Dynamic load self-adaptive adjusting method and system for four-axis picking mechanical arm and computer equipment

Abstract

The invention relates to the technical field of industrial robots, and discloses a dynamic load self-adaptive adjusting method and system for a four-axis picking mechanical arm and computer equipment. By constructing a load parameter response curved surface, load dynamic parameters of the mechanical arm and control parameters of servo systems of all joints are mapped; in this way, in the load grabbing process of the tail end of the four-axis picking mechanical arm, the current load dynamic parameters can be sensed in real time, online adjustment is conducted based on the load parameter response curved surface, and the problems of vibration, overshoot and the like caused by load changes are solved; in the moving process of the mechanical arm, the tail end vibration amplitude and trajectory tracking errors are monitored in real time, a correction instruction is triggered to correct control parameters of all joint servo systems, a load response curved surface is updated on the basis of the corrected control parameters and load data, automation of parameter setting is achieved, and the working efficiency is improved. The four-axis picking mechanical arm is more and more accurate in the load grabbing process, and the self-adaption and intelligentization capacity of the four-axis picking mechanical arm is enhanced.

Description

Technical Field

[0001] This application relates to the field of industrial robot technology, and in particular to a dynamic load adaptive adjustment method, system and computer equipment for a four-axis picking robot arm. Background Technology

[0002] Four-axis robotic arms are widely used in logistics picking due to their compact structure and low cost. However, during the picking process, the weight, size, and center of gravity of the goods grasped by the robotic arm frequently change, causing alterations in the load's dynamic characteristics. For example, increased weight leads to sluggish movement, increased tracking errors, and a need for greater drive torque; increased size results in greater rotational inertia, making it more difficult to start, stop, or change the rotational speed of the arm joints. When the arm stops moving, the load continues to move, pulling on the arm and causing it to vibrate and oscillate. When the load's center of gravity shifts, the arm needs to provide an additional, constant torque to prevent the load from falling, which can easily lead to inaccuracies in the control model and increase control difficulty. While existing technologies can use adaptive control algorithms to control robotic arms, they largely rely on precise dynamic models and parameter adjustments based on a fixed load. When the load changes, the control accuracy and stability of the robotic arm decrease, resulting in low overall picking efficiency. Summary of the Invention

[0003] This application provides a method, system, and computer device for dynamic load adaptive adjustment of a four-axis picking robot arm, which aims to solve the aforementioned technical problems.

[0004] In a first aspect, this application provides a dynamic load adaptive adjustment method for a four-axis picking robot, including: Construct the load parameter response surface of a four-axis picking robot; The dynamic parameters of the load currently grasped by the end effector of the four-axis picking robot are collected in real time, wherein the dynamic parameters of the load include at least the load mass, center of mass position and moment of inertia. The load dynamic parameters are input to the load parameter response surface to dynamically calculate and adjust the control parameters of each joint servo online, so that the motion state of the robotic arm matches the dynamic characteristics of the current load. During the movement of the four-axis picking robot arm, the vibration amplitude and trajectory tracking error of the end effector are monitored in real time. When either the vibration amplitude or the trajectory tracking error fails to meet the preset conditions, a correction command is triggered to correct the control parameters corresponding to each joint servo system. Based on the corrected control parameters and load dynamic parameters corresponding to each joint servo system, the load parameter response surface is iteratively updated to achieve adaptive adjustment of the four-axis picking robot arm.

[0005] In one embodiment, the construction of the load parameter response surface for the four-axis picking robot includes: Configure multiple standardized test loads and construct dynamic parameter information for each test load, wherein the dynamic parameter information includes initial reference values ​​for mass, three-dimensional coordinates of the center of mass, and rotational inertia tensor; Static tests are performed on multiple test loads based on the dynamic parameter information to verify and establish static parameter benchmarks for the mass and three-dimensional coordinates of the centroid of the test loads. Based on the dynamic parameter information, dynamic trajectory tests are performed on multiple test loads to obtain dynamic data; A load parameter response surface is constructed based on the static parameter benchmark and the dynamic data.

[0006] In one embodiment, constructing the load parameter response surface based on the static parameter benchmark and the dynamic data includes: Extract the IMU data from the dynamic data and perform frequency domain analysis to obtain the resonance frequency; Extract the joint data sequence that changes over time from the dynamic data, and combine it with the static parameter benchmark to solve the actual inertial parameter vector of each test load through an identification algorithm; Based on the resonant frequency and the actual inertial parameter vector, find the optimal control parameters for each test load; The static parameter benchmark and the actual inertial parameter vector of each test load are used as inputs, and the corresponding optimal control parameters are used as outputs. A regression algorithm is used to train and generate the load parameter response surface.

[0007] In one embodiment, the step of inputting the load dynamic parameters to a response surface to the load parameters to dynamically calculate and adjust the control parameters of each joint servo online includes: Based on the load parameter response surface, query the control parameter reference value corresponding to the current load, wherein the control parameter reference value includes at least the proportional gain, integral gain, feedforward gain and notch filter frequency of each joint; The real-time speed and acceleration of each joint of the robotic arm are obtained in order to adaptively adjust the reference values ​​of the control parameters; The proportional gain and feedforward gain are scaled according to the real-time speed; The frequency offset compensation of the notch filter is performed based on the acceleration. The adjusted control parameter baseline values ​​are sent as online adjustment commands to the servo systems of each joint.

[0008] In one embodiment, during the movement of the robotic arm, the end-effector vibration amplitude and trajectory tracking error are monitored in real time. When either the vibration amplitude or the trajectory tracking error fails to meet a preset condition, a correction command is triggered to correct the control parameters of each joint servo system, including: During the movement of the robotic arm, the vibration amplitude of the end effector and the trajectory tracking error are monitored in real time; When the vibration amplitude exceeds the first preset threshold and / or the trajectory tracking error exceeds the second preset threshold, the robotic arm is determined to be in an unstable state. The corresponding correction command is triggered based on the unstable state.

[0009] In one embodiment, the iterative update of the load parameter response surface based on the corrected control parameters of each joint servo system and the corresponding load data includes: Based on the corrected control parameters and load dynamic parameters corresponding to each joint servo system, a model update dataset is constructed, wherein the update dataset includes abnormal data samples and successful operation samples; Using the model update dataset as new training samples, the load parameter response surface is retrained using an incremental learning algorithm, and iterative updates are completed.

[0010] In one embodiment, the method further includes: The estimated load data of the next item to be grabbed is obtained, and the subsequent movement trajectory and control parameters to be executed of the next item to be grabbed are optimized in advance based on the estimated load data to suppress vibration caused by load switching.

[0011] In one embodiment, obtaining the estimated load data of the next item to be grabbed, and optimizing the subsequent motion trajectory and control parameters in advance based on the estimated load data, includes: Obtain the estimated load data of the next item to be grabbed, wherein the estimated load data includes at least the estimated mass and estimated size; Based on the estimated mass and estimated size, the subsequent motion trajectory of the four-axis picking robot arm from the current load placement point to the gripping point of the next item to be picked is optimized. Based on the estimated load data, the estimated control parameters corresponding to the cargo to be grabbed are queried from the load response surface, and the estimated control parameters are preset to each joint servo system.

[0012] Secondly, this application also provides a dynamic load adaptive adjustment system for a four-axis picking robot arm, comprising: multiple modules, wherein the multiple modules are used to implement the steps of the dynamic load adaptive adjustment method for the four-axis picking robot arm described in any of the above claims.

[0013] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0014] The beneficial effects of this application are as follows: By constructing a load parameter response surface, the dynamic parameters of the load of the robotic arm are mapped to the control parameters of each joint servo system. In this way, the end effector of the four-axis picking robotic arm can sense the current dynamic parameters of the load in real time and make online adjustments based on the load parameter response surface during the process of grasping the load, thereby reducing problems such as vibration and overshoot caused by load changes. During the movement of the robotic arm, by monitoring the end effector vibration amplitude and trajectory tracking error in real time and triggering correction commands to correct the control parameters of each joint servo system, and updating the load response surface based on the corrected control parameters and load data, the parameter tuning is automated, making the four-axis picking robotic arm increasingly accurate in the process of grasping the load, and enhancing its adaptive and intelligent capabilities. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0016] Figure 2 This is a flowchart illustrating a method in another embodiment of this application.

[0017] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of this application.

[0018] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0020] In one embodiment, such as Figure 1-2 As shown, this application provides a dynamic load adaptive adjustment method for a four-axis picking robot arm, which can be applied to an intelligent warehouse picking workstation. The end effector of the four-axis robot arm in this workstation can be equipped with an adaptive electric gripper and integrates components such as a joint torque sensor, a motor encoder, and a six-axis IMU at the end effector. The six-axis IMU is used for inertial measurement. This embodiment illustrates the method by applying it to a computer device. It is understood that this method can also be applied to a server, and to a system including both computer devices and servers, and can be implemented through the interaction between the computer devices and the server. In this embodiment, the method includes the following steps: S1. Construct the load parameter response surface of the four-axis picking robot to characterize the mapping relationship between the load dynamic parameters of the four-axis picking robot and the control parameters of each joint servo system. Here, the load refers to the goods or workpiece grasped by the end effector of the robotic arm (such as a suction cup or gripper) and that needs to be moved; the load dynamic parameters refer to key physical quantities that describe the inertial characteristics of the load (i.e., the grasped goods) and directly affect the dynamic performance of the robotic arm's motion, such as load mass, center of mass position, and moment of inertia. For example, in this embodiment, during the robotic arm installation and commissioning phase, loads with known masses of, for example, 0.5 kg, 1 kg, 2 kg, and 5 kg can be obtained. The known center of mass is obtained by simulating the offset of the counterweight, and the standard for testing the load is the moment of inertia corresponding to counterweights of different sizes. For example, in the standard test... During the process, the robotic arm grasps loads of different masses, centers of mass, and sizes, and executes standard pick-up and placement trajectories to construct a load parameter response surface. For example, for each load, an automated parameter tuning algorithm can be used to find a set of optimal control parameters that enable the robotic arm to move smoothly and quickly. The optimal control parameters may include the PID gain of each joint servo system, the feedforward gain of speed or acceleration, and the center frequency and depth of the notch filter. The mass, center of mass coordinates, and moment of inertia of all loads are used as inputs, and their corresponding optimal control parameters are used as outputs. A parameter mapping model, i.e., the parameter response surface, is trained in a computer device.

[0021] S2. Real-time acquisition of dynamic load parameters of the current load grasped by the end of the four-axis picking robot arm, wherein the dynamic load parameters include at least the load mass, center of mass position and moment of inertia. In this embodiment, after the gripper successfully grasps the goods, the robotic arm performs a small vertical lifting motion before starting. The joint torque of the joint output shaft can be directly read by the joint torque sensor, i.e., the actual torque value, or it can be indirectly reflected by the current value of the joint motor. Based on the kinematic principle of the robotic arm, the mass of the load can be calculated in real time. When the robotic arm starts to move at the initial low speed along the trajectory, the end effector IMU can collect the initial acceleration and angular velocity data, and combine them with the current (torque) of the joint motor. Based on the extended Kalman filter or the least squares method, the center of mass position and principal rotational inertia of the load can be analyzed online.

[0022] For example, after the gripper successfully grasps the goods, the robotic arm is controlled to perform a small vertical lifting motion. During this process, the influence of acceleration is ignored, and the robotic arm is set to be in quasi-static equilibrium. Then, according to the static equilibrium relationship, the formula for calculating the load mass is: ; Where m represents the load mass, ΔT iThis represents the actual torque difference of the i-th joint torque sensor before and after the lifting action, where i represents the main load-bearing joint. The number of main load-bearing joints can be set according to the specific structure of the robotic arm and is not uniquely limited here; θ i denoted as the joint angular displacement at the i-th joint corresponding to the lifting action; g represents the gravitational acceleration; h represents the effective distance the end effector of the robotic arm moves vertically upward in the opposite direction of gravity during a small-amplitude vertical lifting action.

[0023] For example, when the robotic arm begins its initial low-speed movement along the trajectory, the acceleration 'a' and angular velocity 'ω' data from the end effector IMU, as well as the current of each joint motor, are synchronously acquired to construct a coordinate system based on the load's center of mass [C]. x C y C z ] and principal moment of inertia tensor [I xx ,I yy ,I zz The parameter identification model; When using an extended Kalman filter, the load mass, load centroid coordinates, and load moment of inertia tensor corresponding to the current load are used as state variables x=[m,C]. x C y C z ,I xx ,I yy ,I zz ,I xy ,I xz ,I yz ] T The load mass, load centroid coordinates, and load moment of inertia tensor are all unknown parameters that need to be estimated online by the filter. The current of each joint motor and the acceleration a and angular velocity ω measured by the IMU are used as observation variables. Observation equations can be established based on the observation variables for filtering estimation. The observation equations describe how the joint torque and acceleration change with the motion state and state variables of the robotic arm, including the changes in centroid coordinates and moment of inertia.

[0024] When using the least squares method, a linear equation of the following form is constructed using the collected data sequence: Y=Φ*Θ; Where Y represents the vector consisting of residual torque (total torque minus known dynamics and gravity terms of the robotic arm body), Φ represents the regression matrix consisting of joint angles, velocities, accelerations, IMU data, etc., and Θ represents the parameter vector consisting of the load center of mass position and moment of inertia to be determined.

[0025] S3. Input the load dynamic parameters to the response surface of the load parameters to dynamically calculate and adjust the control parameters of each joint servo online so that the dynamic characteristics of the robotic arm match the motion state of the current load. In this embodiment, the collected dynamic load parameters are input into the load response surface to obtain the corresponding basic optimal control parameters. This allows the robotic arm to automatically call the optimal control parameters output by the parameter response surface during operation, thereby reducing the robotic arm's real-time dependence on on-site personnel. Furthermore, by combining the robotic arm's real-time motion state, such as instantaneous velocity, acceleration, and the curvature of the planned trajectory, the basic optimal control parameters are dynamically calculated and scaled to generate optimized control parameters that match the current motion state of the robotic arm. These control parameters are then sent to the servo systems or drivers of each joint, enabling the servo systems or drivers to operate according to the control parameters. This ensures that the robotic arm's motion state matches the dynamic characteristics of the current load, reducing vibration and trajectory deviation caused by load changes and improving the robotic arm's stability and accuracy.

[0026] S4. During the movement of the robotic arm, the end-effector vibration amplitude and trajectory tracking error are monitored in real time. When either the vibration amplitude or the trajectory tracking error is not satisfied with the preset conditions, a correction command is triggered to correct the control parameters of each joint servo system. Traditional robotic arms require professional engineers to spend a lot of time recalibrating parameters when changing loads or working conditions. As an alternative implementation method, this application allows the end-effector IMU to continuously monitor vibration amplitude and calculate trajectory tracking error throughout the entire movement of the robotic arm. If the vibration amplitude exceeds a preset amplitude or the trajectory tracking error exceeds a preset error, a correction command is triggered to correct the control parameters of each joint servo system. The control parameters corresponding to this correction command can be the safety control parameters in the safety mode, which can force the robotic arm to stabilize.

[0027] S5. Based on the corrected control parameters of each joint servo system and the corresponding load data, the load parameter response surface is iteratively updated to achieve adaptive adjustment of the four-axis picking robot arm.

[0028] In this embodiment, the corrected control parameters and corresponding load dynamic parameters of each joint servo system corresponding to the correction command are marked as abnormal data. The abnormal data includes the load dynamic parameters during the current load grabbing process, the control parameters that cause vibration, and the safety control parameters in the safe mode that stabilizes the load. The abnormal data is stored to form a historical abnormal data sample. At regular intervals, the load parameter response surface can be incrementally learned or retrained based on the new data in the abnormal data sample and the data samples of successful operation, thereby correcting the previous prediction error of the load parameter response surface. This makes the four-axis picking robot arm more and more accurate in grabbing subsequent loads, enhancing its adaptive and intelligent capabilities.

[0029] In the aforementioned dynamic load adaptive adjustment method for a four-axis picking robot, a load parameter response surface is constructed, mapping the robot's load dynamic parameters to the control parameters of each joint servo system. This allows the end effector of the four-axis picking robot to perceive the current load dynamic parameters in real time and make online adjustments based on the load parameter response surface during load grasping, reducing problems such as vibration and overshoot caused by load changes. During the robot's movement, the vibration amplitude and trajectory tracking error at the end effector are monitored in real time, triggering correction commands to adjust the control parameters of each joint servo system. The load response surface is then updated based on the corrected control parameters and load data, achieving automated parameter tuning. This makes the four-axis picking robot increasingly precise during load grasping, enhancing its adaptive and intelligent capabilities.

[0030] In some picking scenarios, considering that the load is in a moving state, dynamic testing is usually performed directly on the load. However, for the load, the change in torque is caused by multiple factors, such as gravity, friction, and noise. Therefore, in order to accurately identify which factors are causing the torque change, the dynamic testing process is quite cumbersome and complex, and usually can only identify a rough and vague torque result. Therefore, the identification accuracy of the dynamic test results in the prior art is not high. Based on this, this application first performs static testing, then dynamic testing, and finally combines static and dynamic testing to improve the identification accuracy of torque changes in dynamic testing.

[0031] In an exemplary embodiment, step S1 above includes: S11, Configure multiple standardized test loads and construct dynamic parameter information for each test load, wherein the dynamic parameter information includes initial reference values ​​for mass, three-dimensional coordinates of the center of mass, and rotational inertia tensor; For example, N standardized test loads can be prepared according to different needs. The mass of the test load is in the range of 0.5kg-5kg. The load has counterweights at different positions inside, and the counterweights at different positions correspond to different center of mass positions and moments of inertia. For each test load, its mass, three-dimensional coordinates of the center of mass (relative to the gripping end) and moment of inertia tensor are measured in advance to form dynamic parameter information.

[0032] S12, Static tests are performed on multiple test loads based on the dynamic parameter information to verify and establish static parameter benchmarks for the mass and three-dimensional coordinates of the centroid of the test loads; In this embodiment, the robotic arm can be placed in at least two different preset static postures and each test load can be grasped in sequence. In each posture, the robotic arm is kept still, and stable readings of the torque sensors of each major load-bearing joint are collected (average value is taken to filter out noise). According to the principle of static equilibrium, a torque balance equation is established to calculate the load parameters. For a load, the torque change caused by its gravity at the joint is related to the load mass and the coordinate of the center of mass. By using torque measurements under multiple (at least two) different postures, a linear equation system can be constructed to solve for these unknown parameters.

[0033] For example, the calculation process may be based on the following principles: The load mass can be initially estimated by considering the relationship between the torque change of a single joint and the lifting height h, as expressed by: Where Δτ represents the joint torque variation vector measured under different postures, g represents gravitational acceleration, h represents height, and m represents the load mass; Calculate the coordinates of the load centroid [C] x C y C z Given the load mass m, a system of equations of the following form can be established by using joint torque readings under two or more different postures: ; in, Represents the gravitational acceleration vector. This represents the transpose of the gravity Jacobian matrix of the robotic arm in the corresponding posture; Solving this system of equations will yield the coordinates of the load's centroid.

[0034] The actual mass and actual centroid position obtained above are cross-validated with the initial reference values ​​in the dynamic parameter information pre-built for the test load. If the error is within an acceptable range, it is used as the static parameter benchmark for the test load; if the error exceeds the limit, the test load or sensor can be calibrated until an accurate static parameter benchmark is obtained.

[0035] S13, Based on the dynamic parameter information, perform dynamic trajectory testing on multiple test loads to obtain dynamic data; The dynamic trajectory test includes linear motion, S-curve motion, and emergency stop motion. In this embodiment, during the entire motion process, the position and speed of each joint encoder, joint torque, motor current value, raw data of the end six-axis IMU, etc. are collected at high speed and synchronously. All the collected data are used as dynamic data. By collecting dynamic data, it is easy to identify the rotational inertia of the test load in the future.

[0036] S14, construct the load parameter response surface based on the static parameter benchmark and the dynamic data.

[0037] In this embodiment, the dynamic data can be analyzed in the frequency domain to obtain the resonance frequency. For example, the raw data collected by the IMU can be extracted. The raw data can be acceleration and angular velocity. Based on the raw data, a fast Fourier transform can be performed to identify the resonance frequency of the robotic arm under the test load. Based on the joint torque, motor current value, position and speed of each joint encoder, and motion data of the robotic arm, combined with static parameter benchmarks, the actual moment of inertia of the load can be identified. Specifically, according to existing technology, during the movement of the robotic arm, the total torque output by any joint motor is the sum of torques used to overcome various forces of different natures. Among them, the torques related to the load mainly include gravitational torque and moment of inertia. Gravitational torque is related to the load mass and center of mass position, and is a static torque related to the robotic arm posture but independent of the motion state. Moment of inertia is related to the load mass, center of mass position, and moment of inertia system, and is proportional to the joint acceleration. Let the load mass be denoted as m, the center of mass position as COM, and the moment of inertia as I. If traditional dynamic testing is used, the total torque of the same joint may be composed of multiple different combinations of {m, COM, I}, which is difficult to effectively identify. Therefore, in this embodiment, the load mass m and the center of mass COM are determined first through static testing. In dynamic testing, the identification of the moment of inertia can be directly focused on, which simplifies the complex dynamic testing, improves the accuracy of identifying changes in dynamic test torque, and reduces the complexity of identification.

[0038] For example, according to the modeling theory of robotic arm dynamics, its core equation is usually expressed as: ; Where τ represents the joint torque vector directly measured from all sampling points in the time series, and q represents the joint position vector collected from all sampling points in the time series. For example, for a four-axis robotic arm, at t k At any moment, there is Then q includes all sampling times t1, t2, ..., tN The set of position data; M(q) represents the inertia matrix vector (containing the inertia of the robotic arm itself and the load). This represents the joint acceleration vector of all sampling points in the time series. Represents the matrix of Coriolis force and centripetal force. Let G(q) represent the joint velocity vector at all sampling points in the time series, and let G(q) represent the gravity term, including the gravity of the robotic arm itself and the load. τ friction Let represent the friction torque vector; since the load has a significant impact on inertia and gravity, the load-related parts of the equation can be separated, resulting in: ; Among them, M arm(q) and G arm(q) These are the inertia and gravity of the robotic arm, respectively, and are known quantities, M. load(q) and G load(q) The inertia and gravity corresponding to the test load are the factors to be solved.

[0039] In this embodiment, firstly, a synchronized, time-varying joint data sequence is extracted from the dynamic trajectory test in S13, including the joint position vector q obtained by the encoder, and the joint velocity vector obtained by differentiating q. and joint acceleration vector And the joint torque vector τ, where τ can be directly measured by the torque sensor installed at the joint output end. Since the output torque of the servo motor is proportional to the winding current, it can also be calculated based on the motor current; and the friction torque vector is obtained based on the pre-calibrated friction force model. Based on the above Substituting the known terms into the formula, the above formula can be transformed into: ; The known quantity that can be calculated on the right can be denoted as the residual torque vector τ. residual This is generated by the inertia and gravity of the robotic arm; Due to the gravitational torque G load(q) It depends only on the mass and center of mass of the load, which have been obtained through static testing. Therefore, for any given q, the moment of inertia M can be calculated. load(q) ,Right now ; Given the moment of inertia M of the load load(q) It is a linear function of its moment of inertia tensor I, therefore a linear equation can be constructed: ; Where, τ inertial The vector of inertial torque is equal to , Representing joint angle q and joint acceleration The regression matrix is ​​formed, where I' represents the actual inertial parameter vector to be solved; Then a system of linear equations representing the time series changes throughout the entire motion process can be constructed. Based on this system of linear equations, the actual inertial parameter vector I' can be solved using standard numerical methods such as the least squares method.

[0040] Based on the identified resonant frequencies and actual inertial parameter vectors, and with the combined objectives of minimizing trajectory tracking error and end-effector vibration amplitude, an optimization algorithm is used to find a set of optimal control parameters for each test load. These optimal control parameters may include the proportional gain, integral gain, feedforward gain, and notch filter frequency for each joint. Since the mass and centroid are precisely known, the optimization algorithm can more efficiently focus on tuning the parameters most relevant to the dynamic response, such as the velocity / acceleration feedforward gain, PID differential gain, and notch filter parameters for the identified resonant frequencies. The optimization algorithm includes, but is not limited to, sequential quadratic programming, genetic algorithms, particle swarm optimization, and online tuning methods; no single method is specified here.

[0041] The data from all the test loads mentioned above are aggregated to form a training sample set. The inputs are the static parameter benchmarks and the actual inertial parameter vectors for each test load, and the output is a set of optimal control parameters. The Gaussian process regression algorithm is used to train this dataset, which ultimately generates a parameter mapping model, i.e., a parameter response surface. More preferably, the load response surface generated by the Gaussian process regression algorithm can provide predicted control parameters and variance corresponding to uncertainty measures for any load dynamic parameter input.

[0042] While the load parameter surface can output the optimal control parameters for the load dynamic parameters, these optimal control parameters are based on static load characteristics. However, the robotic arm is easily affected by instantaneous motion states during actual movement. For example, different gains may be needed to maintain stability during high-speed movement, and the resonant frequency of the robotic arm may shift during high acceleration. Therefore, in one embodiment, this application performs secondary dynamic optimization of the optimal control parameters based on the robotic arm's motion state. Specifically, S3 includes: S31. Based on the load parameter response surface, query the control parameter reference value corresponding to the current load, wherein the control parameter reference value includes at least the proportional gain, integral gain, feedforward gain and notch filter frequency of each joint. S32. Obtain the real-time speed and acceleration of each joint of the robotic arm to adapt and adjust the reference value of the control parameters accordingly; S33. Scale the proportional gain and feedforward gain according to the real-time speed; In this embodiment, the reference value of the control parameter is the optimal control parameter output based on the load parameter response surface in step S14 above. A speed scaling factor can be preset for each joint, and the proportional gain and feedforward gain are scaled according to the speed scaling factor. When the real-time speed of the joint is high, the proportional gain is appropriately reduced to enhance stability. At the same time, the feedforward gain is dynamically adjusted according to the real-time speed to better compensate for friction and inertial forces.

[0043] S34. Perform frequency offset compensation on the notch filter according to the acceleration; For example, the frequency offset of each joint is calculated. The frequency offset is generally the product of a preset offset coefficient and the absolute value of angular acceleration. High acceleration will cause slight deformation of the mechanical structure, thereby causing the system's resonant frequency to shift. To prevent the notch filter from becoming inaccurate, its center frequency can be dynamically compensated.

[0044] S35. The adjusted control parameter reference value is sent to each joint servo system as an online adjustment command.

[0045] In this embodiment, the four-axis picking robot arm can make dual adaptive adjustments according to the load and its own motion state. For example, in the high-speed segment, reducing the proportional gain can effectively suppress overshoot; in the rapid acceleration segment, the dynamic compensation of the notch filter frequency can ensure its accurate suppression of the resonant frequency after drift. This enables the four-axis picking robot arm to exhibit high stability and trajectory tracking accuracy when handling different loads and executing complex trajectories, reducing vibration problems caused by changes in working conditions.

[0046] In one embodiment, step S5, which involves real-time monitoring of end-effector vibration amplitude and trajectory tracking error during robotic arm movement, and triggering a correction command to correct the control parameters of each joint servo system when either the vibration amplitude or the trajectory tracking error fails to meet a preset condition, includes: S51. During the movement of the robotic arm, monitor the end-effector vibration amplitude and trajectory tracking error in real time; For example, the end-effector vibration amplitude and trajectory tracking error can be calculated in real time using joint encoder and IMU data.

[0047] S52. When the vibration amplitude exceeds the first preset threshold and / or the trajectory tracking error exceeds the second preset threshold, it is determined that the robotic arm is in an unstable state. S53. Trigger the corresponding correction command according to the unstable state.

[0048] For example, the instability state is divided into mild instability and severe instability. In the case of mild instability, a first correction instruction is generated. The first correction instruction is used to reduce the speed feedforward gain and proportional gain of each joint servo system according to the first preset information. If it is determined to be severe instability, a second correction instruction is generated. The second correction instruction is used to switch to the preset safety mode control parameters on the basis of reducing the speed feedforward gain and proportional gain, and force the robot arm to decelerate along the current trajectory or execute an emergency stop procedure.

[0049] In one embodiment, step S6, which iteratively updates the load parameter response surface based on the corrected control parameters of each joint servo system and the corresponding load data, includes: S61. Based on the corrected control parameters and load dynamic parameters corresponding to each joint servo system, construct a model update dataset, wherein the update dataset includes abnormal data samples and successful operation samples; For example, abnormal data samples can be the working condition data corresponding to the above-mentioned unstable state and the safety parameters that stabilize it; successful operation samples are data pairs consisting of the load dynamic parameters and the control parameters actually applied by the robotic arm during a stable operation cycle without triggering correction instructions.

[0050] S62. Using the model update dataset as a new training sample, the regression model for constructing the load parameter response surface is retrained using an incremental learning algorithm, and iterative updates are completed.

[0051] In this embodiment, during the retraining process, the abnormal data samples can be assigned a higher weight than the successful running samples, and a forgetting mechanism can be introduced for historical training data.

[0052] Additionally, as an optional implementation, to mitigate and prevent potential instability risks, for example, load condition risk characteristics of robotic arm instability can be extracted based on abnormal data samples. These load condition risk characteristics can be represented by feature vectors composed of load dynamic parameters. These feature vectors can be marked as high-risk points, and the spatial distribution of all high-risk points can be obtained based on the parametric load surface. At least one high-risk condition region can be generated using a clustering algorithm, thereby constructing a load condition risk map. Before the robotic arm performs a task, the estimated load dynamic parameters can be calculated based on the estimated mass and size of the goods to be grasped. These estimated load dynamic parameters are then mapped to the load condition map to determine their spatial location. If a location is found, it is determined that the current task carries an instability risk. At this point, avoidance commands can be automatically triggered, including trajectory avoidance commands, control parameter avoidance commands, and operation avoidance commands. This approach can identify the common characteristics of a certain type of instability risk, provide risk warnings, proactively avoid risks, or enable human-machine collaboration, thereby improving the adaptability of its adjustments.

[0053] In one embodiment, the method further includes: S6, obtaining estimated load data of the next cargo to be grabbed, and optimizing the subsequent motion trajectory and control parameters to be executed in advance based on the estimated load data, so as to suppress the vibration of the robotic arm caused by load switching.

[0054] In this embodiment, when the robotic arm performs the action of grabbing the next item to be grabbed, it can move according to the optimized subsequent motion trajectory and use preset control parameters to coordinately optimize the motion trajectory and control parameters within the same cycle, thereby avoiding vibration and improving the efficiency and stability of the picking operation.

[0055] Meanwhile, to prevent the robotic arm from vibrating during cargo switching and reducing its stability, in one embodiment, step S6 includes: S61. Obtain the estimated load data of the next item to be grabbed, wherein the estimated load data includes at least the estimated mass and estimated size; S62. Based on the estimated mass and estimated size, optimize the subsequent motion trajectory of the four-axis picking robot arm from the current load placement point to the gripping point of the next item to be picked. For example, the optimization process includes smoothing the acceleration of the trajectory and limiting the maximum acceleration of the subsequent motion trajectory.

[0056] S63. Based on the estimated load data, query the estimated control parameters corresponding to the cargo to be grabbed from the load response surface, and preset the estimated control parameters to each joint servo system. For example, to achieve stability, a common technique is to slow down the picking speed of the robotic arm, which results in low picking efficiency. Therefore, in this embodiment, after the current load is picked, the estimated load data for the next item to be picked can be obtained from a pre-entered database containing information on the quality, size, and packaging materials of all goods. This estimated load data includes not only the corresponding quality, size, and packaging materials, but also dynamic parameters of the estimated load. Based on this estimated load data, the upcoming movement trajectory is optimized. Specifically, the accelerometer can be controlled, i.e., the speed of acceleration can be controlled to ensure that the acceleration is sufficient for the desired result. Smooth transitions reduce excitation forces. When optimizing control parameters, the servo filter parameters of each joint servo system need to be preset. The control parameters and servo filter parameters refer to the module parameters in the servo driver used to filter out signals of specific frequencies. For example, this module can be a notch filter. Based on the estimated rotational inertia of the load, the frequency of the notch filter can be preset to suppress the vibration of the robotic arm caused by load switching and filter out the resonant frequencies that may be caused by load inertia. In this way, without causing vibration, the start-up and stop time is shortened, so that the robotic arm can maintain a smooth and precise movement posture when grasping different goods, thereby improving the overall picking efficiency.

[0057] More preferably, in step S61 above, the estimated load data may also include estimated packaging material. The operating parameters of the end effector when the next cargo to be grasped is contacted by the robotic arm can be determined based on the estimated packaging material. For example, the operating parameters include the suction force level for vacuum adsorption and / or the clamping force level for gripping. This allows the end effector to suppress vibration of the body while protecting the safety of the cargo itself. It is suitable for handling fragile items, such as glass bottles, electronic products, soft-packaged products, such as puffed food, or goods with easily damaged surfaces.

[0058] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0059] Based on the same inventive concept, this application also provides a system for implementing the aforementioned dynamic load adaptive adjustment system for a four-axis picking robot. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of the one or more embodiments of the dynamic load adaptive adjustment system for a four-axis picking robot can be found in the limitations of the dynamic load adaptive adjustment method for a four-axis picking robot described above, and will not be repeated here.

[0060] The modules in the aforementioned four-axis picking robot's dynamic load adaptive adjustment system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0061] like Figure 3 As shown, this application also provides a computer device, which can be a server, and its internal structure can be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores all data required for the dynamic load adaptive adjustment method of the four-axis picking robot. The network interface communicates with external terminals via a network connection. When the processor executes the computer program, it implements the dynamic load adaptive adjustment method of the four-axis picking robot.

[0062] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0063] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described four-axis picking robot dynamic load adaptive adjustment methods.

[0064] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0065] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0066] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A dynamic load adaptive adjustment method for a four-axis picking robot arm, characterized by, include: Construct the load parameter response surface of a four-axis picking robot; The dynamic parameters of the load currently grasped by the end effector of the four-axis picking robot are collected in real time, wherein the dynamic parameters of the load include at least the load mass, center of mass position and moment of inertia. The load dynamic parameters are input to the load parameter response surface to dynamically calculate and adjust the control parameters of each joint servo online, so that the motion state of the robotic arm matches the dynamic characteristics of the current load. During the movement of the four-axis picking robot arm, the vibration amplitude and trajectory tracking error of the end effector are monitored in real time. When either the vibration amplitude or the trajectory tracking error fails to meet the preset conditions, a correction command is triggered to correct the control parameters corresponding to each joint servo system. Based on the corrected control parameters and load dynamic parameters corresponding to each joint servo system, the load parameter response surface is iteratively updated to achieve adaptive adjustment of the four-axis picking robot arm.

2. The dynamic load adaptive adjustment method of four-axis picking robot arm according to claim 1, characterized in that, The load parameter response surface for constructing the four-axis picking robot includes: Configure multiple standardized test loads and construct dynamic parameter information for each test load, wherein the dynamic parameter information includes initial reference values ​​for mass, three-dimensional coordinates of the center of mass, and rotational inertia tensor; Static tests are performed on multiple test loads based on the dynamic parameter information to verify and establish static parameter benchmarks for the mass and three-dimensional coordinates of the centroid of the test loads. Based on the dynamic parameter information, dynamic trajectory tests are performed on multiple test loads to obtain dynamic data; A load parameter response surface is constructed based on the static parameter benchmark and the dynamic data.

3. The dynamic load adaptive adjustment method of four-axis picking robot arm according to claim 2, characterized in that, The construction of the load parameter response surface based on the static parameter benchmark and the dynamic data includes: Extract the IMU data from the dynamic data and perform frequency domain analysis to obtain the resonance frequency; Extract the joint data sequence that changes over time from the dynamic data, and combine it with the static parameter benchmark to solve the actual inertial parameter vector of each test load through an identification algorithm; Based on the resonant frequency and the actual inertial parameter vector, find the optimal control parameters for each test load; The static parameter benchmark and the actual inertial parameter vector of each test load are used as inputs, and the corresponding optimal control parameters are used as outputs. A regression algorithm is used to train and generate the load parameter response surface.

4. The dynamic load adaptive adjustment method of four-axis picking robot arm according to claim 1, characterized in that, The step of inputting the load dynamic parameters to the response surface of the load parameters to dynamically calculate and adjust the control parameters of each joint servo online includes: Based on the load parameter response surface, query the control parameter reference value corresponding to the current load, wherein the control parameter reference value includes at least the proportional gain, integral gain, feedforward gain and notch filter frequency of each joint; The real-time speed and acceleration of each joint of the robotic arm are obtained in order to adaptively adjust the reference values ​​of the control parameters; The proportional gain and feedforward gain are scaled according to the real-time speed; The frequency offset compensation of the notch filter is performed based on the acceleration. The adjusted control parameter baseline values ​​are sent as online adjustment commands to the servo systems of each joint.

5. The dynamic load adaptive adjustment method of four-axis picking robot arm according to claim 1, characterized in that, During the movement of the four-axis picking robot arm, the end effector vibration amplitude and trajectory tracking error are monitored in real time. When either the vibration amplitude or the trajectory tracking error fails to meet a preset condition, a correction command is triggered to correct the control parameters of each joint servo system, including: During the movement of the robotic arm, the vibration amplitude of the end effector and the trajectory tracking error are monitored in real time; When the vibration amplitude exceeds the first preset threshold and / or the trajectory tracking error exceeds the second preset threshold, the robotic arm is determined to be in an unstable state. The corresponding correction command is triggered based on the unstable state.

6. The dynamic load adaptive adjustment method of four-axis picking robot arm according to claim 1, characterized in that, The iterative update of the load parameter response surface based on the corrected control parameters of each joint servo system and the corresponding load data includes: Based on the corrected control parameters and load dynamic parameters corresponding to each joint servo system, a model update dataset is constructed, wherein the update dataset includes abnormal data samples and successful operation samples; Using the model update dataset as new training samples, the load parameter response surface is retrained using an incremental learning algorithm, and iterative updates are completed.

7. The dynamic load adaptive adjustment method for a four-axis picking robot arm according to any one of claims 1-5, characterized in that, Also includes: The estimated load data of the next item to be grabbed is obtained, and the subsequent movement trajectory and control parameters to be executed of the next item to be grabbed are optimized in advance based on the estimated load data to suppress vibration caused by load switching.

8. The dynamic load adaptive adjustment method for a four-axis picking robot arm according to claim 7, characterized in that, The step of obtaining the estimated load data of the next item to be grabbed, and optimizing the subsequent motion trajectory and control parameters in advance based on the estimated load data, includes: Obtain the estimated load data of the next item to be grabbed, wherein the estimated load data includes at least the estimated mass and estimated size; Based on the estimated mass and estimated size, the subsequent motion trajectory of the four-axis picking robot arm from the current load placement point to the gripping point of the next item to be picked is optimized. Based on the estimated load data, the estimated control parameters corresponding to the cargo to be grabbed are queried from the load response surface, and the estimated control parameters are preset to each joint servo system.

9. A dynamic load adaptive adjustment system for a four-axis picking robot arm, characterized in that, include: Multiple modules are used to implement the steps of the dynamic load adaptive adjustment method for the four-axis picking robot arm as described in any one of claims 1-6.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

Images