A wheel-foot composite robot, a motion control method thereof, a terminal and a storage medium

Abstract

The application discloses a wheel-foot composite robot, a motion control method thereof, a terminal and a storage medium, and the motion control method of the wheel-foot composite robot is characterized in that the four-foot mode and the two-wheel mode are respectively controlled adaptively under a unified control framework, so that the wheel-foot composite robot has remarkable robustness and dynamic stability in a complex environment. Specifically, the current motion mode is identified, the error between the expected state and the actual state is estimated by using a robust adaptive model predictive control algorithm in the four-foot mode, an equivalent external disturbance vector is obtained and is used for control solution, so as to realize online disturbance compensation and adaptive torque distribution, so that the robot has stronger anti-interference ability to load change and ground disturbance in the four-foot mode, and can maintain stable gait and accurate attitude control under the condition of a higher load.

Description

Technical Field

[0001] This invention relates to the field of robotics, and in particular to a wheel-legged composite robot and its motion control method, terminal, and storage medium. Background Technology

[0002] Currently, mobile robot technology is evolving from single wheeled or legged structures to multimodal motion systems. While traditional wheeled robots offer high speed and energy efficiency, they suffer from poor terrain adaptability and struggle to maintain stable movement in rugged, unstructured environments. Legged robots, on the other hand, can move across complex terrain through leg coordination, but their motion efficiency is low, control complexity is high, and stability under high loads is insufficient. With the increasing demand for service robots, search and rescue robots, and special-purpose robots, balancing motion efficiency and terrain adaptability in complex environments has become a crucial research direction in this field.

[0003] To improve the overall performance of mobile robots, hybrid wheel-legged robot structures have been proposed in recent years. These structures integrate both wheeled and legged motion units within the same robot system, enabling automatic switching between motion modes in different scenarios. On smooth surfaces, the robot uses wheels to improve speed and energy efficiency; on complex or uneven terrain, it switches to legged walking to enhance obstacle-crossing ability and posture stability. However, existing hybrid wheel-legged robots still have significant shortcomings in control. On the one hand, the dynamic models for different motion modes differ considerably, making continuous switching within the same control framework difficult. On the other hand, existing control algorithms are mostly based on ideal dynamics assumptions, lacking real-time estimation and compensation for external disturbances and load changes. This leads to problems such as posture deviations, gait instability, and even tipping over under high loads or impact conditions.

[0004] Traditional model predictive control (MPC), while capable of achieving some trajectory tracking through dynamic constraint optimization, suffers from weak responsiveness to unknown disturbances. When the robot carries external loads or operates in complex terrain environments, deviations between the predictive model and the actual system lead to decreased control accuracy. Furthermore, while whole-body control (WBC) can distribute torques in multi-degree-of-freedom systems, it lacks a disturbance feedback mechanism, failing to actively compensate for external influences and thus limiting the overall system stability. For two-wheeled standing posture control, the commonly used linear inverted pendulum (LIP) model, although computationally simple, neglects the dynamic coupling between the upper body and waist, making it difficult to maintain balance during high-speed movement or impacts. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a wheel-legged composite robot and its motion control method, terminal and storage medium, which can achieve stable gait and precise posture control under high load conditions.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] A motion control method for a wheel-legged hybrid robot, applied to a wheel-legged hybrid robot, wherein the motion modes of the wheel-legged hybrid robot include a quadrupedal mode and a two-wheeled mode, characterized by comprising the following steps:

[0008] S1. Detect the current motion pattern of the wheel-legged composite robot;

[0009] S2. If in a quadrupedal state, obtain the desired state trajectory and actual state trajectory of the wheel-legged composite robot.

[0010] Based on the desired state trajectory and the actual state trajectory, a tracking error vector is calculated and generated. The tracking error vector is then integrated using the RA-MPC model to generate an equivalent external disturbance vector.

[0011] A first torque command is generated based on the external disturbance vector, and the first torque command is used to control the joints of the wheel-legged composite robot to perform actions in the quadrupedal form.

[0012] S3. If it is in a two-wheeled state, obtain the desired state vector of the wheel-foot composite robot;

[0013] Based on the WLIP-WT model which includes the waist and upper body dynamics of the wheel-legged composite robot, the optimal torque required to maintain sagittal plane balance is generated using the desired state vector as input.

[0014] The virtual leg thrust is calculated and generated based on the base height data and roll angle data in the desired state vector.

[0015] The optimal torque and the virtual leg thrust are combined into a virtual leg action vector, and the virtual leg action vector is mapped into a second torque command. The second torque command is used to control the joints of the wheel-leg composite robot to perform actions in the two-wheel configuration.

[0016] To solve the above-mentioned technical problems, the present invention adopts other technical solutions as follows:

[0017] A terminal includes a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that the processor executes the computer program to implement steps in a motion control method for a wheel-legged composite robot.

[0018] A storage medium, characterized in that the storage medium stores a computer program, which, when executed by a processor, implements the steps of a motion control method for a wheel-legged composite robot.

[0019] A wheel-legged composite robot is provided for performing the steps in a motion control method for a wheel-legged composite robot, including:

[0020] Two forelegs formed by point-like foot tips;

[0021] Two rear legs, each of which includes a linkage mechanism and a drive wheel, the linkage mechanism and the drive wheel being kinetically connected;

[0022] The connecting base includes a load-bearing housing and a waist joint; the linkage mechanism is drivenly connected to the load-bearing housing; the waist joint is fixedly connected to the load-bearing housing, and the waist joint is drivenly connected to the two front legs respectively; so as to realize the switching between four forms supported by the two front legs and the two hind legs simultaneously and the two-wheel form supported by the two hind legs.

[0023] The beneficial effects of this invention are as follows: It provides a wheel-legged hybrid robot and its motion control method, terminal, and storage medium. By adaptively controlling both the quadrupedal and two-wheeled configurations under a unified control framework, the wheel-legged hybrid robot exhibits significant robustness and dynamic stability in complex environments. Specifically, by identifying the current motion configuration and using the Robust Adaptive Model Predictive Control (RA-MPC) algorithm in quadrupedal configuration to perform real-time integral estimation of the error between the desired and actual states, an equivalent external disturbance vector is obtained and used for control solution. This achieves online disturbance compensation and adaptive torque allocation, enabling the robot to have stronger anti-interference capabilities against load changes and ground disturbances in quadrupedal configuration, and to maintain stable gait and precise attitude control under high load conditions.

[0024] When the robot is in two-wheeled mode, based on a wheeled linear inverted pendulum model (WLIP-WT) incorporating waist and upper body dynamics, the system uses the desired state vector as input. Through model predictive control, it generates the optimal torque to maintain sagittal plane balance and combines this with virtual leg thrust calculated by virtual model control (VMC) to achieve real-time balance adjustment in the coronal plane. This control method effectively suppresses posture deviations caused by upper body movement or external disturbances during two-wheeled standing motion, maintaining balance and motion continuity. By synthesizing torque and thrust in virtual space and mapping them to physical joint torques, the control signals achieve a smooth transition, enabling the robot to maintain stable and coordinated posture control during mode switching and dynamic movements. Attached Figure Description

[0025] Figure 1This is a flowchart of a motion control method for a wheel-legged composite robot according to an embodiment of the present invention;

[0026] Figure 2 This is a schematic diagram of a quadrupedal posture of a wheel-legged composite robot according to an embodiment of the present invention;

[0027] Figure 3 This is a schematic diagram of the two-wheeled posture of a wheel-legged composite robot according to an embodiment of the present invention;

[0028] Label Explanation:

[0029] 1. Foreleg; 11. Hip joint; 12. Thigh segment; 13. Lower leg segment; 14. Roll motor; 15. Thigh pitch motor; 16. Lower leg pitch motor; 17. Point-type foot;

[0030] 2. Hind leg; 21. Linkage mechanism; 211. Hip joint drive motor; 22. Drive wheel; 221. Hub motor; 222. Pneumatic tire;

[0031] 3. Connecting base; 31. Bearing housing; 32. Waist joint; 321. Waist pitch motor; 322. Carbon fiber tube; 33. Limiting block; 34. Control center. Detailed Implementation

[0032] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0033] Before detailing the embodiments of this application, some related concepts will first be explained:

[0034] Robust Adaptive Model Predictive Control (RA-MPC) is a control method that introduces an adaptive disturbance estimation mechanism on top of traditional Model Predictive Control (MPC). This method introduces an external disturbance estimator into the predictive control framework. By calculating the error between the desired and actual system states in real time, and using integral or observer methods to infer the equivalent external disturbance, it maintains stable control performance even in the presence of uncertainties or external disturbances. This control approach is typically used to improve the system's robustness against parameter variations and external disturbances, ensuring that prediction results remain accurate and continuous under different operating conditions.

[0035] The WLIP-WT model (Wheeled Linear Inverted Pendulum with Waist and Upper Body) is an improved model based on the traditional wheeled inverted pendulum dynamics model, adding waist pitch degrees of freedom and upper body dynamics terms. This model introduces dynamic terms related to waist pitch and upper body mass distribution to more accurately characterize the system's attitude coupling and torque transmission relationships in the sagittal plane. The WLIP-WT model balances linearization of dynamics with structural integrity and is commonly used for attitude control and torque distribution analysis of two-wheeled robots or self-balancing devices.

[0036] Model Predictive Control (MPC) is an optimization control method based on a system predictive model. It solves a constrained optimization problem within a finite prediction time domain to obtain the optimal control input for a future period. This method performs new predictions and optimizations on a rolling basis in each control cycle, enabling the system to dynamically adjust its control strategy according to real-time state changes. Typical characteristics of MPC include its ability to handle multiple variables simultaneously, balance both hard and soft constraints, and maintain high control accuracy and stability.

[0037] Whole-Body Control (WBC) is a coordinated control framework for multi-degree-of-freedom robots or complex mechanical systems. Its core idea is to comprehensively consider the system's dynamic equations, constraints, and task priorities within a single optimization process, using mathematical forms such as quadratic programming to solve for the optimal torque or force distribution at each joint. This method can achieve real-time balance among multiple control objectives, such as posture maintenance, trajectory tracking, and mechanical stability, and is commonly used for the coordinated control of whole-body movements in multi-legged robots, humanoid robots, and exoskeleton systems.

[0038] Virtual Model Control (VMC) is a control method based on virtual mechanical modeling. Its principle is to construct virtual spring, damper, or thrust models corresponding to physical structures within human-machine interface or motion control systems, calculating control inputs in the form of virtual forces. In this way, the controller can generate target forces for joints or actuators in a physically intuitive manner, thereby achieving compliant control, balance maintenance, and posture adjustment. VMC is characterized by its simplicity of implementation, fast response, and strong interpretability, and is widely used in the motion control of legged robots and self-balancing devices.

[0039] A virtual leg is an abstract model of the linkage structure in a mechanical system, connecting the hip and knee joints to end effectors (such as the foot or axle). This model simplifies a multi-link system into a single virtual link with length, angle, and force direction, used to describe the overall support or driving mechanical properties. The virtual leg concept facilitates the representation of support forces and motion states in vector form in control algorithms, thereby simplifying dynamics solutions and control mapping calculations. It is commonly used in the analysis and control design of walking robots, balancing vehicles, and biomimetic structures.

[0040] In existing technologies, wheel-legged hybrid robots, as a novel mobile platform combining the advantages of wheeled and legged locomotion, have been widely used in service robots, rescue robots, and special inspection equipment. These robots typically achieve efficient movement through wheel sets and switch to legged walking in complex terrain or obstacle environments to improve terrain adaptability and obstacle-crossing capabilities. However, existing wheel-legged hybrid robots generally suffer from insufficient control stability. On the one hand, the dynamic models differ significantly under different configurations, making it difficult for the control system to smoothly switch between the two modes, easily leading to abrupt posture changes or discontinuous motion. On the other hand, in complex environments such as load changes, uneven ground, or external disturbances, traditional control methods have limited responsiveness to unknown disturbances and cannot achieve real-time compensation, causing the robot to be prone to gait instability, posture deviation, or even imbalance. Furthermore, while existing model predictive control algorithms can achieve trajectory planning, their robustness to model errors and external disturbances is weak, making it difficult to meet the precise control requirements of high-load, high-dynamic motion scenarios.

[0041] To at least solve the above problems, please refer to Figure 1 This invention provides a motion control method for a wheel-legged composite robot, applicable to a wheel-legged composite robot whose motion modes include a quadrupedal mode and a two-wheeled mode, including the following steps:

[0042] S1. Detect the current motion pattern of the wheel-legged composite robot;

[0043] S2. If in a quadrupedal state, obtain the desired state trajectory and actual state trajectory of the wheel-legged composite robot.

[0044] Based on the desired state trajectory and the actual state trajectory, a tracking error vector is calculated and generated. The tracking error vector is then integrated using the RA-MPC model to generate an equivalent external disturbance vector.

[0045] A first torque command is generated based on the external disturbance vector, and the first torque command is used to control the joints of the wheel-legged composite robot to perform actions in the quadrupedal form.

[0046] S3. If it is in a two-wheeled state, obtain the desired state vector of the wheel-foot composite robot;

[0047] Based on the WLIP-WT model which includes the waist and upper body dynamics of the wheel-legged composite robot, the optimal torque required to maintain sagittal plane balance is generated using the desired state vector as input.

[0048] The virtual leg thrust is calculated and generated based on the base height data and roll angle data in the desired state vector.

[0049] The optimal torque and the virtual leg thrust are combined into a virtual leg action vector, and the virtual leg action vector is mapped into a second torque command. The second torque command is used to control the joints of the wheel-leg composite robot to perform actions in the two-wheel configuration.

[0050] As can be seen from the above description, the beneficial effects of this invention are: achieving stable control and dynamic balance of the robot under different motion modes through perturbation estimation and model decoupling. This method, targeting the structural characteristics of a wheel-legged hybrid robot possessing both quadrupedal support and two-wheeled upright motion modes, designs a unified control flow that automatically switches to the corresponding control branch when different motion modes are detected, thereby achieving adaptive motion adjustment under multimodal conditions. Specifically:

[0051] When the robot starts, it first determines its current motion state through an attitude detection module. Based on attitude angle, angular velocity, and acceleration information collected by the inertial measurement unit (IMU), this module can quickly identify whether the robot is in a quadrupedal support state or a two-wheeled upright state. This detection step provides the basic input for subsequent control, enabling the system to select the corresponding control logic based on the attitude information. This ensures continuous control paths and smooth commands during motion transitions, thus avoiding the delays and instabilities that occur in traditional systems during state transitions.

[0052] When the robot is in quadruped mode, the control system generates the desired trajectory based on the task planning results. Simultaneously, it acquires the actual trajectory outputs from the encoders and IMUs of each joint in real time and calculates the difference between them to obtain the tracking error vector. This error signal is processed by the Robust Adaptive Model Predictive Control (RA-MPC) algorithm. RA-MPC achieves real-time estimation of unknown external forces by introducing disturbance observation and integration, generating an equivalent external disturbance vector. This disturbance vector reflects the effects of uneven ground, friction changes, or load fluctuations experienced by the robot during walking. Based on this vector, the control system generates a first torque command to correct the outputs of each joint. RA-MPC balances predictability and adaptability, enabling the control system to achieve real-time compensation under high load conditions and maintain stable walking.

[0053] In this process, the introduction of external disturbances is not only used for error compensation but also serves as an input signal in torque calculation, achieving a dual utilization of disturbance information. By integrating the disturbance input model predictive controller with the whole-body controller, the system obtains the optimal ground reaction force at the global level and generates real-time feedforward compensation terms at the local level, forming a control structure that combines prediction and compensation. This approach enhances the system's disturbance immunity while ensuring the continuity of torque distribution, enabling the robot to maintain gait stability and precise attitude control under uncertain terrain and load conditions.

[0054] When the detection results indicate that the robot is in a two-wheeled upright position, the system enters the attitude balance control process based on simplified dynamics modeling. Using a wheeled linear inverted pendulum model with waist and upper body dynamics (WLIP-WT), the controller uses the desired state vector as input to predict the robot's attitude change trend in the sagittal plane and solves for the optimal torque to maintain forward and backward balance, including wheel hub driving torque and waist pitch torque. The WLIP-WT model, while retaining linear computational efficiency, introduces waist and upper body coupling characteristics, accurately characterizing the impact of upper body posture on overall balance in the two-wheeled configuration, achieving dynamic balance control in the forward and backward directions.

[0055] Furthermore, based on the base height and roll angle information in the desired state vector, the system calculates the thrust of the left and right virtual legs through the Virtual Model Control (VMC) module. The VMC, based on PID control, outputs virtual support forces on both sides to adjust the balance in the coronal plane and stabilize the base height in real time. When the robot experiences lateral disturbances or changes in ground height on one side, the virtual leg thrusts can instantly adjust the support force difference, achieving a self-balancing effect similar to a suspension system.

[0056] Finally, the control system synthesizes the optimal torque and the virtual leg thrust in the same coordinate system into a virtual leg action vector, and maps it to a physical joint torque through the transpose of the Jacobian matrix. The output is a second torque command used to drive the robot's actuators in two-wheeled mode. This mapping achieves a smooth correspondence between the virtual space and the physical execution layer, ensuring the continuity of the control signal in different modes and avoiding unstable transitions during mode switching.

[0057] In some embodiments, the step S2 of generating a first torque command based on the external disturbance vector specifically includes:

[0058] The external disturbance vector is input into the model predictive controller to generate the optimal ground reaction force;

[0059] The external disturbance vector is input into the whole-body controller to generate a feedforward compensation term;

[0060] The feedforward compensation term is used to compensate for the optimal ground reaction force to generate a first torque command.

[0061] As described above, by inputting the external disturbance vector into the model predictive controller and the whole-body controller respectively, a torque solution process combining dual-channel feedback and feedforward is formed. The model predictive controller is responsible for planning the optimal ground reaction force at the global level, while the whole-body controller generates feedforward compensation terms in real time during local rapid phases. This allows the system to quickly correct the joint output torque when subjected to nonlinear disturbances such as load changes and slope interference, reducing the accumulation of attitude errors. Compared to schemes that rely solely on model prediction or traditional PID control, this combined approach retains the robustness of predictive control while possessing the dynamic response advantage of real-time compensation, significantly improving the stability and load adaptive performance in quadrupedal configuration.

[0062] In some embodiments, inputting the external disturbance vector to the model prediction controller and generating the optimal ground reaction force specifically includes:

[0063] The external disturbance vector is input into the model predictive controller, and the optimal ground reaction force is generated by using the dynamic model, the physical feasibility model and the Lyapunov stability model as constraints.

[0064] As described above, the constraints for the model predictive controller to generate the optimal ground reaction force are further defined by comprehensively introducing the dynamic model, the physical feasibility model, and the Lyapunov stability model. Through this combined constraint, the system considers both the physical limits of the mechanical structure and friction conditions when planning the ground reaction force, and ensures the monotonic decay of error energy through the Lyapunov function. Theoretically, this ensures the asymptotic stability of the system and effectively prevents abrupt changes or out-of-bounds problems in the solution of ground forces without increasing computational complexity, enabling the robot to maintain its posture balance in rugged or unevenly loaded terrain.

[0065] In some embodiments, the step of using the feedforward compensation term to compensate the optimal ground reaction force to generate a first torque command specifically includes:

[0066] The whole-body controller is used to solve a quadratic programming problem based on the robot's complete rigid body east-west mechanical model to accurately track the optimal ground reaction force, and the feedforward compensation term is used to compensate for the optimal ground reaction force to generate a first torque command.

[0067] As described above, by establishing a quadratic programming problem based on whole-body rigid body dynamics within the whole-body controller, the optimal ground reaction force and feedforward compensation term are simultaneously incorporated into the solution. This approach achieves the optimal distribution of force and torque among the joints, compensating for external disturbances while ensuring trajectory accuracy. The quadratic programming solution offers advantages in analytical and real-time performance, ensuring continuous and smooth control signals and avoiding structural vibrations caused by sudden changes in transient torque. This enables high-precision dynamic compensation for complex load disturbances, allowing the quadruped robot to maintain posture stability and trajectory consistency in the high-frequency control loop.

[0068] In some embodiments, predicting the optimal torque required to maintain sagittal plane balance based on the desired state vector using the WLIP-WT model incorporating the waist and upper body dynamics of the wheel-legged composite robot specifically includes:

[0069] The WLIP-WT model is used to describe the dynamic relationship between the waist pitch motion and wheel-leg drive of the wheel-leg hybrid robot in two-wheel configuration.

[0070] Based on the dynamic relationship, the desired state vector is input to the model prediction controller, and the attitude change trend of the wheel-legged composite robot in the prediction time domain is predicted.

[0071] Solving for the attitude change trend yields the optimal torque required to maintain sagittal plane balance. The optimal torque includes the optimal hub drive torque and waist pitch torque for maintaining forward and backward balance.

[0072] As described above, the WLIP-WT model characterizes the coupled dynamic relationship between waist pitch and wheel-leg drive in a two-wheeled configuration. This model balances simplified calculation with physical accuracy, enabling the controller to predict attitude change trends in real time. By minimizing the objective function of attitude deviation and torque change rate, the optimal hub drive torque and waist pitch torque required for sagittal plane balance are obtained, ensuring that the robot can actively resist upper body disturbances and compensate for ground undulations in advance when in a two-wheeled standing posture. Compared with the traditional single pendulum approximation model, this model significantly improves attitude control accuracy and dynamic response capability while maintaining real-time computational efficiency, giving the robot stability performance close to that of a humanoid self-balancing vehicle.

[0073] In some embodiments, the step of calculating and generating the virtual leg thrusts on both sides based on the base height data and roll angle data in the desired state vector specifically includes:

[0074] The base height data and roll angle data are input into the virtual model controller;

[0075] The virtual model controller is used to perform PID adjustments on the base height data and the roll angle data respectively, and outputs the virtual leg thrust on both sides respectively.

[0076] As described above, by introducing a Virtual Model Controller (VMC) to independently adjust the base height and roll angle, stable control in the coronal plane direction is achieved. The VMC employs a dual-loop structure combining PID control and feedforward compensation. When a roll angle deviation is detected, it dynamically adjusts the thrust of the virtual legs on both sides, increasing the thrust of the lower leg and decreasing it on the other, creating an active suspension effect. This maintains overall balance when the robot turns, tilts, or experiences uneven force on one side, significantly reducing the risk of tipping over. By mapping the virtual leg thrust to the actual driving force, the system achieves a self-balancing effect similar to a suspension system without increasing the sensor load, thus improving lateral stability in two-wheel configurations.

[0077] In some embodiments, the step of combining the optimal torque and the virtual leg thrust into a virtual leg action vector, and mapping the virtual leg action vector into a second torque command, specifically includes:

[0078] The optimal torque and the virtual leg thrust are vector-synthesized in the same coordinate system to generate the virtual leg action vector.

[0079] The virtual leg action vector is converted into a torque using a preset Jacobian matrix transpose to obtain the target torque value for each joint, and then output as a second torque command to drive each joint to perform the action.

[0080] As described above, the optimal torque and virtual leg thrust are synthesized into a virtual leg action vector in the same coordinate system, and then mapped to the actual joint torque command via the Jacobian matrix transpose. This design unifies the upper-level MPC output and the lower-level VMC control results into the same mechanical framework, achieving coordination between force and pose control. Through Jacobian mapping, the control quantities in the virtual space are converted into the corresponding physical quantities of the executing joints in real time, ensuring that the torque distribution conforms to the force direction and constraint conditions of the actual structure. When the robot performs dynamic balance in a two-wheel configuration, it can simultaneously satisfy forward and backward balance and left and right stability, ensuring the continuity and smoothness of posture changes.

[0081] A terminal includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement steps in a motion control method for a wheel-legged composite robot.

[0082] A storage medium storing a computer program that, when executed by a processor, implements steps in a motion control method for a wheel-legged composite robot.

[0083] A wheel-legged composite robot is provided for performing the steps in a motion control method for a wheel-legged composite robot, including:

[0084] Two forelegs formed by point-like foot tips;

[0085] Two rear legs, each of which includes a linkage mechanism and a drive wheel, the linkage mechanism and the drive wheel being kinetically connected;

[0086] The connecting base includes a load-bearing housing and a waist joint; the linkage mechanism is drivenly connected to the load-bearing housing; the waist joint is fixedly connected to the load-bearing housing, and the waist joint is drivenly connected to the two front legs respectively; so as to realize the switching between four forms supported by the two front legs and the two hind legs simultaneously and the two-wheel form supported by the two hind legs.

[0087] As described above, the hardware structure provides functional support for the motion control method. By employing point-footed forelegs, hind legs containing linkage mechanisms and drive wheels, and a waist joint structure with pitch freedom, the robot can automatically switch between quadrupedal and biwheeled modes depending on the task.

[0088] Please refer to Figure 1 Embodiment 1 of the present invention is as follows:

[0089] A motion control method for a wheel-legged hybrid robot, applicable to a wheel-legged hybrid robot possessing both quadrupedal and two-wheeled configurations, includes the following steps:

[0090] When the robot starts up, it first detects its current motion state through the attitude detection module. Based on the attitude angle, angular velocity, and acceleration signals acquired by the inertial measurement unit (IMU), the attitude detection module determines whether the robot is currently in a quadrupedal support mode or a two-wheeled upright mode. The system enters the corresponding control branch according to the detection result. If it is in a quadrupedal mode, it executes the quadrupedal control process; if it is in a two-wheeled mode, it executes the two-wheeled control process.

[0091] When the robot is in quadruped mode, the controller generates the desired trajectory based on operation instructions or the task planning module, and collects the actual trajectory output by the encoders and IMU of each joint in real time. The difference between the desired and actual trajectories is calculated to obtain the tracking error vector. The control system integrates the tracking error vector using the Robust Adaptive Model Predictive Control (RA-MPC) algorithm to obtain an equivalent external disturbance vector. This external disturbance vector describes the unknown disturbance force caused by load changes, uneven ground, or friction differences.

[0092] The external disturbance vectors are input to the Model Predictive Controller (MPC) and the Whole Body Controller (WBC). The MPC generates the optimal ground reaction force at the global level based on the disturbance vectors, and the WBC generates feedforward compensation terms in real time based on the disturbance vectors. The results of the two are combined to generate the first torque command, which is used to drive each joint of the robot to perform corresponding actions, thereby realizing disturbance adaptive control.

[0093] The model predictive controller comprehensively considers the dynamic model, the physical feasibility constraint model, and the stability constraint based on the Lyapunov function during the solution process. Based on this, it solves for the optimal ground reaction force sequence in the prediction time domain, ensuring that the error energy monotonically decays at each moment, thus guaranteeing the asymptotic stability of the system. This constraint enables the robot to plan a continuous and smooth ground force output even under uneven loads or rugged terrain, preventing gait tremors caused by sudden changes in ground forces.

[0094] The whole-body controller establishes a quadratic programming problem based on the complete rigid body dynamics equations to achieve optimal distribution of whole-body torque. During the solution process, the controller uses the optimal ground reaction force from the model predictive controller as a reference target and incorporates the feedforward compensation term corresponding to the external disturbance vector into the solution equation. The final output first torque command has both global optimality and local real-time performance, which enables the robot to effectively eliminate the influence of disturbances and maintain attitude stability under high-frequency closed-loop control.

[0095] When the detection results indicate that the robot is in a two-wheel upright position, the desired state vector is acquired, including the base attitude angle, base height, roll angle, and angular velocity information. The controller establishes the dynamic relationship between the two wheels based on the wheeled linear inverted pendulum model with waist and upper body (WLIP-WT), describing the dynamic coupling characteristics between the waist pitch motion and the wheel-leg drive.

[0096] The desired state vector is input into the model predictive controller, and the attitude change trend in the future time domain is predicted by the WLIP-WT model. The objective function is to minimize the attitude deviation and the torque change rate. The optimal torque required to maintain the sagittal plane balance is solved, including the wheel hub drive torque and the waist pitch torque.

[0097] The control system inputs the base height data and roll angle data from the desired state vector into the virtual model controller (VMC) for PID adjustment. The VMC outputs the thrust values ​​of the left and right virtual legs. When the roll angle deviates, the system achieves active balance in the coronal plane direction and stable control of the base height by automatically increasing or decreasing the thrust.

[0098] The control system synthesizes the optimal torque and the thrust of the virtual legs on both sides into a vector in the same coordinate system, generating a virtual leg action vector. This vector is then mapped to the target torque value of the actual physical joint via the Jacobian matrix transpose, and output as a second torque command to drive the robot's joint movements in the two-wheel configuration. By fusing the calculation results of MPC and VMC, the control signal is continuously mapped between virtual and physical spaces, ensuring motion stability and posture smoothness.

[0099] Through the above control process, the comprehensive beneficial effect of this embodiment is that, by constructing a unified adaptive control framework, the wheel-legged hybrid robot can smoothly switch between quadrupedal and two-wheeled modes, and maintain posture stability and motion continuity under different terrain and load conditions. This method uses a posture detection module to determine the robot's form in real time and automatically selects the corresponding control branch to achieve multi-mode adaptive control. In quadrupedal mode, the control system calculates the tracking error based on the desired and actual states, and uses a robust adaptive model predictive control algorithm to integrally estimate the error, obtaining an equivalent external disturbance vector. This disturbance is simultaneously input to the model predictive controller and the whole-body controller. The former plans the optimal ground reaction force at the global level, while the latter generates a feedforward compensation term at the local level, forming a two-layer control mode of predictive and compensation collaboration. During the solution process, the model predictive controller incorporates dynamic constraints, physical feasibility constraints, and stability constraints to ensure monotonically decaying error energy and improve the system's asymptotic stability. The whole-body controller solves for the optimal torque distribution through quadratic programming, achieving smooth control and disturbance rejection compensation at high frequencies, thereby maintaining a stable gait under load changes and uneven terrain. In the two-wheel configuration, a wheeled linear inverted pendulum model with waist and upper body dynamics is used to predict posture and optimize torque. The optimal torque required for front-to-back balance is solved by minimizing posture deviation. The virtual model controller outputs the left and right virtual leg thrusts based on the base height and roll angle data, dynamically adjusting the coronal plane balance. Finally, the optimal torque and virtual leg thrusts are combined into a virtual leg action vector, which is mapped to the actual joint torque via the Jacobian matrix, achieving continuous control in both virtual and physical spaces. This method simultaneously possesses robustness, real-time performance, and high precision, enabling the robot to combine legged stability and wheeled mobility in complex environments.

[0100] Embodiment 2 of the present invention is as follows:

[0101] A terminal includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement steps in a motion control method for a wheel-legged composite robot.

[0102] Embodiment 3 of the present invention is as follows:

[0103] A storage medium storing a computer program that, when executed by a processor, implements steps in a motion control method for a wheel-legged composite robot.

[0104] Please refer to Figure 2 and Figure 3 Embodiment four of the present invention is as follows:

[0105] A wheel-legged composite robot is provided for performing the steps in a motion control method for a wheel-legged composite robot, including:

[0106] The robot comprises two front legs 1, each consisting of a point-foot. Each front leg 1 includes a hip joint 11, a thigh segment 12, a lower leg segment 13, a point-foot 17, and a corresponding drive motor. Specifically, the hip joint 11 houses a roll motor 14 for controlling the abduction and adduction of the front leg 1; the thigh segment 12 is equipped with a thigh pitch motor 15, and the lower leg segment 13 is equipped with a lower leg pitch motor 16, enabling the leg to swing back and forth; the point-foot 17 is mounted at the end of the lower leg, providing a high-friction support point when in contact with the ground for stable walking and foot placement control in quadruped mode. The two front legs 1 are symmetrically arranged at the front of the robot, serving both as frontal support in quadruped mode and folding up when switching to two-wheel mode to avoid interference with the ground.

[0107] The robot includes two hind legs 2, each of which comprises a linkage mechanism 21 and a drive wheel 22. The linkage mechanism 21 is a five-bar parallel structure, controlled by two sets of hip joint drive motors 211 arranged side by side, forming a closed multi-bar support system. The linkage mechanism 21 has the characteristics of high rigidity and high load-bearing capacity, used to maintain stable mechanical support under different postures. The end of each linkage mechanism 21 is connected to the drive wheel 22 for transmission. The drive wheel 22 includes a hub motor 221 and an inflatable tire 222. The hub motor 221 directly provides driving force, while the inflatable tire 222 plays a role in cushioning and improving adhesion. The two hind legs 2 are symmetrically arranged on both sides of the robot base. In quadruped mode, they share the ground support with the front legs 1, and in two-wheel mode, they undertake all support and driving functions.

[0108] The robot also includes a connecting base 3, which consists of a support shell 31 and a waist joint 32. The support shell 31 is the core load-bearing part of the entire robot, housing a main control computer, power module, inertial measurement unit (IMU), and communication module for attitude detection and motion control. Multiple mounting interfaces are provided on the outer surface of the support shell 31 for mechanical fixation to the front and rear leg 2 components and to provide a pivot point for rotation. The waist joint 32 is located above the support shell 31 and includes a waist pitch motor 321 and a reduction mechanism. Its output end is connected to the front leg 1 via a carbon fiber tube 322, enabling the pitch movement of the front leg 1 relative to the base, used to adjust the center of gravity and attitude balance. Simultaneously, the support shell 31 also includes a limiting block 33, which is located on the same side as the linkage mechanism 21 and is used to limit the range of motion of the linkage mechanism 21.

[0109] In terms of connectivity, the two front legs 1 are respectively connected to the output ends on both sides of the waist joint 32, which is connected to the bearing housing 31 via rigid fasteners. The linkage mechanism 21 of the two rear legs 2 is connected to the bottom of the bearing housing 31, forming a movable lower support structure. The end of the linkage mechanism 21 is fixedly connected to the output shaft of the hub motor 221 of the drive wheel 22, realizing the direct transmission of driving force. The bearing housing 31, as the middle structure, connects the front legs 1, rear legs 2, and waist joint 32, providing stable support and torque transmission path for the entire mechanical system. The pitching motion of the waist joint 32 drives the posture change of the front legs 1, enabling center of gravity adjustment; the linkage mechanism 21 of the rear legs 2 and the drive wheel 22 together constitute the main ground support and drive unit.

[0110] Specifically, the connecting base 3 is also fixedly equipped with a control center 34, which is communicatively connected to the drive motors of each joint to realize the steps in the motion control method of a wheel-legged composite robot.

[0111] In quadruped mode, both front legs 1 and both hind legs 2 touch the ground simultaneously, forming a stable four-point support posture. The waist joint 32 maintains a neutral angle, and the robot achieves smooth walking and posture maintenance through the coordinated movements of the four legs. When the front legs 1, waist, and hind legs 2 move in coordination, the center of gravity changes smoothly, and the overall gait is continuous and stable. In two-wheel mode, the two front legs 1 are lifted and tucked off the ground through the waist joint 32. The center of gravity shifts forward through the waist's pitching motion. The robot uses the hind legs 2 as the fulcrum to achieve upright posture. The drive wheels 22 provide forward and backward movement and steering control, and the waist joint 32 adjusts the pitch angle in real time to maintain standing balance.

[0112] In this embodiment, the motion control method is functionally supported by the hardware structure. By employing point-footed forelegs, hind legs containing linkage mechanisms and drive wheels, and a waist joint structure with pitch freedom, the robot can automatically switch between quadrupedal and bipedal configurations according to the task.

[0113] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A motion control method for a wheel-legged composite robot, applied to a wheel-legged composite robot, wherein the motion modes of the wheel-legged composite robot include a quadrupedal mode and a two-wheeled mode, characterized in that, The method comprises the steps of: S1, detecting the current motion form of the wheel-foot composite robot; S2, if in quadruped form, obtaining the desired state trajectory and the actual state trajectory of the wheel-foot composite robot; calculating a tracking error vector based on the desired state trajectory and the actual state trajectory, using a RA-MPC model to integrate the tracking error vector and generating an equivalent external disturbance vector; generating a first torque command according to the external disturbance vector, and using the first torque command to control the joints of the wheel-foot composite robot in quadruped form to execute actions; In step S2, the first torque command is generated according to the external disturbance vector, specifically comprising: inputting the external disturbance vector into a model predictive controller and generating an optimal ground reaction force; inputting the external disturbance vector into a full-body controller to generate a feedforward compensation term; compensating the optimal ground reaction force using the feedforward compensation term to generate the first torque command; S3, if in two-wheel form, obtaining the desired state vector of the wheel-foot composite robot; based on the WLIP-WT model containing the waist and upper body dynamics of the wheel-foot composite robot, taking the desired state vector as input to generate the optimal torque required to maintain the sagittal plane balance; calculating a virtual leg thrust based on the base height data and the roll angle data in the desired state vector; combining the optimal torque and the virtual leg thrust into a virtual leg action vector, and mapping the virtual leg action vector into a second torque command according to the virtual leg action vector, and using the second torque command to control the joints of the wheel-foot composite robot in two-wheel form to execute actions.

2. The motion control method of the wheel-legged hybrid robot according to claim 1, wherein, The optimal ground reaction force is generated by inputting the external disturbance vector into the model predictive controller, specifically comprising: inputting the external disturbance vector into the model predictive controller, and generating the optimal ground reaction force by taking the dynamics model, the physical feasibility model and the Lyapunov stability model as constraints.

3. The motion control method of the wheel-legged hybrid robot according to claim 1, wherein, The first torque command is generated by compensating the optimal ground reaction force using the feedforward compensation term, specifically comprising: solving a quadratic programming problem based on the complete rigid body dynamics model of the robot using the full-body controller to accurately track the optimal ground reaction force, and compensating the optimal ground reaction force using the feedforward compensation term to generate the first torque command.

4. The motion control method of the wheel-legged hybrid robot according to claim 1, wherein, The WLIP-WT model containing the waist and upper body dynamics of the wheel-foot composite robot is used to predict the optimal torque required to maintain the sagittal plane balance according to the desired state vector, specifically comprising: using the WLIP-WT model to describe the dynamic relationship between the waist pitching motion and the wheel-leg drive of the wheel-foot composite robot in two-wheel form; based on the dynamic relationship, inputting the desired state vector into the model predictive controller and predicting the posture change trend of the wheel-foot composite robot in the prediction time domain; solving the posture change trend to obtain the optimal torque required to maintain the sagittal plane balance, the optimal torque including the optimal wheel hub drive torque and the waist pitching torque for maintaining the front-back direction balance.

5. The motion control method of the wheel-legged hybrid robot according to claim 1, wherein, The generating the virtual leg thrusts on both sides according to the base height data and the roll angle data in the desired state vector specifically comprises: inputting the base height data and the roll angle data into a virtual model controller; adjusting the base height data and the roll angle data respectively by the virtual model controller, and outputting the virtual leg thrusts on both sides.

6. The motion control method of the wheel-legged hybrid robot according to claim 1, wherein, The generating the virtual leg thrusts on both sides according to the base height data and the roll angle data in the desired state vector specifically comprises: inputting the base height data and the roll angle data into a virtual model controller; adjusting the base height data and the roll angle data respectively by the virtual model controller, and outputting the virtual leg thrusts on both sides.

7. A terminal comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The synthesizing the optimal moment and the virtual leg thrust into a virtual leg action vector, and mapping the virtual leg action vector into a second moment instruction specifically comprises:

8. A storage medium characterized by, vector synthesizing the optimal moment and the virtual leg thrust in the same coordinate system to generate a virtual leg action vector; 9. A wheel-foot hybrid robot, characterized by performing moment conversion on the virtual leg action vector by using a preset Jacobian matrix transpose to obtain target moment values corresponding to each joint, and outputting the target moment values as the second moment instruction to drive each joint to perform an action. The processor executes the computer program to realize the steps in the motion control method of the wheel-legged robot according to any one of claims 1-6. The storage medium stores a computer program, and the computer program is executed by the processor to realize the steps in the motion control method of the wheel-legged robot according to any one of claims 1-6. The steps in the motion control method of the wheel-legged robot according to any one of claims 1-6 comprise: two front legs composed of point feet; two rear legs, each of which comprises a linkage mechanism and a driving wheel, and the linkage mechanism and the driving wheel are drivingly connected; a connecting base comprising a bearing shell and a waist joint; the linkage mechanism is drivingly connected with the bearing shell; the waist joint is fixedly connected with the bearing shell, and the waist joint is drivingly connected with two front legs respectively; to realize switching between four groups of shapes of two front legs and two rear legs supporting at the same time and two-wheel shape of two rear legs supporting.

Images