Centaur robot dual-arm cooperative anti-recoil force control method and device, electronic equipment, computer readable storage medium, centaur robot and system

Abstract

The application provides a Centaur robot double-arm cooperative anti-rebound force control method and device, electronic equipment, computer readable storage medium, Centaur robot and system. The method obtains an operation task category to be executed by a Centaur robot double-arm, extracts a pre-judgment external force direction feature and a pre-judgment external force amplitude feature; adjusts a quadruped foot bottom support force distribution parameter and a trunk space pose before impact force is generated; collects an actual contact force at the end and obtains a compensation torque through a whole machine dynamics mapping relationship; updates an active impedance control instruction according to the compensation torque, so that the quadruped joint absorbs residual impact force and maintains a zero moment point inside a support area.

Description

Technical Field

[0001] This application relates to the field of robot motion control technology, and more specifically, to a method, device, electronic device, computer-readable storage medium, centaur robot and system for controlling the dual-arm cooperative anti-recoil force of a centaur robot. Background Technology

[0002] Centaur robots combine dual-arm manipulation capabilities with quadrupedal locomotion and support, making them suitable for tasks such as demolition, hammering, tightening high-torque valves, and other operations requiring the upper limbs to perform large contact loads. In these scenarios, the robot not only needs to perform tool operations with its arms but also needs to maintain overall posture stability on complex terrain or in confined spaces. Therefore, load changes during dual-arm operations directly affect the stability of the robot's overall motion control.

[0003] Existing control schemes for centaur robots typically employ force limiting at the upper limbs and deceleration of the operational trajectory. This scheme places force feedback sensors at the ends of both arms and reduces the execution speed of the arms when an increase in contact load is detected. Subsequently, the control system limits the arms from continuing to output large operational forces based on an end-load threshold, while the quadrupedal motion controller maintains the lower limbs' standing or gait execution according to the original support rhythm.

[0004] However, the aforementioned solutions primarily restrict the operation of the dual arms after the contact load has been applied, lacking a coordinated adjustment relationship between the four-legged support side and the force-bearing side of the dual arms for overall force transmission. When the dual arms perform impact operations such as hammering, demolition, or high-torque twisting, the reaction load can easily create sudden disturbances between the torso and lower limbs, reducing the overall posture stability margin and further interfering with the four-legged support state. This results in technical problems such as a high risk of instability, delayed support adjustment, and difficulty in effectively absorbing residual impact during impact operations. Summary of the Invention

[0005] This application provides a method, device, electronic device, computer-readable storage medium, centaur robot and system for controlling the dual-arm cooperative anti-recoil force of a centaur robot, so as to at least alleviate the technical problems of overall instability and disturbance of quadruped support caused by reaction load during the impact operation of the dual arms of the centaur robot.

[0006] A method for controlling the recoil force of a centaur robot with coordinated dual arms, comprising: The task categories that the Centaur robot's arms are about to perform are obtained, and the corresponding predicted external force direction features and predicted external force amplitude features are extracted based on the task categories. Based on the predicted external force direction and amplitude characteristics, the distribution parameters of the support force on the four feet of the Centaur robot are adjusted before the impact force is generated when the two arms perform the operation, so as to coordinate the adjustment of the Centaur robot's torso spatial pose. The actual contact force at the ends of the two arms during operation is collected, and the actual contact force at the ends is converted to each joint of the four legs using a pre-constructed whole-machine dynamics mapping relationship to obtain the compensation torque of each joint. The whole-machine dynamics mapping relationship characterizes the force transmission constraint between the ends of the two arms and each joint of the four legs. The active impedance control commands of each joint of the quadruped are updated based on the compensation torque, and the joints of the quadruped are driven to absorb residual impact force based on the active impedance control commands, so as to keep the zero torque point of the Centaur robot inside the support area enclosed by the support polygon boundary.

[0007] Optionally, the category of the operation task to be performed by the Centaur robot's two arms is obtained, and the corresponding predicted external force direction features and predicted external force amplitude features are extracted based on the operation task category, including: Analyze the task categories to identify the tool contact point trajectories of the two arms during the operation; The rate of change of momentum is extracted along the tangential direction of the tool contact point trajectory to convert the rate of change of momentum into an initial impact force vector; The initial impact force vector is decomposed into the local coordinate system of the Centaur robot to obtain the direction vector and the magnitude value. Then, the direction vector is used to determine the predicted external force direction characteristics, and the magnitude value is used to determine the predicted external force amplitude characteristics.

[0008] Optionally, based on the predicted external force direction characteristics and predicted external force amplitude characteristics, before the impact force is generated by the dual arms performing the operation, the distribution parameters of the foot support force of the Centaur robot's four legs are adjusted to coordinately adjust the spatial pose of the Centaur robot's torso, including: Determine the reverse overturning moment that is opposite to the predicted external force direction characteristics based on the predicted external force direction characteristics; Combining the predicted external force amplitude characteristics and the reverse resistance overturning moment, the expected vertical reaction force of each support foot is calculated, and the distribution parameters of the four-foot support force are adjusted based on the expected vertical reaction force of each support foot. Based on the distribution parameters of the support force on the soles of the four legs, the support output of the corresponding supporting foot is adjusted to coordinate the spatial posture of the torso, thereby causing the center of mass corresponding to the spatial posture of the torso to shift in a direction opposite to the predicted direction of the external force.

[0009] Optionally, the actual end-effector contact forces generated by the two arms during operation are collected, and the actual end-effector contact forces are converted to the joints of the four legs using a pre-constructed whole-machine dynamics mapping relationship to obtain the compensating torque of each joint, including: The actual contact force at the end is obtained by a six-dimensional force sensor, and its force spinor in space is extracted based on the actual contact force at the end. Input the force spinor into the whole machine dynamics mapping relationship, which contains the transmission matrix used to characterize the relative motion between the rigid body links; The force spinor is mapped using the transpose of the conduction matrix to obtain the equivalent joint torque of each joint space of the four legs, and then the compensation torque of each joint is determined based on the equivalent joint torque.

[0010] Optionally, the active impedance control commands for each joint of the four legs are updated based on the compensation torque, including: Obtain the current stiffness and damping parameters of each joint in the quadruped; The compensation torque is introduced into the basic control closed loop to determine the torque compensation requirements of each joint in the corresponding impact direction; Increase the configuration value of the damping parameter in the corresponding impact direction according to the torque compensation requirements; The configuration value of the stiffness parameter in the corresponding impact direction is reduced according to the torque compensation requirements; Based on the configuration values ​​of the increased damping parameter and the decreased stiffness parameter, the target impedance parameter is determined, and an active impedance control command is generated based on the target impedance parameter.

[0011] Optionally, the support output of the corresponding supporting foot among the four legs is adjusted according to the support force distribution parameters of the four feet to coordinately adjust the spatial pose of the torso, thereby causing the center of mass corresponding to the spatial pose of the torso to shift in a direction opposite to the predicted external force direction, including: Obtain the coordinates of the geometric center of the support region enclosed by the support polygon boundary on the horizontal plane; Based on the coordinates of the geometric center and the predicted direction of the external force, an offset vector is established to indicate the direction of the centroid's movement. The direction of the offset vector is opposite to the predicted direction of the external force. The adjustment weights of each supporting foot are configured based on the distribution parameters of the plantar support force of the four legs as the bias vector; The pitch angle of the hip joint of the corresponding supporting leg is adjusted according to the adjustment weight, so as to drive the center of mass corresponding to the spatial pose of the torso to move along the direction of the bias vector, thereby realizing the displacement of the center of mass in a direction opposite to the predicted external force direction.

[0012] Optionally, after mapping the force spinor using the transpose of the conduction matrix to obtain the equivalent joint torques in each joint space of the four legs, and then generating the compensation torques for each joint based on the equivalent joint torques, the method further includes: Extract the current swing state features of the centaur robot to identify the supporting leg that is in a grounded state from the four legs; Identify the target force-bearing joint on the corresponding support leg based on the support leg markings; Select the joint torque to be transferred corresponding to the suspended leg from the equivalent joint torques; The torque of the joint to be transferred is distributed to the target force joint to complete the redistribution of the compensation torque of the target force joint, so that the compensation torque corresponding to the actual contact force at the end is borne by the target force joint in the grounded state.

[0013] Optionally, the joints of the four legs are driven by active impedance control commands to absorb residual impact forces, including: Monitor the actual angular velocity feedback values ​​of each joint of the four legs when executing active impedance control commands; The actual angular velocity feedback value is compared with the mutation threshold to trigger the flexible buffer mode when the actual angular velocity feedback value reaches the mutation threshold. Determine the yield direction of each joint under the flexible buffer mode; Control each joint to perform a limited-angle yielding rotation along the direction of the yielding force, so as to dissipate the energy carried by the residual impact force through yielding rotation.

[0014] Optionally, based on active impedance control commands, each joint of the four legs is driven to absorb residual impact force, so as to keep the zero-moment point of the centaur robot within the support area enclosed by the boundary of the support polygon, including: During the yielding and rotation process, the pressure distribution characteristics at the bottom of each supporting foot are collected in real time; Calculate the actual zero-torque point coordinates for the current control cycle based on the pressure distribution characteristics; Based on the actual zero-moment point coordinates and the boundary of the supporting polygon, calculate the shortest vertical distance between the actual zero-moment point coordinates and the boundary of the supporting polygon; The shortest vertical distance is compared with the critical safety threshold so that when the shortest vertical distance reaches the boundary approach condition defined by the critical safety threshold, a forced restoring torque is injected into the active impedance control command. Based on the active impedance control command after the injection of forced recovery torque, the joints are prevented from continuing to yield and rotate, thereby causing the actual zero torque point coordinates to retract into the support area.

[0015] A centaur robot dual-arm cooperative anti-recoil force control device, comprising: The task external force prediction module is used to obtain the operation task category that the Centaur robot's two arms are about to perform, and extract the corresponding predicted external force direction features and predicted external force amplitude features based on the operation task category; The pre-support adjustment module is used to adjust the support force distribution parameters of the four legs of the Centaur robot before the impact force is generated by the operation of the two arms, based on the predicted external force direction characteristics and predicted external force amplitude characteristics, so as to coordinate the adjustment of the Centaur robot's torso spatial pose. The contact force mapping module is used to collect the actual contact force at the end of the two arms during operation, and to convert the actual contact force at the end to each joint of the four legs using a pre-built whole-machine dynamics mapping relationship, so as to obtain the compensation torque of each joint. The whole-machine dynamics mapping relationship characterizes the force transmission constraint between the end of the two arms and each joint of the four legs. The impedance control update module is used to update the active impedance control commands of each joint of the four legs according to the compensation torque, and drive each joint of the four legs to absorb the residual impact force based on the active impedance control commands, so as to keep the zero torque point of the Centaur robot inside the support area enclosed by the support polygon boundary.

[0016] An electronic device, including a memory and a processor; The memory stores computer programs that can run on the processor; When the processor executes the computer program, it implements the above-mentioned method for controlling the centaur robot's dual-arm cooperative anti-recoil force.

[0017] A computer-readable storage medium storing a computer program; When the computer program is executed by the processor, the above-mentioned method for controlling the counter-recoil force of the centaur robot's two arms is implemented.

[0018] A centaur robot includes a dual-arm mechanism, a quadruped mechanism, a torso mechanism, an end-effector force acquisition component, and a controller; The dual-arm mechanism is used to perform operational tasks and generates actual end contact force during the performance of these tasks. The quadruped mechanism is used to adjust the distribution parameters of the support force on the soles of the four feet through each joint, and to execute active impedance control commands through each joint. The trunk mechanism is used to change the spatial posture of the trunk according to the distribution parameters of the support force on the soles of the four feet; The end contact force acquisition component is used to acquire the actual end contact force generated by the dual-arm mechanism during operation; The controller is connected to the dual-arm mechanism, quadruped mechanism, torso mechanism, and end-effector contact force acquisition component, respectively. The controller is used to obtain the type of operation task that the dual-arm mechanism is about to perform, and extract the corresponding predicted external force direction features and predicted external force amplitude features based on the operation task category; The controller is also used to control the quadruped mechanism to adjust the distribution parameters of the foot support force of the quadruped before the impact force is generated by the dual-arm mechanism, based on the predicted external force direction characteristics and predicted external force amplitude characteristics, so as to coordinate the adjustment of the torso spatial posture of the torso mechanism. The controller is also used to convert the actual contact force at the end of the machine to each joint of the quadruped mechanism using a pre-built whole-machine dynamics mapping relationship, so as to obtain the compensation torque of each joint. The whole-machine dynamics mapping relationship characterizes the force transmission constraint between the end of the double-arm mechanism and each joint of the quadruped mechanism. The controller is also used to update the active impedance control commands of each joint of the quadruped mechanism according to the compensation torque, and drive each joint of the quadruped mechanism to absorb residual impact force based on the active impedance control commands, so as to keep the zero torque point of the Centaur robot inside the support area enclosed by the support polygon boundary.

[0019] A centaur robot dual-arm collaborative anti-recoil force control system includes a task category acquisition unit, a pre-support adjustment unit, an end-effector contact force acquisition unit, a dynamic mapping unit, an active impedance control unit, and a zero-torque point monitoring unit. The task category acquisition unit is used to acquire the operation task category that the Centaur robot's two arms are about to perform, and output the operation task category to the pre-support adjustment unit; The pre-support adjustment unit is used to extract the corresponding predicted external force direction features and predicted external force amplitude features based on the operation task category, and adjust the support force distribution parameters of the four legs of the Centaur robot before the impact force is generated when the two arms perform the operation, so as to coordinate the adjustment of the Centaur robot's torso spatial pose. The end contact force acquisition unit is used to acquire the actual end contact force generated by the two arms during operation and output the actual end contact force to the dynamic mapping unit; The dynamic mapping unit is used to convert the actual contact force at the end of the arm to each joint of the four legs using the pre-built whole-machine dynamic mapping relationship, so as to obtain the compensation torque of each joint. The whole-machine dynamic mapping relationship characterizes the force transmission constraint between the end of the two arms and each joint of the four legs. The active impedance control unit is used to update the active impedance control commands of each joint of the four legs according to the compensation torque, and drive each joint of the four legs to absorb residual impact force based on the active impedance control commands. The zero-moment point monitoring unit is used to monitor the zero-moment point of the Centaur robot during the absorption of residual impact force at each joint of the four legs. When the zero-moment point approaches the boundary of the supporting polygon, it works with the active impedance control unit to adjust the active impedance control command to keep the zero-moment point of the Centaur robot inside the support area enclosed by the boundary of the supporting polygon.

[0020] The technical advantages of the technical solution provided in this application are: This application presents a centaur robot dual-arm cooperative anti-recoil force control method. Addressing the technical shortcomings of existing solutions that rely primarily on contact limiting force and trajectory deceleration during impactful dual-arm operations, leading to lag in quadrupedal support adjustment and reduced overall stability margin, this method first identifies the category of the operational task to be performed by the dual arms. Based on this category, it extracts the predicted external force direction and amplitude characteristics, enabling the robot to obtain control data related to the subsequent external force direction and amplitude before the impact force is generated. Compared to traditional methods that reduce dual-arm output only after contact load is formed, this application proactively links the dual-arm operation task with the overall anti-recoil control, providing a clear source of predicted external force for subsequent quadrupedal support adjustment and torso posture adjustment.

[0021] Furthermore, based on the predicted direction and amplitude characteristics of the external force, this application adjusts the support force distribution parameters of the four legs of the Centaur robot before the impact force is generated by the dual-arm operation, and coordinates the adjustment of the Centaur robot's torso spatial pose. This approach does not simply correct the posture after the reaction load has caused a deviation, but rather changes the support force distribution of the four legs and the torso spatial pose before the impact force is generated, allowing the overall support state to be pre-configured to resist the reaction load from the dual arms. Therefore, compared to the traditional method of maintaining the original support rhythm on the four legs, this application can achieve higher support adaptability before the impact operation of the dual arms, reducing the impact of sudden reaction load transmission on the overall posture stability of the robot.

[0022] Furthermore, this application collects the actual end-effector contact forces generated by the two arms during operation and uses a pre-constructed whole-machine dynamics mapping relationship to convert these actual end-effector contact forces to the joints of the four legs, thereby obtaining the compensation torque for each joint. Since the whole-machine dynamics mapping relationship characterizes the force transmission constraints between the end-effectors of the two arms and the joints of the four legs, the actual contact forces formed at the end-effectors can be converted into executable compensation torques at the joints of the four legs. Compared to traditional methods that only limit force or decelerate at the end-effectors of the two arms, this application allows the force information of the two arms to enter the control loop of the four legs. This enables the four legs to no longer passively bear disturbances according to their original support rhythm, but to participate in recoil force compensation based on force transmission constraints. Therefore, it helps alleviate the support disturbance problem caused by the sudden transmission of reaction loads between the torso and lower limbs.

[0023] Furthermore, this application updates the active impedance control commands of each joint of the quadruped based on the compensated torque, and drives each joint of the quadruped to absorb residual impact force based on the active impedance control commands, so as to maintain the zero-moment point of the Centaur robot within the support area enclosed by the boundary of the support polygon. This processing method connects the compensated torque, active impedance control, and zero-moment point stability constraint into a continuous control link, enabling the quadruped joints to absorb residual impact force while bearing the compensated torque, and using the zero-moment point state within the support area as the overall stability control target. Compared with the traditional approach where upper limb force limiting and lower limb support are separated, this application can form a clearer technical relationship between the impact operation of the two arms, quadruped support adjustment, joint impedance update, and overall stability margin, thereby better mitigating the risk of instability and quadruped gait disturbance caused by reaction loads during hammering, demolition, and high-torque operations of the Centaur robot. Attached Figure Description

[0024] Figure 1 This application provides an embodiment of a centaur robot's dual-arm collaborative anti-recoil control scenario. Figure 2 This application provides an embodiment of a method for controlling the anti-recoil force of a centaur robot with coordinated dual arms. Figure 3 This application provides an embodiment of a centaur robot dual-arm cooperative anti-recoil force control device; Figure 4 An electronic device is described in an embodiment of this application; Figure 5 This application provides an embodiment of a computer-readable storage medium. Figure 6 This application provides an embodiment of a mobile robot. Figure 7 This application provides an embodiment of a centaur robot dual-arm collaborative anti-recoil control system. Detailed Implementation

[0025] like Figure 1 As shown, this is a scenario of dual-arm collaborative anti-recoil control for a centaur robot according to an embodiment of this application; Figure 2 As shown in the figure, a method for controlling the counter-recoil force of a centaur robot with coordinated dual arms is provided in this application, comprising: The operation task category to be performed by the two arms of the centaur robot is obtained, and the corresponding predicted external force direction feature and predicted external force amplitude feature are extracted based on the operation task category; Based on the predicted external force direction characteristics and the predicted external force amplitude characteristics, before the double arms perform the operation and generate impact force, adjust the support force distribution parameters of the four feet of the Centaur robot to coordinately adjust the spatial posture of the Centaur robot's torso. The actual contact force at the ends of the two arms during operation is collected, and the actual contact force at the ends is converted to each joint of the four legs using a pre-constructed whole-machine dynamics mapping relationship to obtain the compensation torque of each joint. The whole-machine dynamics mapping relationship characterizes the force transmission constraint between the ends of the two arms and each joint of the four legs. The active impedance control commands of each joint of the four legs are updated according to the compensation torque, and the joints of the four legs are driven to absorb residual impact force based on the active impedance control commands, so as to keep the zero torque point of the centaur robot inside the support area enclosed by the support polygon boundary.

[0026] Optionally, the step of obtaining the operation task category that the centaur robot's two arms are about to perform, and extracting the corresponding predicted external force direction features and predicted external force amplitude features based on the operation task category, includes: The operation task category is analyzed to identify the tool contact point trajectory of the two arms during the operation process; The rate of change of momentum is extracted tangentially along the trajectory of the tool contact point to convert the rate of change of momentum into an initial impact force vector; The initial impact force vector is decomposed into the local coordinate system of the centaur robot to obtain the direction vector and the magnitude value. Then, the predicted external force direction characteristics are determined based on the direction vector, and the predicted external force amplitude characteristics are determined based on the magnitude value.

[0027] Preferably, the specific implementation process of parsing the operation task category to identify the tool contact point trajectory of the dual arms during operation is as follows: After the Centaur robot enters the dual-arm operation preparation stage, the operation task category corresponding to the current operation task is first read. The operation task category is not a simple text label, but a data-driven task description used to characterize the operation form, tool contact method, operation direction source, and operation duration that the dual arms will perform. The data-driven task description is used to enable subsequent external force prediction processing to obtain data sources from the specific contact process of the dual-arm operation. In order for the operation task category to participate in the external force prediction processing, an operation task parsing and carrying record is first established, and the operation task category, the current pose of the dual arm end, the pose to be reached by the dual arm end, the tool installation pose, the installation offset of the tool contact end relative to the dual arm end, and the operation start time are written into the operation task parsing and carrying record. The operation task parsing record is used to incorporate the task source of the operation task category, the motion source of the current pose and the pose to be reached of the dual arm end caps, the contact end source of the tool mounting pose and the mounting offset of the tool contact end relative to the dual arm end caps, and the time source of the operation start time into the same parsing caliber. This allows the subsequently formed tool contact point trajectory to be traced back to the preparation stage of this dual-arm operation. Subsequently, the operation task category in the operation task parsing record is processed to establish a correspondence between the operation forms such as hammering, demolition, and turning high-torque valves in the operation task category and the motion direction of the tool contact end, the surface orientation of the contact object, and the motion path of the dual arm end caps. The surface orientation of the contact object is used to limit the spatial orientation of the tool contact end when entering the force position during subsequent identification of the tool contact point trajectory. The technical essence of the tool contact point trajectory is a continuous spatial position sequence of the tool contact end relative to the Centaur robot body during dual-arm operation. This continuous spatial position sequence is used to describe the positional change process of the impact force or reaction load entering the dual arm end caps first, rather than to describe the ordinary motion path of the dual arm joints themselves. In specific implementation, firstly, based on the tool installation pose and the installation offset of the tool contact end relative to the ends of the two arms, the current pose of the ends of the two arms is converted into the current spatial position of the tool contact end, and the pose to be reached of the ends of the two arms is converted into the spatial position to be reached of the tool contact end. The current spatial position of the tool contact end and the spatial position to be reached of the tool contact end together define the start and end spatial range of the tool contact point trajectory. Then, according to the operation mode corresponding to the operation task category, the spatial transition position between the current spatial position of the tool contact end and the spatial position to be reached of the tool contact end is sequentially interpolated, and the spatial transition position is arranged into the continuous spatial position sequence according to the operation time sequence to form the tool contact point trajectory.For hammering operations, the tool contact point trajectory primarily reflects a continuous spatial sequence of positions as the tool's contact end moves toward the object being struck and undergoes a velocity change at the moment of contact. For demolition operations, the tool contact point trajectory primarily reflects a continuous spatial sequence of positions as the tool's contact end enters the force-bearing position along the surface or edge of the object being demolished. For turning high-torque valves, the tool contact point trajectory primarily reflects a continuous spatial sequence of positions as the tool's contact end rotates around the valve's axis of rotation. Thus, the tool contact point trajectory inherits the analytical results of the operation task category and serves as the spatial basis for subsequently extracting the momentum change rate along the tangential direction. This allows the predicted external force direction and amplitude characteristics to correspond to the actual contact process of the two arms, avoiding delayed judgments based solely on the end-effector velocity or a single contact threshold.

[0028] Preferably, the specific implementation process for extracting the momentum change rate along the tangential direction of the tool contact point trajectory is as follows: After the tool contact point trajectory has been parsed by the operation task category, the continuous trajectory points in the tool contact point trajectory are read according to the control cycle of the Centaur robot's dual-arm operation, and the displacement directions between adjacent trajectory points are organized into a local trajectory forward direction. This local trajectory forward direction is used to provide a directional basis for the subsequent generation of the trajectory tangential direction. The technical essence of the tangential direction of the tool contact point trajectory is the instantaneous forward direction of the tool contact point trajectory at the current trajectory point, which is used to express the direction in which the tool contact end will transfer the output velocity of the dual arms and the momentum carried by the tool to the contact object. This tangential direction is not an arbitrarily specified direction, nor is it the horizontal direction of the four-legged support surface, but a local direction calculated from the position change sequence of adjacent trajectory points in the tool contact point trajectory. Therefore, it can reflect the real motion trend of the tool contact end when entering the force-bearing position under different operation task categories. In specific implementation, firstly, the current trajectory point corresponding to the current control cycle is selected from the tool contact point trajectory, and the adjacent trajectory points before and after the current trajectory point are read. Then, based on the spatial position changes of the adjacent trajectory points before and after the current trajectory point relative to the current trajectory point, the trajectory tangential direction is generated. The trajectory tangential direction inherits the local forward direction of the trajectory and is used to express the instantaneous motion direction of the tool contact end at the current trajectory point. Subsequently, the trajectory tangential direction is normalized to form a tangential unit direction, which is used to eliminate the influence of the trajectory point spacing on the subsequent momentum change rate calculation without changing the trajectory tangential direction. Then, the equivalent motion mass corresponding to the end of both arms and the tool contact end, the tool contact end velocity corresponding to the current trajectory point, and the tool contact end target velocity corresponding to the next control cycle are read, and the equivalent motion mass, the tool contact end velocity, the tool contact end target velocity, and the tangential unit direction are written into the momentum change rate calculation carrying record. The momentum change rate calculation record is used to incorporate the mass source of the equivalent kinematic mass, the current motion source of the tool contact end velocity, the motion source of the next control cycle of the tool contact end target velocity, and the directional source of the tangential unit direction into the same calculation caliber. This ensures that the subsequently obtained momentum change rate can simultaneously correspond to the motion intensity and motion direction of the tool contact end. The technical essence of the momentum change rate is the amount of momentum change that occurs at the tool contact end tangentially along the trajectory of the tool contact point per unit time, used to characterize the strength of the tool contact end's tendency to exert impact on the contact object during imminent or sustained contact.In the specific calculation, the velocity of the tool contact end is first projected onto the tangential unit direction to obtain the current tangential velocity component; then, the target velocity of the tool contact end is projected onto the tangential unit direction to obtain the target tangential velocity component; subsequently, based on the difference between the current tangential velocity component and the target tangential velocity component, combined with the equivalent motion mass and the control cycle duration, the momentum change rate is obtained. Both the current tangential velocity component and the target tangential velocity component originate from the tangential unit direction; therefore, the difference between them reflects the velocity change of the tool contact end along the tangential direction of the tool contact point trajectory. The equivalent motion mass converts this velocity change into a momentum change, and the control cycle duration converts this momentum change into the momentum change rate per unit time. Since the momentum change rate is extracted along the tangential direction of the tool contact point trajectory, it retains the directional and amplitude sources directly related to the actual contact path, thus providing a calculable force prediction basis for subsequently converting the momentum change rate into an initial impact force vector.

[0029] Preferably, in the specific technical implementation of converting the momentum change rate into an initial impact force vector, the momentum change rate, the tangential unit direction, the operation task category, and the spatial position of the tool contact end corresponding to the current trajectory point are first read from the momentum change rate calculation and bearing record. Based on the above data, an initial impact force vector conversion bearing record is established. This initial impact force vector conversion bearing record is used to incorporate the amplitude source of the momentum change rate, the direction source of the tangential unit direction, the operation form source of the operation task category, and the force source position of the tool contact end spatial position into the same conversion caliber, so that the subsequently formed initial impact force vector can simultaneously express the force source position, direction, and intensity. The momentum change rate itself expresses the intensity of the momentum change occurring tangentially along the trajectory of the tool contact point, while the initial impact force vector is a data structure that integrates the direction and amplitude of the impending impact force in the Centaur robot's dual-arm operation scenario. The technical essence of the initial impact force vector is a spatial force prediction quantity formed by taking the spatial position of the tool contact end as the force source position, the tangential unit direction as the direction source, and the momentum change rate as the amplitude source. This spatial force prediction quantity is used to convert the motion trend of the dual arms before operation into data objects that can be decomposed into coordinates and adjusted for support in the Centaur robot body. In specific implementation, the correspondence between the momentum change direction and the reaction load direction is first determined according to the operation task category. The momentum change direction is expressed by the tangential unit direction, and the reaction load direction is used to express the force direction fed back from the contact object to the ends of the dual arms. The correspondence is used to convert the active motion direction of the tool contact end into the feedback direction of the contact object. In hammering operations, the direction of momentum change of the tool contact end toward the contact object is usually opposite to the direction of the reaction load fed back from the contact object to the ends of the arms. Therefore, the tangential unit direction is reversed to obtain the direction of the reaction load. In demolition operations, the tool contact end may enter the force-bearing position along the surface of the contact object. Therefore, the direction of the reaction load is determined based on the angle between the tangential of the tool contact point trajectory and the orientation of the contact object surface. The orientation of the contact object surface is the same as the orientation of the contact object surface recorded in the analysis of the aforementioned operation task. In high-torque valve turning operations, the tool contact end moves around the valve's rotation axis. Therefore, the direction of the reaction load is determined based on the rotation direction of the tool contact end and the direction of the valve's rotational resistance. Subsequently, the absolute intensity of the momentum change rate is used as the initial amplitude source, and the direction of the reaction load is used as the initial direction source. The initial amplitude source and the initial direction source are combined to form the initial impact force vector. The initial amplitude is derived from the rate of change of momentum and is used to define the intensity of the initial impact force vector; the initial direction is derived from the direction of the reaction load and is used to define the spatial orientation of the initial impact force vector.The initial impact force vector retains the spatial position of the tool's contact end, enabling it to characterize not only the magnitude and direction of the external force but also to correspond to the force transmission entry point between the ends of the arms and the torso. Through this transformation, the momentum change rate is no longer confined to the kinematic change level but is converted into the initial impact force vector that can be used for pre-adjustment of the overall machine attitude and adjustment of the four-legged foot support force distribution parameters. This allows the predicted external force direction characteristics and the predicted external force amplitude characteristics to have a clear data source before the impact force is generated.

[0030] Preferably, the initial impact force vector is decomposed into the local coordinate system of the Centaur robot to obtain the direction vector and magnitude value. The specific implementation process for determining the predicted external force direction characteristics based on the direction vector and the predicted external force amplitude characteristics based on the magnitude value is as follows: After the initial impact force vector is formed, the current torso spatial pose, quadrupedal support state, and the installation position of the arm ends relative to the torso of the Centaur robot are first read. A local coordinate system of the Centaur robot is then established based on the current torso spatial pose. The technical essence of the Centaur robot's local coordinate system is a body coordinate expression aperture that changes synchronously with the torso spatial pose of the Centaur robot. This body coordinate expression aperture is used to convert the predicted impact force in external space into a direction expression that can be directly used for quadrupedal support adjustment and torso spatial pose adjustment. In specific implementation, the forward / backward direction of the centaur robot's local coordinate system is aligned with the forward direction of the centaur robot's torso; the left / right direction of the centaur robot's local coordinate system is aligned with the lateral direction of the centaur robot's torso; and the up / down direction of the centaur robot's local coordinate system is aligned with the vertical direction of the centaur robot's torso. Then, based on the current torso's spatial pose, the initial impact force vector is converted from its external spatial representation into a local impact force vector within the centaur robot's local coordinate system. This local impact force vector includes forward / backward components, left / right components, and up / down components. The forward / backward component is used for subsequent support adjustment to resist forward / backward overturning tendencies; the left / right component is used for subsequent support adjustment to resist lateral overturning tendencies; and the up / down component is used for determining subsequent changes in the vertical reaction force of the supporting feet. The forward / backward, left / right, and up / down components all originate from the decomposition of the initial impact force vector in the local coordinate system of the Centaur robot. Therefore, these components establish a directional correspondence between the initial impact force vector and the subsequent adjustment of the four-legged foot support force distribution parameters. Subsequently, the local impact force vector undergoes directional normalization to obtain the direction vector. The technical essence of this direction vector is to retain the direction of the local impact force vector while removing the amplitude influence, enabling the determination of the four-legged foot support force distribution parameters based on the predicted external force direction characteristics to distinguish whether the primary resistance direction comes from the forward / backward, left / right, or up / down directions. Simultaneously, the length of the local impact force vector is calculated to obtain the modulus value. The modulus value is essentially an intensity expression of the local impact force vector in the local coordinate system of the Centaur robot, enabling the adjustment of the support output based on the predicted external force amplitude characteristics to configure different support force adjustment amplitudes according to the strength of the reaction load.Finally, the direction vector is written into the predicted external force direction feature record, and the modulus value is written into the predicted external force amplitude feature record, to determine the predicted external force direction feature and the predicted external force amplitude feature, respectively. The predicted external force direction feature record is used to carry the direction vector and to serve as the direction basis in the subsequent adjustment of the four-legged foot support force distribution parameters; the predicted external force amplitude feature record is used to carry the modulus value and to serve as the amplitude basis in the subsequent adjustment of the four-legged foot support force distribution parameters. Thus, after the initial impact force vector is decomposed in the local coordinate system of the Centaur robot, the predicted external force direction feature and the predicted external force amplitude feature are formed, which can directly participate in the adjustment of the four-legged foot support force distribution parameters and the coordinated adjustment of the torso spatial pose. This allows the operation task category to be performed by the two arms to be transmitted to the whole machine's anti-recoil force control process through a continuous data processing link.

[0031] Optionally, adjusting the foot support force distribution parameters of the centaur robot's four legs before the impact force is generated by the dual arms performing the operation, based on the predicted external force direction characteristics and the predicted external force amplitude characteristics, to coordinately adjust the centaur robot's torso spatial pose, includes: Based on the predicted external force direction characteristics, determine the opposite overturning moment that is opposite to the predicted external force direction characteristics; Combining the predicted external force amplitude characteristics and the reverse resistance overturning moment, the expected vertical reaction force of each support foot is calculated, so as to adjust the distribution parameters of the four-foot support force based on the expected vertical reaction force of each support foot. The support output of the corresponding supporting foot among the four feet is adjusted according to the support force distribution parameters of the four feet to coordinately adjust the spatial posture of the torso, thereby causing the center of mass corresponding to the spatial posture of the torso to shift in a direction opposite to the predicted external force direction.

[0032] Preferably, the specific implementation process for determining the reverse resistance overturning moment opposite to the predicted external force direction feature based on the predicted external force direction feature is as follows: After the predicted external force direction feature and the predicted external force amplitude feature have been parsed by the operation task category, the predicted external force direction feature, the predicted external force amplitude feature, the current torso spatial pose of the Centaur robot, the installation position of the two arm ends relative to the torso, and the foot position of the four legs currently in contact with the ground are read, and an overturning trend analysis load record is established. The overturning trend analysis load record is used to include the direction source of the predicted external force direction feature, the amplitude source of the predicted external force amplitude feature, the posture source of the current torso spatial pose of the Centaur robot, the force input source of the installation position of the two arm ends relative to the torso, and the ground support source of the foot position of the four legs currently in contact with the ground into the same analysis caliber, so that the subsequently formed reverse resistance overturning moment can correspond to the overall force state of the current two-arm operation. Subsequently, a torso reference center is determined based on the current torso spatial pose of the centaur robot, and an external force prediction entry point is determined based on the installation position of the ends of the arms relative to the torso. This external force prediction entry point indicates the position where the reaction load corresponding to the predicted external force direction and amplitude characteristics enters the centaur robot body. The spatial offset relationship between the external force prediction entry point and the torso reference center is combined with the predicted external force direction characteristics to obtain the predicted overturning direction. The technical essence of this predicted overturning direction is the tendency of the reaction load to drive the centaur robot to tilt forward, backward, or sideways around the torso reference center before the arms generate impact force during operation. The reverse anti-overturning torque is not an arbitrarily increased joint torque, but rather an overall torque adjustment amount opposite to the predicted overturning direction and used to counteract the posture disturbance trend corresponding to the predicted overturning direction. In specific implementation, the direction of the reaction load is determined based on the predicted external force direction characteristics. Then, the predicted overturning direction generated by the reaction load on the torso reference center is determined based on the spatial offset relationship between the predicted external force entry point and the torso reference center. Subsequently, the predicted overturning direction is reversed and set as the resistance direction of the reverse overturning moment. The moment intensity in the resistance direction of the reverse overturning moment is determined in combination with the predicted external force amplitude characteristics, thereby forming the reverse overturning moment. Thus, the reverse overturning moment inherits the predicted external force direction characteristics and the predicted external force amplitude characteristics, and provides a moment adjustment basis with unified direction and intensity for subsequent calculation of the desired vertical reaction force of each support foot.

[0033] Preferably, in the specific technical implementation of calculating the expected vertical reaction force of each supporting foot by combining the predicted external force amplitude characteristics and the reverse resistance overturning moment, the reverse resistance overturning moment, the predicted external force amplitude characteristics, the current foot position of the four legs in contact with the ground, the current center of gravity position of the Centaur robot, and the current plantar reaction force of each supporting foot are first read from the overturning trend analysis load record, and an expected vertical reaction force calculation load record is established. The expected vertical reaction force calculation load record is used to include the torque adjustment source of the reverse resistance overturning moment, the load intensity source of the predicted external force amplitude characteristics, the support geometry source of the current foot position of the four legs in contact with the ground, the mass distribution source of the current center of gravity position of the Centaur robot, and the support state source of the current plantar reaction force of each supporting foot into the same calculation caliber, so that the expected vertical reaction force is not an isolated plantar target quantity, but maintains the same source relationship with the predicted external force direction characteristics and the predicted external force amplitude characteristics generated by the dual-arm operation. The technical essence of the desired vertical reaction force is to assign a vertical support target to each grounded support foot before the impact force is generated. This is used to ensure that the four-legged support surface forms a ground reaction force distribution that matches the reverse overturning moment before the reaction load arrives. This ground reaction force distribution continues to serve as the basis for subsequent adjustments to the foot support force distribution parameters. Specifically, the support polygon boundary is first determined based on the current foot position in contact with the ground, and the current projection position of the centaur's current center of mass within the support polygon boundary is determined based on the current center of mass position of the centaur robot. Subsequently, based on the resistance direction of the reverse overturning moment, it is determined which support feet need to increase their vertical support contribution and which need to decrease it. For scenarios requiring resistance to forward tilting, the front and rear support feet bear different vertical support changes according to the resistance direction of the reverse overturning moment. For scenarios requiring resistance to lateral tilting, the left and right support feet form differentiated left-right vertical support changes according to the resistance direction of the reverse overturning moment. Then, the amplitude of the vertical support change is adjusted according to the predicted external force amplitude characteristics, so that when the predicted external force amplitude characteristics are larger, the amplitude of the vertical support change of each support foot is higher; when the predicted external force amplitude characteristics are smaller, the amplitude of the vertical support change of each support foot is correspondingly reduced. Finally, the current plantar reaction force of each support foot is superimposed and constrained with the corresponding vertical support change, and the total vertical support of the four feet is balanced and corrected to obtain the desired vertical reaction force of each support foot. The desired vertical reaction force continues to participate in the subsequent adjustment of the plantar support force distribution parameters of the four feet, so that the reverse resistance overturning moment can be transmitted to the four-foot support control process through the executable plantar vertical support target.

[0034] Preferably, the specific implementation process of adjusting the foot support force distribution parameters of the four legs based on the expected vertical reaction force of each supporting leg is as follows: After the expected vertical reaction force of each supporting leg is formed, the expected vertical reaction force of each supporting leg, the current foot reaction force of each supporting leg, the current foot position of the four legs in contact with the ground, and the current center of gravity position of the Centaur robot are read from the expected vertical reaction force calculation and bearing record, and a support force distribution adjustment bearing record is established. The support force distribution adjustment bearing record is used to incorporate the target support source of the expected vertical reaction force, the actual support source of the current foot reaction force of each supporting leg, the spatial support source of the foot position of the four legs currently in contact with the ground, and the attitude influence source of the current center of gravity position of the Centaur robot into the same adjustment caliber, so that the foot support force distribution parameters of the four legs can reflect the vertical support change that each supporting leg needs to bear in the current control cycle. The technical essence of the four-legged plantar support force distribution parameters is a parameterized result describing the vertical support force distribution relationship among the four supporting legs. This parameterized result includes both the target vertical support ratio of each supporting leg and the adjustment direction of the transition from the current plantar reaction force of each supporting leg to the desired vertical reaction force of each supporting leg. Specifically, the desired vertical reaction force of each supporting leg is first subtracted from the current plantar reaction force of each supporting leg to obtain the plantar reaction force adjustment amount of each supporting leg. This plantar reaction force adjustment amount represents the vertical support contribution that the corresponding supporting leg needs to increase or decrease before the impact force is generated. Subsequently, based on the plantar reaction force adjustment amount of each supporting leg and the current foot position of the four legs in contact with the ground, the vertical support ratio among the supporting legs is reconfigured to form the four-legged plantar support force distribution parameters. For support feet whose foot position is closer to the boundary corresponding to the predicted overturning direction, the corresponding plantar reaction force adjustment amount will be reflected as a higher vertical support participation in the four-leg plantar support force distribution parameters. For support feet whose foot position is located on the side of the resistance direction against the overturning moment, the corresponding plantar reaction force adjustment amount will be reflected as a support adjustment contribution for pulling back the posture trend in the four-leg plantar support force distribution parameters. In order for the four-leg plantar support force distribution parameters to be used in subsequent support output, a correspondence will be established between the four-leg plantar support force distribution parameters and the corresponding support foot identifier, the current grounding state of the support foot, and the allowable adjustment range of the support foot. The corresponding support foot identifier is used to identify the support foot to which each vertical support ratio in the four-leg plantar support force distribution parameters belongs. The current grounding state of the support foot is used to limit the support feet in the four-leg plantar support force distribution parameters that can participate in the support output. The allowable adjustment range of the support foot is used to limit the adjustment boundary of the four-leg plantar support force distribution parameters when they are subsequently converted into support output.Therefore, the distribution parameters of the support force on the soles of the four legs receive the expected vertical reaction force of each support leg and serve as the direct basis for subsequent adjustment of the support output of the corresponding support leg among the four legs, so that the force prediction result before the double arm operation generates impact force can enter the execution level of the four-leg support side.

[0035] Preferably, the specific implementation process of adjusting the support output of the corresponding supporting foot among the four legs according to the four-legged foot support force distribution parameters to coordinately adjust the torso spatial pose, thereby causing the centroid corresponding to the torso spatial pose to shift in a direction opposite to the predicted external force direction, is as follows: After the four-legged foot support force distribution parameters are formed, the four-legged foot support force distribution parameters, corresponding supporting foot identifiers, current grounding status of the supporting foot, allowable adjustment range of the supporting foot, current torso spatial pose of the Centaur robot, and current centroid position of the Centaur robot are read from the support force distribution adjustment load record, and a torso pose coordinated adjustment load record is established. The torso pose coordinated adjustment load record is used to incorporate the support allocation source of the four-legged foot support force distribution parameters, the execution object source of the corresponding supporting foot identifier, the contact effectiveness source of the current grounding status of the supporting foot, the safe adjustment source of the allowable adjustment range of the supporting foot, the posture source of the current torso spatial pose of the Centaur robot, and the mass distribution source of the current centroid position of the Centaur robot into the same adjustment caliber. The technical essence of the support output is that the corresponding supporting foot forms a vertical support action between the sole and the ground based on the support force distribution parameters of the four feet. This vertical support action is transmitted to the torso through the posture changes of the hip, knee, and ankle joints of the supporting foot, causing the spatial posture of the torso to undergo a pre-adjustment to counteract the reaction load. Specifically, the plantar reaction force adjustment amount of the corresponding supporting foot is first determined based on the support force distribution parameters of the four feet. Then, based on the current grounding state of the supporting foot, the corresponding supporting foot capable of executing the plantar reaction force adjustment amount is selected, and the plantar reaction force adjustment amount is limited according to the allowable adjustment range of the supporting foot to form the support output adjustment amount of the corresponding supporting foot. Subsequently, the support output adjustment amount is converted into the joint posture adjustment amount of the corresponding supporting foot, and the vertical support action of the sole on the ground is changed through the joint posture adjustment amount of the corresponding supporting foot, causing the support output of the corresponding supporting foot among the four feet to change according to the support force distribution parameters of the four feet. When the support output changes, it alters the force distribution between the Centaur robot's torso and the ground support point, thereby causing a pre-adjustment of the torso's spatial pose. The technical essence of this torso spatial pose is the combination of the Centaur robot's torso's spatial position and posture relative to the ground support surface. This combination directly affects the positional relationship between the Centaur robot's current center of mass and the boundary of the supporting polygon.By adjusting the support output based on the four-legged foot support force distribution parameters, the current center of mass position of the centaur robot is guided to displace in a direction opposite to the predicted external force direction. This displacement increases the attitude buffer margin in the direction of the incoming reaction load before the impact force is generated by the arms performing operations, thus establishing a continuous technical interaction between the torso spatial pose, the four-legged foot support force distribution parameters, and the predicted external force direction. Consequently, the four-legged foot support force distribution parameters no longer remain at the support force target level, but are transmitted to the torso spatial pose through the support output of the corresponding supporting legs. This causes the center of mass corresponding to the torso spatial pose to displace in a direction opposite to the predicted external force direction, providing a preliminary attitude basis for subsequently absorbing residual impact force using the overall dynamic mapping relationship and active impedance control commands.

[0036] Optionally, the step of collecting the actual end-effector contact force generated by the two arms during operation and converting the actual end-effector contact force to each joint of the four legs using a pre-constructed whole-machine dynamics mapping relationship to obtain the compensating torque of each joint includes: The actual contact force at the end is obtained by a six-dimensional force sensor, and its force spin in space is extracted based on the actual contact force at the end. The force spinor is input into the overall dynamics mapping relationship, which includes a transmission matrix used to characterize the relative motion between the rigid body links. The force spinor is mapped using the transpose of the conduction matrix to obtain the equivalent joint torque of each joint space of the four legs, and then the compensation torque of each joint is determined based on the equivalent joint torque.

[0037] Preferably, the specific implementation process of acquiring the actual contact force at the end of the arm using a six-dimensional force sensor and extracting its spatial force rotation based on the actual contact force is as follows: After the dual arms perform the operation and the tool contact end enters the force-bearing position, the contact load sampling data output by the six-dimensional force sensor installed at the end of the dual arms or the tool connector is first read. The contact load sampling data, along with the current control cycle marker, the current pose of the dual arm end, the tool mounting pose, and the installation offset of the tool contact end relative to the dual arm end, are written into the actual contact force acquisition and bearing record. The actual contact force acquisition and bearing record is used to include the sampling source of the contact load sampling data, the spatial pose source of the current pose of the dual arm end, the installation source of the tool mounting pose, the contact position source of the installation offset of the tool contact end relative to the dual arm end, and the temporal source of the current control cycle marker into the same reading caliber, so that the subsequently extracted force rotation can correspond to the actual force entry point in this dual-arm operation process. The six-dimensional force sensor does not simply provide a load value detection component; instead, it simultaneously outputs force components and torque components in multiple directions within its own measurement coordinate aperture. The force components characterize the linear reaction load of the contact object on the tool contact end, while the torque components characterize the rotational effect caused by the linear reaction load's point of application deviating from the reference point at the ends of the arms. To convert the contact load sampling data into the actual contact force at the ends, the contact load sampling data is first aligned according to the installation orientation of the six-dimensional force sensor, transforming it from the sensor's own measurement coordinate aperture to the coordinate aperture of the ends of the arms. Subsequently, based on the installation offset of the tool contact end relative to the ends of the arms, the contact load sampling data converted to the coordinate aperture of the ends of the arms undergoes point-of-application compensation processing to convert the contact load at the tool contact end to the reference point at the ends of the arms, thereby forming the actual contact force at the ends. The essence of the force spinor technique is to couple the linear and rotational forces in the actual contact force at the end of the arm with a common spatial reference point, resulting in a data-driven force description. This force spinor allows the actual contact force at the end of the arm to be incorporated into the overall dynamics mapping, rather than remaining only as a local load feedback at the arm's end. Specifically, during extraction, the force components in multiple directions of the actual contact force at the end are treated as spatial force components, and the torque components in multiple directions are treated as spatial torque components. These spatial force components and spatial torque components are then systematically encapsulated according to the same spatial representation caliber at the arm's end reference point to obtain the force spinor. Thus, the force spinor inherits the actual contact force at the end of the arm and provides a force data object that simultaneously expresses the force direction, force intensity, and rotational influence for subsequent input into the overall dynamics mapping.

[0038] Preferably, the force spinor is input into the overall dynamics mapping relationship. The specific implementation process of the overall dynamics mapping relationship, which includes a transmission matrix characterizing the relative motion between the rigid body links, is as follows: After the force spinor has been extracted from the actual contact force at the end, the current spatial pose of the centaur robot's torso, the current angles of each joint of the two arms, the current angles of each joint of the four legs, the installation position of the ends of the two arms relative to the torso, and the connection sequence between the rigid body links in the four legs are read. The above data and the force spinor are written into the overall dynamics mapping input bearing record. The overall dynamics mapping input bearing record is used to include the force source of the force spinor, the attitude source of the current spatial pose of the centaur robot's torso, the arm configuration source of the current angles of each joint of the two arms, the four-leg configuration source of the current angles of each joint of the four legs, and the structural source of the connection sequence between the rigid body links in the four legs into the same input caliber, so that the overall dynamics mapping relationship can be transformed according to the current overall posture rather than a fixed calibration posture. The technical essence of the overall dynamics mapping relationship is to describe how the forces at the ends of the arms are transmitted along the rigid links of the arms, the rigid links of the torso, and the rigid links of the four legs to the joints of the four legs. This force transmission data relationship originates from the link lengths, joint connection sequences, joint rotation directions, and coordinate apertures of each rigid link formed during the Centaur robot's assembly and calibration phase. It is updated in the current control cycle based on the current spatial pose of the Centaur robot's torso, the current angles of each joint of the arms, and the current angles of each joint of the four legs. The technical essence of the transmission matrix is ​​to serve as a parameterized carrier representing the relative motion and force transmission ratio between the rigid links in the overall dynamics mapping relationship. The rows in the transmission matrix correspond to the spatial force components and spatial torque components in the force spinor, and the columns correspond to the joint rotation degrees of freedom in the space of each joint of the four legs. The elements at the intersection of rows and columns represent the transmission ratio of the torque contribution of a certain spatial force component or spatial torque component to a certain joint rotation degree of freedom. During the specific input process, the force spinor is first converted to a unified force expression caliber for the centaur robot body based on its current torso spatial pose. Then, the transmission ratio in the transmission matrix is ​​updated based on the current angles of the joints of the two arms and the joints of the four legs, so that the transmission matrix can reflect the geometric relationship between the ends of the two arms and the joints of the four legs in the current posture. Subsequently, the force spinor converted to the unified force expression caliber of the centaur robot body and the updated transmission matrix are jointly incorporated into the overall dynamics mapping relationship, allowing the force spinor to continue to participate in subsequent mapping processes along the overall dynamics mapping relationship.Therefore, the force spin is no longer an independent force information at the ends of the two arms, but is embedded in the rigid body linkage force transmission link expressed by the whole machine dynamics mapping relationship, providing a structured mapping basis for obtaining the equivalent joint torque of each joint space of the four legs.

[0039] Preferably, the specific implementation process of mapping the force spinor using the transpose of the transmission matrix to obtain the equivalent joint torque of each joint space of the four legs is as follows: After the force spinor, converted to the unified force expression caliber of the centaur robot body, and the updated transmission matrix are entered into the whole-machine dynamics mapping relationship, the transpose of the transmission matrix is ​​first generated according to the correspondence between rows and columns in the transmission matrix. The transpose of the transmission matrix does not reconstruct another set of force transmission relationships, but rather, while keeping the transmission ratio in the transmission matrix unchanged, reverses the correspondence originally used to describe the influence of small joint movements on the spatial changes of the arm ends to a correspondence used to describe the contribution of the force at the arm ends to the torque of each joint space of the four legs. The technical essence of the mapping process is to distribute the spatial force component and spatial torque component in the force spinor to each joint space of the four legs according to the transpose of the transmission matrix, so that the contact load at the arm ends is converted into a force rectangle that each joint of the four legs can execute. In specific implementation, the spatial force components and spatial torque components in the force spinor are first arranged according to the row order of the transmission matrix, so that the spatial force components and spatial torque components correspond to the columns in the transpose of the transmission matrix. Then, according to the transpose of the transmission matrix, the torque contribution of each spatial force component and each spatial torque component to the joint rotational degrees of freedom in each joint space of the four legs is calculated item by item, and the torque contributions belonging to the same joint rotational degree of freedom are superimposed to form the equivalent joint torque of each joint space of the four legs. The technical essence of the equivalent joint torque of each joint space of the four legs is to convert the actual contact force at the ends of the two arms to the equivalent force result obtained after transferring it to each joint space of the four legs. This equivalent force result does not directly represent the motor output command, but rather represents the torque component caused by the contact load at the ends of the two arms that each joint of the four legs needs to bear under the current overall posture and current force transmission constraints. To ensure that the equivalent joint torques of each joint space of the four legs can continue to participate in the subsequent determination of the compensation torque, after mapping processing, a correspondence is established between the equivalent joint torques of each joint space of the four legs and the corresponding joint identifier, the corresponding supporting leg identifier, and the current control cycle marker. This allows the torque component of each joint rotational degree of freedom to be traced back to the spatial force component and the spatial torque component in the force spinor. Thus, the transpose of the transmission matrix transforms the force information at the ends of the two arms into torque distribution information on the joint sides of the four legs, enabling the equivalent joint torques of each joint space of the four legs to serve as a direct basis for subsequently determining the compensation torque of each joint.

[0040] Preferably, the specific implementation process of determining the compensation torque of each joint based on the equivalent joint torque of each joint space of the four legs is as follows: After the equivalent joint torque of each joint space of the four legs is calculated, the equivalent joint torque of each joint space of the four legs, the corresponding joint identifier, the corresponding supporting leg identifier, the current control cycle marker, the current torque feedback value of each joint of the four legs, and the foot position of the four legs currently in contact with the ground are read, and a compensation torque determination load record is established. The compensation torque determination load record is used to include the force mapping source of the equivalent joint torque of each joint space of the four legs, the joint object source of the corresponding joint identifier, the support object source of the corresponding supporting leg identifier, the timing source of the current control cycle marker, the execution state source of the current torque feedback value of each joint of the four legs, and the support contact source of the foot position of the four legs currently in contact with the ground into the same determination caliber, so that the compensation torque formed subsequently can correspond to the current overall machine force state and the current four-leg support state. The technical essence of the compensation torque is that, based on the equivalent joint torque of each joint space of the four legs, a joint-level torque correction amount is formed by combining the current torque feedback value of each joint of the four legs and the current foot position in contact with the ground. The joint-level torque correction amount is used to offset or absorb the residual impact tendency caused by the actual contact force of the end when the active impedance control command is updated. In specific implementation, firstly, the equivalent joint torque of each joint space of the four legs is distributed to each joint of the four legs according to the corresponding joint identifier. Then, based on the corresponding support leg identifier and the current foot position of the four legs in contact with the ground, it is determined whether the support leg of the corresponding joint is in a grounded state. For joints on support legs in a grounded state, the equivalent joint torque of each joint space of the four legs corresponding to that joint is checked for directional consistency with the current torque feedback value of each joint of the four legs corresponding to that joint to determine the resistance torque component that needs to be increased or the compensatory torque component that needs to be reduced for that joint. For joints on support legs that are not in a grounded state, the equivalent joint torque of each joint space of the four legs corresponding to that joint is marked as a source of torque to be transferred. The source of torque to be transferred is used to bear the corresponding torque correction amount in the subsequent grounded support leg. Subsequently, amplitude limiting and continuous smoothing processing are performed on each joint of the four legs on the grounded support legs to keep the compensation torque within an executable range of variation between adjacent control cycles and reduce abrupt conflicts between the compensation torque and the current torque feedback value of each joint of the four legs. Finally, the joint-level torque correction amount after the direction consistency verification, amplitude limiting processing and continuity smoothing processing is used as the compensation torque of each joint, and the compensation torque of each joint, the corresponding joint identifier, the corresponding support leg identifier and the current control cycle mark are written into the compensation torque output record.The compensation torque output record continues to participate in the subsequent active impedance control command update, so that the actual contact force at the end can be converted into the compensation torque that can be used by each joint of the four legs after passing through the force spinor, the whole machine dynamics mapping relationship and the equivalent joint torque of each joint space of the four legs, thereby providing a joint-level control basis for each joint of the four legs to absorb residual impact force.

[0041] Optionally, updating the active impedance control command of each joint of the four legs based on the compensated torque includes: Obtain the current stiffness parameters and damping parameters of each joint of the four legs; The compensation torque is introduced into the basic control closed loop to determine the torque compensation requirements of each joint in the corresponding impact direction; The damping parameter configuration value in the corresponding impact direction is increased according to the torque compensation requirement; The configuration value of the stiffness parameter in the corresponding impact direction is reduced according to the torque compensation requirement; Based on the increased configuration value of the damping parameter and the decreased configuration value of the stiffness parameter, a target impedance parameter is determined, and an active impedance control command is generated based on the target impedance parameter.

[0042] Preferably, the specific implementation process for obtaining the current stiffness parameters and damping parameters of each joint of the four legs is as follows: After the compensation torque of each joint has been formed by the actual contact force at the end through the force spinor, the whole machine dynamics mapping relationship and the equivalent joint torque of each joint space of the four legs, the joint impedance state record corresponding to the current control cycle is read first. The joint impedance state record includes the current stiffness parameters of each joint of the four legs, the current damping parameters of each joint of the four legs, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, the current torque feedback value of each joint of the four legs, the target angle of each joint of the four legs, the corresponding support leg identifier, the foot position of the four legs currently in contact with the ground, and the allowable adjustment range of each joint of the four legs. The joint impedance state record is used to integrate the following sources into a single reading caliber: the source of the restoring capability of the current stiffness parameters of each joint of the four legs; the source of the energy dissipation capability of the current damping parameters of each joint of the four legs; the source of the attitude of the current angle of each joint of the four legs; the source of the velocity of the current angular velocity of each joint of the four legs; the source of the execution status of the current torque feedback value of each joint of the four legs; the source of the target attitude of the target angle of each joint of the four legs; the source of the support object corresponding to the supporting leg identifier; the source of the grounding of the foot end position currently in contact with the ground of each leg; and the source of the adjustment boundary of the allowable adjustment range of each joint of the four legs. The technical essence of the stiffness parameter is an impedance configuration quantity used to represent the ability of each joint of the four legs to generate restoring torque when deviating from the target angle of each joint. The stiffness parameter is not a fixed mechanical constant, but rather joint control configuration data formed within the current control cycle based on the corresponding supporting leg identifier, the current foot end position in contact with the ground of each leg, the torso spatial pose, and the allowable adjustment range of each joint of the four legs. The technical essence of the damping parameter is an impedance configuration amount used to represent the energy dissipation suppression of each joint of the four legs in response to changes in the current angular velocity of each joint of the four legs. The damping parameter is not simply a velocity threshold, but rather joint control configuration data used to control each joint of the four legs to absorb energy with executable angular velocity changes when subjected to residual impact force disturbance. Specifically, during generation, firstly, the joints of the four legs in the support force link are determined based on the corresponding support leg identifier and the current foot position of the four legs in contact with the ground. The support force link is used to represent the joint connection path from the grounded support leg to the torso. Then, the attitude deviation state and velocity change state of each joint of the four legs in the current control cycle are determined based on the current angle and current angular velocity of each joint of the four legs. Subsequently, the current torque feedback value of each joint of the four legs is used to determine whether the existing output of each joint of the four legs is close to the allowable adjustment range of each joint of the four legs, so as to form the current joint impedance reference record.The current joint impedance reference record inherits from the joint impedance state record and serves as the direct basis for reading the stiffness and damping parameters, enabling the understanding of the original recovery and energy dissipation capabilities of each joint in the four legs when the compensation torque is subsequently introduced. Through the above reading and generation method, the stiffness parameter is used to limit the strength of each joint in the four legs in resisting displacement deviation, and the damping parameter is used to limit the strength of each joint in the four legs in absorbing velocity disturbances. Together, they provide a traceable initial impedance state for subsequently determining the torque compensation requirements, adjusting the configuration values ​​of the damping parameters, and adjusting the configuration values ​​of the stiffness parameters.

[0043] Preferably, in the specific technical implementation of introducing the compensation torque into the basic control closed loop to determine the torque compensation requirements of each joint in the corresponding impact direction, the basic control closed loop in this application is specifically manifested as a joint impedance control loop formed by the target angle of each joint of the four legs, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, stiffness parameters, damping parameters, and the current torque feedback value of each joint of the four legs. This joint impedance control loop is used to generate basic joint control quantities based on the attitude deviation of each joint's current angle relative to its target angle and the velocity change corresponding to its current angular velocity. The basic joint control quantities are used to represent the basic joint control quantities formed by the stiffness parameters and damping parameters when the compensation torque is not introduced. The basic joint control quantities continue to participate in the coordination judgment between the compensation torque and the original joint control state. To avoid inconsistencies between the source of the compensation torque and the basic joint control quantity, a torque compensation demand analysis load record is first established. The compensation torque, stiffness parameter, damping parameter, target angle of each joint of the four legs, current angle of each joint of the four legs, current angular velocity of each joint of the four legs, current torque feedback value of each joint of the four legs, basic joint control quantity, corresponding joint identifier, and corresponding support leg identifier are written into this torque compensation demand analysis load record. This torque compensation demand analysis load record is used to incorporate the external impact source of the compensation torque, the source of the recovery capability of the stiffness parameter, the source of the energy dissipation capability of the damping parameter, the source of the target attitude of each joint target angle of the four legs, the source of the attitude of each joint current angle of the four legs, the source of the velocity of each joint current angular velocity of the four legs, the source of the execution status of each joint current torque feedback value of the four legs, the original control source of the basic joint control quantity, the source of the joint object of the corresponding joint identifier, and the source of the support object of the corresponding support leg identifier into the same analysis caliber. In practice, the compensation torque is first matched to each joint of the four legs according to the corresponding joint identifier. Then, the direction of the compensation torque is consistent with the direction of the current torque feedback value of each joint of the four legs to distinguish whether the joints of the four legs need to enhance energy dissipation in the impact direction or only need to maintain the original impedance configuration in the non-impact direction. The technical essence of the corresponding impact direction is that the direction of residual impact force transmission is expressed by the direction of the compensation torque in the joint space. The corresponding impact direction is used to indicate whether the joints of the four legs should mainly dissipate impact energy by increasing damping or allow limited yielding by reducing stiffness.Subsequently, the amplitude of the compensation torque, the restoring torque capability corresponding to the stiffness parameter, the energy dissipation suppression capability corresponding to the damping parameter, the current angular velocity of each joint of the four legs, and the control quantity of the basic joint are proportionalized to form the torque compensation requirement of each joint in the corresponding impact direction. The technical essence of this torque compensation requirement is to convert the compensation torque into a data-driven requirement for adjusting the impedance parameter amplitude. This torque compensation requirement is not a directly issued joint torque command, but rather an intermediate control basis used to guide subsequent increases in the damping parameter configuration value and decreases in the stiffness parameter configuration value. Thus, the compensation torque enters the torque compensation requirement analysis and load record via the joint impedance control loop and is transformed into the torque compensation requirement, enabling the residual impact tendency caused by the actual contact force at the ends of the two arms to enter the four-leg joint impedance parameter update process.

[0044] Preferably, the specific implementation process of increasing the configuration value of the damping parameter in the corresponding impact direction according to the torque compensation requirement is as follows: After the torque compensation requirement is formed, the torque compensation requirement, the damping parameter, the current angular velocity of each joint of the four legs, the current torque feedback value of each joint of the four legs, the allowable adjustment range of each joint of the four legs, the corresponding joint identifier, the corresponding supporting leg identifier, and the corresponding impact direction are read from the torque compensation requirement parsing load record, and a damping parameter configuration value adjustment load record is established. The damping parameter configuration value adjustment load record is used to incorporate the source of the compensation amplitude of the torque compensation requirement, the source of the original energy dissipation capacity of the damping parameter, the source of the velocity change of the current angular velocity of each joint of the four legs, the source of the execution status of the current torque feedback value of each joint of the four legs, the source of the adjustment boundary of the allowable adjustment range of each joint of the four legs, and the source of the direction of the corresponding impact direction into the same adjustment caliber. The technical essence of the configured value of the damping parameter is that it is the value of the damping parameter actually generated by the active impedance control command in the current control cycle. This configured value of the damping parameter is used to determine the energy dissipation suppression strength when the angular velocity of each joint of the four legs changes along the corresponding impact direction. Specifically, during adjustment, the target joint requiring damping adjustment is first determined based on the corresponding joint identifier. The target joint originates from each joint of the four legs pointed to by the corresponding joint identifier and is used to receive the torque compensation requirement. Then, it is determined whether the support leg where the target joint is located is in a grounded state based on the corresponding support leg identifier. For the target joint in a grounded state, the damping increment is determined based on the magnitude of the torque compensation requirement, and the damping increment is added to the damping parameter to obtain the increased configured value of the damping parameter. The damping increment originates from the torque compensation requirement and is used to represent the amount of energy dissipation suppression that needs to be added based on the damping parameter. For the target joint not in a grounded state, the torque compensation requirement is reserved as a source of damping adjustment to be transferred. This source of damping adjustment to be transferred is used to absorb residual impact force in the subsequent operation by the joint on the grounded support leg. To ensure the stable execution of the increased damping parameter configuration value, it is also determined whether there is a significant velocity change in each joint of the four legs based on the current angular velocity of each joint. When the current angular velocity of each joint of the four legs is consistent with the corresponding impact direction, the increased damping parameter configuration value is used to improve the energy dissipation capacity of each joint of the four legs when moving along the corresponding impact direction. When the current angular velocity of each joint of the four legs is opposite to the corresponding impact direction, the increased damping parameter configuration value is used to suppress the secondary disturbance formed during the reverse rebound process of each joint of the four legs.Subsequently, based on the current torque feedback values ​​of each joint of the four legs and the allowable adjustment range of each joint, the increased configuration value of the damping parameter is subjected to amplitude limiting processing to ensure that the increased configuration value of the damping parameter matches the torque range that the joint can currently execute. Thus, the torque compensation requirement is converted into the increased configuration value of the damping parameter through the adjustment of the load record using the damping parameter configuration value. This allows the energy dissipation capacity of the joint in the corresponding impact direction to be adjusted according to the intensity of the residual impact force transmission, providing a damping-side parameter source for the subsequent formation of the target impedance parameter.

[0045] Preferably, the specific implementation process of reducing the configuration value of the stiffness parameter in the corresponding impact direction according to the torque compensation requirement is as follows: After the torque compensation requirement has been used to form the configuration value of the increased damping parameter, the torque compensation requirement, the stiffness parameter, the target angle of each joint of the four legs, the current angle of each joint of the four legs, the current torque feedback value of each joint of the four legs, the allowable adjustment range of each joint of the four legs, the corresponding joint identifier, the corresponding support leg identifier, and the corresponding impact direction are read from the torque compensation requirement analysis load record, and a stiffness parameter configuration value adjustment load record is established. The stiffness parameter configuration value adjustment load record is used to include the source of the yield amplitude of the torque compensation requirement, the source of the original recovery capability of the stiffness parameter, the source of the target attitude of each joint of the four legs, the source of the attitude deviation of the current angle of each joint of the four legs, the source of the execution status of the current torque feedback value of each joint of the four legs, the source of the adjustment boundary of the allowable adjustment range of each joint of the four legs, and the source of the direction of the corresponding impact direction into the same adjustment caliber. The technical essence of the configured stiffness parameter is that it is the stiffness parameter value actually generated by the active impedance control command in the current control cycle. This configured stiffness parameter is used to determine the strength of the restoring torque generated when each joint of the four legs deviates from its target angle. If the configured stiffness parameter remains high when the residual impact force is transmitted to each joint of the four legs, the joints of the four legs will tend to immediately push back to their target angle, thereby transmitting the residual impact force back to the torso. Therefore, this application reduces the configured stiffness parameter in the corresponding impact direction according to the torque compensation requirement, so that each joint of the four legs has limited room for retreat in the impact direction. In practice, firstly, the target joint requiring stiffness adjustment is determined based on the corresponding joint identifier. This target joint originates from each joint of the four legs indicated by the corresponding joint identifier and is used to accommodate the torque compensation requirement. Then, based on the corresponding support leg identifier and the current foot position in contact with the ground, it is determined whether the target joint is located within an effective support link. The effective support link represents the joint connection path in the grounded support leg that can transmit foot support to the torso. For the target joint located within the effective support link, a stiffness reduction is determined based on the torque compensation requirement, and this reduction is subtracted from the stiffness parameters to obtain the reduced configuration value of the stiffness parameters. This stiffness reduction originates from the torque compensation requirement and represents the required reduction in restoring torque capability based on the stiffness parameters.Subsequently, based on the current angles of each joint of the four legs and the allowable adjustment range of each joint, it is determined whether the target joint has approached the allowable adjustment range of each joint of the four legs. When the target joint is close to the allowable adjustment range of each joint of the four legs, the reduced stiffness parameter configuration value retains the minimum recovery capability to reduce the target joint from continuing to excessively yield in the corresponding impact direction. When the target joint is not close to the allowable adjustment range of each joint of the four legs, the reduced stiffness parameter configuration value allows the target joint to generate a limited angular displacement along the corresponding impact direction, in order to cooperate with the increased damping parameter configuration value to consume the energy carried by the residual impact force. Subsequently, based on the current torque feedback value of each joint of the four legs, the reduced stiffness parameter configuration value is boundary-corrected to ensure that the reduced stiffness parameter configuration value maintains a continuous change with the current joint output state. Therefore, the torque compensation requirement results in a reduced stiffness parameter configuration value on the stiffness side, so that each joint of the four legs no longer directly resists the residual impact force with high stiffness in the corresponding impact direction, but instead dissipates energy through limited yielding and damping, providing a source of stiffness side parameters for the formation of subsequent target impedance parameters.

[0046] Preferably, the specific implementation process of determining the target impedance parameter based on the increased configuration value of the damping parameter and the decreased configuration value of the stiffness parameter, and generating the active impedance control command based on the target impedance parameter, is as follows: After the increased configuration value of the damping parameter and the decreased configuration value of the stiffness parameter are formed respectively, a target impedance parameter generation load record is first established, and the increased configuration value of the damping parameter, the decreased configuration value of the stiffness parameter, the corresponding impact direction, the corresponding joint identifier, the corresponding support leg identifier, the target angle of each joint of the four legs, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, and the current torque feedback value of each joint of the four legs are written into the target impedance parameter generation load record. The target impedance parameter generation record is used to incorporate the energy dissipation source of the increased damping parameter configuration value, the yielding source of the decreased stiffness parameter configuration value, the direction source of the corresponding impact direction, the joint object source of the corresponding joint identifier, the support object source of the corresponding supporting leg identifier, the target posture source of each joint target angle of the four legs, and the current joint state source into the same generation caliber. The technical essence of the target impedance parameter is the configuration of stiffness and damping combinations formed for each joint of the four legs within the current control cycle. The target impedance parameter includes both the decreased stiffness parameter configuration value and the increased damping parameter configuration value, as well as the joint object corresponding to the target impedance parameter and the corresponding impact direction. Specifically, the increased damping parameter configuration value and the decreased stiffness parameter configuration value are first combined under the same joint object according to the corresponding joint identifier. Then, directional constraints are configured for the joint object according to the corresponding impact direction, so that the target impedance parameter only changes impedance performance in the direction where the residual impact force is mainly transmitted, while retaining the original impedance continuity in non-impact directions. Subsequently, based on the current angles and angular velocities of each joint of the four legs, the target impedance parameter undergoes continuity processing for adjacent control cycles, enabling a smooth transition from the impedance configuration of the previous control cycle to the impedance configuration of the current control cycle. After the continuity processing is completed, the active impedance control command is generated based on the target impedance parameter. Specifically, the target impedance parameter, the target angles of each joint of the four legs, the current angles of each joint of the four legs, the current angular velocities of each joint of the four legs, and the current torque feedback values ​​of each joint of the four legs are encapsulated together into joint control command data, so that the active impedance control command can simultaneously include joint recovery capability, joint energy dissipation capability, and current joint state.The technical essence of the active impedance control command is to drive each joint of the four legs to execute joint-level control data for impedance adjustment in the impact direction according to the target impedance parameters. The active impedance control command does not simply increase the output torque; rather, after the compensation torque has been converted into the torque compensation requirement, it adjusts the way each joint of the four legs absorbs residual impact force by jointly adjusting the configuration values ​​of the increased damping parameters and decreased stiffness parameters. Thus, the target impedance parameters, corresponding to the torque compensation requirement of the compensation torque, are applied to each joint of the four legs through the active impedance control command, enabling each joint to absorb residual impact force while maintaining a supporting state.

[0047] Optionally, adjusting the support output of the corresponding supporting foot among the four feet according to the distribution parameters of the plantar support force of the four feet to coordinately adjust the spatial pose of the torso, thereby causing the center of mass corresponding to the spatial pose of the torso to shift in a direction opposite to the predicted external force direction, includes: Obtain the geometric center coordinates of the support region enclosed by the boundary of the support polygon in the horizontal plane; Based on the geometric center coordinates and the predicted external force direction characteristics, an offset vector is established to indicate the direction of the centroid movement, and the direction of the offset vector is opposite to the predicted external force direction characteristics. The adjustment weights of each supporting foot are configured for the bias vector based on the distribution parameters of the four-legged foot support force. The hip joint pitch angle of the corresponding supporting leg is adjusted according to the adjustment weight to drive the center of mass corresponding to the torso spatial pose to move along the direction of the offset vector to implement the displacement of the direction.

[0048] Preferably, the specific implementation process for obtaining the geometric center coordinates of the support region enclosed by the support polygon boundary on the horizontal plane is as follows: After the support force distribution parameters of the four legs have been formed according to the expected vertical reaction force of each support leg, the current foot position of the four legs in contact with the ground, the corresponding support leg identifier, the current grounding state of the support leg, the support force distribution parameters of the four legs, and the current centroid position of the Centaur robot are read from the support force distribution adjustment bearing record, and a geometric analysis bearing record of the support region is established. The geometric analysis bearing record of the support region is used to include the spatial support source of the current foot position of the four legs in contact with the ground, the support object source of the corresponding support leg identifier, the contact effectiveness source of the current grounding state of the support leg, the support allocation source of the support force distribution parameters of the four legs, and the attitude influence source of the current centroid position of the Centaur robot into the same analytical caliber. In specific implementation, firstly, based on the current grounding state of the supporting legs, the foot positions currently in contact with the ground are selected from the foot positions of the four legs. These selected grounded foot positions are then projected onto a horizontal plane along the front-back and left-right directions of the Centaur robot body to obtain the horizontal plane foot projection positions. The technical essence of these horizontal plane foot projection positions is the two-dimensional support coordinates of the grounded supporting legs on the ground support surface. These two-dimensional support coordinates, as the coordinate expression of the horizontal plane foot projection positions, describe the actual geometric range within which the four legs can provide support to the torso during the current control cycle. Subsequently, adjacent horizontal plane foot projection positions are connected in a circular sequence on the horizontal plane to form the support polygon boundary. The technical essence of this support polygon boundary is the support stability boundary enclosed by the grounded supporting legs on the horizontal plane, used to constrain the acceptable range of motion of the Centaur robot's current center of mass position and subsequent actual zero-torque point coordinates. Next, the support region is formed by enclosing the boundaries of the supporting polygons. This support region is not an arbitrarily drawn planar area, but rather a horizontal support range defined by the boundaries of the supporting polygons. This support region represents the spatial foundation upon which the Centaur robot can provide stable support for the entire machine through its four legs under the current support configuration. Then, the horizontal coordinates of each boundary vertex of the support region are averaged, and the positions obtained from the averaged horizontal coordinates are verified within the region, taking into account the boundary enclosure shape of the support region, to obtain the geometric center coordinates. The technical essence of these geometric center coordinates is the geometric reference position of the support region on the horizontal plane, used to provide a stable central reference for subsequently establishing the offset vector, rather than directly serving as the final centroid target position.Therefore, the geometric center coordinates inherit the geometric relationship between the boundary of the supporting polygon and the supporting area, and maintain the same support source as the support force distribution parameters of the four feet. This allows the bias vector to be established in the subsequent process based on the geometric center coordinates and the predicted external force direction characteristics, so that the geometric state of the four-foot support and the predicted direction of the reaction load of the double-arm operation can be included in the same adjustment process.

[0049] Preferably, in the specific technical implementation of establishing an offset vector to indicate the direction of the centroid movement based on the geometric center coordinates and the predicted external force direction characteristics, the geometric center coordinates, the support area, the boundary of the support polygon, and the current centroid position of the Centaur robot are first read from the geometric analysis bearing record of the support area. Simultaneously, the predicted external force direction characteristics and the predicted external force amplitude characteristics obtained from the operation task category are read to establish an offset vector to construct the bearing record. The offset vector to construct the bearing record is used to incorporate the source of the support area center (geometric center coordinates), the source of the mass distribution at the current centroid position of the Centaur robot, the source of the external force direction characteristic, and the source of the external force intensity characteristic into the same construction caliber. The technical essence of the predicted external force direction characteristic is to represent the direction of the reaction load in the local coordinate system of the Centaur robot. This direction expression has been formed by decomposing the initial impact force vector through the local coordinate system, and therefore can be directly used to determine the support adjustment direction on the horizontal plane. In specific implementation, the predicted external force direction features are first projected onto the horizontal plane in the local coordinate system of the Centaur robot, along with their forward and backward components and left and right components, to obtain the horizontal external force direction. This horizontal external force direction represents the main direction in which the reaction load tends to overturn the torso within the plane of the support area. Subsequently, the horizontal external force direction is reversed to obtain the horizontal resistance direction. The technical essence of this horizontal resistance direction is a support pre-adjustment direction opposite to the horizontal external force direction. This direction guides the center of mass corresponding to the torso's spatial pose to move to the opposite side of the reaction load's direction, thus creating a posture buffer margin before the arms generate impact force during operation. Next, the offset vector is constructed based on the geometric center coordinates and the horizontal resistance direction. The starting reference position of the offset vector is the geometric center coordinates. The direction of the offset vector is consistent with the horizontal resistance direction and opposite to the projection direction of the predicted external force direction features onto the horizontal plane. The length of the offset vector is jointly defined by the predicted external force amplitude features and the boundary margin of the support area. The boundary margin of the support area originates from the remaining distance between the termination position of the offset vector and the boundary of the support polygon. This boundary margin limits the length of the offset vector, ensuring that the target of the center of mass movement pointed to by the offset vector does not exceed the support area. The technical essence of the offset vector is to convert the predicted direction of external force into a horizontal plane quantity representing the direction of center of mass movement. The offset vector is not a standalone mathematical symbol, but rather serves as the basis for subsequently configuring the adjustment weights of each support foot, adjusting the hip joint pitch angle of the corresponding support leg, and determining the direction that drives the center of mass displacement.To ensure the offset vector remains within the support region, its termination position is constrained within the region based on the boundary of the support polygon, ensuring that the target centroid movement pointed to by the offset vector remains within the support region. Thus, the offset vector incorporates the geometric center coordinates and the predicted external force direction characteristics, linking the quadrupedal support geometry with the predicted direction of the dual-arm operation reaction load. This provides a directional basis for subsequently configuring the adjustment weights of each support foot for the offset vector based on the quadrupedal foot support force distribution parameters.

[0050] Preferably, the specific implementation process of configuring the adjustment weights of each supporting foot for the offset vector based on the four-legged foot support force distribution parameters is as follows: After the offset vector has been established based on the geometric center coordinates and the predicted external force direction characteristics, the offset vector, the support area, the support polygon boundary, the current centroid position of the centaur robot, and the predicted external force amplitude characteristics are read from the offset vector construction bearing record. Then, the four-legged foot support force distribution parameters, corresponding supporting foot identifiers, current grounding status of the supporting feet, allowable adjustment range of the supporting feet, and the foot reaction force adjustment amount of each supporting foot are read from the support force distribution adjustment bearing record to establish an adjustment weight configuration bearing record. The adjustment weight configuration bearing record is used to incorporate the source of the centroid movement direction of the offset vector, the source of the support allocation of the four-legged foot support force distribution parameters, the source of the execution object of the corresponding supporting foot identifier, the source of the contact effectiveness of the current grounding status of the supporting feet, the source of the adjustment boundary of the allowable adjustment range of the supporting feet, and the source of the vertical support change of the foot reaction force adjustment amount of each supporting foot into the same configuration caliber. The technical essence of the adjustment weight is the proportional configuration result used when allocating the bias vector to each support foot in the grounded state. This proportional configuration result is used to explain the support adjustment share that each support foot needs to undertake when driving the center of mass to move along the direction of the bias vector. In specific configuration, firstly, support feet that can participate in the center of mass movement adjustment are screened according to the current grounding state of the support feet, and support feet that cannot form effective surface contact are excluded from the adjustment weight configuration, so that the adjustment weight is only allocated to support feet that can actually change the support output. Subsequently, the spatial orientation relationship of each support foot relative to the geometric center coordinates and the current center of mass position of the Centaur robot is determined according to the direction of the bias vector; support feet located on the side of the bias vector and having a higher foot reaction force adjustment capability are assigned a higher adjustment weight; support feet located on the opposite side of the bias vector, or support feet with a smaller allowable adjustment range, are assigned a lower adjustment weight. Next, the target vertical support ratio in the four-leg foot support force distribution parameters and the foot reaction force adjustment amount of each support foot are jointly normalized to form the adjustment weight of each support foot. The target vertical support ratio is derived from the plantar support force distribution parameters of the four legs. This target vertical support ratio represents the proportion of vertical support each support leg needs to bear in the current control cycle. The plantar reaction force adjustment amount of each support leg represents the increase or decrease in vertical support required for each support leg to transition from the current plantar reaction force to the desired vertical reaction force. Therefore, the adjustment weight formed based on the target vertical support ratio and the plantar reaction force adjustment amount of each support leg reflects both the directional requirement of the bias vector and the already determined support force adjustment capability of the four-leg plantar support force distribution parameters.The adjustment weights do not replace the quadrupedal foot support force distribution parameters, but rather transform the vertical support distribution relationship in the quadrupedal foot support force distribution parameters into a proportional basis for subsequent hip joint pitch angle adjustment. To enable the adjustment weights to continue participating in the hip joint pitch angle adjustment of the corresponding supporting leg, a correspondence is established between the adjustment weights of each supporting foot and the corresponding supporting leg identifier, the corresponding supporting foot identifier, and the allowable adjustment range of the supporting foot. Thus, the adjustment weights inherit the bias vector and the quadrupedal foot support force distribution parameters, and provide an executable proportional allocation basis for subsequent adjustment of the hip joint pitch angle of the corresponding supporting leg based on the adjustment weights. This allows the direction of the center of mass movement to move beyond the direction planning level and enter the joint posture adjustment process on the quadrupedal support side.

[0051] Preferably, the specific implementation process of adjusting the hip joint pitch angle of the corresponding supporting leg according to the adjustment weight to drive the center of mass corresponding to the trunk spatial pose to move along the direction of the bias vector to implement the displacement of the direction is as follows: After the adjustment weight of each supporting foot is formed, the adjustment weight, the bias vector, the corresponding supporting leg identifier, the corresponding supporting foot identifier, the current grounding state of the supporting foot, the allowable adjustment range of the supporting foot, the support force distribution parameters of the four feet, the current trunk spatial pose of the Centaur robot, and the current center of mass position of the Centaur robot are read from the adjustment weight configuration load record, and a hip joint pitch angle adjustment load record is established. The hip joint pitch angle adjustment load record is used to incorporate the following sources into the same adjustment caliber: the proportional allocation of the adjustment weight, the centroid movement direction of the bias vector, the execution object of the corresponding supporting leg identifier, the foot end object of the corresponding supporting foot identifier, the contact effectiveness of the current grounding state of the supporting foot, the adjustment boundary of the allowable adjustment range of the supporting foot, the support force adjustment source of the four-legged foot support force distribution parameters, and the attitude source of the current torso spatial pose of the Centaur robot. The technical essence of the hip joint pitch angle is the joint angle that changes the posture of the supporting leg in the forward-backward direction of the Centaur robot body. The hip joint pitch angle can change the relative positional relationship between the foot end of the supporting leg and the torso, thereby changing the impact of the foot support force transmitted to the torso on the torso spatial pose. In specific implementation, firstly, a hip joint pitch angle adjustment amount is allocated to each corresponding supporting leg according to the adjustment weight. The direction of the hip joint pitch angle adjustment amount is determined by the direction of the offset vector, and the amplitude of the hip joint pitch angle adjustment amount is jointly limited by the adjustment weight and the allowable adjustment range of the supporting foot. The hip joint pitch angle adjustment amount is used to convert the adjustment weight from a proportional allocation result into joint angle adjustment data that the corresponding supporting leg can perform. Subsequently, the corresponding supporting legs that can perform the hip joint pitch angle adjustment amount are selected according to the current grounding state of the supporting foot, and the adjustment share corresponding to the supporting legs that are not in a grounding state is redistributed to the corresponding supporting legs that are in a grounding state and still have adjustment margin. The adjustment margin comes from the difference between the allowable adjustment range of the supporting foot and the allocated hip joint pitch angle adjustment amount, and the adjustment margin is used to prevent the corresponding supporting legs in a grounding state from exceeding the allowable adjustment range of the supporting foot. Next, the hip joint pitch angle of the corresponding supporting leg is adjusted according to the hip joint pitch angle adjustment amount, so that the relative position between the foot support point of the corresponding supporting leg and the torso changes, and the support output corresponding to the four-leg foot support force distribution parameters forms a resultant force adjustment trend along the direction of the offset vector.The technical essence of the movement of the centaur robot's center of mass along the direction of the offset vector, corresponding to its torso spatial pose, is to change the positional relationship of the torso relative to the support area by altering the pitch angle of the hip joint of the corresponding supporting leg. This causes the current center of mass of the centaur robot to undergo a controlled displacement from its original position toward the direction of the offset vector, rather than simply changing the posture of a single supporting leg. To ensure that the center of mass remains within the support area when moving along the direction of the offset vector, the termination position of the centaur robot's current center of mass movement is constrained according to the boundary of the support polygon. Furthermore, the adjustment amount of the hip joint pitch angle of the corresponding supporting leg is reduced when the current center of mass of the centaur robot approaches the boundary of the support polygon. Thus, the adjustment weight is transformed into a pre-adjustment of the torso spatial pose through the adjustment of the hip joint pitch angle of the corresponding supporting leg. This causes the center of mass corresponding to the torso spatial pose to move along the direction of the offset vector, thereby implementing displacement of the center of mass in a direction opposite to the predicted external force direction, providing a pre-positional basis for absorbing residual impact forces.

[0052] Optionally, after mapping the force spinor using the transpose of the conduction matrix to obtain the equivalent joint torques of each joint space of the four legs, and then generating the compensating torques of each joint from the equivalent joint torques, the method further includes: Extract the current swing state features of the centaur robot to identify the supporting leg that is in a grounded state from the four legs; Identify the target force-bearing joint on the corresponding support leg based on the support leg markings; Select the joint torque to be transferred corresponding to the suspended leg from the equivalent joint torques; The torque of the joint to be transferred is distributed to the target force-bearing joint to complete the redistribution of the compensation torque of the target force-bearing joint, so that the compensation torque corresponding to the actual contact force at the end is borne by the target force-bearing joint in the grounded state.

[0053] Preferably, the specific implementation process of extracting the current swing state features of the centaur robot to determine the grounded support leg identifier from the four legs is as follows: After the compensation torque of each joint has been determined by the equivalent joint torque of each joint space of the four legs, first read the compensation torque of each joint, the corresponding joint identifier, the corresponding support leg identifier, and the current control cycle mark in the compensation torque output record, and simultaneously read the foot position of the four legs currently in contact with the ground, the current grounding state of the support leg, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, the current torque feedback value of each joint of the four legs, and the pressure distribution features at the bottom of each support leg, so as to establish a current swing state feature extraction bearing record. The current swing state feature extraction record is used to incorporate the force compensation source of the compensation torque of each joint, the joint object source of the corresponding joint identifier, the support object source of the corresponding supporting leg identifier, the timing source of the current control cycle marker, the spatial contact source of the foot end position of the four legs currently in contact with the ground, the contact effectiveness source of the current grounding state of the supporting leg, the posture source of the current angle of each joint of the four legs, the motion source of the current angular velocity of each joint of the four legs, the execution state source of the current torque feedback value of each joint of the four legs, and the plantar force source of the pressure distribution characteristics of the bottom of each supporting leg into the same extraction caliber. The technical essence of the current swing state feature is a data-driven state description used to characterize whether each supporting leg of the four legs is in a grounded support state, a suspended leg state, or a contact transition state within the current control cycle; the current swing state feature is not a simple leg action name, but a state judgment basis formed by the constraints of foot end spatial height, plantar pressure change, joint angular velocity change, and joint torque feedback change. In specific implementation, firstly, based on the current foot position of the four legs in contact with the ground and the pressure distribution characteristics of the bottom of each supporting leg, it is determined whether the foot of each supporting leg has formed effective support contact with the ground, thus obtaining the foot contact determination result. Then, based on the current angle and angular velocity of each joint of the four legs, it is determined whether each supporting leg is in the lifting swing, landing contact, or stable support phase, thus obtaining the supporting leg movement phase determination result. Next, the foot contact determination result and the supporting leg movement phase determination result are checked for consistency. Supporting legs where plantar pressure has been formed and the joint angular velocity is within the support variation range are classified as grounded support state; supporting legs where plantar pressure has not been formed and the foot position is off the ground are classified as suspended leg state; and supporting legs where plantar pressure is being built but the joint angular velocity is still in the landing variation process are classified as contact transition state. The resulting current swing state characteristics can distinguish which supporting legs of the four legs can bear the compensation torque of each joint and which supporting legs are not suitable to directly bear the compensation torque of each joint.Subsequently, based on the current swing state characteristics, support legs in a grounded support state are selected from the four legs, and the corresponding support leg identifiers of the selected support legs in a grounded support state are identified as grounded support leg identifiers. The technical essence of the grounded support leg identifier is to identify the support leg capable of bearing the compensating torque of each joint through effective support contact between the sole and the ground from the four legs. The grounded support leg identifier continues to participate in subsequent identification of the target force-bearing joint on the corresponding support leg, ensuring that the compensating torque of each joint is not distributed to suspended legs that cannot form effective support contact, thereby allowing the compensating torque corresponding to the actual contact force at the end to be transmitted to the ground support surface along the grounded support leg.

[0054] Preferably, the specific implementation process of identifying the target force-bearing joint on the corresponding support leg based on the support leg identifier in the grounded state is as follows: After the support leg identifier in the grounded state has been determined according to the current swing state characteristics, the support leg identifier in the grounded state, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, the current torque feedback value of each joint of the four legs, the foot position of each foot currently in contact with the ground, and the pressure distribution characteristics of the bottom of each support leg are extracted from the load record, and a target force-bearing joint identification load record is established. The target force-bearing joint identification load record is used to include the grounding support source of the support leg identifier in the grounded state, the attitude source of the current angle of each joint of the four legs, the motion source of the current angular velocity of each joint of the four legs, the execution state source of the current torque feedback value of each joint of the four legs, the support geometry source of the foot position of each foot currently in contact with the ground, and the force distribution source of the pressure distribution characteristics of the bottom of each support leg into the same identification caliber. The technical essence of the target force-bearing joint is that it is a joint on the supporting leg in a grounded support state that can convert the compensation torque of each joint into a foot support function. The target force-bearing joint is not any joint on the supporting leg in a grounded support state, but a joint that is effectively connected to the ground support surface and has the capacity to withstand the torque of the joint to be transferred or the compensation torque adjustment margin of each joint. In specific implementation, firstly, based on the identification of the supporting leg in a grounded state, the joint identifications of the hip, knee, and ankle joints on the corresponding supporting leg are read, and the hip, knee, and ankle joints on the corresponding supporting leg are selected as candidate force-bearing joints. Then, based on the current angle of each joint of the four legs, it is determined whether each candidate force-bearing joint is close to its permissible posture boundary, and based on the current angular velocity of each joint of the four legs, it is determined whether each candidate force-bearing joint is in a motion state that can be impedance adjusted. Finally, combined with the current torque feedback value of each joint of the four legs, it is determined whether each candidate force-bearing joint still has a torque bearing margin, so as to obtain the bearing capacity result of the candidate force-bearing joint. Next, based on the current foot positions of the four legs in contact with the ground and the pressure distribution characteristics of the bottom of each supporting foot, it is determined whether the corresponding supporting leg of each candidate force-bearing joint can transfer the joint torque to the ground support surface through effective support contact between the sole of the foot and the ground, so as to obtain the support effectiveness result of the candidate force-bearing joint. Then, the load-bearing capacity result of the candidate force-bearing joint is cross-filtered with the support effectiveness result of the candidate force-bearing joint, and the candidate force-bearing joints that are in a ground support state, have not approached the allowable posture boundary, still have torque load-bearing margin, and can transfer the support effect through effective support contact between the sole of the foot and the ground are identified as the target force-bearing joints.After establishing a correspondence between the target force-bearing joint and the grounded support leg identifier, the target joint continues to participate in receiving the joint torque to be transferred. This ensures that the joint torque to be transferred, selected from the equivalent joint torques corresponding to the suspended leg, can be allocated to the target force-bearing joint with actual support capabilities. Through this identification method, the grounded support leg identifier is not only used to indicate the support leg status but also further transformed into an identification entry point for the target force-bearing joint, making the compensation torque redistribution clearly defined by the support object and the joint object.

[0055] Preferably, the specific implementation process for selecting the joint torque to be transferred corresponding to the suspended leg from the equivalent joint torques of each joint space of the four legs is as follows: After the target force-bearing joint has been identified based on the grounded support leg identifier, the equivalent joint torques of each joint space of the four legs, the corresponding joint identifier, the corresponding support leg identifier, the current swing state feature, the grounded support leg identifier, and the target force-bearing joint are read, and a joint torque to be transferred screening and load-bearing record is established. The joint torque to be transferred screening and load-bearing record is used to include the force mapping source of the equivalent joint torques of each joint space of the four legs, the joint object source of the corresponding joint identifier, the support leg source of the corresponding support leg identifier, the leg state source of the current swing state feature, the support object source of the grounded support leg identifier, and the load-bearing object source of the target force-bearing joint in the same screening criterion. The technical essence of the suspended leg is that it is a supporting leg that does not form effective support contact with the ground during the current control cycle. The suspended leg is determined by the foot contact determination result and the supporting leg movement phase determination result in the current swing state characteristics. The suspended leg cannot bear the joint torque converted from the actual contact force at the end through effective support contact between the sole of the foot and the ground. The technical essence of the joint torque to be transferred is the torque component in the equivalent joint torque of each joint space of the four legs that was originally mapped to the corresponding joint of the suspended leg, but needs to be transferred to the target force-bearing joint due to the lack of effective support contact of the suspended leg. The joint torque to be transferred is not a newly added torque, nor a residual torque after deletion, but a source of torque to be re-beared selected from the equivalent joint torque of each joint space of the four legs according to the contact state of the supporting leg. In specific implementation, firstly, the equivalent joint moments of each joint space of the four legs are grouped to the corresponding supporting legs according to the corresponding supporting leg identifier. Then, each corresponding supporting leg identifier is matched with the current swing state characteristics to determine whether the supporting leg corresponding to each corresponding supporting leg identifier is a suspended leg. Subsequently, the equivalent joint moments associated with the corresponding supporting leg identifiers of suspended legs are filtered out from the equivalent joint moments of each joint space of the four legs, while maintaining the correspondence between the corresponding joint identifier, the corresponding supporting leg identifier, the moment direction, and the moment amplitude of the equivalent joint moments to form the joint moments to be transferred. Next, according to the moment direction and moment amplitude of the joint moments to be transferred, the transfer priority of the joint moments to be transferred is sorted, so that the joint moments to be transferred that have a high correlation with the current reaction load direction and contribute significantly to the overall attitude disturbance are given priority in the subsequent allocation process.To avoid confusion between the joint torque to be transferred and the equivalent joint torque already borne by the supporting leg in a grounded support state, a source relationship is established between the joint torque to be transferred and the corresponding supporting leg identifier of the suspended leg, and a candidate allocation relationship is established between the joint torque to be transferred and the target force-bearing joint. Thus, the joint torque to be transferred receives the equivalent joint torque of each joint space of the four legs, and the suspended legs that cannot form effective support contact are excluded based on the current swing state characteristics. This allows subsequent compensation torque redistribution to transfer the torque component that the suspended leg side cannot bear to the target force-bearing joint in a grounded support state.

[0056] Preferably, the specific implementation process of allocating the joint torque to be transferred to the target force-bearing joint to complete the redistribution of the compensation torque of the target force-bearing joint, so that the compensation torque corresponding to the actual contact force at the end is borne by the target force-bearing joint in the grounded state, is as follows: After the joint torque to be transferred has been screened out from the equivalent joint torques of each joint space of the four legs, the joint torque to be transferred, the target force-bearing joint, the grounded support leg identifier, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, the current torque feedback value of each joint of the four legs, the pressure distribution characteristics of the bottom of each support leg, and the support polygon boundary are read from the joint torque to be transferred screening and bearing record, and a compensation torque redistribution bearing record is established. The compensation torque redistribution load record is used to incorporate the following sources into the same allocation caliber: the suspended leg source of the joint torque to be transferred, the joint load source of the target force joint, the support object source of the grounded support leg, the attitude source of the current angle of each joint of the four legs, the motion source of the current angular velocity of each joint of the four legs, the execution status source of the current torque feedback value of each joint of the four legs, the plantar force source of the pressure distribution characteristics at the bottom of each support foot, and the support stability source of the support polygon boundary. The technical essence of this compensation torque redistribution is to reconfigure the joint torque to be transferred corresponding to the suspended leg according to the load-bearing capacity of the target force joint in the grounded support state, so that the compensation torque formed by the actual contact force at the end is not allocated to joints lacking effective support contact, but is instead borne by the target force joint that can form a reaction load through the ground support surface. In specific implementation, firstly, based on the identification of the grounded support leg where the target force joint is located, the pressure distribution characteristics at the bottom of each support foot, and the boundary of the support polygon, the support bearing ratio of the support leg corresponding to each target force joint is determined. The support bearing ratio represents the share of the torque that the support leg of each target force joint can bear on the joint torque to be transferred. Subsequently, based on the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, and the current torque feedback value of each joint of the four legs, the joint bearing margin of each joint is determined. The joint bearing margin is used to limit the range of torque variation after the joint torque to be transferred is distributed to the target force joint. Next, the support bearing ratio and the joint bearing margin are used together to distribute the joint torque to be transferred, so that the target force joint with a torque direction consistent with the reaction load resistance direction and a higher bearing margin bears a higher joint torque to be transferred, and the target force joint close to the attitude boundary or with a higher torque feedback value bears a lower joint torque to be transferred.Subsequently, the allocated torque of the joint to be transferred is superimposed in the same direction or canceled out in the opposite direction with the original compensation torque of the target force joint to form the compensation torque of the target force joint, and the compensation torque of the target force joint is written into the compensation torque redistribution record. The compensation torque of the target force joint continues to participate in the subsequent active impedance control command update as the compensation torque of each joint, so that the compensation torque corresponding to the actual contact force at the end is borne by the target force joint in the grounded state. Thus, the torque of the joint to be transferred is redistributed through the target force joint, so that the torque component that the suspended leg cannot bear in the current control cycle is transferred to the support leg side in the grounded support state, thereby enabling each joint of the four legs to have a joint-level compensation basis consistent with the actual support state when absorbing residual impact force.

[0057] Optionally, the step of driving each joint of the four legs to absorb residual impact force based on the active impedance control command includes: Monitor the actual angular velocity feedback value of each joint of the four legs when executing the active impedance control command; The actual angular velocity feedback value is compared with a mutation threshold, so that a flexible buffer mode is triggered when the actual angular velocity feedback value reaches the mutation threshold; The yield direction of the force on each joint is determined under the flexible buffer mode; The joints are controlled to rotate within a limited angle along the direction of the yielding force, so as to dissipate the energy carried by the residual impact force through the rotation.

[0058] Preferably, the specific implementation process for monitoring the actual angular velocity feedback values ​​of each joint of the four legs when executing the active impedance control command is as follows: After the active impedance control command has been formed according to the target impedance parameter, the active impedance control command, the target impedance parameter, the corresponding joint identifier, the corresponding supporting leg identifier, the current control cycle marker, the compensation torque of each joint, and the corresponding impact direction are read from the active impedance control command output record. Simultaneously, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, the current torque feedback value of each joint of the four legs, and the pressure distribution characteristics at the bottom of each supporting leg are read to establish an angular velocity feedback monitoring load record. The angular velocity feedback monitoring load record is used to incorporate the command source of the active impedance control command, the impedance configuration source of the target impedance parameter, the impact compensation source of the compensation torque of each joint, the direction source of the corresponding impact direction, the attitude source of the current angle of each joint of the four legs, the motion source of the current angular velocity of each joint of the four legs, the execution status source of the current torque feedback value of each joint of the four legs, and the foot force source of the pressure distribution characteristics at the bottom of each supporting leg into the same monitoring caliber. The technical essence of the actual angular velocity feedback value is the joint angular velocity response data generated by each joint of the four legs along the joint rotational degrees of freedom after executing the active impedance control command. The actual angular velocity feedback value is not simply used to display the speed of joint movement, but rather to determine whether the residual impact force has been transmitted to each joint of the four legs through changes in joint posture. Specifically, the current angular velocity of each joint of the four legs is first read according to the corresponding joint identifier, and a temporal correspondence is established between the current angular velocity of each joint of the four legs and the current control cycle marker to form the current cycle angular velocity sampling value. The current cycle angular velocity sampling value represents the angular velocity sampling result of each joint of the four legs after executing the active impedance control command within the current control cycle. Subsequently, the historical angular velocity sampling value corresponding to the previous control cycle is read, and the current cycle angular velocity sampling value is continuously compared with the historical angular velocity sampling value to obtain the joint angular velocity change. The joint angular velocity change represents the amplitude of angular velocity change of each joint of the four legs between adjacent control cycles, and together with the current cycle angular velocity sampling value, reflects the degree of disturbance of the joint movement state by the residual impact force. Next, the current cycle angular velocity sample value and the change in joint angular velocity are projected onto the joint rotation direction related to the compensation torque of each joint, according to the corresponding impact direction, to form the actual angular velocity feedback value. The actual angular velocity feedback value inherits the joint angular velocity response data after the execution of the active impedance control command and continues to participate in the subsequent comparison with the mutation threshold, so that whether the residual impact force causes a sudden change in joint velocity can be entered into the flexible buffer mode triggering process in the form of monitorable data.

[0059] Preferably, the specific implementation process of comparing the actual angular velocity feedback value with a mutation threshold to trigger the flexible buffer mode when the actual angular velocity feedback value reaches the mutation threshold is as follows: After the actual angular velocity feedback value has been formed by the angular velocity feedback monitoring load record, the actual angular velocity feedback value, the corresponding joint identifier, the corresponding supporting leg identifier, the corresponding impact direction, the compensation torque of each joint, the target impedance parameter, and the pressure distribution characteristics at the bottom of each supporting foot are read, and a flexible buffer mode trigger load record is established. The flexible buffer mode trigger load record is used to include the joint response source of the actual angular velocity feedback value, the force mapping source of the compensation torque of each joint, the impedance adjustment source of the target impedance parameter, the direction source of the corresponding impact direction, and the support force source of the pressure distribution characteristics at the bottom of each supporting foot into the same comparison caliber. The technical essence of the mutation threshold is to determine whether the change in joint angular velocity has exceeded the boundary of the normal impedance adjustment range. This mutation threshold is not a fixed, single empirical value, but rather a comparison benchmark configured during the assembly calibration or initialization calibration of the Centaur robot. It is based on the allowable rotation range of each joint of the four legs, the torque range that each joint can withstand, the current grounding status of the supporting legs, the target impedance parameters, and the residual impact force absorption requirements. In use, the mutation threshold is associated with the corresponding joint identifier, the corresponding supporting leg identifier, and the corresponding impact direction, allowing different supporting legs, different joints, and different impact directions to use a comparison benchmark that matches their load-bearing capacity. Specifically, during comparison, the mutation threshold corresponding to the joint identifier is first read, and then the actual angular velocity feedback value is confirmed to be an angular velocity response in the direction of residual impact force transmission based on the corresponding impact direction. Subsequently, the amplitude of the actual angular velocity feedback value is compared with the mutation threshold using the same dimensions to obtain the angular velocity mutation determination result. The angular velocity mutation determination result indicates whether the actual angular velocity feedback value has reached the mutation threshold and serves as a direct basis for determining whether to trigger the flexible buffer mode. If the actual angular velocity feedback value does not reach the mutation threshold, the angular velocity mutation determination result is written into the flexible buffer mode trigger bearing record, and the conventional impedance adjustment continues to be executed according to the active impedance control command. If the actual angular velocity feedback value reaches the mutation threshold, the angular velocity mutation determination result, the corresponding joint identifier, the corresponding supporting leg identifier, the corresponding impact direction, and the compensation torque of each joint are written into the flexible buffer mode trigger record to trigger the flexible buffer mode. The technical essence of the flexible buffer mode is to temporarily change the control state of the joint impedance response mode when the residual impact force causes a sudden change in the joint angular velocity, so that the joints of the four legs no longer rely solely on their original stiffness to resist the disturbance, but enter a force processing state that allows limited yielding and damping energy dissipation.Therefore, the comparison between the actual angular velocity feedback value and the sudden change threshold directly determines whether the flexible buffer mode is triggered, so that the absorption process of residual impact force by the active impedance control command can be switched according to the joint angular velocity response.

[0060] Preferably, the specific implementation process for determining the yield direction of each joint under the flexible buffer mode is as follows: After the flexible buffer mode has been triggered by the actual angular velocity feedback value reaching the mutation threshold, the angular velocity mutation judgment result, the corresponding joint identifier, the corresponding supporting leg identifier, the corresponding impact direction, the compensation torque of each joint, the target impedance parameter, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, and the current torque feedback value of each joint of the four legs are read from the flexible buffer mode trigger record, and a bearing record for determining the yield direction is established. The bearing record for determining the yield direction is used to include the trigger source of the flexible buffer mode, the residual impact source of the compensation torque of each joint, the direction source of the corresponding impact direction, the impedance configuration source of the target impedance parameter, the attitude source of the current angle of each joint of the four legs, the motion source of the current angular velocity of each joint of the four legs, and the execution state source of the current torque feedback value of each joint of the four legs into the same determining caliber. The technical essence of the yielding direction is the joint rotation direction that allows each joint of the four legs to yield when subjected to residual impact force. This yielding direction is not an arbitrarily chosen rotation direction, but is jointly defined by the direction of the compensation torque of each joint in the joint space, the direction of the actual angular velocity feedback value, the allowable posture range of the current angle of each joint of the four legs, and the target impedance parameter. In specific implementation, firstly, the direction of the compensation torque after the residual impact force is transmitted to the corresponding joint is determined based on the compensation torque of each joint. Then, the direction of the compensation torque is checked for consistency with the current torque feedback value of each joint of the four legs to obtain the joint output direction verification result. This joint output direction verification result is used to determine whether the current output of the corresponding joint is already resisting the residual impact force and continues to participate in the selection of candidate yielding directions. Subsequently, the angular velocity direction corresponding to the actual angular velocity feedback value is compared with the corresponding impact direction to obtain the joint movement direction verification result. This joint movement direction verification result is used to determine whether the actual movement direction of the corresponding joint is consistent with the direction of residual impact force transmission. If the joint motion direction verification result indicates that the angular velocity direction corresponding to the actual angular velocity feedback value is consistent with the corresponding impact direction, then this direction is taken as the candidate yield direction; if the joint motion direction verification result indicates that the angular velocity direction corresponding to the actual angular velocity feedback value is opposite to the corresponding impact direction, then the candidate yield direction that can consume the residual impact energy is re-determined according to the compensation torque direction.Next, based on the current angle of each joint of the four legs, it is determined whether the candidate yield direction will cause the corresponding joint to approach the allowable attitude boundary, thus forming an attitude boundary constraint; and based on the target impedance parameter, it is determined whether the candidate yield direction can be executed jointly by the increased damping parameter configuration value and the decreased stiffness parameter configuration value, thus forming an impedance parameter constraint; simultaneously, based on the corresponding impact direction, it is determined whether the candidate yield direction is consistent with the residual impact force absorption direction, thus forming an impact direction constraint. After the above screening, the candidate yield direction that simultaneously satisfies the attitude boundary constraint, the impedance parameter constraint, and the impact direction constraint is determined as the force yield direction. Thus, the force yield direction inherits the flexible buffer mode, the compensation torque of each joint, and the actual angular velocity feedback value, and provides a directional basis for subsequently controlling each joint to perform a limited-angle yield rotation along the force yield direction.

[0061] Preferably, the specific implementation process of controlling each joint to perform a limited-angle yielding rotation along the yielding direction to consume the energy carried by the residual impact force through the yielding rotation is as follows: After the yielding direction has been determined by the yielding direction and a bearing record has been formed, the yielding direction, the corresponding joint identifier, the corresponding support leg identifier, the compensation torque of each joint, the target impedance parameter, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, the current torque feedback value of each joint of the four legs, the boundary of the support polygon, and the pressure distribution characteristics of the bottom of each support leg are read from the yielding direction bearing record, and a yielding rotation control bearing record is established. The yielding rotation control bearing record is used to incorporate the directional source of the yielding force, the impact source of the compensating torque of each joint, the impedance configuration source of the target impedance parameter, the attitude source of the current angle of each joint of the four legs, the motion source of the current angular velocity of each joint of the four legs, the execution state source of the current torque feedback value of each joint of the four legs, the stability boundary source of the support polygon boundary, and the plantar force source of the pressure distribution characteristics at the bottom of each supporting foot into the same control caliber. The technical essence of the limited angle is that in the flexible buffer mode, the controlled angle change range of each joint of the four legs along the yielding force is allowed. The limited angle is not an open rotation range, but a yielding amplitude jointly defined by the current angle of each joint of the four legs, the allowed attitude boundary of each joint of the four legs, the support polygon boundary, the target impedance parameter, and the energy carried by the residual impact force. In specific implementation, firstly, the attitude retraction margin is determined based on the distance between the current angle of each joint of the four legs and the allowable attitude boundary of each joint. This attitude retraction margin represents the usable angle space when the corresponding joint undergoes retraction rotation along the yielding direction, and serves as an attitude-side constraint for determining the finite angle. Next, based on the pressure distribution characteristics at the bottom of each supporting leg and the boundary of the supporting polygon, it is determined whether the retraction rotation will weaken the stability margin within the current supporting area, thus forming a support stability constraint. This support stability constraint represents the degree of influence of the retraction rotation on the zero-moment point being within the supporting area, and serves as a support-side constraint for determining the finite angle. Subsequently, based on the compensation torque of each joint and the current angular velocity of each joint of the four legs, the energy transmission intensity carried by the residual impact force on the corresponding joint is estimated. Combined with the increased damping parameter configuration value and the decreased stiffness parameter configuration value in the target impedance parameters, the retraction amplitude is constrained according to the attitude retraction margin and the support stability constraint to determine the finite angle.Subsequently, each joint is controlled to perform yielding rotation within the finite angle according to the yielding direction of the force, so that the corresponding joint transforms the instantaneous torque formed by the residual impact force into a controlled joint movement process through angular yielding. During the yielding rotation, the increased damping parameter suppresses energy dissipation from changes in angular velocity, and the decreased stiffness parameter reduces the tendency of the joint to immediately rebound, thereby consuming the energy carried by the residual impact force through the yielding rotation. The technical essence of the yielding rotation is that the joints of the four legs perform controlled compliant movement along the yielding direction of the force in the flexible buffer mode. The yielding rotation does not abandon the support output, but rather, under the constraints of the support polygon boundary and the target impedance parameter, it transforms the residual impact force from a sudden disturbance into a joint movement that can be absorbed by damping. Thus, the yielding direction, the finite angle, and the yielding rotation form a continuous residual impact force absorption process, enabling the active impedance control command to drive each joint of the four legs to consume the energy carried by the residual impact force while maintaining the support state, and providing a basis for joint movement to maintain the zero torque point of the Centaur robot within the support area.

[0062] Optionally, the step of driving each joint of the four legs to absorb residual impact force based on the active impedance control command to maintain the zero-moment point of the centaur robot within the support area enclosed by the boundary of the support polygon includes: During the retraction and rotation process, the pressure distribution characteristics at the bottom of each supporting foot are collected in real time. Calculate the actual zero-torque point coordinates for the current control cycle based on the pressure distribution characteristics. Based on the actual zero-moment point coordinates and the boundary of the supporting polygon, calculate the shortest vertical distance between the actual zero-moment point coordinates and the boundary of the supporting polygon; The shortest vertical distance is compared with a critical safety threshold, so that when the shortest vertical distance reaches the boundary approach condition defined by the critical safety threshold, a forced restoring torque is injected into the active impedance control command; Based on the active impedance control command after the forced recovery torque is injected, the joints are prevented from continuing the yielding rotation, thereby causing the actual zero torque point coordinates to retract into the support area.

[0063] Preferably, the specific implementation process of real-time acquisition of pressure distribution characteristics at the bottom of each supporting foot during the yielding rotation is as follows: After each joint has entered the yielding rotation of the limited angle according to the force yielding direction, the active impedance control command, the force yielding direction, the limited angle, the compensation torque of each joint, the target impedance parameter, the corresponding joint identifier, the corresponding supporting leg identifier, the supporting polygon boundary, and the current control cycle mark in the yielding rotation control load record are read first, and a pressure distribution characteristic acquisition load record is established. The pressure distribution characteristic acquisition load record is used to include the joint movement source of the yielding rotation, the direction source of the force yielding direction, the yielding amplitude source of the limited angle, the impact source of the compensation torque of each joint, the impedance adjustment source of the target impedance parameter, the joint object source of the corresponding joint identifier, the supporting object source of the corresponding supporting leg identifier, and the stable boundary source of the supporting polygon boundary into the same acquisition caliber, so that the pressure distribution characteristics at the bottom of each supporting foot formed subsequently can correspond to the residual impact force absorption process in this case, rather than to the foot force in a normal standing state. The technical essence of the yielding rotation is the controlled compliant movement of each joint of the four legs along the force yielding direction under the flexible buffer mode. The yielding rotation changes the joint posture of the corresponding supporting leg and further changes the contact pressure distribution between the corresponding supporting foot and the ground. Therefore, it is necessary to synchronously collect the pressure distribution characteristics of the bottom of each supporting foot during the control cycle in which the yielding rotation continues. In specific implementation, the supporting leg in the grounded state is first determined according to the corresponding supporting leg identifier, and the bottom pressure sampling data of the supporting foot corresponding to the grounded supporting leg is read according to the corresponding supporting leg identifier. The bottom pressure sampling data of the supporting foot includes the pressure sampling value of different contact areas of the bottom of the supporting foot, the pressure sampling position, and the pressure sampling time marker. Subsequently, the bottom pressure sampling data of the supporting foot is time-aligned with the current control cycle marker to form the current control cycle pressure sampling data. The current control cycle pressure sampling data is used to represent the change in foot force caused by the yielding rotation during the current control cycle. Next, based on the positional relationship of the contact area at the bottom of the support foot under the foot coordinate caliber, the pressure sampling data of the current control cycle is processed by arranging the contact areas to form the pressure distribution characteristics of the bottom of each support foot. The technical essence of the pressure distribution characteristics of the bottom of each support foot is a data-driven result describing the force position, force intensity, and force offset trend of each support foot on the ground support surface. The pressure distribution characteristics of the bottom of each support foot are not a single pressure value, but rather represent the distribution state of whether the force at the bottom of the support foot is offset forward, backward, left, or right.The pressure distribution characteristics at the bottom of each support foot continue to be used in subsequent calculations of the actual zero torque point coordinates of the current control cycle, so that the change in foot force caused by the yielding rotation can be converted into a geometric position judgment basis for the stable state of the whole machine.

[0064] Preferably, the specific implementation process for calculating the actual zero-torque point coordinates of the current control cycle based on the pressure distribution characteristics at the bottom of each supporting foot is as follows: After the pressure distribution characteristics at the bottom of each supporting foot have been collected and recorded, the pressure distribution characteristics at the bottom of each supporting foot, the foot position currently in contact with the ground, the corresponding supporting foot identifier, the current grounding state of the supporting foot, the current centroid position of the Centaur robot, the current control cycle marker, and the boundary of the supporting polygon are read, and an actual zero-torque point coordinate calculation record is established. The actual zero-torque point coordinate calculation record is used to incorporate the foot force source of the pressure distribution characteristics at the bottom of each supporting foot, the spatial contact source of the foot position currently in contact with the ground, the supporting object source of the corresponding supporting foot identifier, the contact effectiveness source of the current grounding state of the supporting foot, the mass distribution source of the current centroid position of the Centaur robot, and the stable boundary source of the supporting polygon boundary into the same calculation caliber. The technical essence of the actual zero-moment point coordinates of the current control cycle is to calculate the combined action position of the ground reaction force of the entire machine based on the actual pressure distribution between each support foot and the ground within the current control cycle. The actual zero-moment point coordinates of the current control cycle are used to indicate whether the combined action position of the ground support reaction force of the Centaur robot is still located within the support area when it performs the yielding rotation and absorbs the residual impact force. In specific implementation, firstly, based on the current grounding state of the support feet, the pressure distribution characteristics of the bottom of each support foot corresponding to the support foot in the grounding state are selected from the pressure distribution characteristics of the bottom of each support foot, and the pressure distribution characteristics of the bottom of each support foot corresponding to the support foot in the non-grounding state are excluded from the calculation process of the actual zero-moment point coordinates of the current control cycle, so that the actual zero-moment point coordinates of the current control cycle are determined only by the support feet that can form ground support. Subsequently, based on the current foot position of the four feet in contact with the ground, the pressure distribution characteristics of the bottom of each support foot corresponding to the grounding support foot are converted to a horizontal coordinate system, so that the pressure sampling position of the bottom of each support foot can be in the same horizontal reference relationship as the boundary of the support polygon. Next, the pressure distribution characteristics at the bottom of each support foot after conversion are weighted according to their pressure contribution. This ensures that contact areas with higher pressure contribute more to the calculation of the actual zero-moment point coordinates in the current control cycle, while contact areas with lower pressure contribute less, thus forming the location of the ground reaction force synthesis. This location of the ground reaction force synthesis is based on the pressure distribution characteristics at the bottom of each support foot and is further combined with the current center of mass position of the Centaur robot for attitude consistency verification, thereby obtaining the actual zero-moment point coordinates for the current control cycle.The actual zero-moment point coordinates of the current control cycle continue to participate in the calculation of the shortest vertical distance between the current control cycle and the boundary of the supporting polygon, so that the active impedance control command can use the actual zero-moment point coordinates of the current control cycle as the adjustment basis while absorbing the residual impact force.

[0065] Preferably, the specific implementation process for calculating the shortest vertical distance between the actual zero-moment point coordinates of the current control cycle and the support polygon boundary, based on the actual zero-moment point coordinates of the current control cycle and the support polygon boundary, is as follows: After the actual zero-moment point coordinates of the current control cycle have been formed by the load-bearing record calculated from the actual zero-moment point coordinates, the actual zero-moment point coordinates of the current control cycle, the support polygon boundary, the support area, the foot position of the four legs currently in contact with the ground, the corresponding support foot identifier, and the current control cycle mark are read, and a boundary distance calculation load-bearing record is established. The boundary distance calculation load-bearing record is used to incorporate the source of the ground reaction force composition location of the actual zero-moment point coordinates of the current control cycle, the source of the support boundary of the support polygon, the source of the stability range of the support area, and the geometric source of the foot position of the four legs currently in contact with the ground into the same distance calculation caliber. The technical essence of the support polygon boundary is the support stability boundary formed by the support legs in the grounded state enclosing it on the horizontal plane. The support polygon boundary is used to limit the boundary range within which the actual zero-moment point coordinates of the current control cycle can maintain the stability of the entire machine under the current support configuration. The essence of the shortest vertical distance is the vertical distance from the actual zero-moment point coordinates of the current control cycle to the nearest boundary segment in the boundary of the supporting polygon. This shortest vertical distance is not an isolated result of ordinary geometric distance, but rather a measure of the stability margin of the actual zero-moment point coordinates of the current control cycle relative to the edge of the supporting region. Specifically, multiple boundary vertices in the boundary of the supporting polygon are first determined based on the current foot positions of the four legs in contact with the ground, and multiple boundary segments are formed according to the surrounding order of these vertices on the horizontal plane. These multiple boundary segments represent the front, rear, left, and right boundaries of the supporting region on the horizontal plane. Subsequently, the actual zero-moment point coordinates of the current control cycle are projected onto the perpendicular directions corresponding to the multiple boundary segments to obtain multiple candidate vertical distances. When the foot of the perpendicular from the actual zero-moment point coordinates of the current control cycle on a certain boundary segment falls within the range of that boundary segment, the candidate vertical distance is retained as the effective vertical distance. When the foot of the perpendicular from the actual zero-moment point coordinates of the current control cycle on a certain boundary segment does not fall within the range of that boundary segment, the distance from the actual zero-moment point coordinates of the current control cycle to the endpoint of that boundary segment is used as a supplementary distance, and this supplementary distance is included in the distance comparison process. Next, the effective vertical distance and the supplementary distance are subjected to minimum value selection processing to obtain the shortest vertical distance. The shortest vertical distance continues to participate in the subsequent comparison with the critical safety threshold, ensuring that the actual zero-moment point coordinates of the current control cycle are close to the boundary of the supporting polygon and can enter the forced recovery torque injection process with a calculable stability margin.

[0066] Preferably, the specific implementation process of comparing the shortest vertical distance with a critical safety threshold to inject a forced restoring torque into the active impedance control command when the shortest vertical distance reaches the boundary approach condition defined by the critical safety threshold is as follows: After the shortest vertical distance has been calculated and recorded from the boundary distance, the shortest vertical distance, the actual zero-moment point coordinates of the current control cycle, the support polygon boundary, the support area, the active impedance control command, the target impedance parameter, the yielding direction, the compensation torque of each joint, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, and the current torque feedback value of each joint of the four legs are read, and a forced restoring torque injection bearing record is established. The forced restoring torque injection bearing record is used to include the source of the stability margin of the shortest vertical distance, the source of the ground reaction force synthesis position of the actual zero-moment point coordinates of the current control cycle, the source of the joint control of the active impedance control command, the source of the impedance configuration of the target impedance parameter, the source of the yielding direction of the force, and the source of the impact compensation of the compensation torque of each joint into the same comparison and injection caliber. The technical essence of the critical safety threshold is to determine whether the actual zero-torque point coordinates of the current control cycle have approached the stable margin boundary of the supporting polygon. This critical safety threshold is formed when the Centaur robot completes assembly calibration or starts initialization calibration, based on the contact range of the supporting foot sole, the allowable adjustment range of the supporting foot, the boundary dimensions of the supporting area, the current centroid position change range of the Centaur robot, and the yielding amplitude constraint of the active impedance control command. The technical essence of the boundary approach condition is the boundary approach state represented by the shortest vertical distance reaching the critical safety threshold. This boundary approach condition indicates that as the yielding rotation continues, the actual zero-torque point coordinates of the current control cycle will further approach the boundary of the supporting polygon, thus requiring a change in the joint yielding trend of the active impedance control command. In the specific comparison, the shortest vertical distance is first compared with the critical safety threshold using the same dimensions to obtain the boundary approach determination result. When the boundary approach determination result indicates that the shortest vertical distance has not reached the boundary approach condition defined by the critical safety threshold, the yielding rotation continues to be executed according to the active impedance control command. When the boundary approach determination result indicates that the shortest vertical distance has reached the boundary approach condition defined by the critical safety threshold, the recovery direction of the forced recovery torque is determined according to the deviation direction of the actual zero torque point coordinates of the current control cycle relative to the interior of the support area.The technical essence of the forced restoring torque is the joint-level restoring torque correction superimposed on the active impedance control command. This forced restoring torque is used to weaken the continued yielding trend in the yielding direction and guide the actual zero-moment point coordinates of the current control cycle to move inwards towards the support area. Subsequently, the torque amplitude of the forced restoring torque is determined based on the compensation torque of each joint, the current torque feedback value of each joint of the four legs, and the target impedance parameters. The forced restoring torque is then written into the active impedance control command according to the corresponding joint identifier, forming the active impedance control command after the forced restoring torque is injected. Thus, the comparison result between the shortest vertical distance and the critical safety threshold can directly trigger the injection of the forced restoring torque, causing the active impedance control command to switch from a yielding state that simply absorbs the residual impact force to a joint control state that considers both yielding control and zero-moment point retraction control.

[0067] Preferably, the specific implementation process of preventing the joints from continuing the yielding rotation based on the active impedance control command after the injection of the forced restoring torque, thereby causing the actual zero torque point coordinates of the current control cycle to retract into the support area, is as follows: After the active impedance control command after the injection of the forced restoring torque has been formed by the forced restoring torque injection bearing record, the active impedance control command after the injection of the forced restoring torque, the forced restoring torque, the force yielding direction, the finite angle, the actual zero torque point coordinates of the current control cycle, the support area, the support polygon boundary, the current angle of each joint of the four legs, the current angular velocity of each joint of the four legs, the current torque feedback value of each joint of the four legs, and the pressure distribution characteristics at the bottom of each support foot are read, and a zero torque point retraction control bearing record is established. The zero-moment point withdrawal control bearing record is used to incorporate the recovery control source of the active impedance control command after the injection of the forced recovery torque, the torque correction source of the forced recovery torque, the yielding direction source of the yielding direction, the yielding amplitude source of the limited angle, the source of the ground reaction force composite action position of the actual zero-moment point coordinates of the current control cycle, the source of the stability range of the support area, and the boundary constraint source of the support polygon boundary into the same withdrawal control caliber. The technical essence of preventing the joints from continuing the yielding rotation is that, when the boundary approach condition has been met, the active impedance control command after the injection of the forced recovery torque restricts the joints from continuing to increase the yielding angle along the yielding direction, so that the yielding rotation no longer expands the trend of the actual zero-moment point coordinates of the current control cycle approaching the support polygon boundary. In specific implementation, first, directional cancellation processing is performed according to the yielding direction and the recovery direction of the forced recovery torque to obtain a yielding suppression torque; the yielding suppression torque is used to cancel the torque component of the joints continuing to yield along the yielding direction. Subsequently, the yielding suppression torque is continuously corrected with the current torque feedback value of each joint of the four legs to form a retraction execution torque. This retraction execution torque is used to change the yielding tendency of each joint without causing a sudden change in joint torque. Next, the current angle and current angular velocity of each joint of the four legs are adjusted according to the retraction execution torque, so that each joint stops further expanding the yielding rotation within the limited angle, and the pressure at the bottom of the grounded supporting foot is transferred back to the corresponding contact area inside the supporting area.Subsequently, based on the updated pressure distribution characteristics at the bottom of each supporting foot, the actual zero-moment point coordinates for the next control cycle are recalculated, and the positional relationship between the actual zero-moment point coordinates for the next control cycle and the supporting area is verified. When the distance between the actual zero-moment point coordinates for the next control cycle and the boundary of the supporting polygon increases, it is determined that the position of the ground reaction force corresponding to the actual zero-moment point coordinates for the current control cycle has retreated towards the interior of the supporting area. The technical essence of the retreat of the actual zero-moment point coordinates for the current control cycle towards the interior of the supporting area is to move the position of the ground reaction force from a position near the boundary towards the interior of the supporting area through joint retraction suppression and foot pressure redistribution, thereby enabling the Centaur robot to maintain a stable supporting state while absorbing the residual impact force. Therefore, the active impedance control command after injecting the forced recovery torque takes into account the shortest vertical distance, the critical safety threshold, and the boundary approach condition. By limiting the retreat rotation and adjusting the plantar pressure distribution, it causes the actual zero torque point coordinate of the current control cycle to retreat into the support area. This ensures that the process of the active impedance control command driving each joint of the four feet to absorb the residual impact force maintains a continuous technical relationship with the zero torque point stabilization control process.

[0068] like Figure 3 As shown in the figure, a centaur robot dual-arm cooperative anti-recoil force control device according to an embodiment of this application includes: The task external force prediction module is used to obtain the operation task category that the Centaur robot's two arms are about to perform, and extract the corresponding predicted external force direction features and predicted external force amplitude features based on the operation task category. The pre-support adjustment module is used to adjust the support force distribution parameters of the four legs of the Centaur robot before the impact force is generated by the operation of the two arms, based on the predicted external force direction characteristics and the predicted external force amplitude characteristics, so as to coordinately adjust the torso spatial pose of the Centaur robot. The contact force mapping module is used to collect the actual end contact force generated by the two arms during operation, and to convert the actual end contact force to each joint of the four legs using a pre-constructed whole-machine dynamics mapping relationship, so as to obtain the compensation torque of each joint. The whole-machine dynamics mapping relationship characterizes the force transmission constraint between the ends of the two arms and each joint of the four legs. The impedance control update module is used to update the active impedance control commands of each joint of the four legs according to the compensation torque, and drive each joint of the four legs to absorb residual impact force based on the active impedance control commands, so as to keep the zero torque point of the centaur robot inside the support area enclosed by the support polygon boundary.

[0069] like Figure 4 As shown, an electronic device according to an embodiment of this application includes a memory and a processor; The memory stores computer programs that can run on the processor; When the processor executes the computer program, it implements any of the methods described in this application.

[0070] like Figure 5 As shown, this is a computer-readable storage medium according to an embodiment of the present application, wherein the computer-readable storage medium stores a computer program; When the computer program is executed by a processor, it implements any of the methods described in this application.

[0071] like Figure 6 As shown, this is an embodiment of a centaur robot according to this application, which includes a dual-arm mechanism, a quadruped mechanism, a torso mechanism, an end-effector contact force acquisition component, and a controller; The dual-arm mechanism is used to perform operational tasks and generates actual end contact force during the performance of the operational tasks. The quadruped mechanism is used to adjust the distribution parameters of the support force on the soles of the four feet through each joint, and to execute active impedance control commands through each joint. The trunk mechanism is used to change the spatial posture of the trunk according to the distribution parameters of the support force on the soles of the four legs; The end contact force acquisition component is used to acquire the actual end contact force generated by the dual-arm mechanism during operation; The controller is connected to the dual-arm mechanism, the quadruped mechanism, the torso mechanism, and the end contact force acquisition component, respectively. The controller is used to obtain the operation task category that the dual-arm mechanism is about to perform, and extract the corresponding predicted external force direction features and predicted external force amplitude features based on the operation task category; The controller is also used to control the quadruped mechanism to adjust the distribution parameters of the foot support force of the quadrupeds before the double-arm mechanism generates an impact force by performing the operation, based on the predicted external force direction characteristics and the predicted external force amplitude characteristics, so as to coordinately adjust the torso spatial posture of the torso mechanism. The controller is also used to convert the actual contact force at the end of the machine to each joint of the quadruped mechanism using a pre-built whole-machine dynamics mapping relationship, so as to obtain the compensation torque of each joint. The whole-machine dynamics mapping relationship characterizes the force transmission constraint between the end of the double-arm mechanism and each joint of the quadruped mechanism. The controller is also used to update the active impedance control command of each joint of the quadruped mechanism according to the compensation torque, and drive each joint of the quadruped mechanism to absorb residual impact force based on the active impedance control command, so as to maintain the zero torque point of the centaur robot within the support area enclosed by the support polygon boundary.

[0072] like Figure 7 As shown in the figure, a centaur robot dual-arm cooperative anti-recoil force control system according to an embodiment of this application includes a task category acquisition unit, a pre-support adjustment unit, an end contact force acquisition unit, a dynamic mapping unit, an active impedance control unit, and a zero torque point monitoring unit. The task category acquisition unit is used to acquire the operation task category that the Centaur robot's two arms are about to perform, and output the operation task category to the pre-support adjustment unit; The pre-support adjustment unit is used to extract the corresponding predicted external force direction features and predicted external force amplitude features based on the operation task category, and adjust the support force distribution parameters of the four legs of the Centaur robot before the impact force is generated by the operation of the two arms, in order to coordinately adjust the torso spatial pose of the Centaur robot. The end contact force acquisition unit is used to acquire the actual end contact force generated by the two arms during operation, and output the actual end contact force to the dynamic mapping unit; The dynamic mapping unit is used to convert the actual contact force of the end effector to each joint of the four legs using a pre-constructed whole-machine dynamic mapping relationship, so as to obtain the compensation torque of each joint. The whole-machine dynamic mapping relationship characterizes the force transmission constraint between the end effector of the two arms and each joint of the four legs. The active impedance control unit is used to update the active impedance control command of each joint of the four legs according to the compensation torque, and drive each joint of the four legs to absorb residual impact force based on the active impedance control command. The zero-moment point monitoring unit is used to monitor the zero-moment point of the Centaur robot during the process of absorbing the residual impact force at each joint of the four legs, and when the zero-moment point approaches the boundary of the supporting polygon, it cooperates with the active impedance control unit to adjust the active impedance control command to maintain the zero-moment point of the Centaur robot within the support area enclosed by the boundary of the supporting polygon.

[0073] Figures 2-7 For an exemplary description, please refer to the above. Figure 1 This will not be elaborated upon here.

Claims

1. A method for controlling the anti-recoil force of a centaur robot with coordinated dual arms, characterized in that, include: The operation task category to be performed by the two arms of the centaur robot is obtained, and the corresponding predicted external force direction feature and predicted external force amplitude feature are extracted based on the operation task category; Based on the predicted external force direction characteristics and the predicted external force amplitude characteristics, before the double arms perform the operation and generate impact force, adjust the support force distribution parameters of the four feet of the Centaur robot to coordinately adjust the spatial posture of the Centaur robot's torso. The actual contact force at the ends of the two arms during operation is collected, and the actual contact force at the ends is converted to each joint of the four legs using a pre-constructed whole-machine dynamics mapping relationship to obtain the compensation torque of each joint. The whole-machine dynamics mapping relationship characterizes the force transmission constraint between the ends of the two arms and each joint of the four legs. The active impedance control commands of each joint of the four legs are updated according to the compensation torque, and the joints of the four legs are driven to absorb residual impact force based on the active impedance control commands, so as to keep the zero torque point of the centaur robot inside the support area enclosed by the support polygon boundary.

2. The method according to claim 1, characterized in that, The process of obtaining the task category of the operation to be performed by the centaur robot's arms, and extracting the corresponding predicted external force direction features and predicted external force amplitude features based on the task category, includes: The operation task category is analyzed to identify the tool contact point trajectory of the two arms during the operation process; The rate of change of momentum is extracted tangentially along the trajectory of the tool contact point to convert the rate of change of momentum into an initial impact force vector; The initial impact force vector is decomposed into the local coordinate system of the centaur robot to obtain the direction vector and the magnitude value. Then, the predicted external force direction characteristics are determined based on the direction vector, and the predicted external force amplitude characteristics are determined based on the magnitude value.

3. The method according to claim 1, characterized in that, The step of adjusting the foot support force distribution parameters of the centaur robot's four legs before the impact force is generated by the operation of the two arms, based on the predicted external force direction characteristics and the predicted external force amplitude characteristics, to coordinately adjust the torso spatial pose of the centaur robot, includes: Based on the predicted external force direction characteristics, determine the opposite overturning moment that is opposite to the predicted external force direction characteristics; Combining the predicted external force amplitude characteristics and the reverse resistance overturning moment, the expected vertical reaction force of each support foot is calculated, so as to adjust the distribution parameters of the four-foot support force based on the expected vertical reaction force of each support foot. The support output of the corresponding supporting foot among the four feet is adjusted according to the support force distribution parameters of the four feet to coordinately adjust the spatial posture of the torso, thereby causing the center of mass corresponding to the spatial posture of the torso to shift in a direction opposite to the predicted external force direction.

4. The method according to claim 1, characterized in that, The process involves collecting the actual end-effector contact forces generated by the two arms during operation and converting these forces to the joints of the four legs using a pre-built whole-machine dynamics mapping relationship to obtain the compensating torque for each joint. This includes: The actual contact force at the end is obtained by a six-dimensional force sensor, and its force spin in space is extracted based on the actual contact force at the end. The force spinor is input into the overall dynamics mapping relationship, which includes a transmission matrix used to characterize the relative motion between the rigid body links. The force spinor is mapped using the transpose of the conduction matrix to obtain the equivalent joint torque of each joint space of the four legs, and then the compensation torque of each joint is determined based on the equivalent joint torque.

5. The method according to claim 1, characterized in that, The step of updating the active impedance control command of each joint of the four legs based on the compensated torque includes: Obtain the current stiffness parameters and damping parameters of each joint of the four legs; The compensation torque is introduced into the basic control closed loop to determine the torque compensation requirements of each joint in the corresponding impact direction; The damping parameter configuration value in the corresponding impact direction is increased according to the torque compensation requirement; The configuration value of the stiffness parameter in the corresponding impact direction is reduced according to the torque compensation requirement; Based on the increased configuration value of the damping parameter and the decreased configuration value of the stiffness parameter, a target impedance parameter is determined, and an active impedance control command is generated based on the target impedance parameter.

6. A centaur robot dual-arm cooperative anti-recoil force control device, characterized in that, include: The task external force prediction module is used to obtain the operation task category that the Centaur robot's two arms are about to perform, and extract the corresponding predicted external force direction features and predicted external force amplitude features based on the operation task category. The pre-support adjustment module is used to adjust the support force distribution parameters of the four legs of the Centaur robot before the impact force is generated by the operation of the two arms, based on the predicted external force direction characteristics and the predicted external force amplitude characteristics, so as to coordinately adjust the torso spatial pose of the Centaur robot. The contact force mapping module is used to collect the actual end contact force generated by the two arms during operation, and to convert the actual end contact force to each joint of the four legs using a pre-constructed whole-machine dynamics mapping relationship, so as to obtain the compensation torque of each joint. The whole-machine dynamics mapping relationship characterizes the force transmission constraint between the ends of the two arms and each joint of the four legs. The impedance control update module is used to update the active impedance control commands of each joint of the four legs according to the compensation torque, and drive each joint of the four legs to absorb residual impact force based on the active impedance control commands, so as to keep the zero torque point of the centaur robot inside the support area enclosed by the support polygon boundary.

7. An electronic device, characterized in that, Including memory and processor; The memory stores computer programs that can run on the processor; When the processor executes the computer program, it implements the method described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program; When the computer program is executed by a processor, it implements the method described in any one of claims 1-5.

9. A centaur robot, characterized in that, It includes a dual-arm mechanism, a quadruped mechanism, a torso mechanism, an end-effector force acquisition component, and a controller; The dual-arm mechanism is used to perform operational tasks and generates actual end contact force during the performance of the operational tasks. The quadruped mechanism is used to adjust the distribution parameters of the support force on the soles of the four feet through each joint, and to execute active impedance control commands through each joint. The trunk mechanism is used to change the spatial posture of the trunk according to the distribution parameters of the support force on the soles of the four legs; The end contact force acquisition component is used to acquire the actual end contact force generated by the dual-arm mechanism during operation; The controller is connected to the dual-arm mechanism, the quadruped mechanism, the torso mechanism, and the end contact force acquisition component, respectively. The controller is used to obtain the type of operation task that the dual-arm mechanism is about to perform, and extract the corresponding predicted external force direction features and predicted external force amplitude features based on the type of operation task. The controller is also used to control the quadruped mechanism to adjust the distribution parameters of the foot support force of the quadrupeds before the double-arm mechanism generates an impact force by performing the operation, based on the predicted external force direction characteristics and the predicted external force amplitude characteristics, so as to coordinately adjust the torso spatial posture of the torso mechanism. The controller is also used to convert the actual contact force at the end of the machine to each joint of the quadruped mechanism using a pre-built whole-machine dynamics mapping relationship, so as to obtain the compensation torque of each joint. The whole-machine dynamics mapping relationship characterizes the force transmission constraint between the end of the double-arm mechanism and each joint of the quadruped mechanism. The controller is also used to update the active impedance control command of each joint of the quadruped mechanism according to the compensation torque, and drive each joint of the quadruped mechanism to absorb residual impact force based on the active impedance control command, so as to maintain the zero torque point of the centaur robot within the support area enclosed by the support polygon boundary.

10. A centaur robot dual-arm cooperative anti-recoil control system, characterized in that, It includes a task category acquisition unit, a pre-support adjustment unit, an end contact force acquisition unit, a dynamic mapping unit, an active impedance control unit, and a zero torque point monitoring unit; The task category acquisition unit is used to acquire the operation task category that the Centaur robot's two arms are about to perform, and output the operation task category to the pre-support adjustment unit; The pre-support adjustment unit is used to extract the corresponding predicted external force direction features and predicted external force amplitude features based on the operation task category, and adjust the support force distribution parameters of the four legs of the Centaur robot before the impact force is generated by the operation of the two arms, in order to coordinately adjust the torso spatial pose of the Centaur robot. The end contact force acquisition unit is used to acquire the actual end contact force generated by the two arms during operation, and output the actual end contact force to the dynamic mapping unit; The dynamic mapping unit is used to convert the actual contact force of the end effector to each joint of the four legs using a pre-constructed whole-machine dynamic mapping relationship, so as to obtain the compensation torque of each joint. The whole-machine dynamic mapping relationship characterizes the force transmission constraint between the end effector of the two arms and each joint of the four legs. The active impedance control unit is used to update the active impedance control command of each joint of the four legs according to the compensation torque, and drive each joint of the four legs to absorb residual impact force based on the active impedance control command. The zero-moment point monitoring unit is used to monitor the zero-moment point of the Centaur robot during the process of absorbing the residual impact force at each joint of the four legs, and when the zero-moment point approaches the boundary of the supporting polygon, it cooperates with the active impedance control unit to adjust the active impedance control command to maintain the zero-moment point of the Centaur robot within the support area enclosed by the boundary of the supporting polygon.

Images