Motion control method, device and equipment of speed difference steering wheel-legged robot, medium and product

Abstract

The invention discloses a motion control method, device and equipment of a speed difference steering wheel-legged robot, a medium and a product, and relates to the field of motion control. Performing mode switching according to the motion data of the target wheel-legged robot at the current moment, and determining a mode to which the target wheel-legged robot belongs at the current moment; adopting a strategy model to determine joint track information at the current moment according to the motion data at the current moment; and integrating the joint trajectory information at the current moment, optimizing the integrated state trajectory information based on a predictive control model to obtain the optimized expected data at the current moment, determining the expected state quantity at the current moment based on an inverse dynamics method, and further determining the feedback data at the next moment. And taking the motion data as the motion data at the current moment, and returning to carry out the modal switching step until the motion track of the target wheel-legged robot is adjusted to the expected track. The invention aims to improve the steering precision, the energy efficiency ratio and the dynamic stability.

Description

Technical Field

[0001] The present application relates to the field of motion control, and in particular to a motion control method, device, equipment, medium and product for a speed-differential steering wheel-legged robot. Background Art

[0002] With the growing demand for robotic mobility and obstacle-crossing capabilities in unstructured urban environments, hybrid motion systems have attracted widespread attention. Four-wheeled, legged robots, by organically combining the advantages of both wheeled and legged platforms, have become a cutting-edge solution for exploration and transportation tasks in complex indoor and outdoor environments. However, wheeled, legged robots, due to their complex configuration, also face numerous challenges:

[0003] 1. Structural conflicts of traditional wheel-legged robots.

[0004] The inherent high center of gravity of legged architectures requires external joint actuators to output extremely high torque during high-speed cornering, which places stringent demands on the joint motor drive capabilities. Therefore, traditional wheel-legged robots with external swing joints have inherent limitations in handling high-speed turning scenarios.

[0005] 2. Disadvantages of existing reinforcement learning control methods:

[0006] Sim-to-Real Gap: Because simulation environments cannot accurately model complex physical effects such as tire slip and ground friction, the performance of trained policies may degrade when transferred to reality (such as increased trajectory deviation or abnormal leg lift).

[0007] Low training efficiency: Strategy convergence requires massive training samples (usually millions to tens of millions of steps), and the reward function design is sensitive and the parameter adjustment cost is high.

[0008] 3. Disadvantages of existing model predictive control methods:

[0009] Strong model dependence: Control performance is highly dependent on accurate dynamic modeling (such as the tire slip rate-lateral force relationship) and is prone to failure in unknown or time-varying environments (such as sand → ice and snow).

[0010] High computational burden: High-speed maneuvers require high-frequency solving of nonlinear optimization problems (>100Hz), which places stringent requirements on hardware computing power and is difficult to deploy on low-power platforms.

[0011] Traditional wheel-legged robots rely on swing-out joints to generate steering torque, leading to the following core contradictions:

[0012] High energy consumption and low reliability: The outward swing joint needs to frequently output high torque during high-speed steering, resulting in increased power consumption.

[0013] Center of gravity stability defect: The outward swing joint drive is coupled with the high center of gravity configuration, and the lateral acceleration of the center of mass exceeds the limit during high-speed turning, resulting in insufficient joint torque and the risk of roll instability. Summary of the Invention

[0014] The purpose of this application is to provide a motion control method, device, equipment, medium and product for a speed-differential steering wheel-legged robot, which can improve steering accuracy, energy efficiency and dynamic stability.

[0015] To achieve the above objectives, this application provides the following solutions:

[0016] In a first aspect, the present application provides a motion control method for a speed-differential steering wheel-legged robot, comprising:

[0017] Performing mode switching according to the motion data of the target wheel-legged robot at the current moment to determine the mode at the current moment; the mode is a low-speed and obstacle crossing mode, or a high-speed and steering mode;

[0018] Based on the mode at the current moment, a strategy model is used to determine the joint trajectory information at the current moment according to the motion data at the current moment; the strategy model is determined by a reinforcement learning method; the joint trajectory information includes: the expected joint position and the expected wheel speed;

[0019] The joint trajectory information at the current moment is integrated, and the integrated state trajectory information is optimized based on the predictive control model to obtain the optimized expected data at the current moment, and the expected state quantity at the current moment is determined based on the inverse dynamics method; the expected state quantity is used to control the motion of the target wheel-legged robot; the expected state quantity includes: wheel torque and joint torque; the predictive control model is a mathematical model determined based on dynamic characteristic constraints and a cost function to characterize trajectory optimization;

[0020] The joint trajectory information at the current moment, the expected data after optimization at the current moment, and the expected state quantity at the current moment are determined as the feedback data at the next moment, and the feedback data at the next moment is used as the motion data at the current moment, and the function "modal switching according to the motion data of the target wheel-legged robot at the current moment is performed to determine the mode at the current moment" is returned until the motion trajectory of the target wheel-legged robot is adjusted to the expected trajectory.

[0021] In a second aspect, the present application provides a motion control device for a speed-differential steering wheel-legged robot, comprising:

[0022] A mode switching module is used to switch modes according to the motion data of the target wheel-legged robot at the current moment and determine the mode at the current moment; the mode is a low-speed and obstacle-crossing mode, or a high-speed and steering mode;

[0023] A joint trajectory information determination module is configured to determine the joint trajectory information at the current moment based on the current mode and the current motion data using a strategy model; the strategy model is determined using a reinforcement learning method; the joint trajectory information includes: expected joint positions and expected wheel speeds;

[0024] An integration and optimization module is configured to integrate the joint trajectory information at the current moment, optimize the integrated state trajectory information based on a predictive control model, obtain the optimized expected data at the current moment, and determine the expected state quantity at the current moment based on an inverse dynamics method; the expected state quantity is used to control the motion of the target wheel-legged robot; the expected state quantity includes: wheel torque and joint torque; the predictive control model is a mathematical model determined based on dynamic characteristic constraints and a cost function to characterize trajectory optimization;

[0025] The feedback execution module is used to determine the joint trajectory information at the current moment, the expected data after optimization at the current moment, and the expected state quantity at the current moment as the feedback data at the next moment, and use the feedback data at the next moment as the motion data at the current moment, and return it to the "modal switching module" until the motion trajectory of the target wheel-legged robot is adjusted to the expected trajectory.

[0026] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the motion control method of the speed-differential steering wheel-legged robot described above.

[0027] In a fourth aspect, the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the motion control method of the speed-differential steering wheel-legged robot described above is implemented.

[0028] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the motion control method of the speed-differential steering wheel-legged robot described above.

[0029] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0030] The present application provides a motion control method, device, equipment, medium and product for a speed-differential steering wheel-legged robot. The present application can enhance control robustness by combining a strategy model with a predictive control model, and then optimize the state trajectory information to perform motion control on the target wheel-legged robot, thereby adjusting the motion trajectory of the target wheel-legged robot. Since the strategy model is determined by a reinforcement learning method, and the strategy model outputs corresponding joint trajectory information for different modes, it can simultaneously optimize energy efficiency and improve motion stability. In addition, by using the feedback data at the next moment as the motion data at the current moment, returning to the mode switch until the motion trajectory of the target wheel-legged robot is adjusted to the desired trajectory, rolling optimization and feedback correction are achieved, which can improve steering accuracy and dynamic stability. Therefore, the present application can improve steering accuracy, energy efficiency and dynamic stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0032] Figure 1 This is a flow chart of the motion control method for a speed-differential steering wheel-legged robot;

[0033] Figure 2 Schematic diagram of the overall control framework of the motion control method for a speed-differential steering wheel-legged robot;

[0034] Figure 3 This is a structural diagram of the motion control device of the speed-differential steering wheel-legged robot;

[0035] Figure 4 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0037] This application proposes a novel motion control framework for wheel-legged robots. This framework integrates adaptive gait generation based on RL (Reinforcement Learning) with trajectory optimization and control based on MPC (Model Predictive Control), leveraging the strengths of both to enhance the high-speed steering capability and maneuverability of skid-steer wheel-legged robots in complex environments. Furthermore, a mode-switching mechanism is established to enable the robot to execute corresponding motion strategies at different speeds based on controller performance, thereby reducing the training difficulty of a single strategy and improving overall performance.

[0038] Reinforcement learning: A machine learning method that learns optimal decisions through interaction with the environment. Its core principle is that after an agent takes an action in a specific state, it adjusts its strategy based on the rewards provided by the environment to maximize long-term cumulative rewards. Model predictive control: A control method based on rolling optimization and feedback correction. This method solves the optimal control problem online within a finite time domain, generates the control input for the current moment, and then re-predicts and optimizes it at the next moment to achieve effective control of dynamic systems.

[0039] To address the issue of insufficient lateral force generation by external swing motors during high-speed steering, this application addresses the motion control of a speed-differential steering wheel-legged robot without external swing, structurally circumventing this limitation. This configuration enhances the robot-environment coupling effect at high speeds, necessitating the development of a novel control framework for engineering applications.

[0040] Advantages of reinforcement learning: 1. Strong environmental adaptability: RL can optimize motion strategies in complex, unstructured environments (such as rugged terrain and dynamic obstacles) through autonomous learning, without the need for precise dynamic modeling. 2. Multi-objective optimization capability: It can simultaneously optimize multiple objectives such as energy efficiency, motion stability, and speed tracking, and is suitable for high-dimensional action spaces (such as coordinated control of joint torque and wheel speed).

[0041] Advantages of Model Predictive Control: 1. Strong real-time optimization capabilities: Based on rolling horizon optimization, MPC can quickly respond to dynamic changes (such as sudden steering commands) and ensure short-term trajectory tracking accuracy (lateral error is typically <5cm). 2. Explicit processing of physical constraints: Dynamic equations (such as tire force models and center of mass motion constraints) can be embedded to avoid joint overruns or wheel-ground contact force instability.

[0042] This application proposes a non-outward swing joint speed difference steering configuration to resolve structural conflicts in the following ways:

[0043] Mechanical Streamlining: The outward swing joint is eliminated, and the wheel hub motor differential is used to directly generate steering torque, reducing system power consumption. Active Roll Compensation: A center of mass offset mechanism is designed to actively generate centripetal force through leg posture adjustment, making the lateral acceleration of the center of mass more stable during high-speed steering.

[0044] In view of the limitations of reinforcement learning (RL) control methods, existing RL methods have the following problems in the simulation-reality gap and training efficiency bottlenecks in wheeled legged robot control:

[0045] Inaccurate physical constraint modeling: The dynamic characteristics of tire slip steering forces (such as lateral force saturation) are ignored in simulation, resulting in large tracking errors after the strategy is transferred to the actual vehicle. In addition, when turning in the real vehicle, the longitudinal resistance generated by the tires is mistaken for contact with an obstacle, triggering abnormal leg lifting movements.

[0046] Poor adaptability of the motion space: The motion space designed for the outward swing joint (including the swing joint torque command) cannot be directly transferred to the speed difference steering configuration, which reduces the efficiency of strategy exploration.

[0047] Insufficient stability: RL-trained controllers are often only applicable to the scenarios they have been trained on, which may cause them to lose power and robustness in completely new environments.

[0048] This application develops a strategy switching mechanism to achieve the following breakthroughs:

[0049] Develop privileged observation information: Introducing privileged information such as ground contact sensors and elevation measurements during training significantly reduces the incidence of leg lifts after strategy transfer.

[0050] Reward function reconstruction: When training low-speed and obstacle-crossing strategies, the lateral velocity and yaw rate tracking rewards are removed, retaining only the longitudinal velocity tracking target, shortening the strategy convergence period. When training high-speed and turning strategies, a penalty term for leg lift is added, significantly reducing the incidence of leg lift after strategy migration.

[0051] Combined with MPC: After generating the trajectory through RL, it is optimized through MPC to enhance the robustness of the controller.

[0052] Traditional MPC faces bottlenecks in environmental adaptability and computational efficiency in wheeled-legged robot control:

[0053] Insufficient environmental adaptability: When it comes to unknown terrain, the design of a desired trajectory becomes difficult because the environmental information cannot be well modeled and encoded into the controller design.

[0054] Insufficient real-time performance: The trajectory solved by nonlinear programming requires high computing resources and is currently only used offline.

[0055] This application improves performance through the following innovations:

[0056] RL-generated reference trajectories: Dynamically adjust the desired joint trajectories in complex environments through RL training.

[0057] This application systematically solves the technical pain points of traditional wheel-legged robots in terms of structural contradictions, RL simulation migration bottlenecks, and insufficient adaptability to MPC environments through the three-in-one innovation of mechanical configuration innovation, control algorithm collaboration, and training paradigm upgrade, achieving a leap forward in high-speed steering accuracy, energy efficiency, and dynamic stability.

[0058] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0059] In an exemplary embodiment, Figure 1 As shown, a motion control method for a speed-differential steering wheel-legged robot is provided, comprising:

[0060] Step 100: Performing mode switching based on the current motion data of the target wheel-legged robot to determine the current mode, which can be a low-speed and obstacle-crossing mode or a high-speed and steering mode.

[0061] Among them, the mode switching is performed according to the motion data of the target wheel-legged robot at the current moment to determine the mode at the current moment, specifically including:

[0062] According to the longitudinal velocity and yaw angular velocity contained in the motion data of the target wheel-legged robot at the current moment, a modal judgment is performed to obtain a judgment result; the calculation formula used for the modal judgment is:

[0063]

[0064] Wherein, Vel_x is the longitudinal velocity; Vel_x_threshold is the longitudinal velocity threshold; Angular_z is the yaw angular velocity; and Angular_z_threshold is the yaw angular velocity threshold.

[0065] Mode switching is performed based on the judgment result to determine the mode at the current moment; if the judgment result is that the conditions in the calculation formula used for modal judgment are met, the mode at the current moment is the low-speed and obstacle crossing mode; if the judgment result is that the conditions in the calculation formula used for modal judgment are not met, the mode at the current moment is the high-speed and turning mode.

[0066] Step 200: Based on the current mode, the strategy model is used to determine the current joint trajectory information based on the current motion data. The strategy model is determined using reinforcement learning. The joint trajectory information includes the expected joint position and the expected wheel speed.

[0067] Step 300: Integrate the joint trajectory information at the current moment and optimize the integrated state trajectory information based on the predictive control model to obtain the optimized expected data at the current moment. The expected state quantity at the current moment is determined based on inverse dynamics methods. The expected state quantity is used to control the motion of the target wheel-legged robot; the expected state quantity includes wheel torque and joint torque. The predictive control model is a mathematical model that represents trajectory optimization based on dynamic characteristic constraints and a cost function.

[0068] The joint trajectory information at the current moment is integrated, and the integrated state trajectory information is optimized based on the predictive control model to obtain the optimized expected data at the current moment, and the expected state quantity at the current moment is determined based on the inverse dynamics method, specifically including:

[0069] The joint trajectory information at the current moment is integrated to obtain the integrated state trajectory information.

[0070] According to the integrated state trajectory information, the optimization is performed based on the dynamic characteristic constraints and cost function in the predictive control model to obtain the expected data after optimization at the current moment.

[0071] The inverse dynamics method is used to determine the expected state quantity at the current moment based on the wheel dynamics constraints and the expected data after optimization at the current moment.

[0072] The dynamic characteristic constraints include: skid steering kinematic constraints and wheel-ground contact constraints.

[0073] The expression corresponding to the skid steering kinematic constraint is:

[0074]

[0075] The corresponding expression of wheel-ground contact constraint is:

[0076]

[0077] Among them, F i x is the tire longitudinal force; F i y is the tire lateral force; C x is the longitudinal stiffness of the tire; C y is the tire cornering stiffness; α i is the side slip angle of the wheel; s i is the longitudinal slip rate; w 1,3 is the wheel speed of the left wheel; s 1,3 is the longitudinal slip rate of the left wheel; w 2,4 is the wheel speed of the right wheel; s2,4 is the longitudinal slip rate of the right wheel; θ z v is the heading angle of the robot coordinate system relative to the world coordinate system; y is the lateral velocity of the center of mass; v x is the longitudinal velocity of the center of mass; β is the sideslip angle of the center of mass; r w is the wheel radius; ω z is the yaw angular velocity of the robot; B is the wheel track.

[0078] The expression of the cost function is:

[0079]

[0080] Among them, l(x(t),u(t),t) is the cost function; x(t) is the state quantity; u(t) is the input quantity; t is time; is the state vector error; Q is the semi-positive definite Hessian matrix corresponding to the state vector; is the input vector error; R is the semi-positive definite Hessian matrix corresponding to the input vector; T represents the transpose.

[0081] The wheel dynamics constraint expression is:

[0082]

[0083] Among them, I w is the moment of inertia of the wheel; T i is the torque of the i-th wheel; is the wheel acceleration; is the contact force between the wheel and the ground in the x-direction of the robot; yaw is the angle between the wheel motion direction and the world coordinate system; is the contact force between the wheel and the ground in the y direction of the robot.

[0084] Step 400: The joint trajectory information at the current moment, the expected data after optimization at the current moment, and the expected state quantity at the current moment are determined as the feedback data at the next moment, and the feedback data at the next moment is used as the motion data at the current moment, and returns to "modal switching according to the motion data of the target wheel-legged robot at the current moment, and determining the mode at the current moment" until the motion trajectory of the target wheel-legged robot is adjusted to the expected trajectory.

[0085] This application mainly addresses the problem of wheel-legged robots being prone to instability during high-speed steering, and proposes a speed-adaptive RL-MPC collaborative control architecture. The framework mainly consists of five parts: a speed-based mode switching module, a low-speed and obstacle crossing mode, a high-speed and steering mode, a reference trajectory integration module, and a model predictive control module. The overall control framework is as follows: Figure 2 shown.

[0086] The overall operation process of the control system proposed in this application is as follows:

[0087] Step 1: After the user enters the desired speed (longitudinal speed and yaw angular velocity), the robot system will first determine whether the speed is low speed without steering or high speed with steering based on the current state feedback. The judgment formula is as follows:

[0088]

[0089] If all of the above conditions are met, the vehicle is in low-speed, no-steering mode, meaning the current mode is low-speed and obstacle negotiating. Otherwise, the vehicle is in high-speed, steering mode, meaning the current mode is high-speed and steering. If the current mode is low-speed and obstacle negotiating, the strategy model corresponding to low-speed and obstacle negotiating is used, and the reference trajectory integration module receives the trajectory information output by the strategy model for low-speed and obstacle negotiating. Otherwise, the reference trajectory integration module receives the trajectory information output by the strategy model for high-speed and steering.

[0090] Step 2: The current mode is obtained through step 1. If it is the low-speed and obstacle crossing mode, the current state is received (the current state is obtained and fed back based on the sensor equipment carried by the robot system) and the strategy model is used to output the joint trajectory information at the current moment to the reference trajectory integration module.

[0091] The policy model is determined using reinforcement learning by training a target wheeled-legged robot (hereafter referred to as the robot) in obstacle-crossing scenarios. The policy model inputs are the desired velocity and the current robot state, and the output is the desired position of each joint and the desired wheel velocity, i.e., the joint trajectory information. This model enables the robot to overcome complex obstacles.

[0092] The policy model for high-speed and steering mode receives the current state and outputs the policy learned by training the robot on complex road scenarios to the reference trajectory integration module. In this mode, the robot is capable of high-speed driving and steering in complex road environments. The input is the user-entered desired speed, and the output is the desired position of each joint and the desired wheel speed (i.e., joint trajectory information).

[0093] Regarding the policy model determined by the reinforcement learning method, the Proximal Policy Optimization (PPO) algorithm is adopted, the architecture is an actor-critic architecture, and the policy network uses a multilayer perceptron (MLP). Based on the initial policy network, the agent is trained and iteratively updated with the goal of maximizing the cumulative reward. Specifically, the following steps are performed:

[0094] Use simulators such as isaacsim to simulate the robot and obtain data. At each time step, the following operations are performed:

[0095] Determine the state; select an action based on the state; give the action to the robot control. At this time, the current agent is transformed to the new state through physical simulation, and a reward is generated and passed to the agent. Reward function R t The calculation is as follows:

[0096] R t =w1·R1+w2·R2…w n ·R n .

[0097] w i is the weight coefficient corresponding to the i-th action, which needs to be adjusted according to the task. i Calculate the reward function for the i-th action. 1≤i≤n, where n is the total number of actions.

[0098] Based on the two modes, there are the following reward functions:

[0099] 1. Low-speed and obstacle-crossing mode strategy training.

[0100] This mode uses a reinforcement learning algorithm to train the wheel-legged robot's obstacle-crossing ability during linear motion. When setting the control instructions, the random expected linear velocity Vel_x_threshold interval is set to:

[0101] [-Vel_x_threshold, Vel_x_threshold].

[0102] The desired angular velocity Angular_z_threshold interval is set to:

[0103] [-Angular_z_threshold, Angular_z_threshold].

[0104] The reward settings are shown in Table 1.

[0105] Table 1 Reward function setting table for low speed and obstacle crossing mode

[0106]

[0107]

[0108] In addition, in the terrain settings, the settings include terrain such as steps, slopes, and pits.

[0109] 2. High-speed and turning mode strategy training.

[0110] Compared to the Low Speed and Obstacle Climbing modes, this mode adds a steering function. To prevent abnormal leg lift during steering, which can affect vehicle stability by misinterpreting the combined sliding force of the tires as resistance from obstacles, this application adds a wheelie penalty to the rewards in Table 1 to reduce the frequency of abnormal leg lift. Furthermore, the training terrain in this mode mirrors traditional vehicle traversal, including undulating roads, gravel roads, and rolling roads.

[0111] Step 3: After the joint trajectory information (the expected joint position and the expected wheel speed) is calculated by the reinforcement learning strategy, the trajectory needs to be integrated.

[0112] In addition to joint trajectory information, the state trajectory required by MPC also includes: center of mass position, center of mass velocity, center of mass attitude angle, and center of mass angular velocity. The center of mass velocity and angular velocity can be directly substituted into the desired velocity and angular velocity (all other velocity quantities except the longitudinal velocity and yaw angular velocity are set to 0).

[0113] The center of mass position is calculated by the desired velocity and angular velocity after a period of time (the rolling time domain of MPC). The yaw angle is calculated by multiplying the angular velocity by time. The desired pitch angle and roll angle are set to 0.

[0114] The desired position and attitude angle are obtained based on the linear velocity, angular velocity and MPC optimization time domain calculation. The calculation method is as follows:

[0115]

[0116] Where yaw is the heading angle of the robot coordinate system relative to the world coordinate system; yaw_rat is the yaw angular velocity; roll is the roll angle of the robot coordinate system relative to the world coordinate system; and pitch is the pitch angle of the robot coordinate system relative to the world coordinate system.

[0117] Step 4: The model predictive control module optimizes the state trajectory.

[0118] According to the integrated state trajectory information, the optimization is performed based on the dynamic characteristic constraints and cost function in the predictive control model to obtain the expected data after optimization at the current moment.

[0119] Through the kinematic constraints of skid steering and wheel-ground contact constraints, the optimized trajectory is made to meet the dynamic characteristics, and the desired foot-end force, joint velocity, and wheel acceleration are solved.

[0120] When a speed differential skid-steering robot is turning, the tire motion exhibits a combined lateral and longitudinal slip characteristic. If the platform tire slip rate is small and the tire lateral force is small, the nonlinear effect of the tire is relatively weak. In this case, the tire can be approximated as a linear model. Here, the tire model is introduced to constrain the relationship between the tire contact force and the wheel speed, that is, the tire longitudinal force F i x It is linearly related to the slip rate, and the lateral force F i y It is linearly related to the sideslip angle and can be expressed as follows:

[0121]

[0122]

[0123] Among them, V x is the longitudinal velocity in the vehicle coordinate system; V y is the lateral velocity in the vehicle coordinate system; r w is the wheel radius; L 12 is the wheelbase; L 13 is the wheelbase. i is the wheel speed of the i-th wheel.

[0124] In the contact phase, the kinematic model considering tire slip is given by the following formula:

[0125]

[0126] Among them, w 1,3 is the wheel speed of the left wheel, w 2,4 is the wheel speed of the right wheel, s 1,3 is the longitudinal slip rate of the left wheel, s 2,4 is the longitudinal slip rate of the right wheel, is the sideslip angle of the center of mass.

[0127] The motion controller proposed in this application features a universal interface that allows it to track velocity in pure wheeled mode or various standard gaits in legged mode, as well as to track designed wheel-legged gaits and desired trajectories. This is achieved by simply providing different reference trajectories and calculating the cost at time t using the following formula:

[0128]

[0129] in, is the state vector error, is the input vector error, Q and R are the corresponding semi-positive Hessian matrices. ref (t) is the expected state quantity of the reference trajectory integration; u ref (t) is the expected input quantity of the reference trajectory integration.

[0130] Step 5: All the state quantities required for inverse dynamics have been calculated through MPC (state quantities in model predictive control and the desired foot end force, joint velocity, and wheel acceleration obtained by solution). At this point, only some inverse dynamics-based methods, such as whole body control (WBC), need to be used to solve the torque of the corresponding joints and wheels. The wheel dynamics constraints must be met during wheel rotation:

[0131]

[0132] Among them, I w is the moment of inertia of the wheel.

[0133] Step 6: The robot actuators (motors) track the optimized joint positions, speeds, and torques, as well as the optimized wheel speeds and torques, and feed back the current measurements through sensors.

[0134] In an exemplary embodiment, Figure 3 As shown, a motion control device for a speed-differential steering wheel-legged robot is provided, comprising:

[0135] The mode switching module is used to switch modes according to the motion data of the target wheel-legged robot at the current moment and determine the mode at the current moment; the mode is a low-speed and obstacle crossing mode, or a high-speed and steering mode.

[0136] The joint trajectory information determination module is used to determine the joint trajectory information at the current moment based on the current mode and the current motion data using a strategy model; the strategy model is determined using a reinforcement learning method; the joint trajectory information includes: the expected joint position and the expected wheel speed.

[0137] The integrated optimization module is used to integrate the joint trajectory information at the current moment, and optimize the integrated state trajectory information based on the predictive control model to obtain the expected data after optimization at the current moment, and determine the expected state quantity at the current moment based on the inverse dynamics method; the expected state quantity is used to control the motion of the target wheel-legged robot; the expected state quantity includes: wheel torque and joint torque; the predictive control model is a mathematical model determined based on dynamic characteristic constraints and cost function to characterize trajectory optimization.

[0138] The feedback execution module is used to determine the joint trajectory information at the current moment, the expected data after optimization at the current moment, and the expected state quantity at the current moment as the feedback data at the next moment, and use the feedback data at the next moment as the motion data at the current moment, and return it to the "modal switching module" until the motion trajectory of the target wheel-legged robot is adjusted to the expected trajectory.

[0139] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store motion control data of a speed-differential steering wheel-legged robot. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a motion control method of a speed-differential steering wheel-legged robot is implemented.

[0140] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0141] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0142] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0143] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0144] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0145] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0146] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0147] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0148] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A motion control method for a speed-differential steering wheel-legged robot, characterized in that: include: Perform mode switching based on the motion data of the target wheel-legged robot at the current moment to determine the mode it belongs to at the current moment; The mode is a low-speed and obstacle-crossing mode, or a high-speed and turning mode; Based on the current mode, the strategy model is used to determine the joint trajectory information at the current moment according to the current motion data; The strategy model is determined by using a reinforcement learning method; The joint trajectory information includes: expected joint positions and expected wheel speeds; The joint trajectory information at the current moment is integrated, and the integrated state trajectory information is optimized based on the predictive control model to obtain the optimized expected data at the current moment, and the expected state quantity at the current moment is determined based on the inverse dynamics method; the expected state quantity is used to control the motion of the target wheel-legged robot; the expected state quantity includes: wheel torque and joint torque; the predictive control model is a mathematical model determined based on dynamic characteristic constraints and a cost function to characterize trajectory optimization; The joint trajectory information at the current moment, the expected data after optimization at the current moment, and the expected state quantity at the current moment are determined as the feedback data at the next moment. The feedback data at the next moment is used as the motion data at the current moment. The method "Switching modes according to the motion data of the target wheel-legged robot at the current moment to determine the mode at the current moment" is returned until the motion trajectory of the target wheel-legged robot is adjusted to the expected trajectory.

2. The motion control method of the speed-differential steering wheel-legged robot according to claim 1 is characterized in that: Perform mode switching based on the target wheel-legged robot's motion data at the current moment to determine the mode it belongs to at the current moment, specifically including: According to the longitudinal velocity and yaw angular velocity contained in the motion data of the target wheel-legged robot at the current moment, a modal judgment is performed to obtain a judgment result; the calculation formula used for the modal judgment is: Among them, Vel_x is the longitudinal velocity; Vel_x_threshold is the longitudinal velocity threshold; Angular_z is the yaw angular velocity; Angular_z_threshold is the yaw angular velocity threshold; Mode switching is performed based on the judgment result to determine the mode at the current moment; if the judgment result is that the conditions in the calculation formula used for modal judgment are met, then the mode at the current moment is the low speed and obstacle crossing mode; if the judgment result is that the conditions in the calculation formula used for modal judgment are not met, then the mode at the current moment is the high speed and turning mode.

3. The motion control method of the speed-differential steering wheel-legged robot according to claim 1, characterized in that: The joint trajectory information at the current moment is integrated, and the integrated state trajectory information is optimized based on the predictive control model to obtain the optimized expected data at the current moment, and the expected state quantity at the current moment is determined based on the inverse dynamics method, specifically including: Integrate the joint trajectory information at the current moment to obtain the integrated state trajectory information; According to the integrated state trajectory information, the optimization is performed based on the dynamic characteristic constraints and cost function in the predictive control model to obtain the expected data after optimization at the current moment; The inverse dynamics method is used to determine the expected state quantity at the current moment based on the wheel dynamics constraints and the expected data after optimization at the current moment.

4. The motion control method of the speed-differential steering wheel-legged robot according to claim 3 is characterized in that: The dynamic characteristic constraints include: skid steering kinematic constraints and wheel-ground contact constraints; The expression corresponding to the skid steering kinematic constraint is: The expression corresponding to the wheel-ground contact constraint is: Among them, F i x is the tire longitudinal force; F i y is the tire lateral force; C x is the longitudinal stiffness of the tire; C y is the tire cornering stiffness; α i is the side slip angle of the wheel; s i is the longitudinal slip rate; w 1,3 is the wheel speed of the left wheel; s 1,3 is the longitudinal slip rate of the left wheel; w 2,4 is the wheel speed of the right wheel; s 2,4 is the longitudinal slip rate of the right wheel; θ z v is the heading angle of the robot coordinate system relative to the world coordinate system; y is the lateral velocity of the center of mass; v x is the longitudinal velocity of the center of mass; β is the sideslip angle of the center of mass; r w is the wheel radius; ω z is the yaw angular velocity of the robot; B is the wheel track.

5. The motion control method of the speed-differential steering wheel-legged robot according to claim 3 is characterized in that: The expression of the cost function is: Among them, l(x(t),u(t),t) is the cost function; x(t) is the state quantity; u(t) is the input quantity; t is time; is the state vector error; Q is the semi-positive definite Hessian matrix corresponding to the state vector; is the input vector error; R is the semi-positive definite Hessian matrix corresponding to the input vector; T represents the transpose.

6. The motion control method of the speed-differential steering wheel-legged robot according to claim 3, characterized in that: The wheel dynamics constraint is expressed as: Among them, I w is the moment of inertia of the wheel; T i is the torque of the i-th wheel; is the wheel acceleration; is the contact force between the wheel and the ground in the x-direction of the robot; yaw is the angle between the wheel motion direction and the world coordinate system; is the contact force between the wheel and the ground in the y direction of the robot.

7. A motion control device for a speed-differential steering wheel-legged robot, characterized in that: include: A mode switching module is used to switch modes according to the motion data of the target wheel-legged robot at the current moment and determine the mode at the current moment; The mode is a low-speed and obstacle-crossing mode, or a high-speed and turning mode; The joint trajectory information determination module is used to determine the joint trajectory information at the current moment based on the current mode and the current motion data using a strategy model; The strategy model is determined by using a reinforcement learning method; The joint trajectory information includes: expected joint positions and expected wheel speeds; An integration and optimization module is configured to integrate the joint trajectory information at the current moment, optimize the integrated state trajectory information based on a predictive control model, obtain the optimized expected data at the current moment, and determine the expected state quantity at the current moment based on an inverse dynamics method; the expected state quantity is used to control the motion of the target wheel-legged robot; the expected state quantity includes: wheel torque and joint torque; the predictive control model is a mathematical model determined based on dynamic characteristic constraints and a cost function to characterize trajectory optimization; The feedback execution module is used to determine the joint trajectory information at the current moment, the expected data after optimization at the current moment, and the expected state quantity at the current moment as the feedback data at the next moment. The feedback data at the next moment is then used as the motion data at the current moment and returned to the "modal switching module" until the motion trajectory of the target wheel-legged robot is adjusted to the expected trajectory.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the motion control method of the speed-differential steering wheel-legged robot according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the motion control method of the speed-differential steering wheel-legged robot according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the motion control method of the speed-differential steering wheel-legged robot according to any one of claims 1 to 6 is implemented.

Images