State correction method and device of quadruped robot, electronic equipment and storage medium

Abstract

The application discloses a state correction method and device of a quadruped robot, electronic equipment and a storage medium, and is used for solving the problem that the existing quadruped robot is relatively more dependent on external feedback signals than the feedback signals received by itself. The method comprises the following steps: in response to a starting operation of the quadruped robot, initializing an original time sequence of the quadruped robot and generating a target time sequence; obtaining a historical correction data set, and performing data preprocessing on the target time sequence by using the historical correction data set to generate a working condition data set; obtaining redundant mechanical foot parameters of each mechanical foot, and performing clustering calculation on each redundant mechanical foot parameter by using the working condition data set to obtain a clustering center point data set; establishing an actual working condition trajectory of the quadruped robot, performing redundant decomposition on each mechanical foot by using the clustering center point data set to obtain a redundant optimal solution of each mechanical foot, and performing state correction on the quadruped robot according to the redundant optimal solution.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot state correction, and in particular to a state correction method and device for a quadruped robot, an electronic device and a storage medium. BACKGROUND

[0002] MPC (Model Predictive Control) is a special control in the control field, and its working principle is that the current control action is obtained by solving a finite time domain open-loop optimal control problem at each sampling instant. The current state of the MPC control process is used as the initial state of the optimal control problem, and the optimal control sequence obtained is only implemented for the first control action. With the rapid development of artificial intelligence, robots have gradually become a research hotspot in the relevant field, and various research institutions and enterprises have developed and manufactured robots of various shapes. Taking the current popular quadruped robot as an example, the quadruped robot belongs to a floating base robot, that is, a robot with a freely movable base. In other words, the quadruped robot can be regarded as a single rigid body dynamics model in the world coordinate system.

[0003] At present, for the motion control of the quadruped robot, the controller is mainly used to control the quadruped robot to perform tasks such as stepping, turning, climbing and obstacle crossing. Specifically, an MPC controller is mainly used to solve the expected value of the foot reaction force of the quadruped robot to realize the prediction and control of the next motion of the quadruped robot. In the process of motion control of the quadruped robot, in order to keep the motion state at a normal and relatively accurate level, it is usually necessary to correct the state of the mechanical foot of the quadruped robot according to the actual road conditions or the motion state of the quadruped robot. In the current related technology, the expected value of the foot reaction force of the quadruped robot is mainly corrected based on the control task sequence and the control task sequence according to the redundancy characteristics to ensure the smooth running of the quadruped robot. Although this method can control the quadruped robot smoothly to a certain extent, when the robot itself uses external sensors for active obstacle avoidance, the feedback error information cannot be received by the control system in time. In particular, when the robot sensor cannot transmit the feedback signal to the control end, the current state correction method cannot continue to work, that is, the existing state correction method relies relatively more on external feedback signals than on the feedback signals received by itself. SUMMARY

[0004] The present application provides a state correction method and device for a quadruped robot, an electronic device and a storage medium, which solves or partially solves the technical problem that the existing state correction method for a quadruped robot relies relatively more on external feedback signals than on the feedback signals received by itself.

[0005] This invention provides a state correction method for a quadruped robot, the method comprising:

[0006] In response to a startup operation on the quadruped robot, the original time series of the quadruped robot is initialized and a target time series is generated;

[0007] Obtain the historical corrected dataset of the quadruped robot, and use the historical corrected dataset to preprocess the target time series to generate the working condition dataset of the quadruped robot;

[0008] Obtain the redundant mechanical foot parameters of each mechanical foot in the quadruped robot, and use the working condition dataset to perform clustering calculation on each of the redundant mechanical foot parameters to obtain a cluster center point dataset;

[0009] The actual working trajectory of the quadruped robot is established, and the redundancy decomposition of each mechanical leg is performed using the cluster center point dataset to obtain the redundancy optimal solution of each mechanical leg. The state of the quadruped robot is then corrected based on the redundancy optimal solution.

[0010] Optionally, the step of using the historical correction dataset to preprocess the target time series to generate the working condition dataset of the quadruped robot includes:

[0011] The target time series is preprocessed using the historical corrected dataset to obtain the corresponding preprocessed corrected dataset.

[0012] The preprocessed and corrected dataset is integrated, divided, and classified to obtain the working condition process dataset corresponding to the working status, balance control status, and trajectory planning of the quadruped robot. Then, Shannon sampling is performed on the working condition process dataset to obtain the working condition dataset of the quadruped robot.

[0013] Optionally, establishing the actual working trajectory of the quadruped robot, and using the cluster center point dataset to perform redundancy decomposition on each of the mechanical legs to obtain the optimal redundancy solution for each of the mechanical legs, includes:

[0014] Establish the actual working trajectory of the quadruped robot and obtain trajectory planning instructions for the quadruped robot.

[0015] Extract the joint angles and joint angular velocities of the redundant robotic arms in the robotic foot from the cluster center point dataset;

[0016] The Jacobian matrix of the mechanical foot is calculated based on the trajectory planning instructions, the joint angles, and the joint angular velocities. The objective function corresponding to the Jacobian matrix is ​​also calculated, and the redundancy decomposition output of the mechanical foot is generated.

[0017] The redundancy decomposition output is subjected to singularity judgment, and the optimal redundancy solution of the mechanical foot is determined based on the singularity judgment result.

[0018] Optionally, the step of calculating the Jacobian matrix of the mechanical foot based on the trajectory planning instruction, the joint angle, and the joint angular velocity, and calculating the objective function corresponding to the Jacobian matrix to generate the redundancy decomposition output of the mechanical foot includes:

[0019] The trajectory planning instructions are redundantly decomposed to obtain the operation task instructions and extended task instructions of the mechanical foot. The operation task instructions correspond to the operation tasks, and the extended task instructions correspond to the extended tasks.

[0020] The Jacobian matrix of the mechanical foot is calculated based on the operation task, the expansion task, the joint angle, and the joint angular velocity, using the following formula:

[0021]

[0022] The objective function corresponding to the Jacobian matrix is ​​calculated using the following formula:

[0023]

[0024] Based on the Jacobian matrix and the objective function, the redundancy decomposition output of the mechanical foot is generated, and the calculation formula is as follows:

[0025]

[0026] Where q represents the joint angle of the redundant robotic arm. Let Xe be the joint angular velocity, Xc be the operational task, f(*) and g(*) be kinematic functions, X represent the trajectory planning task before redundancy decomposition, Je(q) be the input Jacobian matrix of the operational task Xe, Jc(q) be the input Jacobian matrix of the extended task Xc, and J represent the Jacobian matrix of the trajectory planning task X. This indicates that the Jacobian matrix J is calculated for the trajectory planning task. We represent the self-motion corresponding to operation task Xe, and We represent the weight coefficient corresponding to operation task Xe. Wc represents the self-motion corresponding to the extended task Xc, Wc represents the weight coefficient corresponding to the extended task Xc, and Wv represents the joint angular velocity. The corresponding weighting coefficients are given, where L is the objective control function of trajectory planning task X, Je is the output Jacobian matrix corresponding to operation task Xe, and Jc is the output Jacobian matrix corresponding to augmented task Xc. This is the inverse solution corresponding to the operation task Xe obtained after calculating the Jacobian matrix. This is the inverse solution corresponding to the extended task Xc obtained after calculating the Jacobian matrix. This indicates the redundant output of the mechanical foot. T This indicates the matrix transpose.

[0027] Optionally, the method further includes:

[0028] A corrected state model of the quadruped robot is established based on the objective function;

[0029] The joint angle optimization calculation of the mechanical foot of the quadruped robot is performed based on the modified state model to obtain the joint angle optimization index of the mechanical foot.

[0030] The accuracy of the movement angle of the mechanical foot is optimized based on the joint angle optimization index;

[0031] The joint angular velocity of the mechanical foot is:

[0032]

[0033] The joint angle redundancy of the mechanical foot is calculated using the following formula:

[0034]

[0035] in, For the joint angle redundancy of the mechanical foot, Q is a fixed constant, and I is the joint angle redundancy of J. T The identity matrix corresponding to J, Let S be the gradient operator and S be the optimization index.

[0036] The joint angle optimization index of the mechanical foot is calculated using the following formula:

[0037]

[0038] Where S(w) is the joint angle optimization index, i is the joint dimension of the mechanical foot, and w i Let i be the joint angle redundancy corresponding to the joint dimension i. Let i be the minimum joint angle redundancy corresponding to joint dimension i. This represents the maximum joint angle redundancy corresponding to joint dimension i.

[0039] Optionally, the step of performing singularity judgment on the redundancy decomposition output and determining the optimal redundancy solution of the mechanical foot based on the singularity judgment result includes:

[0040] Perform singular value decomposition on the redundancy decomposition output to obtain the corresponding redundancy decomposition singular values;

[0041] The joint limits of the mechanical foot are set. Combining the joint limits and the singular values ​​of the redundancy decomposition, the redundant mechanical arm joint limit and obstacle avoidance planning of the mechanical foot are completed by selecting parameters in the damped minimum square method, generating the optimal solution task, and determining the redundant optimal solution corresponding to the optimal solution task.

[0042] Optionally, the quadruped robot has a built-in microcontroller, which includes at least a model prediction control module and a state correction memory. The model prediction control module performs MPC correction operations on the quadruped robot, and the state correction memory collects correction data corresponding to the MPC correction operation when performing the MPC correction operation on the quadruped robot, and stores multiple sets of correction data as the quadruped robot's historical correction dataset. The method further includes:

[0043] If the quadruped robot performs a repetitive task and the model prediction control module cannot receive the task feedback signal from the quadruped robot, the historical correction dataset is extracted from the state correction memory and input to the feedback terminal of the model prediction control module so that the quadruped robot can complete the retrieval task.

[0044] The present invention also provides a state correction device for a quadruped robot, comprising:

[0045] A target time series generation module is used to initialize the original time series of the quadruped robot and generate a target time series in response to a start operation for the quadruped robot.

[0046] The working condition dataset generation module is used to obtain the historical corrected dataset of the quadruped robot, and use the historical corrected dataset to perform data preprocessing on the target time series to generate the working condition dataset of the quadruped robot.

[0047] The cluster center point dataset generation module is used to obtain the redundant mechanical foot parameters of each mechanical foot in the quadruped robot, and to perform cluster calculation on each of the redundant mechanical foot parameters using the working condition dataset to obtain the cluster center point dataset.

[0048] The mechanical leg redundancy decomposition module is used to establish the actual working trajectory of the quadruped robot, perform redundancy decomposition on each mechanical leg using the cluster center point dataset, obtain the redundancy optimal solution for each mechanical leg, and perform state correction on the quadruped robot based on each redundancy optimal solution.

[0049] The present invention also provides an electronic device, the device comprising a processor and a memory:

[0050] The memory is used to store program code and transmit the program code to the processor;

[0051] The processor is used to execute the state correction method for the quadruped robot as described above, according to the instructions in the program code.

[0052] The present invention also provides a computer-readable storage medium for storing program code for performing the state correction method for a quadruped robot as described in any of the preceding claims.

[0053] As can be seen from the above technical solutions, the present invention has the following advantages: In the process of motion control of a quadruped robot, in response to the start operation of the quadruped robot, the original time series of the quadruped robot is initialized and a target time series is generated. This new time series ensures that the four mechanical legs of the quadruped robot have a common time reference value, thereby improving the accuracy of motion control and the time synchronization of signal feedback, and reducing delay errors. Next, the historical correction dataset of the quadruped robot is acquired, and the target time series is preprocessed using the historical correction dataset to generate the working condition dataset of the quadruped robot. This preprocessing of the target time series based on the historical correction dataset corresponding to the quadruped robot's own historical correction operations reduces the dependence on external feedback signals and improves the prediction accuracy of motion control. Then, the individual mechanical legs of the quadruped robot are acquired... Redundant mechanical foot parameters are identified, and clustering calculations are performed on each redundant mechanical foot parameter using a working condition dataset to obtain a cluster center point dataset. This clustering calculation automatically groups similar samples in the working condition dataset into a single category. Furthermore, the combination of sampling and clustering minimizes data consumption, improves data processing efficiency and clustering accuracy, and reduces feedback errors during motion control. Next, the actual working trajectory of the quadruped robot is established, and redundancy decomposition is performed on each mechanical foot using the cluster center point dataset to obtain the optimal redundancy solution for each mechanical foot. Based on these optimal redundancy solutions, the quadruped robot's state is corrected. This allows for the rapid acquisition of the optimal redundancy solution corresponding to the task through redundancy decomposition calculations, and timely and accurate state correction of the quadruped robot based on these optimal redundancy solutions. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0055] Figure 1 A flowchart illustrating the steps of a state correction method for a quadruped robot provided in an embodiment of the present invention;

[0056] Figure 2 A schematic diagram of a control framework for state correction of a quadruped robot provided in an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram of the state change of a quadruped robot for state correction, provided in an embodiment of the present invention.

[0058] Figure 4 This is a structural block diagram of a state correction device for a quadruped robot provided in an embodiment of the present invention. Detailed Implementation

[0059] This invention provides a method, apparatus, electronic device, and storage medium for state correction of a quadruped robot, which solves or partially solves the technical problem in existing quadruped robot state correction methods that rely more on external feedback signals than on the feedback signals received by the robot itself.

[0060] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0061] As an example, current motion control of MPC-based quadruped robots primarily involves using a controller to control the robot to perform tasks such as stepping, turning, climbing, and obstacle crossing. Specifically, the MPC controller is used to calculate the expected value of the robot's foot reaction force to predict and control its next movement. During the motion control process, to maintain a normal and relatively accurate motion state, it is usually necessary to perform state corrections on the robot's mechanical feet based on actual road conditions or its own motion state.

[0062] In related technologies, the current main approach is to sort control tasks according to redundancy characteristics and correct the expected value of the foot reaction force of the quadruped robot based on the control task sequence to ensure the smooth operation of the quadruped robot. Although this approach can achieve a certain degree of smooth control of the quadruped robot, when using the robot's external sensors for active obstacle avoidance, the feedback error information cannot be received by the control system in a timely manner. In particular, when the robot's sensors cannot transmit feedback signals to the control end, the current state correction method cannot continue to work. That is, the existing state correction method is more dependent on external feedback signals than on the feedback signals it receives itself.

[0063] Therefore, one of the core inventive points of this invention is as follows: For the motion control process of a quadruped robot, in response to the start operation of the quadruped robot, the original time series of the quadruped robot can be initialized and a target time series can be generated. By generating a new time series, it can be ensured that the four mechanical legs of the quadruped robot have a common time reference value, thereby improving the accuracy of motion control and the time synchronization of signal feedback, and reducing delay errors. Next, the historical correction dataset of the quadruped robot can be obtained, and the target time series can be preprocessed using the historical correction dataset to generate the working condition dataset of the quadruped robot. Thus, by preprocessing the target time series based on the historical correction dataset corresponding to the quadruped robot's own historical correction operations, the dependence on external feedback signals is reduced, and the prediction accuracy of motion control is improved. Then, the redundancy of each mechanical leg in the quadruped robot can be obtained. The quadruped robot's redundant mechanical leg parameters are clustered using a working condition dataset to obtain a cluster center point dataset. This clustering automatically groups similar samples in the working condition dataset into a single category. The combination of sampling and clustering minimizes data consumption, improving data processing efficiency and clustering accuracy, thus reducing feedback errors during motion control. Next, the actual working trajectory of the quadruped robot is established using sensors. Redundancy decomposition of each mechanical leg is performed using the cluster center point dataset to obtain the optimal redundancy solution for each leg. The quadruped robot's state is then corrected based on these optimal solutions. This allows for rapid and accurate state correction of the quadruped robot by quickly obtaining the optimal redundant solution for the task. Furthermore, during active obstacle avoidance using external sensors, even if the external sensors malfunction or fail to transmit feedback signals, simple motion control can be performed based on relevant data from MPC historical corrections when the quadruped robot is performing repetitive tasks, preventing task interruption.

[0064] Reference Figure 1The diagram illustrates a flowchart of a state correction method for a quadruped robot provided by an embodiment of the present invention, which specifically includes the following steps:

[0065] Step 101: In response to the start operation for the quadruped robot, initialize the original time series of the quadruped robot and generate the target time series;

[0066] To achieve motion control of a quadruped robot, a microcontroller is typically installed within it. This microcontroller can house a Model Predictive Control (MPC) module for predictive control. Additionally, a state correction memory can be included within the microcontroller to store relevant data for state correction operations. In a specific implementation, the quadruped robot incorporates a microcontroller containing at least a MPC module and a state correction memory. The MPC module performs MPC correction operations on the quadruped robot, while the state correction memory collects correction data corresponding to the MPC operations, such as joint angles, angular velocities, and redundancy of the redundant robotic arm before and after correction—all processing data related to state correction. Multiple sets of correction data are then stored as the quadruped robot's historical correction dataset.

[0067] When motion control of a quadruped robot is required, the robot must first be started. At this time, the microcontroller containing the model predictive control module can be activated. The model predictive control module of the quadruped robot can then respond to the start operation of the quadruped robot, initialize the original time series of the quadruped robot, and generate a new time series as the target time series. In this step, by generating a new time series, it can be ensured that the four mechanical legs of the quadruped robot can have a common time reference value, thereby improving the accuracy of motion control and the time synchronization of signal feedback, and reducing delay errors.

[0068] The microcontroller's state correction memory stores historical correction datasets corresponding to previous correction operations performed on the quadruped robot. During the quadruped robot's movement, if the model predictive control module (MPC) cannot receive feedback signals from the quadruped robot's vision or attitude sensors, and the quadruped robot is performing a repetitive task, the historical correction dataset stored in the state correction memory can be input to the feedback terminal of the MPC to enable the quadruped robot to complete a simple retrieval task. In specific implementations, if the quadruped robot is performing a repetitive task and the MPC cannot receive task feedback signals, the historical correction dataset is retrieved from the state correction memory and input to the feedback terminal of the MPC to enable the quadruped robot to complete the retrieval task. Therefore, when using external sensors for active obstacle avoidance, even if the quadruped robot's external sensors are damaged or unable to transmit feedback signals, and the quadruped robot is performing a repetitive task, simple motion control can still be performed on the quadruped robot based on the relevant data from the MPC historical correction operations, avoiding task interruption.

[0069] Step 102: Obtain the historical corrected dataset of the quadruped robot, and use the historical corrected dataset to perform data preprocessing on the target time series to generate the working condition dataset of the quadruped robot;

[0070] After the model prediction control module generates a new time series as the target time series, the historical correction dataset of the quadruped robot can be obtained. The target time series can then be preprocessed using the historical correction dataset to generate the working condition dataset of the quadruped robot. The working condition refers to the working state of the quadruped robot under conditions that are directly related to its actions.

[0071] In practical applications, the dynamic changes of quadruped robot parts over time can be recorded using target time series data. The data types corresponding to these dynamic changes can generally be divided into working status, balance status, and trajectory planning. Working status focuses on various time-related motion parameters during movement, such as normal operation, deceleration, acceleration, or other operating states, the speed of movement, the joint angles of the mechanical legs, joint angular velocities, etc. The balance status focuses on the balance situation under different working states, and the trajectory planning focuses on the path corresponding to the actual working condition trajectory. Specifically, the working status of the quadruped robot can be recorded non-linearly using node parameters of a fixed-interval time series, while the robot's balance status is recorded using an attitude sensor composed of the robot's internal gyroscope and gravity sensor, and trajectory planning can be recorded using the robot's vision sensor.

[0072] When the quadruped robot performs forward motion, the parameters executed by its four mechanical legs are preprocessed. At the same time, the target time series can be preprocessed using historical correction datasets. The preprocessed datasets are then integrated, classified, and categorized to generate a working condition process dataset. Based on the working condition process dataset, a working condition dataset is generated. This generated working condition dataset is fed back to the input of the model prediction control module to complete multi-threaded feedback correction. The working condition process dataset is also stored in the state correction memory, serving as the quadruped robot's historical correction dataset and as reference data for the next state correction.

[0073] In the specific implementation, the target time series is preprocessed using a historical correction dataset to generate a working condition dataset for the quadruped robot. This can be done by: preprocessing the target time series using a historical correction dataset to obtain a corresponding preprocessed correction dataset; then integrating, dividing, and classifying the preprocessed correction dataset to obtain a working condition process dataset corresponding to the working status, balance control status, and trajectory planning of the quadruped robot; and finally, performing Shannon sampling on the working condition process dataset to obtain the working condition dataset for the quadruped robot.

[0074] Among them, data preprocessing refers to the processing flow such as data cleaning, data integration, data transformation, and data reduction. Since data preprocessing is a conventional data processing method in existing related technologies and is not the main point of invention to be highlighted in the embodiments of this invention, it will not be described in detail.

[0075] Step 103: Obtain the redundant mechanical foot parameters of each mechanical foot in the quadruped robot, and use the working condition dataset to perform cluster calculation on each redundant mechanical foot parameter to obtain a cluster center point dataset.

[0076] As mentioned above, when a quadruped robot performs forward motion, the parameters executed by its four mechanical legs are preprocessed. Before preprocessing the parameters, the redundant mechanical leg parameters of each mechanical leg in the quadruped robot can be obtained first, and the redundant mechanical leg parameters can be clustered in the processor using the working condition dataset to obtain the cluster center point dataset.

[0077] Redundant mechanical legs, also known as redundant mechanical arms, can be understood as mechanical arms with added degrees of freedom. From a spatial positioning perspective, this means 6+ degrees of freedom, generally referring to 7 degrees of freedom. If we consider the characteristics of human three-dimensional space, we would typically set 6 degrees of freedom. However, due to the limitations of the mechanical structure of quadruped robots, there are several movement points that cannot be reached, or these points can be reached but only in one way. This can cause the mechanical legs of quadruped robots to twist or become uncontrollable during movement, potentially shortening the lifespan of the quadruped robot or even preventing it from completing its task. Therefore, adding redundant degrees of freedom on top of the original degrees of freedom, or understanding it as adding an extra joint, increases the number of movement solutions, which can greatly enhance the flexibility of the mechanical legs of quadruped robots and ensure successful obstacle avoidance and task completion.

[0078] Redundant mechanical leg parameters refer to all relevant parameters that can be executed by the processor, such as joint angles, joint angular velocities, and joint angle redundancy, of the redundant robotic arm when the quadruped robot is moving in the forward direction.

[0079] Meanwhile, clustering is a common method in data statistical analysis, such as the commonly used K-MEANS clustering algorithm, K-MEDOIDS clustering algorithm, or Clara clustering algorithm. As an example, when processing large amounts of data, the Clara algorithm can be used for clustering. Multiple samples are extracted from the actual data, and the K-MEDOIDS algorithm is used to obtain the corresponding class (O1, O2, ... Oi, ..., Ok) for each sample. Then, the class with the smallest consumption E is selected as the final result for output. In this embodiment of the invention, a working condition dataset generated based on a historical corrected dataset and a target time series is used to perform clustering calculations on the redundant mechanical foot parameters of the quadruped robot. Sample data with similar working condition dataset and redundant mechanical foot parameters can be automatically classified into one category. Furthermore, by combining sampling and clustering calculations, data consumption can be minimized, improving data processing efficiency and clustering accuracy. For quadruped robots, improving data processing efficiency means reducing feedback errors during motion control.

[0080] Step 104: Establish the actual working trajectory of the quadruped robot, use the cluster center point dataset to perform redundancy decomposition on each mechanical leg to obtain the redundancy optimal solution of each mechanical leg, and perform state correction on the quadruped robot according to each redundancy optimal solution.

[0081] After obtaining the cluster center point dataset corresponding to the redundant mechanical foot parameters through clustering calculation, the actual working trajectory of the quadruped robot can be established based on the sensors, and the redundancy of the four mechanical feet can be decomposed. Specifically, the actual working trajectory of the quadruped robot can be established, the redundancy of each mechanical foot can be decomposed using the cluster center point dataset, the optimal redundancy solution of each mechanical foot can be obtained, and the state of the quadruped robot can be corrected based on each optimal redundancy solution.

[0082] Redundancy decomposition can include setting joint limits, calculating the Jacobian matrix, performing singularity checks, and finding the optimal solution.

[0083] In vector calculus, the Jacobian matrix is ​​a matrix in which first-order partial derivatives are arranged in a certain way. Its determinant is called the Jacobian determinant. The Jacobian matrix represents the optimal linear approximation of a differentiable equation to a given point. The optimal solution is further obtained through linear approximation. Therefore, the Jacobian matrix is ​​similar to the derivative of a multivariable function.

[0084] For quadruped robots, the phenomenon where a redundant manipulator's joints continue to move even when the end effector's pose is fixed, due to redundancy, is called the manipulator's self-motion, also known as self-motion in the Jacobian matrix null space. Self-motion in the null space does not affect the end effector's pose. Therefore, in practical calculations, the self-motion of the redundant manipulator can be parameterized using a kinematic objective function. The objective function refers to the functional relationship between the target of interest (a certain variable) and related factors (certain variables). In short, the objective function corresponds to the function obtained after solving for the unknown variables. Before solving, the function is unknown; the objective function is obtained by solving for the functional relationship of the unknown variables using known conditions.

[0085] This method utilizes kinematic objective functions to parameterize the self-motion of redundant robotic arms. It is an inverse kinematics module that directly transforms Cartesian space trajectories into joint space trajectories using Jacobian matrices. Typical Cartesian space trajectories include straight lines, circular arcs, and sine curves. The points planned in Cartesian space need to be solved inversely to obtain the corresponding joint angles. Inverse kinematics refers to solving for the positions of each joint given the robot's end-effector pose. It is the foundation of robot motion planning and trajectory control. Inverse kinematics is a key step in planning robot Cartesian space trajectories. Based on the planned spatial points, the desired joint angles are calculated using inverse kinematics. The joint angles solved by inverse kinematics should not change significantly within two adjacent interpolation cycles.

[0086] In the specific implementation, the actual working trajectory of the quadruped robot is established, and the redundancy of each mechanical leg is decomposed using a cluster centroid dataset to obtain the optimal redundancy solution for each mechanical leg. This can be achieved as follows: the actual working trajectory of the quadruped robot is established based on sensors, and trajectory planning instructions for the quadruped robot are obtained; then, the joint angles and joint angular velocities of the redundant mechanical arms in the mechanical legs are extracted from the cluster centroid dataset; then, the Jacobian matrix of the mechanical leg is calculated based on the trajectory planning instructions, joint angles, and joint angular velocities, and the objective function corresponding to the Jacobian matrix is ​​calculated to generate the redundancy decomposition output of the mechanical leg; finally, singularity judgment is performed on the redundancy decomposition output, and the optimal redundancy solution of the mechanical leg is determined based on the singularity judgment results.

[0087] Furthermore, the Jacobian matrix of the robotic foot is calculated based on the trajectory planning instructions, joint angles, and joint angular velocities. The objective function corresponding to the Jacobian matrix is ​​then calculated to generate the redundancy decomposition output of the robotic foot. Specifically, the trajectory planning instructions are first redundantly decomposed to obtain the robotic foot's operation task instructions and extended task instructions. The operation task instructions correspond to the operation tasks, and the extended task instructions correspond to the extended tasks. Then, the Jacobian matrix of the robotic foot is calculated based on the operation tasks, extended tasks, joint angles, and joint angular velocities. The calculation formula is as follows:

[0088]

[0089] Simultaneously, the objective function corresponding to the Jacobian matrix can be calculated, and the calculation formula is as follows:

[0090]

[0091] Next, based on the Jacobian matrix and the objective function, the redundancy decomposition output of the mechanical foot is generated, and the calculation formula is as follows:

[0092]

[0093] Where q represents the joint angle of the redundant robotic arm. Let Xe be the joint angular velocity, Xc be the operational task, f(*) and g(*) be kinematic functions, X represent the trajectory planning task before redundancy decomposition, Je(q) be the input Jacobian matrix of the operational task Xe, Jc(q) be the input Jacobian matrix of the extended task Xc, and J represent the Jacobian matrix of the trajectory planning task X. This indicates that the Jacobian matrix J is calculated for the trajectory planning task. We represent the self-motion corresponding to operation task Xe, and We represent the weight coefficient corresponding to operation task Xe. Wc represents the self-motion corresponding to the extended task Xc, Wc represents the weight coefficient corresponding to the extended task Xc, and Wv represents the joint angular velocity. The corresponding weighting coefficients are given, where L is the objective control function of trajectory planning task X, Je is the output Jacobian matrix corresponding to operation task Xe, and Jc is the output Jacobian matrix corresponding to augmented task Xc. This is the inverse solution corresponding to the operation task Xe obtained after calculating the Jacobian matrix. This is the inverse solution corresponding to the extended task Xc obtained after calculating the Jacobian matrix. This indicates the redundant output of the mechanical foot. T This indicates the matrix transpose.

[0094] After obtaining the redundant decomposition output of the mechanical foot through the above formula, singularity judgment can be performed on the output of the mechanical foot to generate and find the optimal solution for redundancy. Its working principle is to complete the joint limit and obstacle avoidance planning of the redundant mechanical arm by selecting parameters in the damped minimum square method in order to find the optimal solution.

[0095] The damped least squares method, also known as the damped least squares method, primarily involves solving equations. It finds the optimal function match for the data by minimizing the sum of squares of the errors. Using the damped least squares method, unknown data can be easily obtained while minimizing the sum of squares of the errors between the unknown data and the actual data, thus maximizing the reduction of data errors.

[0096] Furthermore, singularity judgment is performed on the redundancy decomposition output, and the optimal redundancy solution of the mechanical foot is determined based on the singularity judgment result. This may include: performing singular value decomposition on the redundancy decomposition output to obtain the corresponding redundancy decomposition singular values; setting the joint limits of the mechanical foot; combining the joint limits and the redundancy decomposition singular values; completing the redundancy mechanical arm joint limit and obstacle avoidance planning of the mechanical foot by selecting parameters in the damped minimum square method; generating the optimal solution task; and determining the optimal redundancy solution corresponding to the optimal solution task.

[0097] This approach eliminates redundancy by increasing the number of operational spaces, treating the redundancy of the redundant robotic arm as a degree of freedom. Furthermore, the arm angle of this redundant robotic arm can be manually set based on the specific application scenario. The arm angle is a good parameter for measuring the overall posture of the robotic arm. Since the arm angle function uses the angles of each joint as independent variables, it can be combined with the T-matrix of the robotic arm to perform position-level inverse kinematics solutions, thereby obtaining its analytical solution.

[0098] To better illustrate the redundancy decomposition process of the mechanical feet of a quadruped robot, exemplarily, refer to... Figure 2 The diagram shows a control framework for state correction of a quadruped robot provided by an embodiment of the present invention.

[0099] When a quadruped robot performs trajectory planning, its input terminal can input trajectory planning instructions. These instructions can include operation task instructions and extended task instructions. After the trajectory planning instructions are redundantly decomposed, operation task instructions and extended task instructions are obtained. The operation task instructions correspond to the operation tasks, and the extended task instructions correspond to the extended tasks. Then, the operation tasks and extended tasks are input to the calculus unit for calculus conversion, and the corresponding results are output to the computational torque controller for calculation and processing, and the control values ​​are output. After that, the control values ​​are used to make the quadruped robot execute the processed operation tasks and extended tasks by moving the mechanical legs in the forward direction.

[0100] Furthermore, the control value output by the torque controller is fed back to the calculus unit, and simultaneously fed back to the input of the redundancy decomposition processing module via IC bus (Inter Integrated Circuit Bus) 1. The feedback of the forward motion of the mechanical foot is also subjected to redundancy decomposition after passing through IC bus 2. The output parameters of the forward motion of the mechanical foot are fed back to the input of the forward motion of the mechanical foot again after passing through IC bus 3.

[0101] As an optional embodiment, after obtaining the objective function through calculation, a corrected state model of the quadruped robot can be established based on the objective function. Then, the joint angle optimization calculation of the mechanical foot of the quadruped robot is performed based on the corrected state model to obtain the joint angle optimization index of the mechanical foot. Finally, the accuracy of the movement angle of the mechanical foot is optimized based on the joint angle optimization index.

[0102] The joint angular velocity of the mechanical foot is:

[0103]

[0104] The joint angle redundancy of the mechanical foot is calculated using the following formula:

[0105]

[0106] in, For the joint angle redundancy of the mechanical foot, Q is a fixed constant, and I is the joint angle redundancy of J. T The identity matrix corresponding to J, Let S be the gradient operator and S be the optimization index.

[0107] The optimal joint angle index for the mechanical foot is calculated using the following formula:

[0108]

[0109] Where S(w) is the joint angle optimization index, i is the joint dimension of the mechanical foot, and w iLet i be the joint angle redundancy corresponding to the joint dimension i. Let i be the minimum joint angle redundancy corresponding to joint dimension i. This represents the maximum joint angle redundancy corresponding to joint dimension i.

[0110] Therefore, by establishing a corresponding corrected state model for the quadruped robot through the objective function of kinematics, the state of the quadruped robot can be more model-based corrected based on the optimization index obtained by calculation. This makes the joint angles of each mechanical leg of the quadruped robot more matched with the current motion state after correction during the movement process, so as to maximize the achievement of the optimal task operation state.

[0111] For better explanation, refer to Figure 3 This illustration shows a schematic diagram of state changes for state correction of a quadruped robot according to an embodiment of the present invention. Taking one of the mechanical legs of the quadruped robot as an example, the mechanical leg can correspond to three joint dimensions, namely joint 1, joint 2, and joint 3. When joint angle optimization calculation is performed through the correction state model, the joint angle redundancy w1 corresponding to joint 1, the joint angle redundancy w2 corresponding to joint 2, and the joint angle redundancy w3 corresponding to joint 3 can be output. Furthermore, the joint angles of each corresponding joint can be adjusted based on the joint angle redundancy to further correct the state of the mechanical leg of the quadruped robot and achieve a better state correction optimization effect.

[0112] In this embodiment of the invention, during the motion control of the quadruped robot, in response to the start operation of the quadruped robot, the original time series of the quadruped robot is initialized and a target time series is generated. This new time series ensures that the four mechanical legs of the quadruped robot have a common time reference value, thereby improving the accuracy of motion control and the time synchronization of signal feedback, and reducing delay errors. Next, the historical correction dataset of the quadruped robot is acquired, and the target time series is preprocessed using this historical correction dataset to generate the quadruped robot's working condition dataset. This preprocessing of the target time series based on the historical correction dataset corresponding to the quadruped robot's own historical correction operations reduces dependence on external feedback signals and improves the prediction accuracy of motion control. Finally, the redundant mechanical legs of each mechanical leg in the quadruped robot are acquired. The parameters of the quadruped robot are clustered using a working condition dataset to obtain a cluster center point dataset. This clustering automatically groups similar samples in the working condition dataset into a single category. The combination of sampling and clustering minimizes data consumption, improving data processing efficiency and clustering accuracy, thus reducing feedback errors during motion control. Next, the actual working trajectory of the quadruped robot is established. Redundancy decomposition of each mechanical leg is performed using the cluster center point dataset to obtain the optimal redundancy solution for each leg. The quadruped robot's state is then corrected based on these optimal solutions. This allows for rapid acquisition of the optimal redundant solution corresponding to the task through redundancy decomposition calculations, enabling timely and accurate state correction. Furthermore, when using external sensors for active obstacle avoidance, even if the external sensors malfunction or fail to transmit feedback signals, simple motion control can be performed based on relevant data from MPC historical correction operations when the quadruped robot is performing repetitive tasks, preventing task interruption.

[0113] Reference Figure 4 The diagram illustrates a structural block diagram of a state correction device for a quadruped robot provided in an embodiment of the present invention, which may specifically include:

[0114] The target time series generation module 401 is used to initialize the original time series of the quadruped robot and generate a target time series in response to the start operation of the quadruped robot;

[0115] The working condition dataset generation module 402 is used to obtain the historical corrected dataset of the quadruped robot, and use the historical corrected dataset to perform data preprocessing on the target time series to generate the working condition dataset of the quadruped robot.

[0116] The cluster center point dataset generation module 403 is used to obtain the redundant mechanical foot parameters of each mechanical foot in the quadruped robot, and to perform cluster calculation on each of the redundant mechanical foot parameters using the working condition dataset to obtain the cluster center point dataset.

[0117] The mechanical leg redundancy decomposition module 404 is used to establish the actual working trajectory of the quadruped robot, perform redundancy decomposition on each mechanical leg using the cluster center point dataset, obtain the redundancy optimal solution of each mechanical leg, and perform state correction on the quadruped robot based on each redundancy optimal solution.

[0118] In one optional embodiment, the working condition dataset generation module 402 includes:

[0119] The preprocessing and correction dataset generation module is used to preprocess the target time series using the historical correction dataset to obtain the corresponding preprocessing and correction dataset.

[0120] The working condition dataset generation submodule is used to integrate, divide, and classify the preprocessed and corrected dataset to obtain the working condition process dataset corresponding to the working condition, balance control condition, and trajectory planning of the quadruped robot. Then, Shannon sampling is performed on the working condition process dataset to obtain the working condition dataset of the quadruped robot.

[0121] In one optional embodiment, the mechanical foot redundancy decomposition module 404 includes:

[0122] The trajectory planning instruction acquisition module is used to establish the actual working trajectory of the quadruped robot and acquire trajectory planning instructions for trajectory planning of the quadruped robot.

[0123] A redundant robotic arm data extraction module is used to extract the joint angles and joint angular velocities of the redundant robotic arms in the robotic foot from the cluster center point dataset.

[0124] The redundancy decomposition output generation module is used to calculate the Jacobian matrix of the mechanical foot based on the trajectory planning instruction, the joint angle, and the joint angular velocity, and to calculate the objective function corresponding to the Jacobian matrix, thereby generating the redundancy decomposition output of the mechanical foot.

[0125] The singularity detection module is used to perform singularity detection on the redundancy decomposition output and determine the optimal redundancy solution of the mechanical foot based on the singularity detection result.

[0126] In one optional embodiment, the redundancy decomposition output generation module includes:

[0127] The trajectory planning instruction redundancy decomposition module is used to perform redundancy decomposition on the trajectory planning instruction to obtain the operation task instruction and the extended task instruction of the mechanical foot. The operation task instruction corresponds to the operation task, and the extended task instruction corresponds to the extended task.

[0128] The Jacobian matrix calculation module is used to calculate the Jacobian matrix of the mechanical foot based on the operation task, the extended task, the joint angle, and the joint angular velocity. The calculation formula is as follows:

[0129]

[0130] The objective function calculation module is used to calculate the objective function corresponding to the Jacobian matrix. The calculation formula is as follows:

[0131]

[0132] The redundancy decomposition output generation submodule is used to generate the redundancy decomposition output of the mechanical foot based on the Jacobian matrix and the objective function. The calculation formula is as follows:

[0133]

[0134] Where q represents the joint angle of the redundant robotic arm. Let Xe be the joint angular velocity, Xc be the operational task, f(*) and g(*) be kinematic functions, X represent the trajectory planning task before redundancy decomposition, Je(q) be the input Jacobian matrix of the operational task Xe, Jc(q) be the input Jacobian matrix of the extended task Xc, and J represent the Jacobian matrix of the trajectory planning task X. This indicates that the Jacobian matrix J is calculated for the trajectory planning task. We represent the self-motion corresponding to operation task Xe, and We represent the weight coefficient corresponding to operation task Xe. Wc represents the self-motion corresponding to the extended task Xc, Wc represents the weight coefficient corresponding to the extended task Xc, and Wv represents the joint angular velocity. The corresponding weighting coefficients are given, where L is the objective control function of trajectory planning task X, Je is the output Jacobian matrix corresponding to operation task Xe, and Jc is the output Jacobian matrix corresponding to augmented task Xc. This is the inverse solution corresponding to the operation task Xe obtained after calculating the Jacobian matrix. This is the inverse solution corresponding to the extended task Xc obtained after calculating the Jacobian matrix. This indicates the redundant output of the mechanical foot. T This indicates the matrix transpose.

[0135] In one alternative embodiment, the device further includes:

[0136] The corrected state model establishment module is used to establish a corrected state model of the quadruped robot based on the objective function.

[0137] The joint angle optimization calculation module is used to perform joint angle optimization calculations on the mechanical feet of the quadruped robot according to the corrected state model, and obtain the joint angle optimization index of the mechanical feet.

[0138] The motion angle accuracy optimization module is used to optimize the accuracy of the motion angle of the mechanical foot based on the joint angle optimization index.

[0139] The joint angular velocity of the mechanical foot is:

[0140]

[0141] The redundancy calculation module is used to calculate the joint angle redundancy of the mechanical foot. The calculation formula is as follows:

[0142]

[0143] in, For the joint angle redundancy of the mechanical foot, Q is a fixed constant, and I is the joint angle redundancy of J. T The identity matrix corresponding to J, Let S be the gradient operator and S be the optimization index.

[0144] The joint angle optimization index calculation module is used to calculate the joint angle optimization index of the mechanical foot. The calculation formula is as follows:

[0145]

[0146] Where S(w) is the joint angle optimization index, i is the joint dimension of the mechanical foot, and w i Let i be the joint angle redundancy corresponding to the joint dimension i. Let i be the minimum joint angle redundancy corresponding to joint dimension i. This represents the maximum joint angle redundancy corresponding to joint dimension i.

[0147] In one optional embodiment, the singularity determination module includes:

[0148] The singular value decomposition module is used to perform singular value decomposition on the redundancy decomposition output to obtain the corresponding redundancy decomposition singular values.

[0149] The optimal solution task generation module is used to set the joint limits of the mechanical foot, combine the joint limits and the redundancy decomposition singular values, and complete the redundant mechanical arm joint limit and obstacle avoidance planning of the mechanical foot by selecting parameters in the damped minimum square method, generate the optimal solution task, and determine the redundant optimal solution corresponding to the optimal solution task.

[0150] In one optional embodiment, the quadruped robot has a built-in microcontroller, which includes at least a model prediction control module and a state correction memory. The model prediction control module performs MPC correction operations on the quadruped robot, and the state correction memory collects correction data corresponding to the MPC correction operation when performing the MPC correction operation on the quadruped robot, and stores multiple sets of correction data as the quadruped robot's historical correction dataset. The device further includes:

[0151] The historical correction dataset extraction module is used to extract the historical correction dataset from the state correction memory and input the historical correction dataset to the feedback terminal of the model prediction control module if the quadruped robot is performing a repetitive operation task and the model prediction control module cannot receive the task feedback signal from the quadruped robot, so that the quadruped robot can complete the retrieval task.

[0152] As the device embodiment is basically similar to the method embodiment, it is described in a relatively simple way. For relevant details, please refer to the description of the method embodiment above.

[0153] This invention also provides an electronic device, which includes a processor and a memory:

[0154] The memory is used to store program code and transfer the program code to the processor;

[0155] The processor is used to execute the state correction method of the quadruped robot according to the instructions in the program code of any embodiment of the present invention.

[0156] This invention also provides a computer-readable storage medium for storing program code for executing the state correction method for a quadruped robot according to any embodiment of this invention.

[0157] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0158] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0159] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0160] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0161] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0162] The above-described 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for state correction of a quadruped robot, characterized in that, include: In response to a startup operation on the quadruped robot, the original time series of the quadruped robot is initialized and a target time series is generated; Obtain the historical corrected dataset of the quadruped robot, and use the historical corrected dataset to preprocess the target time series to generate the working condition dataset of the quadruped robot; Obtain the redundant mechanical foot parameters of each mechanical foot in the quadruped robot, and use the working condition dataset to perform cluster calculation on each of the redundant mechanical foot parameters to obtain a cluster center point dataset; The actual working trajectory of the quadruped robot is established, and the redundancy decomposition of each mechanical leg is performed using the cluster center point dataset to obtain the redundancy optimal solution of each mechanical leg. The state of the quadruped robot is then corrected based on the redundancy optimal solution. The step of establishing the actual working trajectory of the quadruped robot, and using the cluster centroid dataset to perform redundancy decomposition on each of the mechanical legs to obtain the optimal redundancy solution for each mechanical leg, includes: establishing the actual working trajectory of the quadruped robot; obtaining trajectory planning instructions for trajectory planning of the quadruped robot; extracting the joint angles and joint angular velocities of the redundant robotic arms in the mechanical legs from the cluster centroid dataset; calculating the Jacobian matrix of the mechanical legs based on the trajectory planning instructions, the joint angles, and the joint angular velocities, and calculating the objective function corresponding to the Jacobian matrix to generate the redundancy decomposition output of the mechanical legs; performing singularity judgment on the redundancy decomposition output, and determining the optimal redundancy solution of the mechanical legs based on the singularity judgment result; The step of calculating the Jacobian matrix of the mechanical foot based on the trajectory planning instruction, the joint angle, and the joint angular velocity, and calculating the objective function corresponding to the Jacobian matrix to generate the redundancy decomposition output of the mechanical foot includes: The trajectory planning instructions are redundantly decomposed to obtain the operation task instructions and extended task instructions of the mechanical foot. The operation task instructions correspond to the operation tasks, and the extended task instructions correspond to the extended tasks. The Jacobian matrix of the mechanical foot is calculated based on the operation task, the expansion task, the joint angle, and the joint angular velocity, using the following formula: ； The objective function corresponding to the Jacobian matrix is ​​calculated using the following formula: ； Based on the Jacobian matrix and the objective function, the redundancy decomposition output of the mechanical foot is generated, and the calculation formula is as follows: ； Where q represents the joint angle of the redundant robotic arm. Let Xe be the joint angular velocity, Xc be the operational task, f(*) and g(*) be kinematic functions, X represent the trajectory planning task before redundancy decomposition, Je(q) be the input Jacobian matrix of the operational task Xe, Jc(q) be the input Jacobian matrix of the extended task Xc, and J represent the Jacobian matrix of the trajectory planning task X. This indicates that the Jacobian matrix J is calculated for the trajectory planning task. We represent the self-motion corresponding to operation task Xe, and We represent the weight coefficient corresponding to operation task Xe. Wc represents the self-motion corresponding to the extended task Xc, Wc represents the weight coefficient corresponding to the extended task Xc, and Wv represents the joint angular velocity. The corresponding weighting coefficients are given, where L is the objective control function of trajectory planning task X, Je is the output Jacobian matrix corresponding to operation task Xe, and Jc is the output Jacobian matrix corresponding to augmented task Xc. This is the inverse solution corresponding to the operation task Xe obtained after calculating the Jacobian matrix. This is the inverse solution corresponding to the extended task Xc obtained after calculating the Jacobian matrix. This indicates the redundant output of the mechanical foot. T Indicates matrix transpose; The step of performing singularity judgment on the redundant decomposition output and determining the redundancy optimal solution of the mechanical foot based on the singularity judgment result includes: performing singular value decomposition on the redundant decomposition output to obtain the corresponding redundant decomposition singular values; setting the joint limits of the mechanical foot, and combining the joint limits and the redundant decomposition singular values ​​to complete the redundant mechanical arm joint limit and obstacle avoidance planning of the mechanical foot by selecting parameters in the damped minimum square method, generating the optimal solution task, and determining the redundancy optimal solution corresponding to the optimal solution task.

2. The state correction method for a quadruped robot according to claim 1, characterized in that, The step of preprocessing the target time series using the historical correction dataset to generate the working condition dataset of the quadruped robot includes: The target time series is preprocessed using the historical corrected dataset to obtain the corresponding preprocessed corrected dataset. The preprocessed and corrected dataset is integrated, divided, and classified to obtain the working condition process dataset corresponding to the working status, balance control status, and trajectory planning of the quadruped robot. Then, Shannon sampling is performed on the working condition process dataset to obtain the working condition dataset of the quadruped robot.

3. The state correction method for a quadruped robot according to claim 1, characterized in that, include: A corrected state model of the quadruped robot is established based on the objective function; The joint angle optimization calculation of the mechanical foot of the quadruped robot is performed based on the modified state model to obtain the joint angle optimization index of the mechanical foot. The accuracy of the movement angle of the mechanical foot is optimized based on the joint angle optimization index; The joint angular velocity of the mechanical foot is: ； The joint angle redundancy of the mechanical foot is calculated using the following formula: ； in, For the joint angle redundancy of the mechanical foot, Q is a fixed constant, and I is the joint angle redundancy of J. T The identity matrix corresponding to J, Let S be the gradient operator and S be the optimization index. The joint angle optimization index of the mechanical foot is calculated using the following formula: ； in, For the joint angle optimization index, i is the joint dimension of the mechanical foot, and w i Let i be the joint angle redundancy corresponding to the joint dimension i. Let i be the minimum joint angle redundancy corresponding to joint dimension i. This represents the maximum joint angle redundancy corresponding to joint dimension i.

4. The state correction method for a quadruped robot according to claim 1, characterized in that, The quadruped robot has a built-in microcontroller, which includes at least a model prediction and control module and a state correction memory. The model prediction and control module performs MPC correction operations on the quadruped robot. The state correction memory collects correction data corresponding to the MPC correction operation when performing the MPC correction operation on the quadruped robot, and stores multiple sets of correction data as the quadruped robot's historical correction dataset. The method further includes: If the quadruped robot performs a repetitive task and the model prediction control module cannot receive the task feedback signal from the quadruped robot, the historical correction dataset is extracted from the state correction memory and input to the feedback terminal of the model prediction control module so that the quadruped robot can complete the retrieval task.

5. A state correction device for a quadruped robot, characterized in that, include: A target time series generation module is used to initialize the original time series of the quadruped robot and generate a target time series in response to a start operation for the quadruped robot. The working condition dataset generation module is used to obtain the historical corrected dataset of the quadruped robot, and use the historical corrected dataset to perform data preprocessing on the target time series to generate the working condition dataset of the quadruped robot. The cluster center point dataset generation module is used to obtain the redundant mechanical foot parameters of each mechanical foot in the quadruped robot, and to perform cluster calculation on each of the redundant mechanical foot parameters using the working condition dataset to obtain the cluster center point dataset. The mechanical leg redundancy decomposition module is used to establish the actual working trajectory of the quadruped robot, use the cluster center point dataset to perform redundancy decomposition on each mechanical leg, obtain the redundancy optimal solution of each mechanical leg, and perform state correction on the quadruped robot based on each redundancy optimal solution. The mechanical leg redundancy decomposition module includes: a trajectory planning instruction acquisition module, used to establish the actual working trajectory of the quadruped robot and acquire trajectory planning instructions for trajectory planning of the quadruped robot; a redundant robotic arm data extraction module, used to extract the joint angles and joint angular velocities of the redundant robotic arms in the mechanical leg from the cluster center point dataset; a redundancy decomposition output generation module, used to calculate the Jacobian matrix of the mechanical leg based on the trajectory planning instructions, the joint angles, and the joint angular velocities, and calculate the objective function corresponding to the Jacobian matrix to generate the redundancy decomposition output of the mechanical leg; and a singularity judgment module, used to perform singularity judgment on the redundancy decomposition output and determine the optimal redundancy solution of the mechanical leg based on the singularity judgment result. The redundancy decomposition output generation module includes: The trajectory planning instruction redundancy decomposition module is used to perform redundancy decomposition on the trajectory planning instruction to obtain the operation task instruction and the extended task instruction of the mechanical foot. The operation task instruction corresponds to the operation task, and the extended task instruction corresponds to the extended task. The Jacobian matrix calculation module is used to calculate the Jacobian matrix of the mechanical foot based on the operation task, the extended task, the joint angle, and the joint angular velocity. The calculation formula is as follows: ； The objective function calculation module is used to calculate the objective function corresponding to the Jacobian matrix. The calculation formula is as follows: ； The redundancy decomposition output generation submodule is used to generate the redundancy decomposition output of the mechanical foot based on the Jacobian matrix and the objective function. The calculation formula is as follows: ； Where q represents the joint angle of the redundant robotic arm. Let Xe be the joint angular velocity, Xc be the operational task, f(*) and g(*) be kinematic functions, X represent the trajectory planning task before redundancy decomposition, Je(q) be the input Jacobian matrix of the operational task Xe, Jc(q) be the input Jacobian matrix of the extended task Xc, and J represent the Jacobian matrix of the trajectory planning task X. This indicates that the Jacobian matrix J is calculated for the trajectory planning task. We represent the self-motion corresponding to operation task Xe, and We represent the weight coefficient corresponding to operation task Xe. Wc represents the self-motion corresponding to the extended task Xc, Wc represents the weight coefficient corresponding to the extended task Xc, and Wv represents the joint angular velocity. The corresponding weighting coefficients are given, where L is the objective control function of trajectory planning task X, Je is the output Jacobian matrix corresponding to operation task Xe, and Jc is the output Jacobian matrix corresponding to augmented task Xc. This is the inverse solution corresponding to the operation task Xe obtained after calculating the Jacobian matrix. This is the inverse solution corresponding to the extended task Xc obtained after calculating the Jacobian matrix. This indicates the redundant output of the mechanical foot. T Indicates matrix transpose; The singularity judgment module includes: a singular value decomposition module, used to perform singular value decomposition on the redundant decomposition output to obtain the corresponding redundant decomposition singular values; and an optimal solution task generation module, used to set the joint limits of the mechanical foot, combine the joint limits and the redundant decomposition singular values, complete the redundant mechanical arm joint limit and obstacle avoidance planning of the mechanical foot by selecting parameters in the damped minimum square method, generate the optimal solution task, and determine the redundant optimal solution corresponding to the optimal solution task.

6. An electronic device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the state correction method for the quadruped robot according to any one of claims 1-4, based on the instructions in the program code.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for executing the state correction method for the quadruped robot according to any one of claims 1-4.

Images