Distributed robust model predictive control method and device based on bionic foot optimization
By collecting terrain point cloud and foot contact force data for preprocessing and feature extraction, the optimal foothold is determined and global optimization is performed, which solves the problem of posture instability of quadruped robots on complex terrain and achieves higher control precision and motion stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU VOCATIONAL COLLEGE OF TECH & BUSINESS
- Filing Date
- 2026-02-12
- Publication Date
- 2026-05-19
AI Technical Summary
Traditional control methods struggle to achieve precise posture and trajectory control for quadruped robots on complex, unstructured terrain, and posture instability is particularly prone to occur during dynamic walking.
Terrain point cloud data and foot contact force data are collected by a data acquisition device installed on the quadruped robot. Data preprocessing and feature extraction are performed to obtain target terrain feature data. The optimal foothold is determined according to preset constraints and evaluation functions. Global optimization is performed, and the initial control parameters are adjusted to improve control accuracy.
It improves the control precision of quadruped robots, reduces the risk of posture imbalance, and enhances movement stability on complex terrain.
Smart Images
Figure CN122063879A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of robotics, and also to a distributed robust model predictive control method and device based on bionic foot optimization. Background Technology
[0002] In extreme operational environments such as search and rescue exploration, industrial inspection, and military reconnaissance, the requirements for terrain adaptability and motion stability of mobile robots are becoming increasingly stringent. With their unique biomimetic leg structures, quadruped robots exhibit stronger terrain-crossing capabilities and load-bearing potential than traditional wheeled or tracked robots, making them one of the core research directions in the field of special-purpose robots. However, when facing typical unstructured and complex terrains such as stairs, slopes, and gravel roads, the control systems of quadruped robots still face severe challenges. These terrains typically involve abrupt changes in geometric parameters, unstable ground support, and unknown and variable physical properties (such as the coefficient of friction), requiring robots to possess rapid dynamic response, precise posture adjustment, and efficient constraint handling capabilities.
[0003] Traditional control methods, while satisfying complex kinematic constraints, struggle to achieve precise control of the posture and trajectory of quadruped robots. Especially during dynamic walking, posture instability can easily occur due to model simplification or control delays. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to provide a distributed robust model predictive control method and device based on bionic foot optimization, so as to improve the control accuracy of quadruped robots.
[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A first aspect of the present invention provides a distributed robust model predictive control method based on biomimetic foot optimization, comprising: Terrain point cloud data and foot contact force data are collected using a data acquisition device installed on the quadruped robot; The terrain point cloud data and the foot contact force data are preprocessed to obtain preprocessed data. The preprocessed data is subjected to terrain feature extraction to obtain target terrain feature data; Based on the target terrain feature data and preset constraints, the actual support point data is obtained; The optimal foothold is obtained based on the actual support point data and the preset evaluation function; Based on the optimal foothold and the target terrain feature data, global optimization processing is performed to obtain initial control parameters; The initial control parameters are adjusted to obtain the target control parameters for the quadruped robot; The joint angles of the quadruped robot are controlled according to the target control parameters.
[0006] Optionally, terrain point cloud data and foot contact force data are collected using a data acquisition device mounted on the quadruped robot, including: The surrounding terrain is scanned by a lidar mounted on a quadruped robot to obtain terrain point cloud data; Tactile sensors installed on the feet of the quadruped robot collect data on the contact force at the feet.
[0007] Optionally, the terrain point cloud data and the foot contact force data are preprocessed to obtain preprocessed data, including: The terrain point cloud data and the foot contact force data are aligned to obtain aligned data. The aligned data is then denoised to obtain denoised data. The denoised data is fused to obtain preprocessed data.
[0008] Optionally, terrain feature extraction is performed on the preprocessed data to obtain target terrain feature data, including: pass The terrain slope is obtained; among which, For terrain slope, This is the unit normal vector of the optimal ground plane obtained from the preprocessed data. This is the unit vector of gravity direction obtained from the preprocessed data; pass The step height is obtained; where, The height of the step. The average elevation of the ground plane obtained from the preprocessed data, The average elevation of the plane above the step, obtained from the preprocessed data; pass The ground friction coefficient is obtained; where, The coefficient of friction of the ground. The tangential force of the ground contact force obtained from the preprocessed data. The normal force of the ground contact force obtained from the preprocessed data; The target terrain feature data are obtained based on the terrain slope, the step height, and the ground friction coefficient.
[0009] Optionally, based on the target terrain feature data and preset constraints, the actual support point data is obtained, including: Obtain preset constraints; the preset constraints include the range of hip joint yaw angle, the range of hip joint roll angle, and the range of knee joint pitch angle. Based on the target terrain feature data and the preset constraints, the actual support point data is obtained; the actual support point data includes: stability priority support points, efficiency priority support points, and balance priority support points.
[0010] Optionally, based on the actual support point data and the preset evaluation function, the optimal foothold is obtained, including: pass The preset evaluation function value is obtained; where, To preset the evaluation function value, , , All are dynamic weights. To support the stability index, through get, The angle between the normal vector of the foot point and the direction of gravity. For trajectory tracking error, through get, For actual support point data, For the ideal fulcrum position, This is the error attenuation coefficient. The leg energy expenditure index is calculated by... get, Let i be the angle of the i-th joint. This represents the maximum angle of the joint. The actual support point data is filtered based on the preset evaluation function value to obtain the optimal foothold.
[0011] Optionally, based on the optimal foothold and the target terrain feature data, global optimization processing is performed to obtain initial control parameters, including: Acquire real-time status data of the quadruped robot; the real-time status data includes the position of the torso center of mass, posture angles, joint angles, and velocity; Map the target terrain feature data to constraint parameters; Based on the optimal foothold and the target terrain feature data, the trunk reference state data and leg reference state data are obtained; Global optimization is performed based on the trunk reference state data and the leg reference state data to obtain initial control parameters; the initial control parameters include joint angle command data.
[0012] Optionally, the initial control parameters are adjusted to obtain the target control parameters for the quadruped robot, including: Obtain error data; The initial control parameters are adjusted based on the error data and the preset correction strategy to obtain the target control parameters for the quadruped robot.
[0013] A second aspect of the present invention provides a distributed robust model predictive control device based on biomimetic foot optimization, comprising: The data acquisition module is used to acquire terrain point cloud data and foot contact force data through a data acquisition device installed on the quadruped robot; The processing module is used to preprocess the terrain point cloud data and the foot contact force data to obtain preprocessed data; extract terrain features from the preprocessed data to obtain target terrain feature data; obtain actual support point data based on the target terrain feature data and preset constraints; obtain the optimal foothold based on the actual support point data and preset multi-target data; perform global optimization processing based on the optimal foothold to obtain initial control parameters; adjust the initial control parameters to obtain the target control parameters of the quadruped robot; and control the joint angles of the quadruped robot based on the target control parameters.
[0014] A third aspect of the present invention provides a computing device, comprising: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described in the first aspect.
[0015] The above-described solution of the present invention has at least the following beneficial effects: The above-described solution of the present invention collects terrain point cloud data and foot contact force data through a data acquisition device installed on a quadruped robot. The terrain point cloud data and foot contact force data are preprocessed to obtain preprocessed data. Then, terrain features are extracted from the preprocessed data to obtain target terrain feature data. Based on the target terrain feature data and preset constraints, actual support point data is obtained. Based on the actual support point data and a preset evaluation function, an optimal foothold is obtained. Global optimization processing is performed based on the optimal foothold and the target terrain feature data to obtain initial control parameters. Finally, the initial control parameters are adjusted to obtain the target control parameters for the quadruped robot. Controlling the joint angles of the quadruped robot based on these target control parameters can improve the accuracy of quadruped robot control and reduce the risk of posture imbalance. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating the distributed robust model predictive control method based on bionic foot optimization in an embodiment of the present invention. Figure 2 This is a schematic diagram of the distributed robust model predictive control device based on bionic foot optimization in an embodiment of the present invention. Detailed Implementation
[0017] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
[0018] like Figure 1 As shown, an embodiment of the present invention proposes a distributed robust model predictive control method based on biomimetic foot optimization, comprising the following steps: Step 101: Collect terrain point cloud data and foot contact force data using the data acquisition device installed on the quadruped robot; Step 102: Perform data preprocessing on the terrain point cloud data and the foot contact force data to obtain preprocessed data; Step 103: Extract terrain features from the preprocessed data to obtain target terrain feature data; Step 104: Obtain actual support point data based on the target terrain feature data and preset constraints; Step 105: Based on the actual support point data and the preset evaluation function, obtain the optimal foothold. Step 106: Perform global optimization processing based on the optimal foothold and the target terrain feature data to obtain initial control parameters; Step 107: Adjust the initial control parameters to obtain the target control parameters for the quadruped robot; Step 108: Control the joint angles of the quadruped robot according to the target control parameters.
[0019] The distributed robust model predictive control method based on biomimetic foot optimization in this invention collects terrain point cloud data and foot contact force data through a data acquisition device installed on a quadruped robot. The data is preprocessed to obtain preprocessed data, followed by terrain feature extraction to obtain target terrain feature data. Based on the target terrain feature data and preset constraints, actual support point data is obtained. Based on the actual support point data and a preset evaluation function, an optimal foothold is determined. Global optimization is performed based on the optimal foothold and the target terrain feature data to obtain initial control parameters. Finally, the initial control parameters are adjusted to obtain the target control parameters for the quadruped robot. Controlling the joint angles of the quadruped robot based on these target control parameters improves the control accuracy of the quadruped robot and reduces the risk of posture imbalance.
[0020] In an optional embodiment of the present invention, step 101, which involves collecting terrain point cloud data and foot contact force data using a data acquisition device mounted on the quadruped robot, may include: Step 1011: Scan the surrounding terrain using a lidar mounted on the quadruped robot to obtain terrain point cloud data; Specifically, the terrain around the quadruped robot is scanned at a fixed frequency (e.g., 10Hz) by a lidar installed on the robot to obtain terrain point cloud data (three-dimensional point cloud data, which includes the spatial coordinate information of the terrain).
[0021] Step 1012: Collect contact force data at the foot end using tactile sensors installed at the foot end of the quadruped robot.
[0022] Specifically, uSkin tactile sensors (a flexible, thin, distributed triaxial force tactile sensing system designed specifically for robots) installed on the feet of the quadruped robot collect foot contact force data in real time (including tactile data such as normal force, tangential force, and contact area during the foot-to-ground contact process), with a sampling frequency ≥100Hz.
[0023] In an optional embodiment of the present invention, step 102, which involves preprocessing the terrain point cloud data and the foot contact force data to obtain preprocessed data, may include: Step 1021: Perform data alignment processing on the terrain point cloud data and the foot contact force data to obtain aligned data; Specifically, timestamps are used to align terrain point cloud data and foot contact force data to ensure spatiotemporal consistency.
[0024] Step 1022: Denoise the aligned data to obtain denoised data; Specifically, the aligned data can be denoised by filtering or removing outliers to obtain denoised data. Denoising can improve data accuracy and provide an accurate data foundation for subsequently determining the robot's control parameters.
[0025] Step 1023: Perform data fusion on the denoised data to obtain preprocessed data.
[0026] Specifically, based on the robot's pose information, the foot contact force data (contact force) in the denoised data is associated and fused with the corresponding terrain point cloud data (terrain geometry) in the denoised data to obtain preprocessed data.
[0027] In an optional embodiment of the present invention, step 103, which involves extracting terrain features from the preprocessed data to obtain target terrain feature data, may include: Step 1031, through The terrain slope is obtained; among which, For terrain slope, This is the unit normal vector of the optimal ground plane obtained from the preprocessed data. This is the unit vector of gravity direction obtained from the preprocessed data; Specifically, three non-collinear points are randomly selected from the preprocessed data as fitting points to fit the initial plane; through... The distance from the fitted point to the initial plane is calculated; when the distance from the point to the plane is less than a distance threshold, the corresponding point is confirmed as an interior point; the plane with the most interior points is taken as the optimal plane; the terrain slope is obtained based on the unit normal vector and the unit vector of gravity direction of the optimal plane, i.e., through... The terrain slope is obtained. Let be the distance from the i-th (i=1,2,3) fitted point to the initial plane. Let be the coordinates of the i-th point, and a, b, c, and d be the coefficients of the three-dimensional ground plane equation.
[0028] Step 1032, through The step height is obtained; where, The height of the step. To determine the average elevation of the ground plane based on the preprocessed data, by... get, To determine the average elevation of the upper plane of the step based on the preprocessed data, by... get, This represents the total number of points within the horizontal plane region. This represents the total number of points within the area on the upper surface of the step. A horizontal plane area The upper surface area of the step. Points within a horizontal plane region, For points within the area on the upper surface of the step, for The height value, for The height value.
[0029] Specifically, the normal vector of all points in the preprocessed data is calculated (which can be obtained by fitting a local plane with neighborhood points); points within the ground plane are selected as initial seed points, and growth is performed according to preset criteria to obtain the ground region; unclassified points are traversed, and region growth is repeated to obtain the upper plane region of the steps, the vertical surface region of the steps, etc.; the vertical surface region is removed (the angle between the normal vector and the gravity direction is >60°), and the horizontal plane region (ground) and the upper surface region of the steps are retained. Here, the preset criteria may include: for seed points and neighborhood points, when the seed points and neighborhood points satisfy... and At that time, the seed point and its neighboring points are merged. Among them, The angle between the seed point normal vector and the neighboring point normal vectors. Seed point The local normal vector, For neighborhood points The local normal vector, The threshold for the angle between the normal vectors (ranging from 10° to 15°). Seed point The height value, For neighborhood points The height value, The height difference threshold (with a value range of 0.02 to 0.05).
[0030] Step 1033, through The ground friction coefficient is obtained; where, The coefficient of friction of the ground. The tangential force of the ground contact force obtained from the preprocessed data. The normal force of the ground contact force obtained from the preprocessed data; Specifically, ground contact force is obtained by analyzing foot contact force data from preprocessed data. (Normal force) and (Tangential force), based on the Coulomb friction model Real-time estimation of friction coefficient The sampling frequency is 100Hz, and the error is controlled within... Within the range.
[0031] Step 1034: Obtain target terrain feature data based on the terrain slope, the step height, and the ground friction coefficient.
[0032] Specifically, the target terrain feature data consists of feature vectors including terrain slope, step height, and ground friction coefficient. ,in Indicates the slope of the terrain. Indicates the height of the step (when there are no steps). ), This represents the coefficient of friction of the ground.
[0033] In an optional embodiment of the present invention, step 104, obtaining the actual support point data based on the target terrain feature data and preset constraints, may include: Step 1041: Obtain preset constraints; the preset constraints include the range of hip joint yaw angle, the range of hip joint roll angle, and the range of knee joint pitch angle. Specifically, preset constraints may include: hip joint yaw angle range range of hip joint rolling angle Knee joint pitch angle range Depending on the actual situation, the preset constraints can be adjusted to adapt to specific application scenarios.
[0034] Step 1042: Based on the target terrain feature data and the preset constraints, obtain the actual support point data; the actual support point data includes: stability priority support points, efficiency priority support points, and balance priority support points.
[0035] Specifically, based on preset constraints and the forward kinematics of the robot's legs, by traversing all legal joint angles, the set of all spatial positions reachable by the robot's feet can be obtained. Within this set, the angle between the ground normal vector and the direction of gravity is defined as... Points are prioritized for stability support. These points are close to the topographical convexity or the center of the plane to ensure maximum foot support area and improve support stability, making them suitable for unstable terrains such as steep slopes and gravel roads. Efficiency-priority support points are generated along the ideal movement trajectory (preset torso center of mass trajectory) to minimize the degree to which the foot deviates from the ideal trajectory and reduce leg movement energy consumption, making them suitable for gentle slopes or fast-moving scenarios. When the target terrain feature data... or When stability priority support points are used as balance priority support points, the target terrain feature data... or The system will switch from prioritizing efficiency to prioritizing efficiency, thereby achieving a dynamic balance between stability and energy consumption.
[0036] In an optional embodiment of the present invention, step 105, obtaining the optimal foothold based on the actual support point data and the preset evaluation function, may include: Step 1051, through The preset evaluation function value is obtained; where, To preset the evaluation function value, , , All are dynamic weights. To support the stability index, , The angle between the normal vector of the foot point and the direction of gravity. For trajectory tracking error, , For actual support point data, For the ideal fulcrum position, This is the error attenuation coefficient. This represents the energy expenditure index of the legs. , Let i be the angle of the i-th joint. This represents the maximum angle of the joint. Step 1052: Filter the actual support point data according to the preset evaluation function value to obtain the optimal foothold.
[0037] Specifically, the actual support points in the actual support point data are sorted from largest to smallest according to the preset evaluation function value, and the actual support point with the largest preset evaluation function value is taken as the optimal foothold.
[0038] In an optional embodiment of the present invention, step 106 involves performing global optimization based on the optimal foothold and the target terrain feature data to obtain initial control parameters, including: Step 1061: Obtain real-time status data of the quadruped robot; the real-time status data includes the position of the torso center of mass, posture angles, joint angles, and velocity; Specifically, real-time status data may include the position of the trunk's center of mass, trunk posture angle, center of mass velocity, posture angular velocity, the position of the foot of the i-th leg, the foot velocity of the i-th leg, and the angle of the i-th joint at time k.
[0039] Step 1062: Map the target terrain feature data to constraint parameters; Specifically, the foot friction cone constraint parameters are obtained based on the ground friction coefficient in the target terrain feature data, i.e. , The ground friction coefficient in the target terrain feature data. The tangential force of the ground contact force obtained from the preprocessed data. The normal force is obtained from the ground contact force based on the preprocessed data; the reference position parameters of the centroid are obtained from the slope in the target terrain feature data. , The reference position parameters for the centroid of the main trunk. This serves as the nominal reference location for the trunk's center of mass on flat terrain. The slope compensation coefficient is used; the leg reference trajectory parameters are obtained from the step height in the target terrain feature data, i.e. ,in, Let be the reference position of the foot of the j-th leg. Let j be the optimal footing point for the j-th leg. This is the step height compensation coefficient. The step height is the target terrain feature data.
[0040] Step 1063: Based on the optimal foothold and the target terrain feature data, obtain the trunk reference state data and the leg reference state data; Specifically, the backbone reference state data is obtained through... We obtained, among which, Let k be the reference state vector of the main subsystem predicted at time k. The reference position of the torso's center of mass at time k is... Let the torso reference posture angle be at time k. For the centroid reference speed, The attitude reference angular velocity. Leg reference state data (i=2 to 5), through... We obtained, among which, Let be the reference state vector for the i-th leg at time k, predicting the step. Let the reference position of the foot of the i-th leg at time k be denoted as . Let be the reference velocity of the foot tip of the i-th leg at time k.
[0041] Step 1064: Perform global optimization processing based on the trunk reference state data and the leg reference state data to obtain initial control parameters; the initial control parameters include joint angle command data.
[0042] Specifically, the main trunk is designated as subsystem 1, and the four legs as subsystems 2 through 5. The global optimization problem for each subsystem is solved by predicting the time domain. and control time domain Construct, its objective function as follows:
[0043] in, Let i be the cost function of the i-th subsystem. To predict the time domain, To control the time domain, Let i be the state of the i-th subsystem at time k. Let i be the reference state of the i-th subsystem at time k. This is the control input for the i-th subsystem at time k. This is the reference input for the i-th subsystem at time k. Here is the state error weight matrix. To control the input weight matrix, This is the terminal state weight matrix.
[0044] Subsystem State and Input: For the main subsystem ( ), its state Input the centroid reference speed. For attitude reference angular velocity; for the leg subsystem ( ), its state Input the joint angle. For joint angles; constraints include foot friction cone constraints, joint torque and angle limitations, and swing leg trajectory constraints.
[0045] The global optimization problem is decomposed into five sub-problems, and information collaboration is achieved through iterative solutions. The iterative process is as follows:
[0046]
[0047]
[0048] in, To augment the Lagrangian function, Z is a globally shared variable. As dual variables, For penalty parameters, Let i be the optimization variable for the i-th subsystem. This represents the mapping matrix between subsystems and global variables. By solving subproblems in parallel and iteratively updating shared and dual variables, the system achieves efficient collaboration between subsystems while ensuring global optimality.
[0049] Through the above iterative solution, input parameters that satisfy the constraint parameters are obtained, such as attitude reference angular velocity and joint angles, which are used as initial control parameters. The initial control parameters include joint angle command data, such as trunk attitude control quantity and leg joint angle command data.
[0050] In an optional embodiment of the present invention, step 107, adjusting the initial control parameters to obtain the target control parameters for the quadruped robot, may include: Step 1071, Obtain error data; Step 1072: Adjust the initial control parameters according to the error data and the preset correction strategy to obtain the target control parameters of the quadruped robot.
[0051] Specifically, using the error data between the system's measured output and predicted output, the EID estimator (a real-time signal processing module) is based on equivalent interference, through... The compensation control quantity is calculated and superimposed on the initial control parameters to obtain the target control parameters of the quadruped robot. This achieves feedforward compensation for disturbances and errors, improving the accuracy of the control parameters. The compensation control quantity is the compensation control input at time k, where B is the control input matrix and D is the disturbance input matrix. This represents the equivalent disturbance at time k.
[0052] In an optional embodiment of the present invention, in step 108, the joint angles of the quadruped robot are controlled according to the target control parameters. By converting the target control parameters into angle commands for the hip / knee joints, the leg joint angle commands are verified by inverse kinematics and then directly drive the joint motors through the "position closed-loop controller". This ensures that the commands comply with kinematic constraints and are tracked in real time throughout the process, thereby achieving control of the quadruped robot.
[0053] A specific embodiment of the distributed robust model predictive control method based on biomimetic foot optimization according to the present invention includes: Step 111: Collect terrain point cloud data and foot contact force data using the data acquisition device installed on the quadruped robot; The surrounding terrain is scanned by a lidar installed on the quadruped robot to obtain terrain point cloud data; the contact force data of the foot end is collected in real time by tactile sensors installed on the foot end of the quadruped robot.
[0054] Step 112: Perform data preprocessing on the terrain point cloud data and the foot contact force data to obtain preprocessed data; The terrain point cloud data and foot contact force data are subjected to data alignment, noise reduction and data fusion to obtain preprocessed data.
[0055] Step 113: Extract terrain features from the preprocessed data to obtain target terrain feature data; The target terrain features, such as slope, step height, and ground friction coefficient, are calculated using the corresponding formulas.
[0056] Step 114: Obtain actual support point data based on the target terrain feature data and preset constraints; Based on the target terrain feature data, multiple actual support points that meet the preset constraints are obtained.
[0057] Step 115: Based on the actual support point data and the preset evaluation function, obtain the optimal foothold. The preset evaluation function value of each actual support point is calculated based on the preset evaluation function, and the actual support point with the largest preset evaluation function is taken as the optimal footing point.
[0058] Step 116: Perform global optimization processing based on the optimal foothold and the target terrain feature data to obtain initial control parameters; The overall optimization problem is decomposed into five independent subsystems (Subsystem 1: the backbone; Subsystems 2-5: the four legs). Each subsystem solves the optimization problem independently based on local state and interaction information, thus achieving subsystem collaboration.
[0059] Step 117: Adjust the initial control parameters to obtain the target control parameters for the quadruped robot; The compensation control quantity is calculated based on the error data and equivalent disturbance, and then superimposed on the initial control parameters to obtain the target control parameters of the quadruped robot.
[0060] Step 118: Control the joint angles of the quadruped robot according to the target control parameters.
[0061] By converting the target control parameters into angle commands for the hip / knee joints, and then verifying the leg joint angle commands through inverse kinematics, the joint motors are directly driven by the "position closed-loop controller," ensuring that the commands conform to kinematic constraints and are tracked in real time throughout the process, thus achieving control of the quadruped robot.
[0062] This invention presents a distributed robust model predictive control method based on biomimetic foot optimization. This method optimizes the biomimetic foot using terrain feature encoding, reducing the risk of posture imbalance at its source. By constructing a "torso-quadruped" distributed control architecture, parallel solution and information coordination of subsystems are achieved, significantly reducing computational complexity. Simultaneously, an Eid estimator is embedded to capture external disturbances and modeling errors in real time, and a feedforward compensation mechanism is used to correct initial control parameters to improve system robustness.
[0063] like Figure 2 As shown, an embodiment of the present invention proposes a distributed robust model prediction control device 200 based on biomimetic foot optimization, comprising: The acquisition module 201 is used to acquire terrain point cloud data and foot contact force data through an acquisition device installed on the quadruped robot; The processing module 202 is used to preprocess the terrain point cloud data and the foot contact force data to obtain preprocessed data; extract terrain features from the preprocessed data to obtain target terrain feature data; obtain actual support point data based on the target terrain feature data and preset constraints; obtain the optimal foothold based on the actual support point data and preset multi-target data; perform global optimization processing based on the optimal foothold to obtain initial control parameters; adjust the initial control parameters to obtain the target control parameters of the quadruped robot; and control the joint angles of the quadruped robot based on the target control parameters.
[0064] Optionally, terrain point cloud data and foot contact force data are collected using a data acquisition device mounted on the quadruped robot, including: The surrounding terrain is scanned by a lidar mounted on a quadruped robot to obtain terrain point cloud data; Tactile sensors installed on the feet of the quadruped robot collect data on the contact force at the feet.
[0065] Optionally, the terrain point cloud data and the foot contact force data are preprocessed to obtain preprocessed data, including: The terrain point cloud data and the foot contact force data are aligned to obtain aligned data. The aligned data is then denoised to obtain denoised data. The denoised data is fused to obtain preprocessed data.
[0066] Optionally, terrain feature extraction is performed on the preprocessed data to obtain target terrain feature data, including: pass The terrain slope is obtained; among which, For terrain slope, This is the unit normal vector of the optimal ground plane obtained from the preprocessed data. This is the unit vector of gravity direction obtained from the preprocessed data; pass The step height is obtained; where, The height of the step. The average elevation of the ground plane obtained from the preprocessed data, The average elevation of the plane above the step, obtained from the preprocessed data; pass The ground friction coefficient is obtained; where, The coefficient of friction of the ground. The tangential force of the ground contact force obtained from the preprocessed data. The normal force of the ground contact force obtained from the preprocessed data; The target terrain feature data are obtained based on the terrain slope, the step height, and the ground friction coefficient.
[0067] Optionally, based on the target terrain feature data and preset constraints, the actual support point data is obtained, including: Obtain preset constraints; the preset constraints include the range of hip joint yaw angle, the range of hip joint roll angle, and the range of knee joint pitch angle. Based on the target terrain feature data and the preset constraints, the actual support point data is obtained; the actual support point data includes: stability priority support points, efficiency priority support points, and balance priority support points.
[0068] Optionally, based on the actual support point data and the preset evaluation function, the optimal foothold is obtained, including: pass The preset evaluation function value is obtained; where, To preset the evaluation function value, , , All are dynamic weights. To support the stability index, , The angle between the normal vector of the foot point and the direction of gravity. For trajectory tracking error, , For actual support point data, For the ideal fulcrum position, This is the error attenuation coefficient. This represents the energy expenditure index of the legs. , Let i be the angle of the i-th joint. This represents the maximum angle of the joint. The actual support point data is filtered based on the preset evaluation function value to obtain the optimal foothold.
[0069] Optionally, based on the optimal foothold and the target terrain feature data, global optimization processing is performed to obtain initial control parameters, including: Acquire real-time status data of the quadruped robot; the real-time status data includes the position of the torso center of mass, posture angles, joint angles, and velocity; Map the target terrain feature data to constraint parameters; Based on the optimal foothold and the target terrain feature data, the trunk reference state data and leg reference state data are obtained; Global optimization is performed based on the trunk reference state data and the leg reference state data to obtain initial control parameters; the initial control parameters include joint angle command data.
[0070] Optionally, the initial control parameters are adjusted to obtain the target control parameters for the quadruped robot, including: Obtain error data; The initial control parameters are adjusted based on the error data and the preset correction strategy to obtain the target control parameters for the quadruped robot.
[0071] The distributed robust model predictive control device based on biomimetic foot optimization in this invention collects terrain point cloud data and foot contact force data through a data acquisition device installed on a quadruped robot. The device preprocesses the terrain point cloud data and foot contact force data to obtain preprocessed data. Then, terrain features are extracted from the preprocessed data to obtain target terrain feature data. Based on the target terrain feature data and preset constraints, actual support point data is obtained. Based on the actual support point data and a preset evaluation function, the optimal foothold is obtained. Global optimization is performed based on the optimal foothold and the target terrain feature data to obtain initial control parameters. Finally, the initial control parameters are adjusted to obtain the target control parameters for the quadruped robot. This method can improve the accuracy of the quadruped robot's control parameters and reduce the risk of posture imbalance.
[0072] It should be noted that this device corresponds to the method described above, and all implementations in the method embodiments described above are applicable to the embodiments of this device and can achieve the same technical effect. Further details will not be provided in this embodiment.
[0073] This invention also provides a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method as described in any of the above embodiments. All implementations in the above method embodiments are applicable to the embodiments of this device and can achieve the same technical effects. Further details are omitted in this embodiment.
[0074] This invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method as described in any of the above embodiments. All implementations in the above method embodiments are applicable to the embodiments of this device and can achieve the same technical effects. Further details are omitted in this embodiment.
[0075] It should be noted that in the apparatus and method of the present invention, the components or steps can obviously be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Furthermore, the steps for performing the above series of processes can naturally be performed in the order described and in chronological order, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel, overlapping, or independently of each other.
[0076] It should be noted that in the above embodiments, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments described above is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.
[0077] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A distributed robust model predictive control method based on biomimetic foot optimization, characterized in that, include: Terrain point cloud data and foot contact force data are collected using a data acquisition device installed on the quadruped robot; The terrain point cloud data and the foot contact force data are preprocessed to obtain preprocessed data. The preprocessed data is subjected to terrain feature extraction to obtain target terrain feature data; Based on the target terrain feature data and preset constraints, the actual support point data is obtained; The optimal foothold is obtained based on the actual support point data and the preset evaluation function; Based on the optimal foothold and the target terrain feature data, global optimization processing is performed to obtain initial control parameters; The initial control parameters are adjusted to obtain the target control parameters for the quadruped robot; The joint angles of the quadruped robot are controlled according to the target control parameters.
2. The distributed robust model predictive control method based on bionic foot optimization according to claim 1, characterized in that, Terrain point cloud data and foot contact force data are collected using a data acquisition device mounted on the quadruped robot, including: The surrounding terrain is scanned by a lidar mounted on a quadruped robot to obtain terrain point cloud data; Tactile sensors installed on the feet of the quadruped robot collect data on the contact force at the feet.
3. The distributed robust model predictive control method based on bionic foot optimization according to claim 1, characterized in that, The terrain point cloud data and the foot contact force data are preprocessed to obtain preprocessed data, including: The terrain point cloud data and the foot contact force data are aligned to obtain aligned data. The aligned data is then denoised to obtain denoised data. The denoised data is fused to obtain preprocessed data.
4. The distributed robust model predictive control method based on bionic foot optimization according to claim 1, characterized in that, The preprocessed data is subjected to terrain feature extraction to obtain target terrain feature data, including: pass The terrain slope is obtained; among which, For terrain slope, This is the unit normal vector of the optimal ground plane obtained from the preprocessed data. This is the unit vector of gravity direction obtained from the preprocessed data; pass The step height is obtained; where, The height of the step. The average elevation of the ground plane obtained from the preprocessed data, This refers to the average elevation of the plane above the step, obtained from the preprocessed data. pass The ground friction coefficient is obtained; where, The coefficient of friction of the ground. The tangential force of the ground contact force obtained from the preprocessed data. The normal force of the ground contact force obtained from the preprocessed data; The target terrain feature data are obtained based on the terrain slope, the step height, and the ground friction coefficient.
5. The distributed robust model predictive control method based on bionic foot optimization according to claim 1, characterized in that, Based on the target terrain feature data and preset constraints, the actual support point data is obtained, including: Obtain preset constraints; the preset constraints include the range of hip joint yaw angle, the range of hip joint roll angle, and the range of knee joint pitch angle. Based on the target terrain feature data and the preset constraints, the actual support point data is obtained; the actual support point data includes: stability priority support points, efficiency priority support points, and balance priority support points.
6. The distributed robust model predictive control method based on biomimetic foot optimization according to claim 1, characterized in that, Based on the actual support point data and the preset evaluation function, the optimal foothold is obtained, including: pass The preset evaluation function value is obtained; where, To preset the evaluation function value, , , All are dynamic weights. To support the stability index, through get, The angle between the normal vector of the foot point and the direction of gravity. For trajectory tracking error, through get, For actual support point data, For the ideal fulcrum position, This is the error attenuation coefficient. The leg energy expenditure index is calculated by... get, Let i be the angle of the i-th joint. This represents the maximum angle of the joint. The actual support point data is filtered based on the preset evaluation function value to obtain the optimal foothold.
7. The distributed robust model predictive control method based on bionic foot optimization according to claim 1, characterized in that, Based on the optimal foothold and the target terrain feature data, global optimization is performed to obtain initial control parameters, including: Acquire real-time status data of the quadruped robot; the real-time status data includes the position of the torso center of mass, posture angles, joint angles, and velocities; Map the target terrain feature data to constraint parameters; Based on the optimal foothold and the target terrain feature data, the trunk reference state data and leg reference state data are obtained; Global optimization is performed based on the trunk reference state data and the leg reference state data to obtain initial control parameters; the initial control parameters include joint angle command data.
8. The distributed robust model predictive control method based on bionic foot optimization according to claim 1, characterized in that, The initial control parameters are adjusted to obtain the target control parameters for the quadruped robot, including: Obtain error data; The initial control parameters are adjusted based on the error data and the preset correction strategy to obtain the target control parameters for the quadruped robot.
9. A distributed robust model predictive control device based on biomimetic foot optimization, characterized in that, include: The data acquisition module is used to acquire terrain point cloud data and foot contact force data through a data acquisition device installed on the quadruped robot; The processing module is used to preprocess the terrain point cloud data and the foot contact force data to obtain preprocessed data; The preprocessed data is subjected to terrain feature extraction to obtain target terrain feature data; Based on the target terrain feature data and preset constraints, the actual support point data is obtained; Based on the actual support point data and preset multi-target data, the optimal foothold is obtained; global optimization is performed based on the optimal foothold to obtain initial control parameters; the initial control parameters are adjusted to obtain the target control parameters of the quadruped robot; and the joint angles of the quadruped robot are controlled based on the target control parameters.
10. A computing device, characterized in that, include: A processor, a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method as described in any one of claims 1 to 8.