environment are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent force control technology for robots, and in particular to a torque compensation calibration method and system based on articulated robotic arms. Background Technology

[0002] With the continuous development of industrial automation and intelligent manufacturing, articulated robotic arms are increasingly used in precision assembly, flexible grinding, and human-robot collaboration, placing higher demands on force control performance. Traditional force control solutions often rely on external six-dimensional force / torque sensors, which suffer from high cost, susceptibility to interference, and complex installation. In recent years, sensorless force control technology based on joint torque observation has gradually become a research hotspot. Its core idea is to reconstruct the joint output torque through motor current and dynamic model, and combine residual analysis to estimate the external contact force. This type of method relies on high-precision encoders and real-time control architectures and has been applied in some industrial robot platforms. To improve the adaptability of force control, researchers have introduced online parameter identification techniques, such as recursive least squares (RLS) and Kalman filtering, to dynamically estimate environmental stiffness and damping characteristics, thereby achieving self-tuning of compliant control parameters. Combined with frequency sweep excitation, state filtering, and friction compensation, the accuracy and robustness of parameter identification have been significantly improved, providing a theoretical basis and technical path for achieving high-precision torque compensation. Existing residual-based torque compensation methods still have limitations. Several key technical bottlenecks exist. Most solutions rely on offline calibrated dynamic parameters, which are difficult to cope with model mismatch caused by load changes, temperature drift, or mechanical wear, resulting in high residual noise and high contact detection misjudgment rate. Secondly, the online identification process often lacks a complete verification mechanism, failing to systematically verify the physical rationality of the identification parameters, residual quality, and dynamic smoothness, which easily introduces unstable control commands. Existing technologies are relatively crude in their post-contact parameter update strategies, lacking a re-identification mechanism with clear trigger conditions and safety backoff logic, making it difficult to adapt to complex and ever-changing operating environments. During data acquisition and parameter fusion, the setting of time synchronization, filtering strategies, and fusion coefficients lacks engineering considerations, affecting control continuity and restricting the practical application of sensorless force control technology in high-dynamic and high-reliability scenarios. Summary of the Invention

[0003] In view of the aforementioned existing problems, the present invention is proposed.

[0004] Therefore, this invention provides a torque compensation calibration method and system based on articulated robotic arms, which solves the problems of poor parameter adaptability, low identification reliability, and lack of safety mechanisms in existing force control technologies based on joint torque observation.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a torque compensation calibration method based on an articulated robotic arm, comprising, A joint angle position encoder and a current measurement module are configured on the articulated robotic arm, and the operating parameters are set. The original joint angle position is generated based on the joint angle position encoder, and the joint angular velocity, joint angular acceleration and motor torque are estimated based on the original joint angle position. Using joint angular velocity and acceleration as well as motor torque estimation, the external torque residual of the joint is calculated, and a real-time residual sequence is generated based on the external torque residual. Contact is determined based on real-time residual sequences, and the joint angle is recorded as the contact equilibrium point after contact. Data is collected based on the contact equilibrium point to generate a labeled dataset. Based on the labeled dataset, stiffness and damping parameters are obtained through the least squares algorithm, and fused parameters are obtained based on the stiffness and damping parameters. Environmental torque compensation is calculated based on fused parameters, and then superimposed with nominal dynamics feedforward to generate torque commands. The RLS algorithm is then run to update parameters and logs are recorded.

[0006] As a preferred embodiment of the torque compensation calibration method based on articulated robotic arms described in this invention, the method of configuring a joint encoder and a current measurement module on the articulated robotic arm and setting operating parameters refers to installing and enabling the joint angular position encoder and the motor current measurement module on the real-time motion controller of the industrial robot, recording the torque constant on the motor nameplate into the controller, loading the nominal dynamic model of the robotic arm into the real-time motion controller, and setting deterministic parameters in the real-time motion controller.

[0007] As a preferred embodiment of the torque compensation calibration method for articulated robotic arms described in this invention, the following steps are described: generating the original joint angle position based on the joint angle position encoder, calculating the joint angular velocity and joint angular acceleration based on the original joint angle position, and estimating the motor torque refer to the following steps: based on the joint angle position encoder and motor current measurement module configured in the controller, the original joint angle position is sampled and read every millisecond; the original instantaneous joint angular velocity is calculated using the center difference method; then, the velocity value is applied to the exponential smoothing filter to obtain the joint velocity estimate; the acceleration is calculated in a differential manner and a smoother is applied to obtain the joint angular acceleration estimate; the motor current signal is first subjected to a small bandwidth low-pass filter to suppress high-frequency noise; the filtered current is multiplied by the torque constant and the drive correction coefficient to obtain the real-time motor output torque estimate; and the four sets of data are stored in a high-speed circular cache with timestamps.

[0008] As a preferred embodiment of the torque compensation calibration method for articulated robotic arms described in this invention, the following steps are taken: The calculation of the external torque residual of the joint using joint angular velocity and acceleration, as well as motor torque estimation, and the generation of a real-time residual sequence based on the external torque residual refer to the following steps in the controller: The nominal dynamic model is used in conjunction with the current angular acceleration and angular velocity to calculate the expected driving torque requirement of the model; then, the expected value of the model and the friction torque estimated by the friction model are subtracted from the real-time motor torque estimation. The result is the external torque residual of each joint. Under non-contact safety conditions, the system runs for several seconds to record the baseline noise variance of the residual as a subsequent threshold reference. The external torque residual values ​​are arranged by time to form a real-time residual sequence, which is then saved and the time period is marked.

[0009] As a preferred embodiment of the torque compensation calibration method for the articulated robotic arm described in this invention, the following steps are taken: Contact is determined based on a real-time residual sequence, and the joint angle is recorded as the contact equilibrium point after contact. Data is collected based on the contact equilibrium point to generate a labeled dataset. Specifically, based on the real-time residual sequence, if the estimated value of a joint exceeds a set threshold and continues to exceed a set de-jittering time, contact is determined to have occurred. After contact is determined, the control loop of the joint is switched to a low-gain steady-state mode, and contact equilibrium recording is performed to achieve near-static equilibrium. When the environmental torque fluctuation amplitude of the contact segment is lower than the baseline setting, the current joint angle is recorded as the contact equilibrium point of that joint. Frequency sweep angle excitation is performed sequentially on each joint to be identified. During the excitation process, the joint displacement change with reference to the equilibrium point, the corresponding velocity change, and the simultaneous environmental torque estimate are recorded. The data are then aligned by time and labeled with joint numbers and contact segment identifiers to generate a labeled dataset.

[0010] As a preferred embodiment of the torque compensation calibration method for articulated robotic arms described in this invention, the following steps are performed: Based on a labeled dataset, stiffness and damping parameters are obtained using a least squares algorithm. The fusion parameters are then obtained by running a recursive least squares algorithm on the identification dataset of each joint within the controller, recursively updating the stiffness and damping parameter estimates. After identification, the stiffness and damping parameters are verified, including checking that they are non-negative and within pre-defined upper and lower limits in engineering, calculating the variance of the residuals after identification and comparing it with the saved baseline noise variance, requiring the residual variance to be less than a preset multiple of the baseline to determine the identification quality, performing exponential smoothing fusion on the parameters, combining the new parameters with the controller's currently used parameters and fusion coefficients to generate fusion parameters, marking any verification failure as unreliable and triggering a re-excitation process, and finally storing the smoothed and verified stiffness and damping parameters of each joint in the controller's operating parameter table.

[0011] As a preferred embodiment of the torque compensation calibration method for articulated robotic arms described in this invention, the following steps are taken: Environmental torque compensation is calculated based on fused parameters and superimposed with nominal dynamics feedforward to generate a torque command. The RLS algorithm is run to update parameters and log the process. The fused and verified stiffness and damping parameters are used. The difference between the current joint angle and the contact equilibrium point is calculated, multiplied by the stiffness parameter, and multiplied by the joint velocity and damping parameter to obtain an estimated environmental torque model for the joint. The estimated value is taken in the opposite direction as the compensation torque, and superimposed with the existing nominal dynamics feedforward to form the final torque command, which is then sent to the motor drive. During execution, a saturation check is performed on the applied torque, including gradually reducing the environmental compensation component according to a linear retention strategy when the torque approaches the motor's rated capacity, and recording the event. Simultaneously, a rolling window monitors the residual variance between the environmental torque estimate and the model output. If the residual variance exceeds three times the initial identification baseline, a re-identification process is automatically triggered. At the moment of triggering, the controller is switched to a safe mode that only performs nominal dynamics compensation and an alarm is sent to the host computer. During each contact, the RLS algorithm is continuously run at a low frequency to update the environmental parameters, and the updated environmental parameters are smoothly replaced in the running parameter table.

[0012] Secondly, the present invention provides a torque compensation calibration system based on an articulated robotic arm, comprising, The sensor configuration and parameter initialization module is responsible for deploying the joint angle position encoder and motor current measurement module on the controller and configuring the operating parameters. The real-time state estimation module is used to sample and calculate joint angles, angular velocities, angular accelerations, and motor output torque based on encoder and current signals. The dynamic residual calculation module is used to obtain the external environment torque residual using the nominal dynamic model and the friction model, and store the residual sequence by time stamp; The contact detection module is used to determine contact by monitoring whether the environmental torque residual exceeds a threshold and continuously de-jittering for a certain period of time, and to generate a labeled dataset. The online parameter identification module is used to run a recursive least squares algorithm on the labeled dataset to identify the environmental stiffness and damping parameters of each joint.

[0013] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the torque compensation calibration method for articulated robotic arms as described in the first aspect of the present invention.

[0014] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the torque compensation calibration method for articulated robotic arms as described in the first aspect of the present invention.

[0015] The beneficial effects of this invention are as follows: By configuring an encoder and a current measurement module on the articulated robotic arm, combined with high-frequency state estimation, residual-driven contact detection, equilibrium-point-based labeled excitation acquisition, recursive least squares identification with multiple verifications, and a smooth parameter fusion mechanism, it effectively overcomes the defects of traditional sensorless force control, such as model mismatch, unreliable identification, parameter mutations, and lack of safety closed loop. It significantly improves the identification accuracy and robustness of environmental stiffness and damping parameters, realizes online self-calibration and dynamic updating of torque compensation, enhances the adaptability of the articulated robotic arm and its supporting real-time motion controller to unknown or time-varying environments, and ensures operational reliability through residual monitoring and safety backoff strategies. It has good engineering practicality and promotion value. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the torque compensation calibration method based on the articulated robotic arm in Example 1.

[0018] Figure 2 This is a schematic diagram of the torque compensation calibration system based on the articulated robotic arm in Example 1.

[0019] Figure 3 This is the parameter identification and compensation control logic diagram in Example 1. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, referring to Figures 1 to 3 This is the first embodiment of the present invention, which provides a torque compensation calibration method based on an articulated robotic arm, including the following steps: S1. Configure a joint angle position encoder and a current measurement module on the articulated robotic arm, and set the operating parameters. Generate the original joint angle position based on the joint angle position encoder, and calculate the joint angular velocity, joint angular acceleration and motor torque estimation based on the original joint angle position. Specifically, a joint encoder and a current measurement module are configured on the articulated robotic arm, and the operating parameters are set. This involves installing and enabling the joint angular position encoder and the motor current measurement module on the real-time motion controller of the industrial robot (articulated robotic arm), setting the controller sampling period to 1 millisecond, recording the torque constant from the motor nameplate (the factory parameter label on the motor housing, which contains the key electrical and mechanical characteristics of the motor) into the controller, loading the nominal dynamic model of the robotic arm (inertia matrix, Coriolis / eccentric terms, and gravity terms obtained from CAD parameters or previously offline identification) into the real-time motion controller, and setting deterministic parameters in the real-time motion controller, including: excitation amplitude of 1 degree, sweep frequency of 0.5 to 8 Hz, sweep frequency duration per joint of 4 seconds, parameter fusion coefficient of 0.01, de-jitter time of 5 milliseconds, and contact balance recording wait of 0.3 seconds.

[0024] By deploying a high-precision encoder and current measurement module in the real-time motion controller and setting a 1-millisecond sampling period, the high dynamic response capability of the state signal is ensured. By inputting the motor torque constant and loading the nominal dynamic model, an accurate physical basis for torque observation is provided. At the same time, by setting key parameters such as excitation amplitude, sweep frequency range, de-jittering time, and balance waiting time, a standardized identification process and anti-interference mechanism are constructed. This effectively solves problems such as contact misjudgment and parameter identification instability caused by sampling delay, model mismatch, and noise interference. It significantly improves the system's perception accuracy and response reliability of environmental interaction forces, provides a high-quality data foundation and controllable operating conditions for subsequent online identification and torque compensation, and enhances the robustness and engineering applicability of the overall force control system.

[0025] Furthermore, the original joint angle position is generated based on the joint angle position encoder. The joint angular velocity, joint angular acceleration, and motor torque estimation are calculated based on the original joint angle position. The original joint angle position is sampled and read every millisecond using the joint angle position encoder and motor current measurement module configured in the controller. The original instantaneous joint angular velocity is calculated using the center difference method. Then, an exponential smoothing filter (smoothing coefficient of 0.8) is applied to the velocity value to obtain a stable joint velocity estimate. Similarly, the acceleration is calculated using the differential method and the same smoother is applied to obtain the joint angular acceleration estimate. The motor current signal is first subjected to a small bandwidth low-pass filter to suppress high-frequency noise. Then, the filtered current is multiplied by the torque constant and the drive correction coefficient to obtain the real-time motor output torque estimate. The four sets of data (joint angle, angular velocity, angular acceleration, and motor torque) are stored in a high-speed ring cache with timestamps.

[0026] By sampling the original joint angle position at a high frequency every millisecond and using a method combining central difference and exponential smoothing filtering (smoothing coefficient 0.8) to estimate velocity and acceleration, the noise amplification problem of numerical differentiation and dynamic delay are effectively balanced, improving the accuracy and stability of motion state estimation. At the same time, the motor current signal is low-pass filtered and the motor output torque is calculated by combining the torque constant and drive correction coefficient, enhancing the accuracy of torque observation. The four sets of data—joint angle, velocity, acceleration, and torque—are synchronously written into a high-speed ring buffer with timestamps, ensuring the time consistency and continuity of multi-source signals, avoiding residual distortion caused by data asynchrony, and solving common problems such as high noise in state estimation, torque observation distortion, and signal asynchrony. This provides a reliable data foundation for subsequent high-precision external force sensing, contact detection, and parameter identification, significantly improving the dynamic response capability and robustness of the force control system.

[0027] S2. Using joint angular velocity and acceleration as well as motor torque estimation, calculate the joint external torque residual and generate a real-time residual sequence based on the external torque residual. Contact is determined based on real-time residual sequences, and the joint angle is recorded as the contact equilibrium point after contact. Data is collected based on the contact equilibrium point to generate a labeled dataset. Specifically, using joint angular velocity and acceleration, as well as motor torque estimation, the external torque residual of the joint is calculated. Based on the external torque residual, a real-time residual sequence is generated. In the controller, the expected drive torque requirement (including inertial, Coriolis / centrifugal, and gravitational components) is calculated using the nominal dynamic model combined with the current angular acceleration and angular velocity. Then, the expected value of the model and the friction torque estimated by the friction model are subtracted from the real-time motor torque estimate. The result is the external torque residual of each joint (called the environmental torque estimate). Under non-contact safety conditions, the system is run for several seconds to record the baseline noise variance of the residual as a reference for subsequent thresholds. The external torque residual values ​​are arranged by time to form a real-time residual sequence, which is saved and the time period is marked (for identifying the time period division).

[0028] By combining the nominal dynamics model with real-time angular velocity and angular acceleration, the expected driving torque demand of the model is accurately calculated. Inertia, Coriolis / centrifugal, gravity, and friction model compensation terms are deducted from the measured motor torque estimate, effectively separating the torque residual reflecting external interaction and significantly improving the accuracy of environmental force perception. By recording the residual baseline noise variance in the non-contact state, a dynamic judgment threshold benchmark is established, overcoming the false triggering or missed detection problems caused by fixed thresholds in traditional methods. The residual values ​​are organized by time series and marked with time periods, providing a reliable time label for subsequent contact state division and identification data alignment. This solves key defects such as inaccurate external force estimation and unreliable contact detection caused by insufficient model compensation, large noise interference, and rigid threshold settings, enhancing the sensitivity and robustness of the force control system.

[0029] Furthermore, contact is determined based on the real-time residual sequence, and the joint angle is recorded as the contact equilibrium point after contact. Data is collected based on the contact equilibrium point to generate a labeled dataset. Based on the real-time residual sequence, if the estimated value of a joint exceeds a set threshold G and remains so for more than 5 milliseconds beyond the de-jitter time, contact is determined to have occurred (a joint of the robotic arm makes physical contact with the external environment (such as a workpiece), causing the environmental torque residual of that joint to exceed the set threshold and persist for a certain period). After determining contact, the control loop of that joint is switched to a low-gain steady-state mode, and contact equilibrium recording is performed. The process waits for more than 0.3 seconds to reach near-static equilibrium. When the environmental torque fluctuation amplitude of the contact segment is less than 5% of the baseline, the contact point is recorded. The joint angle before the contact equilibrium point of the joint is used as the contact equilibrium point of the joint. Then, small-amplitude frequency sweep angle excitation is performed on each joint that needs to be identified in sequence, including: each joint is individually subjected to a frequency sweep excitation with an amplitude of 1 degree, the frequency is smoothly swept from 0.5Hz to 8Hz, the frequency sweep duration lasts for 4 seconds, and other joints are kept stationary or under low rigidity constraints to ensure safety. During the excitation process, three columns of data are recorded, including the joint displacement change with reference to the equilibrium point, the corresponding velocity change, and the simultaneous environmental torque estimate. The data are aligned by time and labeled with joint number and contact segment identifier to generate a labeled dataset. Safety monitoring is always running during the excitation. If the command torque approaches the motor limit at any time, the excitation is immediately stopped and it reverts to nominal compensation only.

[0030] S3. Based on the labeled dataset, stiffness and damping parameters are obtained through the least squares algorithm, and fusion parameters are obtained based on the stiffness and damping parameters; Environmental torque compensation is calculated based on fused parameters, and then superimposed with nominal dynamics feedforward to generate torque commands. The RLS algorithm is then run to update parameters and logs are recorded.

[0031] Specifically, based on the labeled dataset, stiffness and damping parameters are obtained using the least squares algorithm. The fused parameters are then derived by running the recursive least squares (RLS) algorithm on the identified dataset (displacement, velocity, and environmental torque) for each joint within the controller. This involves recursively updating the joint stiffness and damping parameter estimates (the damping parameter is the environmental drag coefficient related to joint velocity identified by the RLS algorithm) using the collected displacement and velocity as regression values ​​and the environmental torque as the target value. During the update, a forgetting factor (e.g., 0.995) from the RLS algorithm is used to quickly track environmental changes. A large initial covariance is set for the RLS to ensure responsiveness to initial data. The identification process concludes. Next, the stiffness and damping parameters are verified, including checking that the stiffness and damping parameters are non-negative and within the upper and lower limits set in the engineering. The variance of the residual after identification is calculated and compared with the saved baseline noise variance. The residual variance is required to be within a preset multiple of the baseline (such as 3 times) to determine the identification quality. The parameters are exponentially smoothed and fused. The new parameters (referring to the original parameter estimates that have just been completed by RLS identification and have not been smoothed) are combined with the current parameters used by the controller with a fusion coefficient of 0.01 to generate fused parameters to avoid abrupt changes. If any verification fails, it is marked as unreliable and a re-excitation process is triggered. Finally, the smoothed and verified stiffness and damping parameters of each joint are stored in the controller's operating parameter table.

[0032] By running the Recursive Least Squares (RLS) algorithm on the displacement, velocity, and environmental torque dataset of each joint within the controller, online identification of stiffness and damping parameters is achieved. Introducing a forgetting factor (e.g., 0.995) and a large initial covariance significantly improves the algorithm's response speed to initial data and its ability to track time-varying environments. A multi-verification mechanism effectively prevents system instability caused by abnormal parameters being injected into the control loop. If any verification fails, a re-excitation process is triggered, ensuring the reliability of the identification results. This method systematically solves common problems such as slow convergence of parameter identification, poor noise resistance, lack of verification mechanisms, and oscillations caused by parameter mutations. It achieves high-precision, highly robust online modeling of environmental characteristics, providing crucial support for safe and reliable torque compensation control.

[0033] Furthermore, environmental torque compensation is calculated based on the fused parameters and superimposed with the nominal dynamics feedforward to generate a torque command. The RLS algorithm is run to update parameters and log the results. The fused and verified stiffness and damping parameters are used. The difference between the current joint angle and the contact equilibrium point is calculated and multiplied by the stiffness parameter, and then multiplied by the joint velocity and the damping parameter to obtain an estimated environmental torque model for the joint. The estimated value is taken in the opposite direction as the compensation torque and superimposed with the existing nominal dynamics feedforward (inertia, Coriolis, gravity, and friction compensation) to form the final torque command, which is then sent to the motor drive. During execution, a saturation check is performed on the applied torque. If it approaches the motor's rated capacity, a linear retention strategy is adopted to gradually reduce the torque. The system reduces the environmental compensation component and records events. Simultaneously, it monitors the residual variance between the environmental torque estimate and the model output using a rolling window. If the residual variance exceeds three times the initial identification baseline, it automatically triggers the re-identification process (returning to frequency sweep excitation and re-estimating parameters). At the moment of triggering, it switches the controller to a safe mode that only performs nominal dynamic compensation and alarms the host computer. During each contact, it continuously runs the RLS algorithm at a low frequency (per second or according to the forgetting factor) to update the environmental parameters (stiffness and damping), and smoothly replaces the updated environmental parameters in the running parameter table. At the same time, it records all identification, residual, and compensation command logs for subsequent offline analysis and maintenance.

[0034] An environmental torque model is constructed by fusing verified stiffness and damping parameters. This model combines joint pose deviations with velocity to generate compensating torque, which is then superimposed with nominal dynamics feedforward to achieve precise torque compensation output. This effectively improves the compliance and adaptability of the robotic arm in contact tasks. Torque saturation checks and linear retention strategies prevent actuator overload, ensuring system safety. A rolling window residual monitoring mechanism is introduced, automatically triggering re-identification and switching to a safe mode when residual variance increases significantly or environmental forces change abruptly. This solves the problems of poor adaptability to environmental changes and delayed fault response in traditional methods. Simultaneously, the RLS algorithm is continuously run at low frequency during contact, smoothly updating parameters to achieve dynamic tracking of slowly changing environmental characteristics. Identification, residual, and control logs are recorded throughout the process, supporting subsequent analysis and optimization. This scheme systematically overcomes defects such as low compensation accuracy, lack of abnormal response mechanisms, discontinuous parameter updates, and insufficient safety, significantly enhancing the robustness, adaptability, and engineering reliability of the force control system.

[0035] This embodiment also provides a torque compensation calibration system based on an articulated robotic arm, including: The sensor configuration and parameter initialization module is responsible for deploying the joint angle position encoder and motor current measurement module on the controller and configuring the operating parameters. The real-time state estimation module is used to sample and calculate joint angles, angular velocities, angular accelerations, and motor output torque based on encoder and current signals. The dynamic residual calculation module is used to obtain the external environment torque residual using the nominal dynamic model and the friction model, and store the residual sequence by time stamp; The contact detection module is used to determine contact by monitoring whether the environmental torque residual exceeds a threshold and continuously de-jittering for a certain period of time, and to generate a labeled dataset. The online parameter identification module is used to run a recursive least squares algorithm on the labeled dataset to identify the environmental stiffness and damping parameters of each joint.

[0036] This embodiment also provides a computer device applicable to the torque compensation calibration method based on articulated robotic arms, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the torque compensation calibration method based on articulated robotic arms as proposed in the above embodiment.

[0037] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0038] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the torque compensation calibration method and system for an articulated robotic arm as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0039] In summary, this invention, by configuring an encoder and current measurement module on the articulated robotic arm, and combining high-frequency state estimation, residual-driven contact detection, equilibrium-point-based labeled excitation acquisition, recursive least squares identification with multiple verifications, and a smooth parameter fusion mechanism, effectively overcomes the defects of traditional sensorless force control, such as model mismatch, unreliable identification, parameter mutations, and lack of safety closed loops. It significantly improves the identification accuracy and robustness of environmental stiffness and damping parameters, realizes online self-calibration and dynamic updating of torque compensation, enhances the adaptability of the articulated robotic arm and its supporting real-time motion controller to unknown or time-varying environments, and ensures operational reliability through residual monitoring and safety backoff strategies. It has good engineering practicality and promotion value.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A torque compensation calibration method based on an articulated robotic arm, characterized in that: The application relates to a method for training a robot joint torque controller, comprising the following steps: including, configuring joint angle position encoders and current measurement modules on a joint robot arm, setting operation parameters, generating original joint angle positions based on the joint angle position encoders, calculating joint angular velocity and joint angular acceleration and motor torque estimation based on the original joint angle positions; using joint angular velocity and acceleration and motor torque estimation, calculating joint external torque residual, generating real-time residual sequence based on external torque residual; determining contact based on the real-time residual sequence, and recording the joint angle as a contact equilibrium point after contact, collecting data based on the contact equilibrium point, and generating a labeled data set; based on the labeled data set, obtaining stiffness and damping parameters through a least squares algorithm, and obtaining fusion parameters based on the stiffness and damping parameters; 2.The joint-based robot arm torque compensation calibration method of claim 1, wherein: based on the fusion parameters, calculating environmental torque compensation, and superimposing torque commands with nominal dynamics feedforward to generate torque commands, running the RLS algorithm to update parameters and record logs. 3.The method of claim 2, wherein: The application relates to a method for training a robot joint torque controller, comprising the following steps:

4. The joint-based robot arm torque compensation calibration method of claim 3, wherein: configuring joint angle position encoders and current measurement modules on a joint robot arm, setting operation parameters, generating original joint angle positions based on the joint angle position encoders, calculating joint angular velocity and joint angular acceleration and motor torque estimation based on the original joint angle positions; using joint angular velocity and acceleration and motor torque estimation, calculating joint external torque residual, generating real-time residual sequence based on external torque residual; determining contact based on the real-time residual sequence, and recording the joint angle as a contact equilibrium point after contact, collecting data based on the contact equilibrium point, and generating a labeled data set; based on the labeled data set, obtaining stiffness and damping parameters through a least squares algorithm, and obtaining fusion parameters based on the stiffness and damping parameters; based on the fusion parameters, calculating environmental torque compensation, and superimposing torque commands with nominal dynamics feedforward to generate torque commands, running the RLS algorithm to update parameters and record logs. The application relates to a method for training a robot joint torque controller, comprising the following steps: configuring joint angle position encoders and current measurement modules on a joint robot arm, setting operation parameters, generating original joint angle positions based on the joint angle position encoders, calculating joint angular velocity and joint angular acceleration and motor torque estimation based on the original joint angle positions; using joint angular velocity and acceleration and motor torque estimation, calculating joint external torque residual, generating real-time residual sequence based on external torque residual; determining contact based on the real-time residual sequence, and recording the joint angle as a contact equilibrium point after contact, collecting data based on the contact equilibrium point, and generating a labeled data set; based on the labeled data set, obtaining stiffness and damping parameters through a least squares algorithm, and obtaining fusion parameters based on the stiffness and damping parameters; based on the fusion parameters, calculating environmental torque compensation, and superimposing torque commands with nominal dynamics feedforward to generate torque commands, running the RLS algorithm to update parameters and record logs. The application relates to a method for training a robot joint torque controller, comprising the following steps: configuring joint angle position encoders and current measurement modules on a joint robot arm, setting operation parameters, generating original joint angle positions based on the joint angle position encoders, calculating joint angular velocity and joint angular acceleration and motor torque estimation based on the original joint angle positions; using joint angular velocity and acceleration and motor torque estimation, calculating joint external torque residual, generating real-time residual sequence based on external torque residual; determining contact based on the real-time residual sequence, and recording the joint angle as a contact equilibrium point after contact, collecting data based on the contact equilibrium point, and generating a labeled data set; based on the labeled data set, obtaining stiffness and damping parameters through a least squares algorithm, and obtaining fusion parameters based on the stiffness and damping parameters; based on the fusion parameters, calculating environmental torque compensation, and superimposing torque commands with nominal dynamics feedforward to generate torque commands, running the RLS algorithm to update parameters and record logs.

5. The joint-based robot arm torque compensation calibration method of claim 4, wherein: The process of determining contact based on real-time residual sequences and recording joint angles as contact equilibrium points after contact, and generating labeled datasets based on contact equilibrium points, refers to the following: if the estimated value of a joint exceeds a set threshold and continues to exceed a set de-jittering time, contact is determined to have occurred. After contact is determined, the control loop of the joint is switched to a low-gain steady-state mode, and contact equilibrium recording is performed to achieve near-static equilibrium. When the environmental torque fluctuation amplitude of the contact segment is lower than the baseline setting, the current joint angle is recorded as the contact equilibrium point of the joint. Frequency sweep angle excitation is performed on each joint that needs to be identified in sequence. During the excitation process, the joint displacement change with reference to the equilibrium point, the corresponding velocity change, and the simultaneous environmental torque estimate are recorded. The data are then aligned by time and labeled with joint numbers and contact segment identifiers to generate a labeled dataset.

6. The joint-based robot arm torque compensation calibration method of claim 5, wherein: The process involves obtaining stiffness and damping parameters based on a labeled dataset using a least squares algorithm. The fusion parameters are then derived by running a recursive least squares algorithm on the identification dataset for each joint within the controller, recursively updating the estimated stiffness and damping parameters. After identification, the stiffness and damping parameters are validated, including checking that they are non-negative and within pre-defined upper and lower limits. The variance of the residuals after identification is calculated and compared with the saved baseline noise variance. The residual variance is required to be less than a preset multiple of the baseline to determine the identification quality. The parameters are then exponentially smoothed and fused. The new parameters are combined with the controller's currently used parameters and fusion coefficients to generate fusion parameters. If any validation fails, it is marked as unreliable, triggering a re-excitation process. Finally, the smoothed and validated stiffness and damping parameters for each joint are stored in the controller's operating parameter table.

7. The joint-based robot arm torque compensation calibration method of claim 6, wherein: The process involves calculating environmental torque compensation based on fused parameters, superimposing it with nominal dynamics feedforward to generate a torque command, running the RLS algorithm to update parameters and recording logs. This refers to using the fused and verified stiffness and damping parameters, calculating the difference between the current joint angle and the contact equilibrium point, multiplying it by the stiffness parameter, and multiplying it by the joint velocity and damping parameter to obtain an estimated environmental torque model for the joint. The estimated value is then used as the compensation torque in the opposite direction, superimposed with the existing nominal dynamics feedforward to form the final torque command, which is then sent to the motor drive. During execution, a saturation check is performed on the applied torque, including gradually reducing the environmental compensation component according to a linear retention strategy if it approaches the motor's rated capacity, and recording the event. Simultaneously, a rolling window monitors the residual variance between the environmental torque estimate and the model output. If the residual variance exceeds three times the initial identification baseline, a re-identification process is automatically triggered. At the moment of triggering, the controller switches to a safe mode that only performs nominal dynamics compensation and alarms the host computer. During each contact, the RLS algorithm is continuously run at a low frequency to update environmental parameters, and the updated environmental parameters are smoothly replaced in the running parameter table.

8. A jointed robot arm torque compensation calibration system based on any one of claims 1 to 7, characterized in that: include, The sensor configuration and parameter initialization module is responsible for deploying the joint angle position encoder and motor current measurement module on the controller and configuring the operating parameters. The real-time state estimation module is used to sample and calculate joint angles, angular velocities, angular accelerations, and motor output torque based on encoder and current signals. The dynamic residual calculation module is used to obtain the external environment torque residual using the nominal dynamic model and the friction model, and store the residual sequence by time stamp; The contact detection module is used to determine contact by monitoring whether the environmental torque residual exceeds a threshold and continuously de-jittering for a certain period of time, and to generate a labeled dataset. The online parameter identification module is used to run a recursive least squares algorithm on the labeled dataset to identify the environmental stiffness and damping parameters of each joint. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: When the processor executes the computer program, it implements the steps of the torque compensation calibration method for the articulated robotic arm as described in any one of claims 1 to 7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: When the computer program is executed by the processor, it implements the steps of the torque compensation calibration method for the articulated robotic arm as described in any one of claims 1 to 7.

Images