Centaur robot operation end and foot end trajectory cooperative planning method, device, equipment, medium, robot and system

Abstract

The application provides a trajectory cooperative planning method and device for a Centaur robot work end and foot end, an electronic device, a computer readable storage medium, a Centaur robot and a system. Physical joint connection information of the Centaur robot is acquired to construct a unified kinematics representation matrix containing quadruped state data, torso state data and double-arm state data, an initial foot end, torso and double-arm trajectory is generated by analyzing a work target pose and a moving target pose, is mapped to a self-collision three-dimensional space grid for spatial interference checking and state parameter adjustment, and a non-interference cooperative motion trajectory is generated to separate control instructions for executing cooperative work.

Description

Technical Field

[0001] This application relates to the field of robot motion control technology, and more specifically, to a method, apparatus, electronic device, computer-readable storage medium, centaur robot and system for collaborative planning of end-effector and foot trajectories of a centaur robot. Background Technology

[0002] Centaur robots typically possess both lower limb mobility and upper limb operational capabilities, enabling them to move to target areas in complex terrain environments and perform tasks such as handling, grasping, assembling, and maintenance during or after movement. Because these robots contain multiple sets of support and execution components, torso-bearing components, and operational components, changes in posture during movement affect the upper limb's working range, and vice versa, which in turn affect the robot's overall stability. Therefore, these robots place high demands on overall motion planning in scenarios involving both movement and manipulation.

[0003] In existing composite robot control schemes, motion control and operation control are typically separated into two relatively independent control processes. This scheme first generates a bottom motion path based on the target position using the motion control unit, then generates an upper operation path based on the work object using the operation control unit. Finally, during the execution phase, both types of paths are sent to the corresponding actuators in chronological order. This scheme focuses on satisfying the basic requirements of movement and operation respectively, while paying less attention to the spatial occupancy and posture linkages between different actuators under the same planning framework.

[0004] However, the above-mentioned solutions have certain technical drawbacks. Since the movement path and the operation path are generated separately, the spatial relative relationships between different execution components are difficult to fully constrain during the planning stage when the robot is performing compound tasks. When the robot is moving and performing end-effector operations at the same time, it is easy for the main body components to approach each other, the range of motion to overlap, or the posture compensation to lag. This leads to an increased risk of collision or a greater degree of deviation in the posture of the end-effector, making it difficult to meet the requirements of the Centaur robot for collaborative planning when performing operation tasks while moving. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, computer-readable storage medium, centaur robot, and system for collaborative planning of end-effector and foot trajectories of a centaur robot, in order to at least alleviate the aforementioned technical problems.

[0006] A method for collaborative trajectory planning between the end effector and feet of a centaur robot includes: Obtain the physical joint connection information of the centaur robot, and construct a unified kinematic representation matrix containing quadruped state data, torso state data and arm state data based on the physical joint connection information. The operation command contains the target pose and the moving target pose. Based on the unified kinematic representation matrix, the target pose and the moving target pose are transformed by pose constraint to generate the initial foot landing point trajectory sequence, torso displacement trajectory sequence and arm end displacement trajectory sequence. A self-collision three-dimensional space mesh is established for the centaur robot. The trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement, and the trajectory sequence of the end displacement of the two arms are mapped to the self-collision three-dimensional space mesh for spatial interference verification to obtain the interference verification result. The state parameters in the unified kinematic representation matrix are adjusted according to the interference verification result to generate an interference-free cooperative motion trajectory. Based on the interference-free cooperative motion trajectory, foot control commands and dual-arm control commands are separated to control the centaur robot to perform cooperative tasks.

[0007] Optionally, the step of acquiring the physical joint connection information of the centaur robot, and constructing a unified kinematic representation matrix including quadrupedal state data, trunk state data, and arm state data based on the physical joint connection information, includes: Read the degree-of-freedom parameters of each joint position of the centaur robot and the link length data of adjacent joints from the physical joint connection information; Based on the degree of freedom parameters and the link length data, determine the independent motion parameters of the four legs corresponding to the four-legged state data, the single-axis rotation parameters of the waist corresponding to the trunk state data, and the coordinated motion parameters of the two arms corresponding to the two-arm state data. According to the joint connection order defined by the physical joint connection information, the quadrupedal independent motion parameters, the waist uniaxial rotation parameters, and the biarm coordinated motion parameters are configured into a data structure with a Jacobian matrix format to generate the unified kinematic representation matrix.

[0008] Optionally, the step of parsing the global operation instructions includes the target pose and the moving target pose, and performing pose constraint transformation on the target pose and the moving target pose based on the unified kinematic representation matrix to generate initial foot landing point trajectory sequences, torso displacement trajectory sequences, and arm end-effector displacement trajectory sequences, including: Extract the ground projection coordinates of the moving target pose in the global coordinate system, and plan multiple sets of off-road dynamic support points based on the ground projection coordinates, so as to generate the trajectory sequence of the foot landing point based on the multiple sets of off-road dynamic support points; Based on the centroid offset constraints corresponding to the torso state data in the unified kinematic representation matrix, the centaur robot's centroid balance position at each of the off-road state support points is calculated, so as to generate the torso displacement trajectory sequence based on each of the centroid balance positions. Extract the spatial coordinates of the target pose in the local space, and combine them with the torso displacement trajectory sequence to perform inverse coordinate transformation on the spatial coordinates to generate the end-arm displacement trajectory sequence.

[0009] Optionally, establishing the self-collision three-dimensional space mesh of the centaur robot includes: Obtain the geometric bounding box size data of each physical mechanical component of the centaur robot; A three-dimensional voxel space is established with the centaur robot's center of mass as the origin; In the three-dimensional voxel space, the geometric bounding box size data is divided into multiple three-dimensional occupancy nodes according to a pre-configured voxel resolution corresponding to the geometric bounding box size data, so as to construct the self-collision three-dimensional space mesh representing the robot body's occupied area using the multiple three-dimensional occupancy nodes.

[0010] Optionally, mapping the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence to the self-collision three-dimensional space grid for spatial interferometry verification to obtain the interferometry verification result includes: The foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence are respectively converted into space occupancy voxel blocks at the corresponding control time. The coordinates of the space-occupying voxel block are compared with the coordinates of the three-dimensional occupying node in the self-collision three-dimensional space mesh to obtain the coordinate overlap state between the space-occupying voxel block and the three-dimensional occupying node. When the coordinate coincidence state indicates that there are overlapping coordinate points, spatial interference is determined to have occurred, and the overlapping coordinate points are extracted as interference collision points to generate the interference verification result based on the interference collision points.

[0011] Optionally, adjusting the state parameters in the unified kinematic representation matrix based on the interference verification result to generate an interference-free cooperative motion trajectory includes: When the interference verification result indicates that spatial interference has occurred, the corresponding interference joint identifier is obtained based on the interference collision point; Calculate the escape vector of the interference collision point along the normal direction, and generate an interference avoidance correction term based on the escape vector; Based on the interference avoidance correction term, the state parameters corresponding to the interference joint identifier in the unified kinematic representation matrix are updated in reverse iteration until there is no spatial interference in the self-collision three-dimensional space grid, so as to obtain the updated state parameters. The motion paths of the feet, torso, and arms are reconstructed based on the updated state parameters, and the reconstructed motion paths are used as the interference-free cooperative motion trajectories.

[0012] Optionally, separating the foot control command and the two arm control commands based on the interference-free coordinated motion trajectory includes: The interference-free cooperative motion trajectory is discretized in time dimension according to the control cycle to obtain the synchronous pose data corresponding to each control cycle. Identify the actuator label category contained in the synchronized pose data, and separate and parse the synchronized pose data according to the foot drive category and the arm drive category to obtain the data belonging to the foot drive category and the data belonging to the arm drive category; The data belonging to the foot drive category is converted into a first torque pulse signal required by the underlying actuator motor, and the foot control command is generated based on the first torque pulse signal. The data belonging to the arm drive category is converted into a second torque pulse signal required by the underlying actuator motor, and the dual-arm control command is generated based on the second torque pulse signal.

[0013] Optionally, after separating the foot control commands and the dual-arm control commands based on the interference-free cooperative motion trajectory to control the centaur robot to perform cooperative work, the method further includes: Read the end contact force data collected by the six-dimensional force sensor installed at the end of the two arms during the collaborative operation; Extract the deformation components corresponding to each coordinate axis direction of the end contact force data; Determine whether the deformation component deviates from the pre-configured safe operating range corresponding to the end contact force data; When the deformation component deviates from the safe working range, the deviation value of the deformation component relative to the safe working range is determined, and the deviation value is fed back to the unified kinematic representation matrix to readjust the trajectory sequence of the foot landing point and the posture of the two arm ends, thereby compensating for the force deviation of the two arm ends.

[0014] Optionally, the step of calculating the centaur robot's center of mass equilibrium position at each of the off-road dynamic support points based on the center of mass offset constraints corresponding to the torso state data in the unified kinematic representation matrix, so that the torso displacement trajectory sequence is formed by each of the center of mass equilibrium positions, includes: Based on the off-walking support points in the foot landing point trajectory sequence, a polygonal support area is established by the off-walking support points; Read the dual-arm posture corresponding to the dual-arm state data in the unified kinematic representation matrix, and determine the mechanical arm weight offset data caused by the Centaur robot in different operation stages based on the dual-arm posture; By combining the weight offset data of the robotic arm and the mass distribution data of the centaur robot, the overall centroid coordinates in the cooperative motion state are determined. The composite centroid coordinates are projected onto the polygonal support area, and the pitch angle in the torso state data is adjusted so that the projection point of the composite centroid coordinates is located in the center area of ​​the polygonal support area, so that the adjusted state is determined as the centroid equilibrium position.

[0015] A trajectory collaborative planning device for the end effector and foot of a centaur robot includes: The representation matrix construction module is configured to acquire the physical joint connection information of the centaur robot, and construct a unified kinematic representation matrix containing quadruped state data, trunk state data and arm state data based on the physical joint connection information. The initial trajectory generation module is configured to parse the operation target pose and the moving target pose contained in the global operation instructions, and perform pose constraint transformation on the operation target pose and the moving target pose based on the unified kinematic representation matrix to generate the initial foot landing point trajectory sequence, torso displacement trajectory sequence and arm end displacement trajectory sequence. The interference trajectory generation module is configured to establish a self-collision three-dimensional space mesh of the centaur robot, map the trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement and the trajectory sequence of the end displacement of the two arms to the self-collision three-dimensional space mesh for spatial interference verification, so as to obtain the interference verification result, and adjust the state parameters in the unified kinematic representation matrix according to the interference verification result, thereby generating an interference-free cooperative motion trajectory. The control command separation module is configured to separate foot control commands and dual-arm control commands based on the interference-free cooperative motion trajectory, so as to control the centaur robot to perform cooperative operations.

[0016] An electronic device, comprising a memory and a processor; The memory stores computer programs that can run on the processor; When the processor executes the computer program, it performs the above-mentioned trajectory collaborative planning method for the end effector and foot of any of the centaur robots.

[0017] A computer-readable storage medium storing a computer program; When the computer program is executed by the processor, it performs the above-mentioned trajectory collaborative planning method for the end effector and foot effect of any half-centrod robot.

[0018] A centaur robot includes a robot body, a quadruped actuator, a torso actuator, a dual-arm actuator, and a controller; The quadruped actuator, the torso actuator, and the dual-arm actuator are respectively connected to the robot body; The controller is communicatively connected to the quadruped actuator, the torso actuator, and the dual-arm actuator, respectively. The controller is configured to acquire the physical joint connection information of the centaur robot, and to construct a unified kinematic representation matrix containing quadrupedal state data, torso state data and arm state data based on the physical joint connection information. The controller is also configured to parse the target pose and the moving target pose contained in the global operation instructions, and perform pose constraint transformation on the target pose and the moving target pose based on the unified kinematic representation matrix to generate an initial foot landing point trajectory sequence, a torso displacement trajectory sequence and a double arm end displacement trajectory sequence. The controller is also configured to establish a self-collision three-dimensional space mesh of the centaur robot, map the trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement and the trajectory sequence of the end arms displacement to the self-collision three-dimensional space mesh for spatial interference verification, so as to obtain the interference verification result, and adjust the state parameters in the unified kinematic representation matrix according to the interference verification result, thereby generating an interference-free cooperative motion trajectory. The controller is also configured to separate foot control commands and dual-arm control commands based on the interference-free cooperative motion trajectory, and to control the quadruped actuator to perform foot movements based on the foot control commands, and to control the dual-arm actuator to perform end-effector operations based on the dual-arm control commands, so that the centaur robot can perform cooperative operations.

[0019] A trajectory collaborative planning system for the end effector and foot of a centaur robot includes a global operation instruction generation device, a trajectory collaborative planning device, and a centaur robot. The global operation instruction generation device is communicatively connected to the trajectory collaborative planning device, and the trajectory collaborative planning device is communicatively connected to the centaur robot; The global operation instruction generation device is configured to generate global operation instructions that include the pose of the task target and the pose of the moving target, and send the global operation instructions to the trajectory collaborative planning device. The trajectory cooperative planning device is configured to acquire the physical joint connection information of the centaur robot, and to construct a unified kinematic representation matrix containing quadruped state data, trunk state data and arm state data based on the physical joint connection information. The trajectory collaborative planning device is also configured to parse the operation target pose and the moving target pose contained in the global operation instruction, and perform pose constraint transformation on the operation target pose and the moving target pose based on the unified kinematic representation matrix, so as to generate an initial foot landing point trajectory sequence, torso displacement trajectory sequence and arm end displacement trajectory sequence. The trajectory cooperative planning device is also configured to establish a self-collision three-dimensional space grid for the centaur robot, map the trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement, and the trajectory sequence of the end displacement of the two arms to the self-collision three-dimensional space grid for spatial interference verification, so as to obtain the interference verification result, and adjust the state parameters in the unified kinematic representation matrix according to the interference verification result, thereby generating an interference-free cooperative motion trajectory. The trajectory cooperative planning device is also configured to separate foot control commands and dual-arm control commands based on the interference-free cooperative motion trajectory, and send the foot control commands and dual-arm control commands to the centaur robot. The centaur robot is configured to perform foot movements according to the foot control commands and to perform end-effector tasks according to the arm control commands, in order to complete collaborative work.

[0020] The technical advantages of the technical solution provided in this application are: This application presents a collaborative trajectory planning method for the end effector and feet of a centaur robot. Addressing the technical shortcomings of existing methods where movement and manipulation paths are generated separately, leading to insufficient constraints on the spatial relationships between different actuators during the planning stage, this method acquires the physical joint connection information of the centaur robot and constructs a unified kinematic representation matrix containing quadrupedal state data, torso state data, and arm state data. This prevents the quadrupeds, torso, and arms from being treated as isolated planning objects. Compared to traditional methods that process movement and manipulation paths separately, this application first converts the overall joint connection relationships of the robot into a unified kinematic representation basis. This allows subsequent trajectory generation for the feet, torso, and arms to be performed using the same data, reducing the possibility of insufficient posture coordination between different actuators due to inconsistent planning sources.

[0021] Furthermore, this application analyzes the target pose and moving target pose contained in the global operation instructions, and performs pose constraint transformation on the target pose and moving target pose respectively based on a unified kinematic representation matrix to generate initial foot landing point trajectory sequences, torso displacement trajectory sequences, and arm end-effector displacement trajectory sequences. This processing method does not generate paths separately for the moving target or the target pose, but processes the moving target pose and the target pose simultaneously under the same unified kinematic representation matrix, ensuring that the foot landing point, torso displacement, and arm end-effector displacement have a mutual correspondence in the initial generation stage. Compared to traditional methods that splice and distribute different paths during the execution stage, this application can form a more continuous collaborative trajectory foundation in the planning stage, thereby reducing the risk of significant deviations in the end-effector pose due to torso movement.

[0022] Furthermore, this application establishes a self-collision 3D spatial mesh for the centaur robot and maps the foot landing point trajectory sequence, torso displacement trajectory sequence, and arm end-effector displacement trajectory sequence onto the self-collision 3D spatial mesh for spatial interference verification to obtain the interference verification results. This processing method ensures that the initial trajectories of the feet, torso, and arms not only meet their respective motion objectives but also undergo spatial interference verification within the same 3D spatial occupancy relationship. Compared to the traditional approach of generating movement paths and operation paths separately and then directly issuing them for execution, this application can detect potential spatial overlaps or interference relationships between different execution components during trajectory formation, thereby allowing for more thorough planning-level handling of self-collision risks before execution.

[0023] Furthermore, this application adjusts the state parameters in the unified kinematic representation matrix based on the interference verification results, thereby generating an interference-free cooperative motion trajectory. Foot control commands and dual-arm control commands are then separated from this trajectory to control the Centaur robot to perform cooperative tasks. Since the interference verification results are fed back to the state parameters in the unified kinematic representation matrix, the foot control commands and dual-arm control commands are not directly generated based on independent original paths, but rather based on the interference-free cooperative motion trajectory that has undergone spatial interference verification and state parameter adjustment. Compared to traditional methods where motion control and operation control easily interfere with each other during execution, this application enables foot movement and dual-arm end-effector operations to be executed along a more consistent cooperative motion trajectory, thereby reducing the probability of interference between body components and improving the stability of the operational pose in complex motion and operation scenarios. Attached Figure Description

[0024] Figure 1 This application provides an embodiment of a trajectory collaborative planning scenario for the end effector and foot of a centaur robot. Figure 2 This application provides an embodiment of a trajectory collaborative planning method for the end effector and foot of a centaur robot. Figure 3 This application provides an embodiment of a trajectory collaborative planning device for the end effector and foot of a centaur robot. 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 trajectory collaborative planning system for the end effector and foot of a centaur robot. Detailed Implementation

[0025] like Figure 1 The image shows a scenario of trajectory collaborative planning between the end effector and the feet of a centaur robot according to an embodiment of this application; as shown... Figure 2 As shown in the figure, a method for collaborative trajectory planning between the end effector and the foot of a centaur robot according to an embodiment of this application includes: Obtain the physical joint connection information of the centaur robot, and construct a unified kinematic representation matrix containing quadruped state data, torso state data and arm state data based on the physical joint connection information. The operation command contains the target pose and the moving target pose. Based on the unified kinematic representation matrix, the target pose and the moving target pose are transformed by pose constraint to generate the initial foot landing point trajectory sequence, torso displacement trajectory sequence and arm end displacement trajectory sequence. A self-collision three-dimensional space mesh is established for the centaur robot. The trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement, and the trajectory sequence of the end displacement of the two arms are mapped to the self-collision three-dimensional space mesh for spatial interference verification to obtain the interference verification result. The state parameters in the unified kinematic representation matrix are adjusted according to the interference verification result to generate an interference-free cooperative motion trajectory. Based on the interference-free cooperative motion trajectory, foot control commands and dual-arm control commands are separated to control the centaur robot to perform cooperative tasks.

[0026] Optionally, the step of acquiring the physical joint connection information of the centaur robot, and constructing a unified kinematic representation matrix including quadrupedal state data, trunk state data, and arm state data based on the physical joint connection information, includes: Read the degree-of-freedom parameters of each joint position of the centaur robot and the link length data of adjacent joints from the physical joint connection information; Based on the degree of freedom parameters and the link length data, determine the independent motion parameters of the four legs corresponding to the four-legged state data, the single-axis rotation parameters of the waist corresponding to the trunk state data, and the coordinated motion parameters of the two arms corresponding to the two-arm state data. According to the joint connection order defined by the physical joint connection information, the quadrupedal independent motion parameters, the waist uniaxial rotation parameters, and the biarm coordinated motion parameters are configured into a data structure with a Jacobian matrix format to generate the unified kinematic representation matrix.

[0027] Preferably, the specific implementation process of reading the degree-of-freedom parameters of each joint position of the Centaur robot and the link length data of adjacent joints from the physical joint connection information is as follows: When the Centaur robot completes assembly calibration or starts initialization calibration, the mechanical connection relationship between the robot body, quadruped actuator, torso actuator, and dual-arm actuator is first described in a node-based manner to form the physical joint connection information. The technical essence of the physical joint connection information is not a simple list of structural names, but a data-based connection description that can be calculated and read. The data-based connection description includes at least the joint node identifier, the component type to which the joint node belongs, the identifier of adjacent joint nodes, the direction of the joint axis, the range of motion of the joint, the link end point identifier, and the link calibration length field. The joint node identifiers are used to distinguish the joint positions of the Centaur robot. The component type to which the joint node belongs is used to map each joint position to a quadruped actuator, a torso actuator, or a dual-arm actuator. The adjacent joint node identifiers are used to express the direct connection relationship between one joint position and another. The joint axis direction and the joint range of motion jointly contribute to forming the degree of freedom parameters. The link endpoint identifier and the link calibration length field jointly contribute to forming the link length data. When reading the physical joint connection information, the joint node records corresponding to the joint node identifiers are accessed one by one according to the joint node identifiers. Then, the joint axis direction and the joint range of motion are extracted from the joint node records to form the degree of freedom parameters of the corresponding joint positions. Subsequently, the joint node records connecting the two ends of the same mechanical segment are located according to the adjacent joint node identifiers, and the link endpoint identifier and the link calibration length field are read from the joint node records at both ends of the same mechanical segment to form the link length data. The degree-of-freedom parameters are used to express the number of directions, directional constraints, and angular boundaries that can change the posture of each joint position during motion planning. The link length data is used to express the geometric distance between adjacent joints in the coordinate reference corresponding to the robot's center of mass. The degree-of-freedom parameters and the link length data together determine the reachable space of the feet, torso, and arms under the same kinematic caliber. Since the centaur robot's overall posture is simultaneously altered by the foot landing, torso displacement, and arm end-effector displacement when performing end-effector tasks during movement, reading only the joint data of one type of actuator makes it difficult to reflect the spatial coupling relationship between the quadrupedal state data, torso state data, and arm state data when generating a unified kinematic representation matrix. Therefore, this application reads both the degree-of-freedom parameters and the link length data simultaneously through the physical joint connection information. This allows the determination of quadrupedal independent motion parameters, waist single-axis rotation parameters, and arm cooperative motion parameters to utilize the same joint node source and the same link length source, avoiding trajectory coordination errors caused by inconsistent source calibers in the motion data of different actuators.

[0028] Preferably, in the specific technical implementation of determining the independent motion parameters of the four legs corresponding to the four-legged state data, the single-axis rotation parameters of the waist corresponding to the torso state data, and the coordinated motion parameters of the two arms corresponding to the two-arm state data based on the degree-of-freedom parameters and the link length data, the degree-of-freedom parameters are first grouped according to the component type to which the joint nodes belong, to form a grouping result of the component type to which the joint nodes belong. This allows the degree-of-freedom parameters belonging to the four-legged actuator to enter the four-legged parameter reading path, the degree-of-freedom parameters belonging to the torso actuator to enter the torso parameter reading path, and the degree-of-freedom parameters belonging to the two-arm actuator to enter the two-arm parameter reading path. Subsequently, the link length data is associated with the grouping result of the component type to which the joint nodes belong, according to the adjacent joint node identifiers, so that each group of degree-of-freedom parameters in the four-legged parameter reading path, the torso parameter reading path, and the two-arm parameter reading path has a corresponding link length data. For the quadrupedal parameter reading path, the motion direction and boundary of the hip joint, knee joint, and related joints of the foot are read based on the degree of freedom parameters on each foot kinematic chain. The geometric transmission distance between adjacent foot joints is read based on the link length data on the same foot kinematic chain to form the quadrupedal independent motion parameters. The technical essence of these quadrupedal independent motion parameters is that each foot kinematic chain can independently describe the change in foot landing position without relying on the displacement of the arms. This includes both the joint angle boundaries of each foot kinematic chain and the foot reachable area boundary defined by the link length data. When subsequently generating the foot landing point trajectory sequence, the quadrupedal independent motion parameters can directly determine whether the change in foot landing point falls within the foot reachable area boundary. For the trunk parameter reading path, the single-axis motion direction and range of motion of the lumbar rotation joint are read based on the degree-of-freedom parameters in the trunk actuator. The spatial offset of the lumbar rotation joint relative to the robot body's center of mass is determined based on the link length data between the trunk actuator and adjacent joints of the robot body, thus forming the lumbar single-axis rotation parameters. The technical essence of these lumbar single-axis rotation parameters is that they describe the single-axis posture adjustment data that affects the spatial position of the arm ends and the overall center of mass position during pitch or rotation adjustments. When generating the trunk displacement trajectory sequence, the lumbar single-axis rotation parameters can constrain the trunk posture change amplitude and express the transmission influence of trunk posture changes on the arm end-of-arm displacement trajectory sequence. For the dual-arm parameter reading path, the motion direction and boundary of the shoulder joint, elbow joint, and wrist joint are read based on the degree-of-freedom parameters of the left and right arms in the dual-arm actuator. The spatial transmission distance of the arm ends relative to the trunk actuator is determined based on the link length data between adjacent joints in the dual-arm actuator, thus forming the dual-arm cooperative motion parameters.The technical essence of the dual-arm coordinated motion parameters is to incorporate the pose changes of the left and right arm end-effectors into a data combination with the same dual-arm end-effector operational caliber. When generating the subsequent dual-arm end-effector displacement trajectory sequence, the reachability and mutual avoidance relationships of the two arm end-effectors can be simultaneously constrained based on the dual-arm coordinated motion parameters. This mutual avoidance relationship reduces spatial conflicts within the dual-arm end-effector displacement trajectory sequence. Therefore, the quadrupedal independent motion parameters, the waist single-axis rotation parameters, and the dual-arm coordinated motion parameters correspond to the quadrupedal state data, the torso state data, and the dual-arm state data, respectively. All three originate from the same set of degree-of-freedom parameters and link length data, enabling a unified kinematic representation matrix to express the correlation between the centaur robot's movement and operations from the same data source.

[0029] Preferably, the specific implementation process of configuring the independent motion parameters of the four legs, the single-axis rotation parameters of the waist, and the cooperative motion parameters of the two arms into a data structure with a Jacobian matrix format to generate the unified kinematic representation matrix according to the joint connection order defined by the physical joint connection information is as follows: First, the identifiers of adjacent joint nodes are read from the physical joint connection information, and the connection direction from the robot's center of mass to the four-legged actuator, the torso actuator, and the two-arm actuator is determined based on the identifiers of adjacent joint nodes. The technical essence of the joint connection order is the sequential arrangement of each joint node in the mechanical transmission path. It is not an arbitrary numbering order, but a data arrangement order determined by the physical connection path extending outward from the robot's center of mass to the feet and the ends of the two arms. When determining the joint connection sequence, the torso center node corresponding to the robot's center of mass is first used as the starting node. The lumbar rotation joint, quadrupedal root joints, and arm root joints directly connected to the torso center node are then searched along the adjacent joint node identifiers. The next-level adjacent joint node is then searched along each connection path until the foot end nodes and arm end nodes are reached, thus forming the joint connection sequence from the torso center node to each end node. After the joint connection sequence is formed, the quadrupedal independent motion parameters are written into the quadrupedal state area of ​​the data structure according to the positions of the four foot kinematic chains in the joint connection sequence. The lumbar single-axis rotation parameters are written into the torso state area of ​​the data structure according to the positions of the lumbar rotation joints in the joint connection sequence. The arm cooperative motion parameters are written into the arm state area of ​​the data structure according to the positions of the left and right arm kinematic chains in the joint connection sequence. The technical essence of the Jacobian matrix format data structure is a data structure arranged according to the correspondence between joint state changes and end-effector spatial changes. Its row direction is used to express the displacement or posture response of the feet, torso, and end-effectors in the spatial direction, and its column direction is used to express the joint state parameters arranged according to the joint connection order. The elements at the intersection of rows and columns are used to express the numerical contribution of a change in a certain joint state parameter to the response of the feet, torso, or end-effectors in a certain spatial direction. Since the foot landing point trajectory sequence, torso displacement trajectory sequence, and end-effector displacement trajectory sequence need to be generated under the same spatial and temporal caliber during collaborative operation of the centaur robot, the Jacobian matrix format data structure can place the independent motion parameters of the four legs, the single-axis rotation parameters of the waist, and the collaborative motion parameters of the two arms into the same computable arrangement, so that subsequent pose constraint transformations do not require the separate establishment of independent quadrupedal calculation calibers, torso calculation calibers, and end-effector calculation calibers.In the specific configuration, firstly, column indexes are established according to the joint connection order, so that each column index corresponds one-to-one with the joint state parameters at each joint position; then, row indexes are established according to the spatial response directions of the feet, trunk, and arms, so that each row index corresponds to the spatial displacement direction, spatial posture direction, or end-effector response direction; subsequently, the values ​​related to joint state changes in the quadrupedal independent motion parameters, the lumbar uniaxial rotation parameters, and the arm coordinated motion parameters are written into the corresponding row-column intersection elements, thereby forming the Jacobian matrix format data structure. After the Jacobian matrix format data structure is formed, the quadrupedal state region, the trunk state region, and the arm state region within it can be associated through the row-column intersection elements, thus providing a structured data foundation for generating the unified kinematic representation matrix.

[0030] Preferably, when generating the unified kinematic representation matrix, the data structure with the Jacobian matrix format described above is used as the matrix framework. The independent motion parameters of the four legs, the single-axis rotation parameters of the waist, and the coordinated motion parameters of the two arms are written into the corresponding positions in the matrix framework according to the joint connection order, so as to form the unified kinematic representation matrix containing the state data of the four legs, the state data of the torso, and the state data of the two arms. The technical essence of the unified kinematic representation matrix is ​​to express the data of the entire kinematic chain of the centaur robot under the same coordinate reference and the same joint arrangement. It is not a simple splicing of the state data of the four legs, the state data of the torso, and the state data of the two arms, but rather to uniformly map the joint state parameters of the four-legged actuator, the torso actuator, and the two-arm actuator to the same matrix framework, so that the influence of a joint state parameter on the spatial response of the foot, torso, or the end of the two arms can be read in the elements at the intersection of the rows and columns. The rows of the unified kinematic representation matrix are used to express the response dimensions of different moving parts of the Centaur robot in space. The rows corresponding to the quadruped actuators express the responses of the feet in the ground support direction, forward and backward movement direction, and lateral movement direction. The rows corresponding to the torso actuators express the responses corresponding to torso displacement and torso posture changes. The rows corresponding to the dual-arm actuators express the responses corresponding to the position and posture changes of the dual-arm end-effectors in the work space. The columns of the unified kinematic representation matrix are used to express the joint state parameters arranged according to the joint connection order. The list from the quadrupedal independent motion parameters reaches the movable states of each joint position of the quadrupeds. The list from the waist single-axis rotation parameters reaches the movable states of the waist rotation joint. The list from the dual-arm cooperative motion parameters reaches the movable states of each joint position of the dual arms. The row-column intersection elements in the unified kinematic representation matrix are used to express the influence relationship between the joint state parameters in the corresponding column and the spatial response dimension in the corresponding row. For example, the row-column intersection element at the intersection of the column corresponding to the foot root joint and the row corresponding to the fore-and-aft movement direction of the foot is used to express the contribution of the foot root joint change to the fore-and-aft movement response of the foot; the row-column intersection element at the intersection of the column corresponding to the waist rotation joint and the row corresponding to the working space position of the two arms is used to express the influence of the waist single-axis rotation parameter change on the displacement trajectory sequence of the two arms; the row-column intersection element at the intersection of the column corresponding to the shoulder joint of the two arms and the row corresponding to the trunk posture response is used to express the linkage influence of the change of the coordinated motion parameters of the two arms on the trunk state data.When generating the unified kinematic representation matrix, the coordinate references of the quadrupedal independent motion parameters, the waist single-axis rotation parameters, and the dual-arm coordinated motion parameters are first made consistent, so that the quadrupedal independent motion parameters, the waist single-axis rotation parameters, and the dual-arm coordinated motion parameters all adopt the coordinate reference corresponding to the robot's body center of mass. Then, according to the joint connection order, the quadrupedal independent motion parameters, the waist single-axis rotation parameters, and the dual-arm coordinated motion parameters after the coordinate reference consistency processing are written into the matrix frame, and the partitioning relationship of the quadrupedal state data, the torso state data, and the dual-arm state data to which the parameters belong is preserved during the writing process. Finally, the elements at the intersections of each row and column in the matrix frame are associated and filled, so that the motion transmission relationship between the feet, torso, and dual arms can be directly read by the unified kinematic representation matrix. The unified kinematic representation matrix formed in this way can participate in the pose constraint transformation of the task target pose and the moving target pose when parsing global operation instructions. It also provides the same data basis for generating foot landing point trajectory sequences, torso displacement trajectory sequences and arm end-effector displacement trajectory sequences, thus enabling the collaborative planning between the centaur robot's movement and end-effector operations to have a continuous data source and a clear computational transfer relationship.

[0031] Optionally, the step of parsing the global operation instructions includes the target pose and the moving target pose, and performing pose constraint transformation on the target pose and the moving target pose based on the unified kinematic representation matrix to generate initial foot landing point trajectory sequences, torso displacement trajectory sequences, and arm end-effector displacement trajectory sequences, including: Extract the ground projection coordinates of the moving target pose in the global coordinate system, and plan multiple sets of off-road dynamic support points based on the ground projection coordinates, so as to generate the trajectory sequence of the foot landing point based on the multiple sets of off-road dynamic support points; Based on the centroid offset constraints corresponding to the torso state data in the unified kinematic representation matrix, the centaur robot's centroid balance position at each of the off-road state support points is calculated, so as to generate the torso displacement trajectory sequence based on each of the centroid balance positions. Extract the spatial coordinates of the target pose in the local space, and combine them with the torso displacement trajectory sequence to perform inverse coordinate transformation on the spatial coordinates to generate the end-arm displacement trajectory sequence.

[0032] Preferably, the specific implementation process of parsing the target pose and moving target pose included in the global operation instruction, and performing pose constraint transformation on the target pose and moving target pose based on the unified kinematic representation matrix is ​​as follows: After the global operation instruction is read, the target pose and moving target pose are first separated from the global operation instruction, and the target pose and moving target pose are written into the pose parsing carrying record, so that the same global operation instruction source can be used when generating the foot landing point trajectory sequence, the torso displacement trajectory sequence and the arm end displacement trajectory sequence. The pose parsing carrying record is also used to save the reading correspondence between the target pose, the moving target pose and the unified kinematic representation matrix, so that the pose constraint transformation can read the target pose and moving target pose under the same global operation instruction source from the pose parsing carrying record. The technical essence of the moving target pose is a description of the spatial posture that the Centaur robot needs to reach or traverse during a movement task. The moving target pose includes a moving target position component and a moving target orientation component. The moving target position component is used to express the target arrival position of the Centaur robot in the global coordinate system, and the moving target orientation component is used to express the body orientation constraint corresponding to the Centaur robot when it reaches the target arrival position. The moving target pose is used to guide the generation of the foot landing point trajectory sequence and the torso displacement trajectory sequence. The technical essence of the working target pose is a description of the spatial posture that the end-effectors need to approach or reach when performing end-effector tasks. The working target pose includes a working target position component and a working target orientation component. The working target position component is used to express the target working position of the end-effectors in the working space, and the working target orientation component is used to express the end-effector orientation constraint corresponding to the end-effectors when they reach the target working position. The working target pose is used to guide the generation of the end-effector displacement trajectory sequence. To enable the moving target pose and the task target pose to enter the same pose constraint transformation process, a global coordinate system is first constructed. The global coordinate system uses the ground reference plane in the scene where the Centaur robot is performing collaborative work as the spatial positioning reference, and the position corresponding to the centroid of the Centaur robot body when it starts planning as the coordinate reference point. A longitudinal coordinate direction is established along the ground forward direction, a lateral coordinate direction is established along the ground lateral direction, and a height coordinate direction is established along the vertical ground direction. The ground reference plane is used to carry the subsequent ground projection coordinates and multiple sets of discrete dynamic support points. The coordinate reference point is used to define the spatial reading starting point of the moving target pose and the task target pose in the global coordinate system. The longitudinal coordinate direction, the lateral coordinate direction, and the height coordinate direction together constitute the spatial measurement caliber of the global coordinate system.The global coordinate system provides a unified spatial metric for the moving target pose, the task target pose, the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the end-effector displacement trajectory sequence, preventing the movement path from using one spatial reference while the end-effector operation uses another. The technical essence of the pose constraint transformation is to convert the moving target pose and the task target pose into spatial response requirements that can be read by the unified kinematic representation matrix. These spatial response requirements include ground movement spatial response requirements corresponding to the moving target pose and end-effector operation spatial response requirements corresponding to the task target pose. This ensures that the ground movement spatial response requirements correspond to the quadrupedal state data and torso state data in the unified kinematic representation matrix, and that the end-effector operation spatial response requirements correspond to the end-effector state data and torso state data in the unified kinematic representation matrix. Specifically, the moving target pose is first written into the global coordinate system to extract the ground projection coordinates corresponding to the moving target pose in the global coordinate system; then, the task target pose is written into a local space associated with the torso state data to extract the spatial coordinates corresponding to the task target pose in the local space. Through the above processing, a computable transfer relationship is established between the moving target pose, the task target pose, the unified kinematic representation matrix, the ground projection coordinates, and the spatial coordinates. This computable transfer relationship is further preserved by the pose analysis bearing record and can be used by subsequent multiple sets of off-axis support points, center of mass equilibrium positions, and end-arm displacement trajectory sequences.

[0033] Preferably, in the specific technical implementation of extracting the ground projection coordinates of the moving target pose in the global coordinate system and planning multiple sets of off-road dynamic support points based on the ground projection coordinates to generate the trajectory sequence of the foot landing point based on the multiple sets of off-road dynamic support points, the moving target position component and the moving target orientation component in the global coordinate system are first read from the pose analysis bearing record, and the moving target position component is projected onto the ground reference plane along the height coordinate direction of the global coordinate system to form the ground projection coordinates. The technical essence of the ground projection coordinates is to convert the three-dimensional spatial data in the moving target pose used to express the target's arrival position into two-dimensional ground position data that can guide the foot to contact the ground; the ground projection coordinates do not replace the moving target pose, but rather separate the ground position constraints directly related to the foot landing planning from the moving target pose, so that the subsequent multiple sets of off-road dynamic support points can be generated around the ground movement process of the Centaur robot. The two-dimensional ground position data and the ground position constraints are both expressed by the ground projection coordinates. After extracting the ground projection coordinates, a ground movement reference line is constructed based on the current position of the Centaur robot's center of mass and the ground projection coordinates. The movement distance from the current position of the robot's center of mass to the ground projection coordinates is then divided into multiple discrete planning moments based on this ground movement reference line. The ground movement reference line represents the ground direction reference for the Centaur robot's movement from the current position of its center of mass to the ground projection coordinates. The discrete planning moments represent the sequential relationship between adjacent foot landing points in the foot landing point trajectory sequence. Subsequently, at each discrete planning moment, the reachable range of the quadruped actuators is read by combining the quadruped state data in the unified kinematic representation matrix, and the reachable region boundaries of each foot kinematic chain are read by combining the quadruped independent motion parameters, thereby generating a set of discrete walking dynamic support points at the corresponding discrete planning moment. The technical essence of the multiple sets of discrete-time support points is the combination of foot support positions that the quadruped actuator can choose at different discrete planning moments. The reason for using the multiple sets of discrete-time support points is that the Centaur robot does not only experience one footing state during movement, but needs to continuously change the foot support position at multiple discrete planning moments. The multiple sets of discrete-time support points corresponding to multiple discrete planning moments can express the continuous support change from the position corresponding to the current center of mass of the robot body to the ground projection coordinates.When generating the multiple sets of discrete dynamic support points, each set of discrete dynamic support points includes candidate foot support positions corresponding to the quadruped actuator. These candidate foot support positions must fall within the reachable area boundary of the corresponding foot kinematic chain and maintain a spatial correspondence with the ground movement reference line. This ensures that the candidate foot support positions are constrained by the quadruped state data and the quadruped independent motion parameters, while also corresponding to the movement direction defined by the ground projection coordinates. Subsequently, according to the chronological order of the discrete planning moments, the candidate foot support positions corresponding to the same foot in the multiple sets of discrete dynamic support points are arranged sequentially to form the foot landing point trajectory sequence. The technical essence of the foot landing point trajectory sequence is the sequence of foot support positions of the quadruped actuator at multiple discrete planning moments. This sequence subsequently participates in the calculation of the centroid equilibrium position of the torso displacement trajectory sequence and in spatial interference verification within the self-collision three-dimensional spatial mesh, ensuring that foot movement is no longer a separate path independent of the torso and the end-effector operations.

[0034] Preferably, the specific implementation process of calculating the centaur robot's center of mass equilibrium position at each of the off-walk dynamic support points based on the center of mass offset constraints corresponding to the torso state data in the unified kinematic representation matrix, and generating the torso displacement trajectory sequence based on each of the center of mass equilibrium positions, is as follows: After the foot landing point trajectory sequence has been formed based on the multiple sets of off-walk dynamic support points, the rows and columns corresponding to the torso state data are first read from the unified kinematic representation matrix, and the influence relationship between the torso displacement response, torso posture response, and waist single-axis rotation parameters on the torso posture response is extracted from the rows and columns corresponding to the torso state data to form the center of mass offset constraints. The technical essence of the torso state data is that the centaur robot's torso actuator can change the robot's center of mass position and torso posture during coordinated movement. The torso state data is associated with the quadrupedal state data and the bi-arm state data through the unified kinematic representation matrix, so that the torso displacement trajectory sequence can be constrained by both the foot support position and the bi-arm end-effector working posture. The technical essence of the centroid offset constraint is to limit the allowable offset direction, allowable offset distance, and torso posture adjustment range of the robot body's centroid within the ground support range formed by each of the discrete-time support points. When constructing the centroid offset constraint, the foot support area is first determined based on a set of discrete-time support points corresponding to the current discrete planning time. Then, the response dimensions of the torso state data in the forward / backward movement direction, lateral movement direction, and height direction are read from the unified kinematic representation matrix, and the response dimensions are spatially mapped to the foot support area to form the centroid offset constraint at the current discrete planning time. Subsequently, combining the centroid offset constraint at the current discrete planning time, the set of discrete-time support points of the quadruped actuator at the current discrete planning time, and the end-arm displacement requirements corresponding to the arm state data, the centaur robot's centroid balance position at each of the discrete-time support points is calculated. The technical essence of the centroid balance position is the torso spatial position that allows the robot's centroid to maintain a matching relationship with the foot support area within the current discrete planning time. This centroid balance position is not an isolated coordinate point, but rather a torso displacement result jointly determined by a set of discrete-time support points, torso state data, and centroid offset constraints corresponding to the discrete planning time. Since the multiple sets of discrete-time support points correspond to multiple discrete planning times, and each discrete planning time forms a corresponding centroid balance position, it is necessary to generate a torso displacement trajectory sequence based on each centroid balance position.When generating the torso displacement trajectory sequence, the centroid equilibrium positions are arranged sequentially according to the order of the discrete planning moments. The continuity of the torso displacement direction and torso posture change direction is checked based on the change in centroid equilibrium position between adjacent discrete planning moments, forming a trajectory continuity check result. This result is used to determine whether the torso displacement direction and torso posture change direction between adjacent discrete planning moments match the centroid offset constraint. It is also used to correct torso spatial positions in the torso displacement trajectory sequence that do not satisfy the centroid offset constraint, ensuring a temporal correspondence between the torso displacement trajectory sequence and the foot landing point trajectory sequence. The torso displacement trajectory sequence subsequently participates in the inverse coordinate transformation of the arm end-effector displacement trajectory sequence and in the spatial interference verification within the self-collision three-dimensional spatial mesh. This ensures that the torso displacement is not merely a local balance adjustment around the foot support but also provides a working space transformation benchmark when generating the arm end-effector displacement trajectory sequence.

[0035] Preferably, the specific implementation process of extracting the spatial coordinates of the target pose in the local space and performing inverse coordinate transformation on the spatial coordinates in conjunction with the torso displacement trajectory sequence to generate the end-effector displacement trajectory sequence is as follows: After the torso displacement trajectory sequence has been formed based on each of the centroid equilibrium positions, a local space corresponding to each discrete planning moment is constructed based on the torso position and torso posture corresponding to each discrete planning moment in the torso displacement trajectory sequence. The technical essence of the local space is a working space expression caliber established based on the torso position and torso posture of the torso actuator at the current discrete planning moment. The working space expression caliber is used to express the working position of the end-effectors relative to the torso actuator. The working space expression caliber is different from the global coordinate system. The global coordinate system is used to express the overall movement position of the Centaur robot in the working scene, and the local space is used to express the working position of the end-effectors relative to the torso actuator. Since the dual-arm actuators are mounted on the torso actuators, the reachable range of the arm ends varies relative to the torso position and posture in the torso displacement trajectory sequence. Therefore, directly processing the target pose in the global coordinate system makes it difficult to reflect the influence of the torso displacement trajectory sequence on the dual-arm end-effector displacement trajectory sequence. Based on this relationship, the target pose is first transformed from the global coordinate system to the local space corresponding to the current discrete programming moment to extract the spatial coordinates of the target pose within the local space. The technical essence of these spatial coordinates is the relative position and orientation data of the target pose relative to the torso actuators. These spatial coordinates can directly correspond to the dual-arm state data and dual-arm cooperative motion parameters in the unified kinematic representation matrix. Subsequently, the spatial coordinates are transformed by inverse kinematic transformation in conjunction with the torso displacement trajectory sequence. The essence of the inverse coordinate transformation is to convert the spatial coordinates within the local space into end-effector displacement data executable by the dual-arm actuator. This process is not a simple coordinate conversion, but rather involves simultaneously reading the dual-arm state data, the dual-arm cooperative motion parameters, and the torso state data from the unified kinematic representation matrix. This allows the dual-arm end-effectors to utilize the local spatial reference provided by the torso displacement trajectory sequence when approaching the target pose. Specifically, during the inverse coordinate transformation, the spatial differences between the spatial coordinates and the left and right arm end-effectors are first read based on the local space at the current discrete planning time. Then, the reachable range and mutual avoidance relationships of the left and right arm kinematic chains are read based on the dual-arm cooperative motion parameters. Finally, the spatial differences are converted into the end-effector displacement requirements of the left and right arms at the current discrete planning time.Subsequently, according to the chronological order of the discrete planning moments, the end-effector displacement requirements at each discrete planning moment are arranged sequentially to form the dual-arm end-effector displacement trajectory sequence. The technical essence of the dual-arm end-effector displacement trajectory sequence is the end-effector spatial motion sequence of the dual-arm actuators approaching the target pose under the constraint of the torso displacement trajectory sequence. The dual-arm end-effector displacement trajectory sequence maintains a correspondence with the foot landing point trajectory sequence and the torso displacement trajectory sequence at the discrete planning moments. This ensures that when the foot landing point trajectory sequence, torso displacement trajectory sequence, and dual-arm end-effector displacement trajectory sequence are subsequently mapped to the self-collision three-dimensional spatial mesh for spatial interference verification, the spatial occupancy relationship between the feet, torso, and dual-arm end-effectors can be checked in the same temporal order.

[0036] Optionally, establishing the self-collision three-dimensional space mesh of the centaur robot includes: Obtain the geometric bounding box size data of each physical mechanical component of the centaur robot; A three-dimensional voxel space is established with the centaur robot's center of mass as the origin; In the three-dimensional voxel space, the geometric bounding box size data is divided into multiple three-dimensional occupancy nodes according to a pre-configured voxel resolution corresponding to the geometric bounding box size data, so as to construct the self-collision three-dimensional space mesh representing the robot body's occupied area using the multiple three-dimensional occupancy nodes.

[0037] Preferably, the specific implementation process of obtaining the geometric bounding box size data of each physical mechanical component of the Centaur robot is as follows: When the Centaur robot completes the assembly calibration or starts the initialization calibration, the component calibration record corresponding to the physical joint connection information is read first. The component calibration record is formed by the robot body, quadruped actuator, torso actuator and dual-arm actuator during the assembly calibration stage. The component calibration record includes physical mechanical component identification, component type of physical mechanical component, local coordinate reference of physical mechanical component, outer contour sampling point of physical mechanical component, adjacent joint node identification and installation pose of physical mechanical component. The physical mechanical component identifier is used to distinguish the physical mechanical components involved in space occupancy calculations among the robot body, quadruped actuators, torso actuators, and dual-arm actuators; the component type of the physical mechanical component is used to map the physical mechanical component identifier to the robot body, quadruped actuator, torso actuator, or dual-arm actuator; the local coordinate reference of the physical mechanical component is used to describe the coordinate reading aperture used when the corresponding physical mechanical component samples its own outer contour; the outer contour sampling point of the physical mechanical component is used to express the spatial boundary of the outer surface of the corresponding physical mechanical component under the local coordinate reference of the physical mechanical component; the installation pose of the physical mechanical component is used to convert the outer contour sampling point of the physical mechanical component to the coordinate reference corresponding to the center of mass of the robot body. When obtaining the geometric bounding box size data, the outer contour sampling points of the physical mechanical component are first read one by one according to the physical mechanical component identifier, and then the outer contour sampling points of the physical mechanical component are converted to the coordinate reference corresponding to the center of mass of the robot body according to the installation pose of the physical mechanical component. Then, the boundary value processing of the converted outer contour sampling points of the physical mechanical component is performed to form the geometric bounding box boundary of the corresponding physical mechanical component in the forward and backward movement direction, the lateral movement direction, and the height direction. The bounding box boundaries are used to form the bounding box size data, which is represented in the bounding box size data as the length boundary in the forward and backward movement direction, the width boundary in the lateral movement direction, and the height boundary in the height direction of the corresponding physical mechanical component. The technical essence of the bounding box size data is the minimum circumscribed cuboid boundary data occupied by each physical mechanical component under the coordinate reference corresponding to the robot's center of mass. The bounding box size data includes the length boundary in the forward and backward movement direction, the width boundary in the lateral movement direction, and the height boundary in the height direction of the corresponding physical mechanical component. The bounding box size data is not simply a record of external dimensions, but rather a basis for subsequently mapping the physical mechanical components into three-dimensional voxel space. Through the bounding box size data, the spatial occupancy range of the robot body, quadruped actuators, torso actuators, and dual-arm actuators can be converted into discretely readable boundary data without directly processing complex curved surfaces.Because the support position of the feet of the quadruped actuator, the posture of the torso actuator, and the end positions of the arms of the dual-arm actuator all change simultaneously when the centaur robot performs end-effector tasks while moving, it is easy to overlook the spatial proximity relationship between the outer contours of the physical mechanical components if only the joint positions are used to determine whether there is interference. Therefore, this application first obtains the geometric bounding box size data of each physical mechanical component, so that when establishing the three-dimensional voxel space and constructing the self-collision three-dimensional space mesh, the spatial occupancy range expressed by the geometric bounding box size data can be used as the basis for spatial interference verification.

[0038] Preferably, the specific implementation process of establishing a three-dimensional voxel space with the centaur robot's center of mass as the origin is as follows: After the geometric bounding box size data of each physical mechanical component has been formed, the position of the robot's center of mass corresponding to the torso state data in the unified kinematic representation matrix is ​​first read, and the position of the robot's center of mass is used as the origin of the three-dimensional voxel space, so that the three-dimensional voxel space and the aforementioned unified kinematic representation matrix adopt the same coordinate reference corresponding to the robot's center of mass. The centaur robot's center of mass is not an arbitrarily selected spatial point, but a spatial reference position jointly determined by the mass distribution and installation posture of the robot body, quadruped actuators, torso actuators, and dual-arm actuators in the current planned state. The centaur robot's center of mass can reflect the central reference relationship of the spatial occupancy range when the centaur robot's overall posture changes. The central reference relationship is used to make the geometric bounding box size data, foot landing point trajectory sequence, torso displacement trajectory sequence, and dual-arm end displacement trajectory sequence unfold around the same spatial reference position in the subsequent mapping process. When establishing the three-dimensional voxel space, the origin is first established at the spatial position corresponding to the centaur robot's center of mass. Then, the three spatial axes of the three-dimensional voxel space are established along the forward / backward movement direction, the lateral movement direction, and the height direction in the global coordinate system. This allows the three-dimensional voxel space to simultaneously accommodate the spatial occupancy changes corresponding to the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end-effector displacement trajectory sequence. The technical essence of the three-dimensional voxel space is to divide the continuous three-dimensional space around the centaur robot into discrete spatial carrying areas that can be read according to rows, columns, and layers. The row direction corresponds to the forward / backward movement direction, the column direction corresponds to the lateral movement direction, and the layer direction corresponds to the height direction. The spatial units formed by the intersection of rows, columns, and layers are used to carry subsequent three-dimensional occupancy nodes. The difference between the three-dimensional voxel space and the ordinary three-dimensional coordinate space is that the ordinary three-dimensional coordinate space mainly expresses continuous positional relationships, while the three-dimensional voxel space further converts continuous positional relationships into discrete spatial unit relationships, so that the geometric bounding box size data of each physical mechanical component can be divided into multiple three-dimensional occupancy nodes. By establishing the three-dimensional voxel space with the centaur robot's center of mass as the origin, the spatial occupancy data of the quadruped actuator, torso actuator, and dual-arm actuator can be unfolded around the same central reference position. When the torso displacement trajectory sequence changes the corresponding position of the robot's center of mass, the three-dimensional voxel space can synchronously update the spatial reading reference with the corresponding position of the robot's center of mass, so that the subsequent foot landing point trajectory sequence, torso displacement trajectory sequence, and dual-arm end displacement trajectory sequence can be mapped to the three-dimensional voxel space with a consistent spatial measurement caliber.Thus, the three-dimensional voxel space provides a spatial support framework for the self-collision three-dimensional spatial mesh. The spatial support framework is subsequently used to receive multiple three-dimensional occupancy nodes obtained by dividing the geometric bounding box size data, and further form the self-collision three-dimensional spatial mesh representing the occupancy range of the robot body.

[0039] Preferably, in the specific technical implementation of dividing the geometric bounding box size data into multiple three-dimensional occupancy nodes in the three-dimensional voxel space according to a pre-configured voxel resolution corresponding to the geometric bounding box size data, a voxel configuration record is first read. This voxel configuration record is formed during assembly calibration or startup initialization calibration and establishes a correspondence with the component calibration record and the three-dimensional voxel space. The voxel configuration record includes voxel resolution, voxel side length reading aperture, voxel axial arrangement order, voxel occupancy determination aperture, and voxel boundary expansion margin. The voxel resolution is used to express the spatial division granularity of a single voxel unit in the three-dimensional voxel space in the forward and backward movement direction, the lateral movement direction, and the height direction. The voxel side length reading aperture is used to limit the reading method of the voxel resolution in the three spatial axes. The voxel axial arrangement order is used to limit the arrangement direction of multiple three-dimensional occupancy nodes in the three-dimensional voxel space. The voxel occupancy determination aperture is used to determine whether a voxel unit is occupied by a physical mechanical component. The voxel boundary expansion margin is used to leave space for outer contour sampling error and attitude change during the movement of the physical mechanical component. The essence of the pre-configured technology is to establish a voxel division correspondence between the voxel resolution and the size range of the geometric bounding box size data before performing spatial interferometry verification, and write the voxel division correspondence into the voxel configuration record. This allows subsequent division of the geometric bounding box size data to be performed directly based on the voxel configuration record, rather than generating the division criteria temporarily during each spatial interferometry verification. When configuring the voxel resolution, the geometric bounding box size data of each physical mechanical component is first read, and the minimum boundary scale of the corresponding physical mechanical component in the forward / backward movement direction, lateral movement direction, and height direction is determined from the geometric bounding box size data. Then, combining the spatial variation amplitude of the foot landing point trajectory sequence, torso displacement trajectory sequence, and arm end-effector displacement trajectory sequence during the Centaur robot's collaborative operation, a voxel resolution capable of distinguishing the occupied area of ​​adjacent physical mechanical components is determined, and this voxel resolution is written into the voxel configuration record. The minimum boundary scale is used to limit the spatial division granularity of the voxel resolution, ensuring that the voxel resolution and the geometric bounding box size data maintain the voxel division correspondence. The correspondence between the voxel resolution and the geometric bounding box size data means that the spatial division granularity of the voxel unit should be able to express the boundary changes of the geometric bounding box size data, so that smaller physical and mechanical components are not ignored in the division process, while larger physical and mechanical components do not generate too many invalid 3D occupancy nodes.When dividing the geometric bounding box size data into multiple 3D occupancy nodes, the geometric bounding box size data is first written into the 3D voxel space at a position matching the installation pose of the corresponding physical mechanical component. Then, the geometric bounding box size data is axially segmented according to the voxel resolution in the voxel configuration record, so that the geometric bounding box size data is divided into multiple voxel units in the forward / backward movement direction, the lateral movement direction, and the height direction, respectively. Subsequently, based on the voxel occupancy determination criteria, it is determined whether each voxel unit falls within the boundary range defined by the geometric bounding box size data of the corresponding physical mechanical component, and the voxel units falling within the boundary range are marked as the 3D occupancy nodes. The technical essence of the 3D occupancy node is that it is a discrete spatial occupancy unit of the physical mechanical component in the 3D voxel space. The 3D occupancy node stores the physical mechanical component identifier, the row position, column position, and layer position in the 3D voxel space, as well as the corresponding occupancy state, so that when constructing a self-collision 3D spatial mesh, the robot body occupancy range can be expressed according to the spatial positional relationship between the 3D occupancy nodes.

[0040] Preferably, the specific implementation process of constructing the self-collision three-dimensional spatial mesh representing the robot body's occupied area using the multiple three-dimensional occupancy nodes is as follows: After the geometric bounding box size data of each physical mechanical component is divided into multiple three-dimensional occupancy nodes, the multiple three-dimensional occupancy nodes are first assigned according to the physical mechanical component identifier, so that the three-dimensional occupancy nodes corresponding to the robot body, quadruped actuator, torso actuator, and dual-arm actuator can be distinguished and read in the same three-dimensional voxel space; then, according to the voxel axial arrangement order, the multiple three-dimensional occupancy nodes that have completed the assignment marking are written into the corresponding row position, column position, and layer position in the three-dimensional voxel space to form the robot body occupancy node arrangement. The technical essence of the robot body occupancy node arrangement is the spatial arrangement result of multiple three-dimensional occupancy nodes in the three-dimensional voxel space. The robot body occupancy node arrangement is used to express the spatial occupancy distribution of each physical mechanical component of the Centaur robot in the current planning state. The spatial occupancy distribution is subsequently entered into the mesh occupancy record through the robot body occupancy node arrangement. After forming the robot body's occupancy node arrangement, joint connection associations are further established for the three-dimensional occupancy nodes belonging to adjacent physical mechanical components based on the adjacent joint node identifiers in the physical joint connection information. This ensures that the robot body, quadruped actuators, torso actuators, and dual-arm actuators not only have discrete spatial occupancy relationships but also retain corresponding mechanical connection relationships. The joint connection associations are used to express the source of mechanical connections between adjacent physical mechanical components. These associations subsequently participate in determining the coordinate overlap state between the three-dimensional occupancy nodes, enabling spatial interference verification to distinguish between adjacent contact relationships generated by normal connections and spatial overlap relationships between non-adjacent physical mechanical components. Subsequently, the robot body's occupancy node arrangement and the joint connection associations are jointly written into a mesh occupancy record, and the self-collision three-dimensional spatial mesh is constructed based on the mesh occupancy record. The mesh occupancy record is used to store the robot body's occupancy node arrangement, the joint connection associations, and the occupancy state of the three-dimensional occupancy nodes, allowing the self-collision three-dimensional spatial mesh to read the robot body's occupancy range from the mesh occupancy record. The technical essence of the self-collision 3D spatial mesh is a data structure that expresses the spatial occupancy range of the Centaur robot's physical and mechanical components in the 3D voxel space using multiple 3D occupancy nodes. This self-collision 3D spatial mesh is not simply a 3D display mesh, but rather an occupancy relationship data structure that can be read and compared in subsequent spatial interference verification. The self-collision 3D spatial mesh is used to represent the robot's body occupancy range because it incorporates the 3D occupancy nodes of the robot body, quadrupedal actuators, torso actuators, and dual-arm actuators into the same spatial framework, and retains the mechanical connection source corresponding to each 3D occupancy node through the joint connection association.Subsequently, when mapping the foot landing point trajectory sequence, torso displacement trajectory sequence, and arm end-effector displacement trajectory sequence onto the self-collision 3D spatial mesh for spatial interference verification, the coordinate points of the space-occupying voxel block at the corresponding control moment can be compared with the 3D occupying nodes in the self-collision 3D spatial mesh to obtain the coordinate overlap state between the space-occupying voxel block and the 3D occupying nodes. Thus, the self-collision 3D spatial mesh connects the geometric bounding box size data, 3D voxel space, voxel resolution, and multiple 3D occupying nodes into a single spatial occupying representation, enabling the Centaur robot's self-collision risk in combined mobility and end-effector operation scenarios to be incorporated into the planning-level data verification process before trajectory execution.

[0041] Optionally, mapping the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence to the self-collision three-dimensional space grid for spatial interferometry verification to obtain the interferometry verification result includes: The foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence are respectively converted into space occupancy voxel blocks at the corresponding control time. The coordinates of the space-occupying voxel block are compared with the coordinates of the three-dimensional occupying node in the self-collision three-dimensional space mesh to obtain the coordinate overlap state between the space-occupying voxel block and the three-dimensional occupying node. When the coordinate coincidence state indicates that there are overlapping coordinate points, spatial interference is determined to have occurred, and the overlapping coordinate points are extracted as interference collision points to generate the interference verification result based on the interference collision points.

[0042] Preferably, the specific implementation process of converting the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence into spatially occupied voxel blocks at the corresponding control time is as follows: After the self-collision three-dimensional space mesh has been constructed based on the geometric bounding box size data, the three-dimensional voxel space, the voxel resolution, and the multiple three-dimensional occupied nodes, the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence are first read, and the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence are aligned with the control time according to the control cycle to form a control time alignment record. The control time alignment record is used to store the foot landing point, torso spatial position, torso posture, and arm end-effector spatial positions at the same control time. The foot landing point is derived from the foot landing point trajectory sequence, the torso spatial position and torso posture are derived from the torso displacement trajectory sequence, and the arm end-effector spatial positions are derived from the arm end-effector displacement trajectory sequence. This ensures that subsequent spatial interference verification does not check the isolated positions of the foot, torso, and arm end-effectors separately, but rather checks the spatial occupancy relationship between the foot landing point, torso spatial position, torso posture, and arm end-effector spatial positions at the same control time. After the control time alignment record is formed, the foot landing point trajectory sequence segment, torso displacement trajectory sequence segment, and arm end-effector displacement trajectory sequence segment corresponding to each control time are read according to the control time alignment record. The read foot landing point trajectory sequence segment, torso displacement trajectory sequence segment, and arm end-effector displacement trajectory sequence segment are then converted to the coordinate reference corresponding to the robot's body centroid in the three-dimensional voxel space to form trajectory space alignment data. The trajectory space alignment data is used to convert the continuous spatial positions of the foot landing point trajectory sequence segment, the torso displacement trajectory sequence segment, and the arm end displacement trajectory sequence segment at the same control moment into spatial position data that can be compared with the self-collision three-dimensional spatial mesh. Subsequently, based on the voxel resolution in the voxel configuration record, voxel boundary positioning processing is performed on the trajectory space alignment data to determine the row position, column position, and layer position of each of the foot landing point trajectory sequence segment, torso displacement trajectory sequence segment, and arm end displacement trajectory sequence segment in the three-dimensional voxel space at each control moment, and the row positions, column positions, and layer positions with adjacent occupancy relationships are combined into the space-occupying voxel block.The technical essence of the space-occupying voxel block is that it represents the discrete spatially occupied region after the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the end-arm displacement trajectory sequence are mapped onto the three-dimensional voxel space at the corresponding control moment. This space-occupying voxel block differs from a single coordinate point; it comprises multiple voxel units covered by the foot landing point trajectory sequence segment, the torso displacement trajectory sequence segment, or the end-arm displacement trajectory sequence segment at the corresponding control moment. The reason for generating this space-occupying voxel block is that when the Centaur robot performs end-effector tasks while moving, the actual object of spatial interference is usually not an abstract trajectory point, but rather the discrete spatially occupied region corresponding to the foot landing point trajectory sequence segment, the torso displacement trajectory sequence segment, and the end-arm displacement trajectory sequence segment, which have spatial dimensions. By converting the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence into the spatial occupancy voxel blocks at the corresponding control time, subsequent spatial interference verification can be performed directly around the discrete spatial occupancy region, rather than just around the abstract trajectory points. This allows the spatial occupancy voxel blocks to establish a comparable data relationship with the three-dimensional occupancy nodes in the self-collision three-dimensional spatial mesh.

[0043] Preferably, in the specific technical implementation of comparing the coordinates of the spatially occupied voxel block with the coordinates of the three-dimensional occupied nodes in the self-collision three-dimensional space mesh to obtain the coordinate overlap state between the spatially occupied voxel block and the three-dimensional occupied nodes, the coordinates of the spatially occupied voxel block are first read from the spatially occupied voxel block at the corresponding control time. The coordinates of the spatially occupied voxel block include the row position, column position, and layer position of the spatially occupied voxel block in the three-dimensional voxel space. At the same time, the coordinates of the three-dimensional occupied nodes are read from the mesh occupancy record of the self-collision three-dimensional space mesh. The coordinates of the three-dimensional occupied nodes also include the row position, column position, and layer position of the three-dimensional occupied nodes in the three-dimensional voxel space. Since both the spatially occupied voxel block and the three-dimensional occupied nodes use the row position, column position, and layer position in the three-dimensional voxel space to express spatial position, the coordinates of the spatially occupied voxel block and the coordinates of the three-dimensional occupied nodes have the same spatial measurement caliber, and the coordinates can be directly compared. When performing the coordinate point comparison, firstly, the trajectory source of the space-occupying voxel block is read according to the control time. The trajectory source includes a foot landing point trajectory sequence, a torso displacement trajectory sequence, or a double arm end displacement trajectory sequence. The trajectory source is used to identify that the space-occupying voxel block originates from the foot landing point trajectory sequence, the torso displacement trajectory sequence, or the double arm end displacement trajectory sequence. Then, the physical mechanical component identifier and joint connection association of the 3D occupying node are read according to the grid occupancy record. The physical mechanical component identifier is used to confirm the physical mechanical component corresponding to the 3D occupying node, and the joint connection association is used to confirm whether the 3D occupying node belongs to an adjacent physical mechanical component with a mechanical connection relationship with the current trajectory source. Subsequently, the coordinate points of the space-occupying voxel block are compared one by one with the coordinate points of the 3D occupying node in three directions: row position, column position, and layer position. When the row position, column position, and layer position all correspond to the same position, the corresponding coordinate point is recorded as a candidate coordinate point of overlap. When at least one direction of the row position, column position, and layer position does not correspond to the same position, the corresponding coordinate point is recorded as a candidate coordinate point of non-overlap. The three-way consistency comparison is used to simultaneously confirm whether the space-occupying voxel block and the three-dimensional occupying node occupy the same discrete spatial position in the row, column, and layer directions of the three-dimensional voxel space. The coincident candidate coordinate points are used to store coordinate points where the row, column, and layer positions are all consistent in the three-way consistency comparison, and the non-coincident candidate coordinate points are used to store coordinate points where the row, column, and layer positions are not all consistent in the three-way consistency comparison. The coincident candidate coordinate points and the non-coincident candidate coordinate points together participate in forming the coordinate coincidence state.The technical essence of the coordinate coincidence state is a determination of whether the space-occupying voxel block and the 3D-occupying node occupy the same discrete spatial position in the same 3D voxel space. It is not a simple conclusion of numerical equality, but rather a spatial occupancy comparison result formed by combining the corresponding control time, trajectory source, physical mechanical component identification, and joint connection associations. Through the coordinate coincidence state, it is possible to distinguish whether the space-occupying voxel block and the 3D-occupying node do not coincide, whether they have an adjacent contact relationship caused by mechanical connection, or whether there is a spatial overlap relationship between non-adjacent physical mechanical components. Therefore, the coordinate coincidence state can subsequently serve as a basis for determining whether spatial interference has occurred, so that spatial interference verification is not based solely on whether the coordinate points are consistent, but can be judged by combining the joint connection associations stored in the self-collision 3D spatial mesh.

[0044] Preferably, the specific implementation process of determining spatial interference and extracting the overlapping coordinate points as interference collision points when the coordinate coincidence state indicates the existence of overlapping coordinate points is as follows: When the coordinate coincidence state has been formed by comparing the coordinate points of the spatially occupied voxel block with the coordinate points of the three-dimensional occupied node, the overlapping candidate coordinate points in the coordinate coincidence state are first read, and the trajectory source, physical mechanical component identifier, and joint connection association corresponding to the overlapping candidate coordinate points are read simultaneously. If the spatially occupied voxel block corresponding to the overlapping candidate coordinate point originates from the trajectory sequence of the foot landing point, and the three-dimensional occupied node corresponding to the overlapping candidate coordinate point belongs to a physical mechanical component in the same foot kinematic chain that has a mechanical connection relationship with the foot, then the overlapping candidate coordinate point is marked as a connecting contact coordinate point according to the joint connection association. The connecting contact coordinate point is used to represent the spatial adjacency relationship caused by the mechanical connection relationship between adjacent physical mechanical components. The connecting contact coordinate point is not used as the data source for the interference collision point. If the space-occupying voxel block corresponding to the coincident candidate coordinate point originates from the torso displacement trajectory sequence or the end-arm displacement trajectory sequence, and the three-dimensional occupancy node corresponding to the coincident candidate coordinate point does not belong to an adjacent physical mechanical component with a mechanical connection relationship to the current trajectory source, then the coincident candidate coordinate point is marked as an interference candidate coordinate point according to the joint connection association. The technical essence of the interference candidate coordinate point is that the space-occupying voxel block and the three-dimensional occupancy node corresponding to the non-adjacent physical mechanical component occupy the same discrete spatial position at the same control time. The interference candidate coordinate point is used to express the spatial occupancy conflict between different moving parts. Subsequently, a continuous voxel adjacency check is performed on the interference candidate coordinate points. The continuous voxel adjacency check is used to determine whether multiple interference candidate coordinate points form a continuous adjacent occupancy region in the three-dimensional voxel space. If multiple interference candidate coordinate points have a continuous adjacent relationship in row position, column position, or layer position, then multiple interference candidate coordinate points are merged into the same interference region record. If a certain interference candidate coordinate point is not continuously adjacent to other interference candidate points, then the interference candidate coordinate point is written into the interference region record separately. The interference region record is used to store candidate coordinate points of spatial occupancy conflicts and provides a data source for extracting the interference collision points. When there is at least one candidate coordinate point in the interference region record, spatial interference is determined to have occurred, and the overlapping coordinate points that can represent the location of the spatial occupancy conflict are extracted from the interference region record as the interference collision points. The technical essence of the interference collision points is that, at the corresponding control moment, the spatial occupancy voxel block and the three-dimensional occupancy node in the self-collision three-dimensional spatial mesh have non-adjacent spatial overlaps, and the interference collision points retain both row position, column position, and layer position, as well as the corresponding trajectory source and physical mechanical component identifier.By using the aforementioned interference collision points, when adjusting the state parameters in the unified kinematic representation matrix based on the interference verification results, the spatial interference can be located to a specific control moment, a specific trajectory source, and a specific physical mechanical component. This avoids obtaining only a higher-level conclusion on whether interference occurs without being able to adjust the state parameters.

[0045] Preferably, the specific implementation process of generating the interference verification result based on the interference collision points is as follows: After the interference collision points have been extracted from the interference region record, an interference verification bearing record is first established, and the interference collision points, corresponding control time, trajectory source, physical mechanical component identifier, joint connection association, and coordinate coincidence state are written into the interference verification bearing record. The interference verification bearing record is used to incorporate the coordinate point comparison process between the spatially occupied voxel block and the three-dimensional occupied node, the spatial interference determination process, and the interference collision point extraction process into the same data source, so that the spatial occupancy relationship at the corresponding control time can be traced when the interference verification result is subsequently generated. When generating the interference verification result, the interference collision points at each corresponding control moment are first read according to the interference verification bearing record, and the interference collision points are classified into the foot trajectory interference category, trunk trajectory interference category, or arm end trajectory interference category according to the trajectory source. The foot trajectory interference category is used to store interference collision points from the foot landing point trajectory sequence, the trunk trajectory interference category is used to store interference collision points from the trunk displacement trajectory sequence, and the arm end trajectory interference category is used to store interference collision points from the arm end displacement trajectory sequence. Subsequently, the physical mechanical components associated with the interference collision points are read according to the physical mechanical component identifier, and it is determined whether the physical mechanical components associated with the interference collision points are non-adjacent physical mechanical components according to the joint connection association. If the physical mechanical components associated with the interference collision points are non-adjacent physical mechanical components, the interference collision points are written into the valid interference field in the interference verification result. The valid interference field is used to store the interference collision points participating in subsequent state parameter adjustments, so that the interference verification result can directly provide a spatial location source for the generation of subsequent interference avoidance correction items. If the interference collision point does not exist at a certain control moment, a non-overlapping state field is written into the interference verification result. The non-overlapping state field is used to express that no non-adjacent spatial overlap is formed between the space-occupying voxel block and the three-dimensional occupying node at the corresponding control moment. The technical essence of the interference verification result is the data result formed after verifying the spatial occupancy status of the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence in the self-collision three-dimensional space grid. The interference verification result includes not only the state of whether spatial interference has occurred, but also the interference collision point corresponding to the occurrence of spatial interference, the corresponding control moment, the trajectory source, and the physical mechanical component identification.Therefore, the interference verification result can directly provide the source of the spatial position and trajectory to be adjusted when adjusting the state parameters in the unified kinematic representation matrix according to the interference verification result, so that the state parameter adjustment can correspond to the spatial occupancy conflict between the foot landing point trajectory sequence, the torso displacement trajectory sequence and the arm end displacement trajectory sequence.

[0046] Optionally, adjusting the state parameters in the unified kinematic representation matrix based on the interference verification result to generate an interference-free cooperative motion trajectory includes: When the interference verification result indicates that spatial interference has occurred, the corresponding interference joint identifier is obtained based on the interference collision point; Calculate the escape vector of the interference collision point along the normal direction, and generate an interference avoidance correction term based on the escape vector; Based on the interference avoidance correction term, the state parameters corresponding to the interference joint identifier in the unified kinematic representation matrix are updated in reverse iteration until there is no spatial interference in the self-collision three-dimensional space grid, so as to obtain the updated state parameters. The motion paths of the feet, torso, and arms are reconstructed based on the updated state parameters, and the reconstructed motion paths are used as the interference-free cooperative motion trajectories.

[0047] Preferably, when the interference verification result indicates that spatial interference has occurred, the specific implementation process of obtaining the corresponding interference joint identifier based on the interference collision point is as follows: After the interference verification result has been formed by comparing the coordinate points between the spatially occupied voxel block and the three-dimensional occupied node, the effective interference field in the interference verification result is first read, and the interference collision point, corresponding control time, trajectory source, physical mechanical component identifier, and joint connection association are extracted from the effective interference field to establish an interference joint backtracking record. The interference joint backtracking record is used to include the interference collision point, the trajectory source that generated the interference collision point, the physical mechanical component identifier, and the joint connection association in the same backtracking caliber, so that the subsequently obtained interference joint identifier is not a separate joint position identifier detached from the interference collision point, but can correspond to the corresponding control time and physical mechanical component where the spatial interference occurred. The technical essence of the interference joint identifier is that it is a joint position identifier in the unified kinematic representation matrix that can generate a spatial response influence relationship on the spatial position of the interference collision point. The interference joint identifier is used to backtrack the interference collision point in the interference verification result to the state parameter that needs to be adjusted. Specifically, the process involves first searching for the joint node identifier corresponding to the physical mechanical component that caused the spatial interference in the physical joint connection information, based on the physical mechanical component identifier. Then, based on the joint connection association, the upper-level and lower-level joint node identifiers that have a mechanical connection relationship with the physical mechanical component are read to form candidate joint identifiers. After forming the candidate joint identifiers, the column indices arranged in the joint connection order in the unified kinematic representation matrix are read, and the candidate joint identifiers are matched with the column indices to filter out candidate joint identifiers that do not participate in the spatial response of the current trajectory source. For interference collision points originating from the trajectory sequence of the foot landing point, candidate joint identifiers that can affect the spatial response of the foot are preferentially retained; for interference collision points originating from the trajectory sequence of the trunk displacement, candidate joint identifiers that can affect the spatial response of the trunk are preferentially retained; for interference collision points originating from the trajectory sequence of the displacement of the two arms ends, candidate joint identifiers that can affect the spatial response of the two arms ends are preferentially retained. After the above-described correspondence lookup, the retained candidate joint identifiers are written into the interference joint field of the interference joint backtracking record, and the candidate joint identifiers in the interference joint field are used as the interference joint identifiers. The resulting interference joint identifiers inherit both the spatial source of the interference collision point and the column index source of the unified kinematic representation matrix. When subsequently performing reverse iterative updates on the state parameters corresponding to the interference joint identifiers in the unified kinematic representation matrix, the system can directly locate the state parameters that need adjustment, avoiding ineffective adjustments to state parameters unrelated to the spatial interference.

[0048] Preferably, in the specific technical implementation of calculating the escape vector of the interference collision point along the normal direction and generating an interference avoidance correction term based on the escape vector, the interference collision point is first read from the interference joint backtracking record, and simultaneously the interference region record where the interference collision point is located, the space-occupying voxel block corresponding to the interference collision point, and the three-dimensional occupying node coinciding with the interference collision point are read. The technical essence of the normal direction is the separation direction from the discrete spatial position where non-adjacent spatial overlap occurs to the adjacent unoccupied voxel unit. The normal direction is used to express the spatial adjustment direction in which the space-occupying voxel block should exit from the three-dimensional occupying node coinciding with the interference collision point. When determining the normal direction, firstly, using the row, column, and layer positions of the interference collision point as the center, read the voxel units adjacent to the interference collision point in the three-dimensional voxel space, and determine whether the adjacent voxel units are occupied by the three-dimensional occupancy nodes corresponding to non-adjacent physical mechanical parts based on the mesh occupancy record; then, select the direction from the interference collision point to the adjacent voxel unit not occupied by the three-dimensional occupancy nodes corresponding to non-adjacent physical mechanical parts as the normal direction. If the interference region record contains multiple consecutive adjacent interference candidate coordinate points, first determine the outer boundary of the interference region based on the distribution range of the multiple interference candidate coordinate points in the row, column, and layer positions, and then determine the normal direction from the outer boundary of the interference region to the adjacent unoccupied voxel unit, so that the normal direction can express the spatial exit direction of the entire interference region record, rather than only reflecting the local position of a single interference candidate coordinate point. After determining the normal direction, read the discrete spatial distance between the interference collision point and the adjacent unoccupied voxel unit according to the voxel resolution, and combine the normal direction and the discrete spatial distance into the escape vector. The technical essence of the escape vector is to use the direction and distance data required to move the interference collision point from a non-adjacent spatial overlapping position to an adjacent unoccupied voxel unit. The escape vector is used to convert discrete spatial conflicts in spatial interference verification into spatial correction requirements that can participate in state parameter adjustments. Subsequently, the spatial response influence relationship corresponding to the interference joint identifier in the unified kinematic representation matrix is ​​read, and the escape vector is converted into an interference avoidance correction term corresponding to the interference joint identifier according to the spatial response influence relationship. The technical essence of the interference avoidance correction term is to allocate the spatial exit requirement expressed by the escape vector to the correction data of the state parameters corresponding to the interference joint identifier in the unified kinematic representation matrix. The interference avoidance correction term retains both the spatial direction source of the escape vector and the state parameter source of the interference joint identifier, thus providing an adjustment basis for subsequent reverse iterative updates.

[0049] Preferably, the specific implementation process of performing reverse iterative updates on the state parameters corresponding to the interference joint identifier in the unified kinematic representation matrix according to the interference avoidance correction term until there is no spatial interference within the self-collision 3D spatial grid to obtain the updated state parameters is as follows: After the interference avoidance correction term has been formed based on the escape vector and the interference joint identifier, the column index corresponding to the interference joint identifier in the unified kinematic representation matrix is ​​read first, and the state parameter corresponding to the interference joint identifier is extracted from the column index. The technical essence of the state parameter is the adjustable data in the unified kinematic representation matrix used to express the motion state of the joint position at the current control moment. The state parameter can affect the response position of the foot, torso, or arm ends in space. When determining the state parameters corresponding to the interference joint identifier, the interference joint identifier is first mapped one-to-one with the column index in the unified kinematic representation matrix. Then, the elements at the row and column intersections of the column index and the corresponding foot spatial response, trunk spatial response, or arm end spatial response are read to determine the influence relationship of the interference joint identifier on the spatial response of the current trajectory source. Subsequently, the state parameters with spatial response influence relationships are written into the state parameter update record. The state parameter update record is used to store the interference joint identifier, the state parameters, the interference avoidance correction term, and the corresponding control time, so that subsequent reverse iteration updates can be performed around the spatial interference under the same control time. When performing reverse iteration updates, the state parameters in the state parameter update record are first reverse-corrected according to the interference avoidance correction term, so that the state parameters are adjusted in the direction of reducing the spatial overlap of the interference collision points. Then, the spatial occupancy voxel block under the corresponding control time is regenerated based on the adjusted state parameters, and the regenerated spatial occupancy voxel block is mapped again to the self-collision 3D spatial mesh for spatial interference verification to obtain the iterative interference verification result. The essence of the reverse iterative update technique is to propagate the spatial occupancy conflict corresponding to the interference collision point to the state parameters in the unified kinematic representation matrix through the interference avoidance correction term. After each state parameter adjustment, a new spatial occupancy comparison is performed to ensure the state parameter adjustment result is acceptable for further verification by the self-collision 3D spatial mesh. If the iterative interference verification result still indicates the existence of spatial interference, a new escape vector is calculated based on the new interference collision point, and a new interference avoidance correction term is generated based on the new escape vector. Then, the state parameters corresponding to the new interference joint identifier are updated again using the reverse iterative update technique. If the iterative interference verification result indicates that the spatial interference does not exist within the self-collision 3D spatial mesh, the reverse iterative update stops, and the current state parameters are written into the updated state parameter field of the state parameter update record to obtain the updated state parameters.The updated state parameters are not simply replacement values ​​for joint angles, but state parameters that have been reconstructed using space-occupying voxel blocks and re-verified using self-collision 3D spatial meshes. They are subsequently used to reconstruct the motion paths of the feet, torso, and arms.

[0050] Preferably, the specific implementation process of reconstructing the motion paths of the feet, torso, and arms based on the updated state parameters, and using the reconstructed motion paths as the interference-free cooperative motion trajectory, is as follows: After the updated state parameters have been written into the state parameter update record, the corresponding control time, the interference joint identifier, the updated state parameters, and the trajectory source are read according to the state parameter update record. The updated state parameters are then written back to the column index position corresponding to the interference joint identifier in the unified kinematic representation matrix to form an updated unified kinematic representation matrix. The updated unified kinematic representation matrix is ​​used to receive the updated state parameters and re-express the spatial response relationship between the quadrupedal state data, torso state data, and arm state data under the same matrix arrangement. Subsequently, based on the updated unified kinematic representation matrix, the spatial response positions of the feet, torso, and arms are recalculated according to the order of the control times to form the reconstructed motion path. The reconstructed motion path includes a foot reconstructed motion path corresponding to the foot landing point trajectory sequence, a trunk reconstructed motion path corresponding to the trunk displacement trajectory sequence, and a double arm reconstructed motion path corresponding to the double arm end displacement trajectory sequence. The foot reconstructed motion path is used to express the changes in the foot support position of the quadruped actuator at each control moment. The trunk reconstructed motion path is used to express the changes in the trunk spatial position and trunk posture of the trunk actuator at each control moment. The double arm reconstructed motion path is used to express the changes in the spatial position of the double arm end of the double arm actuator at each control moment. When reconstructing the foot reconstructed motion path, the foot support position at each control moment is regenerated based on the rows and columns corresponding to the quadrupedal state data in the updated unified kinematic representation matrix, and the regenerated foot support position continues to maintain the control moment correspondence with the trunk reconstructed motion path; when reconstructing the trunk reconstructed motion path, the trunk spatial position and trunk posture at each control moment are regenerated based on the rows and columns corresponding to the trunk state data in the updated unified kinematic representation matrix, and the regenerated trunk spatial position and trunk posture continue to serve as the spatial reference for the transformation of the arm end position; when reconstructing the arm reconstructed motion path, the arm end position at each control moment is regenerated based on the rows and columns corresponding to the arm state data in the updated unified kinematic representation matrix, and the regenerated arm end position continues to maintain the spatial correspondence with the foot reconstructed motion path and the trunk reconstructed motion path at the same control moment.After the reconstructed motion path is formed, the reconstructed motion paths of the feet, torso, and arms are then converted back into spatial voxel blocks at the corresponding control moments. These reconstructed spatial voxel blocks are then mapped to the self-collision 3D spatial mesh for spatial interference verification. When the re-obtained interference verification result indicates that there is no spatial interference within the self-collision 3D spatial mesh, the reconstructed motion path is used as the interference-free cooperative motion trajectory. The technical essence of the interference-free cooperative motion trajectory is the reconstructed cooperative motion path of the feet, torso, and arms based on the updated state parameters and verified by the self-collision 3D spatial mesh. This interference-free cooperative motion trajectory is subsequently used to separate foot control commands and arm control commands, enabling the Centaur robot to use the trajectory source that has already eliminated spatial interference when performing cooperative tasks.

[0051] Optionally, separating the foot control command and the two arm control commands based on the interference-free coordinated motion trajectory includes: The interference-free cooperative motion trajectory is discretized in time dimension according to the control cycle to obtain the synchronous pose data corresponding to each control cycle. Identify the actuator label category contained in the synchronized pose data, and separate and parse the synchronized pose data according to the foot drive category and the arm drive category to obtain the data belonging to the foot drive category and the data belonging to the arm drive category; The data belonging to the foot drive category is converted into a first torque pulse signal required by the underlying actuator motor, and the foot control command is generated based on the first torque pulse signal. The data belonging to the arm drive category is converted into a second torque pulse signal required by the underlying actuator motor, and the dual-arm control command is generated based on the second torque pulse signal.

[0052] Preferably, the specific implementation process of discretizing the non-interference cooperative motion trajectory according to the control cycle to obtain the synchronous pose data corresponding to each control cycle is as follows: After the non-interference cooperative motion trajectory has been reconstructed by the updated state parameters and verified by the self-collision three-dimensional spatial mesh, the control cycle configuration record related to the centaur robot's cooperative operation is first read. The control cycle configuration record is formed when the centaur robot completes the startup initialization calibration and establishes a correspondence with the instruction refresh rhythm of the quadruped actuator, the posture update rhythm of the torso actuator, and the end-effector operation update rhythm of the dual-arm actuator. The technical essence of the control cycle is to divide the continuous non-interference cooperative motion trajectory into time segments that can issue foot control commands and dual-arm control commands, so that foot movement and dual-arm end-effector operation are not taken separately in different time segments, but are read within the same control cycle, including the corresponding foot spatial position, torso spatial position, torso posture, and dual-arm end-effector spatial position. The control cycle configuration record includes at least the control cycle start time, control cycle duration, instruction refresh order, trajectory sampling timestamp, and the correspondence between the trajectory sampling timestamp and the actuator. When determining the control cycle, the instruction refresh cycle of the quadruped actuator, the posture update cycle of the torso actuator, and the end-effector update cycle of the dual-arm actuator are first read. Then, a common control time segment is selected that allows the quadruped actuator, torso actuator, and dual-arm actuator to complete pose reading under the same trajectory sampling timestamp. This common control time segment is written into the control cycle configuration record, so that subsequent foot control commands and dual-arm control commands come from the same trajectory source in the same time dimension. Subsequently, according to the trajectory sampling timestamp in the control cycle configuration record, the interference-free cooperative motion trajectory is discretized in the time dimension. The discretization in the time dimension is not a simple interception of trajectory endpoints, but rather, based on the control cycle, the corresponding trajectory time position is located in the interference-free cooperative motion trajectory. When there is a time interval between adjacent trajectory time positions, the spatial positions of the feet, torso, torso posture, and dual-arm end-effectors are read in time alignment to form synchronous pose data corresponding to each control cycle. The technical essence of the synchronized pose data is a combination of data that simultaneously records the spatial states of the feet, torso, and arms under the same control cycle. The synchronized pose data includes at least a control cycle marker, foot spatial position, torso spatial position, torso posture, arm end spatial position, and trajectory source marker derived from the interference-free cooperative motion trajectory.The synchronized pose data is subsequently used for actuator label category identification and separation parsing, so that foot control commands and dual-arm control commands are not generated based on different time calibers, but are generated based on synchronized pose data under the same control cycle. This ensures that foot movement and dual-arm end-effector operations maintain a time correspondence consistent with the interference-free coordinated motion trajectory during the execution phase.

[0053] Preferably, the specific implementation process of identifying the actuator label category contained in the synchronized pose data and separating and parsing the synchronized pose data according to the foot drive category and the arm drive category to obtain data belonging to the foot drive category and data belonging to the arm drive category is as follows: After the synchronized pose data corresponding to each control cycle is formed, the trajectory source mark in the synchronized pose data is first read, and the trajectory fragment from the foot landing point trajectory sequence, the torso displacement trajectory sequence, or the double arm end displacement trajectory sequence is traced according to the trajectory source mark. The technical essence of the actuator label category is the actuator affixing mark attached to the synchronized pose data, which is used to express whether different pose fields in the synchronized pose data should enter the control command generation path of the quadruped actuator or the control command generation path of the double arm actuator. The actuator label category is not an independently generated classification result. Instead, it is derived from the trajectory source markers of the reconstructed foot, trunk, and arm motion paths during the reconstruction of the interference-free cooperative motion trajectory. These markers are then discretized and sampled in the time dimension and written into the synchronous pose data, ensuring consistency between the actuator label category and the source relationship of the interference-free cooperative motion trajectory. When identifying the actuator label category, the control cycle marker, trajectory source marker, and pose field name are first read from each synchronous pose data entry. Then, the pose field corresponding to the foot's spatial position is identified as the foot drive category, the pose field corresponding to the spatial positions of the arm ends is identified as the arm drive category, and the pose field corresponding to the trunk's spatial position and posture is used as trunk cooperative reference data jointly referenced by the foot drive category and the arm drive category. The trunk collaborative reference data originates from the trunk spatial position and trunk posture in the same synchronized pose data. This trunk collaborative reference data is subsequently written into both the foot drive analysis record and the arm drive analysis record, ensuring that data belonging to the foot drive category and data belonging to the arm drive category retain the same trunk spatial position and trunk posture after separate analysis. The technical essence of the foot drive category is the data classification used to drive the quadrupedal actuator to perform foot movements. This foot drive category enables the foot spatial position to be converted into a first torque pulse signal that can be received by each underlying actuator motor in the quadrupedal actuator. The technical essence of the arm drive category is the data classification used to drive the dual-arm actuator to perform end-effector operations. This arm drive category enables the dual-arm end-effector spatial position to be converted into a second torque pulse signal that can be received by each underlying actuator motor in the dual-arm actuator.During the separation and parsing process, the synchronized pose data is first split into foot-driven parsing records and arm-driven parsing records according to the actuator label category. The trunk coordination reference data is then simultaneously written into both the foot-driven and arm-driven parsing records, ensuring that the foot movements of the quadrupedal actuator and the end-effector operations of the dual-arm actuator still utilize the same trunk spatial position and posture. The data belonging to the foot-driven category includes control cycle markers, foot spatial positions, and trunk coordination reference data, which is subsequently used to generate the first torque pulse signal. Similarly, the data belonging to the arm-driven category includes control cycle markers, dual-arm end-effector spatial positions, and trunk coordination reference data, which is subsequently used to generate the second torque pulse signal. Through this separation and parsing, different pose fields in the synchronized pose data can enter different control command generation paths according to the actuator affiliation markers, while preserving the source relationship of the same control cycle and the same trunk coordination reference data. This prevents the loss of the time and posture correspondence required for coordinated operation between foot-driven and dual-arm control commands due to separation and parsing.

[0054] Preferably, the specific implementation process of converting data belonging to the foot drive category into the first torque pulse signal required by the underlying actuator motor, and generating the foot control command based on the first torque pulse signal, is as follows: After the foot drive parsing record has been formed according to the actuator label category, the control cycle marker, foot spatial position, and trunk coordination reference data in the foot drive parsing record are read first. The foot spatial position is then read in correspondence with the quadrupedal independent motion parameters in the unified kinematic representation matrix corresponding to the quadrupedal state data to determine the foot joint state requirements of each foot kinematic chain under the current control cycle. The underlying actuator motor is not limited to a single physical motor, but is a general term for motor units in quadrupedal actuators and dual-arm actuators that can receive pulse control signals and output joint drive torque. When generating the first torque pulse signal, the underlying actuator motor specifically corresponds to the motor unit at each foot joint position in the quadrupedal actuator. When converting data belonging to the foot drive category into the first torque pulse signal, the relative pose data of the foot end relative to the torso spatial position is first calculated based on the foot end spatial position and the torso collaborative reference data. This relative pose data is used to convert the foot end spatial position in the synchronous pose data into a relative spatial state that the quadruped actuator can read along the torso spatial position. Then, based on the quadruped independent motion parameters, the movable direction, motion boundary, and linkage relationship of each foot kinematic chain in the current control cycle are read, and the foot joint displacement requirements of each foot joint position are determined based on the relative pose data. The foot joint displacement requirements are used to express which foot joint positions need to be jointly realized in the corresponding foot kinematic chain for the foot end spatial position. The foot joint displacement requirements continue to enter the torque conversion process to form the first torque distribution data corresponding to the quadruped actuator. The technical essence of the first torque allocation data is to convert the change in the spatial position of the foot into data results of the driving torque required for each foot joint position. The first torque allocation data retains the correspondence between the control cycle marker and the foot joint position, so that when the first torque pulse signal is generated subsequently, it can be issued according to the foot movement requirements under the current control cycle. Subsequently, according to the pulse reading aperture of each underlying actuator motor in the quadruped actuator, the first torque allocation data is converted into the first torque pulse signal with pulse amplitude, pulse duration, and pulse issuance sequence. The technical essence of the first torque pulse signal is that the underlying actuator motor in the quadruped actuator can read and convert it into a motor control signal that drives the foot joints. The first torque pulse signal is different from the upper-level foot spatial position; it has already converted the data belonging to the foot drive category into pulsed control data that can drive the foot joint positions to perform movements.When generating the foot control command, a foot control command carrying record is first established. The control cycle marker, the underlying actuator motor identifier corresponding to the quadruped actuator, the first torque pulse signal, and the foot movement execution sequence are written into this record. The foot movement execution sequence expresses the order in which each foot joint in the quadruped actuator receives the first torque pulse signal within the same control cycle. Subsequently, the foot control command is encapsulated based on the foot control command carrying record to form the foot control command. The foot control command is used to control the quadruped actuator to perform foot movements. The foot control command follows the control cycle marker in the interference-free cooperative motion trajectory, ensuring that the foot movements of the quadruped actuator maintain the same temporal origin as the subsequently generated dual-arm control command.

[0055] Preferably, the specific implementation process of converting data belonging to the arm drive category into the second torque pulse signal required by the underlying actuator motor, and generating the dual-arm control command based on the second torque pulse signal, is as follows: After the arm drive parsing record has been formed according to the actuator label category, the control cycle marker, the spatial position of the two arm ends, and the trunk coordination reference data in the arm drive parsing record are read first. The spatial position of the two arm ends is then read in correspondence with the dual-arm coordination motion parameters in the unified kinematic representation matrix corresponding to the dual-arm state data to determine the dual-arm joint state requirements of the left arm kinematic chain and the right arm kinematic chain under the current control cycle. Here, the underlying actuator motor is still a general term for motor units that can receive pulse control signals and output joint drive torque. However, when generating the second torque pulse signal, the underlying actuator motor specifically corresponds to the motor units at the relevant joint positions of the shoulder, elbow, and wrist in the dual-arm actuator, and is not confused with the underlying actuator motor in the quadruped actuator corresponding to the generation of the first torque pulse signal. When converting data belonging to the arm drive category into the second torque pulse signal, the relative pose data of the arm ends relative to the torso is first calculated based on the spatial positions of the arm ends and the torso collaborative reference data. This relative pose data is used to convert the spatial positions of the arm ends in the synchronous pose data into a relative spatial state that the arm actuators can read along the torso spatial position. Then, the movable directions, motion boundaries, and mutual avoidance relationships of the left and right arm kinematic chains are read based on the arm collaborative motion parameters. The mutual avoidance relationships are used to limit the spatial motion boundaries of the left and right arm kinematic chains in the same control cycle. Finally, the arm joint displacement requirements are determined based on the relative pose data. These joint displacement requirements express which arm joint positions are needed to achieve the spatial positions of the arm ends in the left and right arm kinematic chains. These joint displacement requirements then proceed to torque conversion processing to form the second torque distribution data corresponding to the arm actuators. The technical essence of the second torque allocation data is to convert the spatial position changes of the two arm ends into data results of the driving torque required for the joint positions of the two arms. The second torque allocation data retains the correspondence between the control cycle marker, the left arm kinematic chain, the right arm kinematic chain, and the spatial positions of the two arm ends, so that when the second torque pulse signal is subsequently generated, it can be issued according to the end-effector operation requirements under the current control cycle. Subsequently, based on the pulse reading aperture of each underlying actuator motor in the dual-arm actuator, the second torque allocation data is converted into the second torque pulse signal with pulse amplitude, pulse duration, and pulse issuance sequence.The technical essence of the second torque pulse signal is that the underlying actuator motor in the dual-arm actuator can read and convert it into a motor control signal that drives the dual-arm joints. This second torque pulse signal differs from the upper-level dual-arm end-effector spatial position; it has already converted data belonging to the arm drive category into pulsed control data that can drive the dual-arm joints to perform end-effector operations. When generating the dual-arm control command, a dual-arm control command carrying record is first established, and the control cycle marker, the underlying actuator motor identifier corresponding to the dual-arm actuator, the second torque pulse signal, and the end-effector operation execution order are written into the dual-arm control command carrying record. The end-effector operation execution order is used to express the order in which each dual-arm joint position in the dual-arm actuator receives the second torque pulse signal within the same control cycle. Subsequently, the dual-arm control command is encapsulated based on the dual-arm control command carrying record to form the dual-arm control command. The dual-arm control commands are used to control the dual-arm actuators to perform end-effector operations. The dual-arm control commands and the foot control commands jointly utilize the control cycle markers and torso coordination reference data in the synchronous pose data, enabling the Centaur robot to perform foot movements and end-effector operations according to the time correspondence and posture correspondence in the interference-free cooperative motion trajectory when performing collaborative operations.

[0056] Optionally, after separating the foot control commands and the dual-arm control commands based on the interference-free cooperative motion trajectory to control the centaur robot to perform cooperative work, the method further includes: Read the end contact force data collected by the six-dimensional force sensor installed at the end of the two arms during the collaborative operation; Extract the deformation components corresponding to each coordinate axis direction of the end contact force data; Determine whether the deformation component deviates from the pre-configured safe operating range corresponding to the end contact force data; When the deformation component deviates from the safe working range, the deviation value of the deformation component relative to the safe working range is determined, and the deviation value is fed back to the unified kinematic representation matrix to readjust the trajectory sequence of the foot landing point and the posture of the two arm ends, thereby compensating for the force deviation of the two arm ends.

[0057] Preferably, the specific implementation process of reading the end-effector contact force data collected by the six-dimensional force sensors installed at the ends of the two arms during the collaborative operation is as follows: After the centaur robot has executed foot movements according to the foot control commands and performed end-effector operations according to the arm control commands, the control execution record corresponding to the current collaborative operation process is read first. The control execution record includes a control cycle marker, the foot control commands, the arm control commands, the foot movement execution state, the end-effector operation execution state, the end-effector pose, and torso collaborative reference data. The end-effector pose includes the end-effector spatial position and the end-effector posture. The end-effector spatial position is used to express the spatial coordinates of the end-effectors in the current control cycle, and the end-effector posture is used to express the orientation state of the end-effectors in the current control cycle. The control execution record is used to correlate the control source obtained by separating the interference-free collaborative motion trajectory with the force source collected during the collaborative operation, so that the subsequently read end-effector contact force data can be correlated with the foot movements, end-effector operations, and end-effector poses in the same control cycle, rather than isolated sensor data detached from the trajectory execution process. The six-dimensional force sensor is installed between the end effector of the dual arms and the end effector component. The sensor's data acquisition range includes contact force data formed along the front-back, lateral, and height directions of the local coordinate reference of the dual arm end effector, as well as contact torque data formed around these directions. The end effector component is used to contact the work object's contact surface, which forms the external force source corresponding to the contact force and torque data. The technical essence of the end effector contact force data is the multi-directional force state data formed after the dual arm end effector contacts the work object's contact surface during end-effector operation. This data is not simply a sensor reading for display, but rather a source of data reflecting whether the dual arm end effector posture matches the work target posture, the torso displacement trajectory sequence, and the foot landing point trajectory sequence. When reading the end-effector contact force data, the contact force data and contact torque data corresponding to the current control cycle are first extracted from the six-dimensional force sensor data acquisition cache according to the control cycle marker. The six-dimensional force sensor data acquisition cache is used to save the contact force data and contact torque data continuously acquired by the six-dimensional force sensor according to the control cycle, so that the contact force data and contact torque data can maintain a correspondence with the control cycle marker in the control execution record. Subsequently, the contact force data, contact torque data, end-effector pose of both arms, torso collaborative reference data, and control cycle marker are written into the end-effector contact acquisition record.The end-effector contact acquisition record is used to store the acquisition source, time source, and pose source of the end-effector contact force data. When extracting the deformation components later, the end-effector contact force data can be converted to a spatial reading aperture consistent with the unified kinematic representation matrix based on the end-effector contact acquisition record. Since the contact state between the end-effectors and the work object is affected by changes in foot support and torso posture when the Centaur robot performs end-effector tasks while moving, reading the end-effector contact force data during the collaborative operation allows the force deviation formed during the execution phase to be reintroduced into the unified kinematic representation matrix used for trajectory planning. This provides a contact force source for subsequent readjustments of the foot landing point trajectory sequence and the end-effector pose.

[0058] Preferably, in the specific technical implementation of extracting the deformation components corresponding to the end-effector contact force data in each coordinate axis direction, the end-effector contact force data, the end-effector poses, the torso collaborative reference data, and the control cycle marker in the end-effector contact acquisition record are first read, and the end-effector force deformation mapping record formed during the initialization calibration is also read. The end-effector force deformation mapping record includes the end-effector local coordinate reference, the six-dimensional force sensor installation offset, the six-dimensional force sensor acquisition coordinate aperture, the end-effector stiffness mapping data, the contact force reading direction, the contact torque reading direction, and the deformation component output direction. The end-effector force deformation mapping record is used to convert the end-effector contact force data from the six-dimensional force sensor acquisition coordinate aperture into a deformation reading aperture that can participate in trajectory adjustment. The technical essence of the deformation components is the positional and attitude offset of the distal ends of the two arms along various coordinate axes under the combined influence of the contact force data and the contact torque data. Specifically, the deformation component corresponding to the contact force data expresses the linear offset of the distal ends of the two arms along the forward / backward, lateral, and height force directions, while the deformation component corresponding to the contact torque data expresses the attitude offset of the distal ends of the two arms around the forward / backward, lateral, and height force directions. When extracting the deformation components, firstly, based on the installation offset of the six-dimensional force sensor, the distal end contact force data is converted from the coordinate aperture of the six-dimensional force sensor to the local coordinate reference of the distal ends of the two arms. Then, based on the torso collaborative reference data, the contact force data and contact torque data under the local coordinate reference of the distal ends of the two arms are converted to the spatial response direction corresponding to the unified kinematic representation matrix, ensuring that the deformation components maintain consistency with the row and column correspondence between the dual arm state data and the torso state data. After completing the spatial reading aperture conversion, based on the end effector stiffness mapping data, the contact force data in each coordinate axis direction is subjected to force-displacement mapping processing to obtain the linear deformation components corresponding to each coordinate axis direction; and based on the end effector stiffness mapping data, the contact torque data in each coordinate axis direction is subjected to force-attitude mapping processing to obtain the attitude deformation components corresponding to each coordinate axis direction. The linear deformation components and the attitude deformation components together constitute the deformation components and are written into the deformation component reading record. The deformation component reading record continues to retain the control cycle marker, the end effector pose of both arms, and the torso collaborative reference data, so that when subsequently determining whether the deformation components deviate from the safe operating range, data comparison can be performed under the same control cycle, the same end effector pose of both arms, and the same torso collaborative reference data.Through the above processing, the end contact force data is no longer limited to the readings of the contact force data and the contact torque data, but is converted into the deformation components that can be fed back to the unified kinematic representation matrix and participate in the readjustment of the foot landing point trajectory sequence and the end pose of the arms.

[0059] Preferably, the specific implementation process of determining whether the deformation component deviates from the pre-configured safe operating range corresponding to the end-effector contact force data is as follows: After the deformation component has been written into the deformation component reading record, the safe operating range configuration record is read first. The safe operating range configuration record is formed when the Centaur robot completes the startup initialization calibration or executes the target pose analysis, and establishes a correspondence with the target pose, the end-effector pose, the end-effector contact force data, and the end-effector force deformation mapping record. The essence of the pre-configuration is that before entering the collaborative operation process, based on the end-effector operation type, the end-effector stiffness mapping data, the six-dimensional force sensor acquisition range, and the target contact direction corresponding to the target pose, the direction boundary and amplitude boundary that the deformation component is allowed to fall into are pre-determined, and the direction boundary and amplitude boundary are written into the safe operating range configuration record, so that the execution stage can directly read the safe operating range for judgment, without having to temporarily determine the judgment criteria after the end-effector contact force data has been generated. The technical essence of the operational safety zone is the allowable deformation range corresponding to the end-effector contact force data. This includes the allowable linear deformation range along each coordinate axis and the allowable attitude deformation range around each coordinate axis. The operational safety zone is not a single threshold, but rather multi-directional range data associated with the target pose, the end-effector poses, and the torso collaborative reference data. When configuring the operational safety zone, first, the target contact direction when the end-effectors contact the target object's contact surface through the end-effector is read based on the target pose. Then, the primary and secondary force directions are determined based on the target contact direction. Subsequently, based on the end-effector stiffness mapping data, a relatively wide allowable linear deformation range is configured for the primary force direction, and a relatively narrow allowable linear deformation range is configured for the secondary force direction. Attitude deformation allowable ranges are configured for each coordinate axis direction based on the end-effector pose. Finally, the allowable linear deformation range, the allowable attitude deformation range, the target contact direction, and the corresponding control cycle reading caliber are written into the operational safety zone configuration record. When determining whether the deformation component deviates from the safe operating range, first read the deformation component corresponding to the current control cycle based on the control cycle marker, then read the safe operating range corresponding to the end contact force data from the safe operating range configuration record; subsequently, compare the linear deformation component with the allowable linear deformation range in the same direction, and compare the attitude deformation component with the allowable attitude deformation range in the same direction.If the deformation component corresponding to a certain coordinate axis direction falls within the safe operating range, the state within the range is written into the deformation judgment record; if the deformation component corresponding to a certain coordinate axis direction exceeds the safe operating range, the state of deviation is written into the deformation judgment record. The deformation judgment record is used to store the comparison relationship between the deformation component and the safe operating range, and to provide a direct data source for subsequently determining the deviation value.

[0060] Preferably, the specific implementation process of determining the deviation value of the deformation component relative to the safe working range when the deformation component deviates from the safe working range, and feeding the deviation value back to the unified kinematic representation matrix, is as follows: When the deformation judgment record indicates that the deformation component corresponding to at least one coordinate axis direction deviates from the safe working range, the deviation state in the deformation judgment record is first read, and the control cycle marker, coordinate axis direction, deformation component, safe working range, end-effector pose, and torso collaborative reference data corresponding to the deviation state are read simultaneously to establish a force deviation feedback record. The force deviation feedback record is used to connect the deformation component deviation state formed during the execution phase with the unified kinematic representation matrix used during the planning phase, so that the deviation value subsequently fed back to the unified kinematic representation matrix can be traced back to the specific control cycle, the specific end-effector pose, and the specific coordinate axis direction. When determining the deviation value, first determine whether the deformation component is located outside the upper or lower boundary of the safe working range. When the deformation component exceeds the upper boundary of the safe working range, the boundary offset formed by the deformation component relative to the upper boundary of the safe working range is taken as the deviation value. When the deformation component is below the lower boundary of the safe working range, the boundary offset formed by the deformation component relative to the lower boundary of the safe working range is taken as the deviation value. When the deviation exists in multiple coordinate axis directions, the deviation value corresponding to each coordinate axis direction is determined separately, and the deviation values ​​corresponding to each coordinate axis direction are written into the same force deviation feedback record. The technical essence of the deviation value is the boundary offset of the deformation component relative to the safe working range, which is used to express the degree and direction of deviation of the contact force state at the end of the double arms relative to the allowable deformation range. When feeding the deviation value back to the unified kinematic representation matrix, firstly, based on the end-arm poses and the trunk coordinated reference data, the row and column corresponding to the arm state data are searched in the unified kinematic representation matrix, and the row and column corresponding to the trunk state data are read simultaneously; then, the deviation value is written into the force deviation feedback field in the unified kinematic representation matrix according to the coordinate axis direction, so that the force deviation feedback field corresponds to the arm state data and the trunk state data.The technical essence of the unified kinematic representation matrix is ​​a matrix structure that expresses the spatial response relationship between quadrupedal state data, trunk state data, and arm state data under the same coordinate reference and the same joint connection sequence. Feeding the deviation value back to the unified kinematic representation matrix does not change the source of the unified kinematic representation matrix, but rather adds force deviation information of the execution stage to the unified kinematic representation matrix, so that the unified kinematic representation matrix can simultaneously read the kinematic relationship and the end contact force relationship when the trajectory sequence of the foot landing point and the end pose of the arms are subsequently readjusted.

[0061] Preferably, the specific implementation process of readjusting the foot landing point trajectory sequence and the arm end pose to compensate for the force deviation of the arm end is as follows: After the deviation value has been fed back to the force deviation feedback field of the unified kinematic representation matrix, the arm end pose in the deviation state is first read according to the control cycle marker, and the arm joint state parameters that affect the arm end pose are determined according to the row and column of the unified kinematic representation matrix corresponding to the arm state data; at the same time, the trunk state parameters that affect the trunk spatial position and trunk posture are determined according to the row and column of the unified kinematic representation matrix corresponding to the trunk state data, and the amount of support position adjustment that the change in trunk spatial position needs to be undertaken by the foot landing point trajectory sequence is determined according to the row and column of the unified kinematic representation matrix corresponding to the quadruped state data. The support position adjustment amount is used to express the displacement direction and displacement amplitude of the off-road dynamic support point that needs to be readjusted in the foot landing point trajectory sequence within the ground support range. The support position adjustment amount is subsequently incorporated into the readjustment process of the foot landing point trajectory sequence, so that the foot support change can maintain a corresponding relationship with the force compensation process of the end-arm pose. When readjusting the end-arm pose, the end-arm pose is first corrected by reverse force posture according to the coordinate axis direction where the deviation value is located, so that the end-arm pose is adjusted in the direction of reducing the deviation of the deformation component. Then, the reachable range and mutual avoidance relationship of the left arm kinematic chain and the right arm kinematic chain are read according to the arm cooperative motion parameters, and the end-arm pose after reverse force posture correction is remapped to the arm joint state parameters corresponding to the left arm kinematic chain and the right arm kinematic chain to form the end-arm pose after force compensation. When readjusting the foot landing point trajectory sequence, first read the centroid offset constraint under the current control cycle based on the torso state parameters, then recalculate the polygonal support region based on the off-track dynamic support points in the foot landing point trajectory sequence, and project the torso spatial position change caused by the deviation value onto the polygonal support region; then adjust the off-track dynamic support points corresponding to the current control cycle within the polygonal support region according to the support position adjustment amount, so that the adjusted off-track dynamic support points can maintain the same torso collaborative reference data with the force-compensated end-arm poses.After completing the above adjustments, the adjusted foot landing point trajectory sequence, torso spatial position, torso posture, and force-compensated end-effector poses are rewritten into the synchronous pose data. The rewritten synchronous pose data is then mapped to the self-collision 3D spatial mesh for spatial interference verification. When the re-obtained interference verification result indicates that there is no spatial interference within the self-collision 3D spatial mesh, and the re-extracted deformation components fall within the operational safety zone, the adjusted foot landing point trajectory sequence and the force-compensated end-effector poses are used as the trajectory source for continuing collaborative work. Through this processing, the deviation value can originate from the end-effector contact force data, pass through the deformation components, the operational safety zone, and the unified kinematic representation matrix, and further propagate to the foot landing point trajectory sequence and the end-effector poses. This allows the force deviation at the end of the arms to be compensated through the joint adjustment of the foot support position and the spatial state of the end-effectors.

[0062] Optionally, the step of calculating the centaur robot's center of mass equilibrium position at each of the off-road dynamic support points based on the center of mass offset constraints corresponding to the torso state data in the unified kinematic representation matrix, so that the torso displacement trajectory sequence is formed by each of the center of mass equilibrium positions, includes: Based on the off-walking support points in the foot landing point trajectory sequence, a polygonal support area is established by the off-walking support points; Read the dual-arm posture corresponding to the dual-arm state data in the unified kinematic representation matrix, and determine the mechanical arm weight offset data caused by the Centaur robot in different operation stages based on the dual-arm posture; By combining the weight offset data of the robotic arm and the mass distribution data of the centaur robot, the overall centroid coordinates in the cooperative motion state are determined. The composite centroid coordinates are projected onto the polygonal support area, and the pitch angle in the torso state data is adjusted so that the projection point of the composite centroid coordinates is located in the center area of ​​the polygonal support area, so that the adjusted state is determined as the centroid equilibrium position.

[0063] Preferably, the specific implementation process of establishing a polygonal support area enclosed by the off-walk dynamic support points based on the off-walk dynamic support points in the foot landing point trajectory sequence is as follows: After the foot landing point trajectory sequence has been generated by the ground projection coordinates corresponding to the moving target pose and multiple sets of off-walk dynamic support points, a set of off-walk dynamic support points corresponding to the current control cycle in the foot landing point trajectory sequence is read according to the control cycle, and the foot marker, foot contact state, and torso spatial position in the torso displacement trajectory sequence corresponding to the set of off-walk dynamic support points corresponding to the current control cycle are read simultaneously. The technical essence of the off-walk dynamic support point is the foot spatial position in which the quadruped actuator can form support contact with the ground reference plane within the corresponding control cycle. The off-walk dynamic support point is not a simple path point, but foot support position data formed by constraining the foot reachable area boundary, the ground movement reference line, and the sequence of control cycles. After reading the detached dynamic support points, foot positions not in support contact are first filtered out based on the foot contact state. Then, the detached dynamic support points in support contact are projected onto the ground reference plane in the global coordinate system to form a ground support point record. The ground support point record continues to retain the control cycle, the foot identifier, and the torso spatial position, so that the subsequently established polygonal support region can correspond to the foot support relationship formed by the detached dynamic support points under the current control cycle. When establishing the polygonal support region, the outer boundaries of each detached dynamic support point are sorted based on the forward and backward movement coordinates and lateral movement coordinates of each detached dynamic support point on the ground reference plane in the ground support point record to form a detached dynamic support point outer boundary sorting result. Subsequently, adjacent detached dynamic support points are connected according to the detached dynamic support point outer boundary sorting result to form the polygonal support region enclosed by the detached dynamic support points in support contact. The technical essence of the polygonal support region is the two-dimensional ground area within which the quadrupedal actuator can transmit support to the ground during the current control cycle. The polygonal support region represents the support boundary within which the centaur robot's center of mass projection falls during the current control cycle. The polygonal support region maintains a direct source relationship with the foot landing point trajectory sequence because it is enclosed by the off-center dynamic support points in the foot landing point trajectory sequence. The polygonal support region subsequently participates in the projection judgment of the comprehensive center of mass coordinates and the adjustment of the pitch angle, so that the torso displacement trajectory sequence is no longer determined solely by the torso movement itself, but is constrained by the foot support relationship formed by the off-center dynamic support points.

[0064] Preferably, in the specific technical implementation of reading the bi-arm postures corresponding to the bi-arm state data in the unified kinematic representation matrix and determining the robotic arm weight offset data caused by the Centaur robot in different operation stages based on the bi-arm postures, the unified kinematic representation matrix corresponding to the current control cycle is first read, and the rows and columns corresponding to the bi-arm state data are located in the unified kinematic representation matrix. The rows in the unified kinematic representation matrix are used to express the foot spatial response, trunk spatial response, and bi-arm end-effector spatial response, and the columns in the unified kinematic representation matrix are used to express the joint state parameters arranged according to the physical joint connection information; the rows and columns corresponding to the bi-arm state data are used to record the mapping relationship between the shoulder joint state, elbow joint state, wrist joint state, and bi-arm end-effector spatial positions of the left and right arm kinematic chains in the current control cycle. When reading the dual-arm posture, the joint angle changes, end-effector orientation changes, and spatial positions of the dual-arm end-effectors are first extracted based on the rows and columns corresponding to the dual-arm state data. Then, the joint angle changes, end-effector orientation changes, and spatial positions of the dual-arm end-effectors are correlated with the torso spatial position to form a dual-arm posture reading record. This record saves the correspondence between the dual-arm posture and the torso spatial position, ensuring that the subsequently determined robotic arm counterweight offset data corresponds to the torso bearing state in the current control cycle. The technical essence of the dual-arm posture is the spatial deployment state of the dual-arm actuators relative to the torso actuators in the current control cycle. The dual-arm posture includes the left-arm kinematic chain posture, the right-arm kinematic chain posture, and the dual-arm end-effector postures. The dual-arm posture expresses the spatial source of the counterweight changes generated by the dual-arm actuators relative to the robot's center of mass. The technical essence of the different operation stages is that the operation state segments are divided according to the sequence of control cycles during the execution of the end-effector displacement trajectory sequence. These different operation stages include approaching the target pose, maintaining end-effector contact, performing end-effector displacement, and exiting the target pose, all determined by the end-effector displacement trajectory sequence. When determining the robot arm counterweight offset data, the operation stage to which the current control cycle belongs is first determined based on the control cycle marker. Then, the extension direction, extension distance, and end-effector posture changes of the left and right arm kinematic chains relative to the torso's spatial position are read from the dual-arm posture reading records. Combined with the robot arm component mass records formed during the assembly and calibration phase, the counterweight offset direction and magnitude of the dual-arm actuators relative to the robot's center of mass are calculated to form the robot arm counterweight offset data. The robot arm component mass records provide the mass source of each physical mechanical component in the left and right arm kinematic chains. The robot arm component mass records and the dual-arm posture reading records jointly contribute to the formation of the robot arm counterweight offset data.The technical essence of the robotic arm counterweight offset data is the mass distribution offset data caused by the posture of the two arms at different operation stages. The robotic arm counterweight offset data and the body mass distribution data are then used together to determine the comprehensive center of mass coordinates, so that the influence of the end-effector operation on the torso balance can be included in the calculation process of the center of mass balance position.

[0065] Preferably, the specific implementation process of determining the comprehensive center of mass coordinates in the cooperative motion state by combining the robotic arm counterweight offset data and the centaur robot's body mass distribution data is as follows: After the robotic arm counterweight offset data has been formed based on the dual-arm posture and the different operation stages, the body mass distribution record formed during the assembly calibration stage is first read. The body mass distribution record includes robot body mass data, quadruped actuator mass data, torso actuator mass data, dual-arm actuator mass data, the installation posture of each physical mechanical component relative to the robot body's center of mass, and the local center of mass position of each physical mechanical component. The centaur robot's body mass distribution data originates from the body mass distribution record. The body mass distribution data is used to express the mass positional relationship of the robot body, quadruped actuator, torso actuator, and dual-arm actuator in the assembly calibration state. The body mass distribution data is not simply a weight record, but rather a data source used to determine the basic center of mass coordinates when subsequently calculating the comprehensive center of mass coordinates. The robotic arm counterweight offset data is used to introduce the mass distribution offset of the dual-arm actuators due to the current operation stage and the current dual-arm posture, based on the basic center of mass coordinates. When determining the comprehensive center of mass coordinates, the torso spatial position and torso posture under the current control cycle are first read based on the torso displacement trajectory sequence, and the off-center support point under the current control cycle is read based on the foot landing point trajectory sequence. Then, the body mass distribution data is converted to the robot body center of mass coordinate reference corresponding to the current control cycle to form the basic center of mass coordinates. The basic center of mass coordinates are used to express the overall center of mass position without the superimposed robotic arm counterweight offset data. After forming the basic center of mass coordinates, the robotic arm counterweight offset data is superimposed onto the basic center of mass coordinates according to the spatial direction corresponding to the arm postures to form the comprehensive center of mass coordinates. The technical essence of the comprehensive center of mass coordinates is that, under coordinated motion, the robot's overall center of mass coordinates are formed by comprehensively considering the foot support position, torso spatial position, torso posture, arm postures, and robotic arm counterweight offset data. The coordinated motion state is used to express the state in which foot movement, torso displacement, and end-effector operations simultaneously participate in motion planning within the same control cycle. Therefore, the comprehensive center-of-gravity coordinates are not the center-of-gravity coordinates in a static assembly state, but rather dynamic center-of-gravity coordinates that maintain a control cycle correspondence with the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the end-effector displacement trajectory sequence. The comprehensive center-of-gravity coordinates are subsequently projected onto the polygonal support area, and their positional relationship with the center area of ​​the polygonal support area is determined to ascertain whether the pitch angle in the torso state data needs adjustment. Thus, a continuous data transfer relationship is formed between the basic center-of-gravity coordinates, the robotic arm counterweight offset data, and the comprehensive center-of-gravity coordinates.

[0066] Preferably, the specific implementation process of projecting the composite centroid coordinates onto the polygonal support region and adjusting the pitch angle in the torso state data so that the projection point of the composite centroid coordinates is located in the center region of the polygonal support region is as follows: After the composite centroid coordinates have been formed, the composite centroid coordinates are first projected onto the ground reference plane according to the height coordinate direction in the global coordinate system to form a composite centroid projection point; the composite centroid projection point is used to express the corresponding position of the composite centroid coordinates within the polygonal support region. Then, the polygonal support region is read, and it is determined whether the composite centroid projection point is located within the polygonal support region; if the composite centroid projection point is located outside the polygonal support region, the torso spatial position in the torso state data is first adjusted in the opposite direction according to the deviation direction of the composite centroid projection point relative to the boundary of the polygonal support region, and the composite centroid coordinates and the composite centroid projection point are recalculated under the adjusted torso spatial position; if the composite centroid projection point is located within the polygonal support region, the central region configuration record is read again. The central region configuration record is generated during the initial calibration or gait planning parameter configuration. This record includes the support boundary inward reduction diameter, the central region generation direction, and the central region judgment caliber. The technical essence of the central region is the internal support range formed within the polygonal support region according to the inward reduction diameter of the support boundaries. The central region represents the positional range of the comprehensive centroid projection point relative to each support boundary, allowing for balance adjustment. The support boundary inward reduction diameter originates from the central region configuration record and is read during the central region generation process, ensuring that the central region is not a temporarily given abstract region but is formed by the polygonal support region according to the generation caliber in the central region configuration record. When adjusting the pitch angle in the torso state data, first read the rows and columns corresponding to the torso state data in the unified kinematic representation matrix, and extract the torso spatial position, torso posture, and pitch angle under the current control cycle from the rows and columns corresponding to the torso state data. The technical essence of the pitch angle is the posture parameter of the torso actuator tilting back and forth around the lateral coordinate direction. The pitch angle can change the mass distribution of the torso actuator and the two arm actuators relative to the foot support range. Then, based on the offset direction of the comprehensive center of mass projection point relative to the central region, adjust the pitch angle so that the forward and backward tilt direction of the torso actuator is opposite to the direction in which the comprehensive center of mass projection point deviates from the central region. After each adjustment of the pitch angle, recalculate the comprehensive center of mass coordinates and the comprehensive center of mass projection point.If the recalculated composite centroid projection point is located within the central region, the current pitch angle and the current torso spatial position are written into the centroid balance adjustment record. If the recalculated composite centroid projection point is still not located within the central region, the pitch angle continues to be adjusted based on the central region determination caliber until the composite centroid projection point is located within the central region, or the torso attitude adjustment range under the current control cycle reaches the adjustment boundary defined by the centroid offset constraint. The centroid balance adjustment record continues to store the correspondence between the composite centroid coordinates, the composite centroid projection point, the central region, and the adjusted pitch angle, so that it can be directly read when determining the centroid balance position later.

[0067] Preferably, the specific implementation process of determining the adjusted state as the centroid balance position is as follows: After the comprehensive centroid projection point is located in the central region of the polygonal support area, or the pitch angle has completed the attitude adjustment allowed by the current control cycle according to the centroid offset constraint, the control cycle marker, the polygonal support area, the central region, the comprehensive centroid coordinates, the comprehensive centroid projection point, the adjusted pitch angle, and the adjusted torso spatial position in the centroid balance adjustment record are read first, and the above data are established in correspondence with the discrete dynamic support point corresponding to the current control cycle. The technical essence of the adjusted state is the torso state combination formed after the comprehensive centroid coordinate projection judgment, the central region position judgment, and the pitch angle adjustment within the current control cycle. The adjusted state includes the adjusted torso spatial position, the adjusted torso posture, and the corresponding comprehensive centroid projection point position relationship. When determining the centroid balance position, the positional relationship between the composite centroid projection point and the central region is first determined. When the composite centroid projection point is located within the central region, the adjusted torso spatial position is taken as the centroid balance position corresponding to the current control cycle, and the adjusted pitch angle is taken as the torso attitude parameter corresponding to the centroid balance position. When the composite centroid projection point is not located within the central region but has already met the adjustment boundary defined by the centroid offset constraint, the adjusted torso spatial position that allows the composite centroid projection point to approach the central region under the current control cycle is taken as the centroid balance position, and the positional offset relationship of the composite centroid projection point relative to the central region is retained in the centroid balance adjustment record, so that subsequent control cycles can continue to perform torso displacement connection based on the positional offset relationship of the composite centroid projection point relative to the central region. The technical essence of the centroid balance position is the torso spatial position jointly determined by the polygonal support area enclosed by the staggered support points, the robotic arm counterweight offset data caused by the dual-arm posture, the body mass distribution data, the comprehensive centroid coordinates, and the pitch angle. The centroid balance position is not an isolated coordinate point, but rather a torso displacement result reflecting the mutual constraints between foot support, torso posture, and dual-arm working counterweights under the current control cycle. After forming the centroid balance positions corresponding to each control cycle, the centroid balance positions are arranged sequentially according to the order of the control cycles. A continuity check is performed by combining the direction of change in pitch angle, the direction of change in torso spatial position, and the change relationship of the polygonal support area between adjacent control cycles, so that the torso displacement trajectory sequence is formed by the various centroid balance positions.The torso displacement trajectory sequence subsequently participates in the coordinate inverse transformation of the dual-arm end-effector displacement trajectory sequence and the spatial interference verification in the self-collision three-dimensional space grid, enabling the centaur robot to determine the source of torso displacement in a combined scenario of movement and end-effector operation by using the foot-supported state, torso posture state, and dual-arm operation state.

[0068] like Figure 3 As shown, this embodiment of the present application discloses a trajectory collaborative planning device for the end effector and foot of a centaur robot, comprising: The representation matrix construction module is configured to acquire the physical joint connection information of the centaur robot, and construct a unified kinematic representation matrix containing quadruped state data, trunk state data and arm state data based on the physical joint connection information. The initial trajectory generation module is configured to parse the operation target pose and the moving target pose contained in the global operation instructions, and perform pose constraint transformation on the operation target pose and the moving target pose based on the unified kinematic representation matrix to generate the initial foot landing point trajectory sequence, torso displacement trajectory sequence and arm end displacement trajectory sequence. The interference trajectory generation module is configured to establish a self-collision three-dimensional space mesh of the centaur robot, map the trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement and the trajectory sequence of the end displacement of the two arms to the self-collision three-dimensional space mesh for spatial interference verification, so as to obtain the interference verification result, and adjust the state parameters in the unified kinematic representation matrix according to the interference verification result, thereby generating an interference-free cooperative motion trajectory. The control command separation module is configured to separate foot control commands and dual-arm control commands based on the interference-free cooperative motion trajectory, so as to control the centaur robot to perform cooperative operations.

[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 performs 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 performs any of the methods described in this application.

[0071] like Figure 6As shown, this is an embodiment of a centaur robot according to this application, including a robot body, a quadruped actuator, a torso actuator, a dual-arm actuator, and a controller; The quadruped actuator, the torso actuator, and the dual-arm actuator are respectively connected to the robot body; The controller is communicatively connected to the quadruped actuator, the torso actuator, and the dual-arm actuator, respectively. The controller is configured to acquire the physical joint connection information of the centaur robot, and to construct a unified kinematic representation matrix containing quadrupedal state data, torso state data and arm state data based on the physical joint connection information. The controller is also configured to parse the target pose and the moving target pose contained in the global operation instructions, and perform pose constraint transformation on the target pose and the moving target pose based on the unified kinematic representation matrix to generate an initial foot landing point trajectory sequence, a torso displacement trajectory sequence and a double arm end displacement trajectory sequence. The controller is also configured to establish a self-collision three-dimensional space mesh of the centaur robot, map the trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement and the trajectory sequence of the end arms displacement to the self-collision three-dimensional space mesh for spatial interference verification, so as to obtain the interference verification result, and adjust the state parameters in the unified kinematic representation matrix according to the interference verification result, thereby generating an interference-free cooperative motion trajectory. The controller is also configured to separate foot control commands and dual-arm control commands based on the interference-free cooperative motion trajectory, and to control the quadruped actuator to perform foot movements based on the foot control commands, and to control the dual-arm actuator to perform end-effector operations based on the dual-arm control commands, so that the centaur robot can perform cooperative operations.

[0072] like Figure 7 As shown, this application embodiment provides a trajectory collaborative planning system for the end effector and foot of a centaur robot, which includes a global operation instruction generation device, a trajectory collaborative planning device, and a centaur robot. The global operation instruction generation device is communicatively connected to the trajectory collaborative planning device, and the trajectory collaborative planning device is communicatively connected to the centaur robot; The global operation instruction generation device is configured to generate global operation instructions that include the pose of the task target and the pose of the moving target, and send the global operation instructions to the trajectory collaborative planning device. The trajectory cooperative planning device is configured to acquire the physical joint connection information of the centaur robot, and to construct a unified kinematic representation matrix containing quadruped state data, trunk state data and arm state data based on the physical joint connection information. The trajectory collaborative planning device is also configured to parse the operation target pose and the moving target pose contained in the global operation instruction, and perform pose constraint transformation on the operation target pose and the moving target pose based on the unified kinematic representation matrix, so as to generate an initial foot landing point trajectory sequence, torso displacement trajectory sequence and arm end displacement trajectory sequence. The trajectory cooperative planning device is also configured to establish a self-collision three-dimensional space grid for the centaur robot, map the trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement, and the trajectory sequence of the end displacement of the two arms to the self-collision three-dimensional space grid for spatial interference verification, so as to obtain the interference verification result, and adjust the state parameters in the unified kinematic representation matrix according to the interference verification result, thereby generating an interference-free cooperative motion trajectory. The trajectory cooperative planning device is also configured to separate foot control commands and dual-arm control commands based on the interference-free cooperative motion trajectory, and send the foot control commands and dual-arm control commands to the centaur robot. The centaur robot is configured to perform foot movements according to the foot control commands and to perform end-effector tasks according to the arm control commands, in order to complete collaborative work.

[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 collaborative trajectory planning between the end effector and the feet of a centaur robot, characterized in that, include: Obtain the physical joint connection information of the centaur robot, and construct a unified kinematic representation matrix containing quadruped state data, torso state data and arm state data based on the physical joint connection information. The operation command contains the target pose and the moving target pose. Based on the unified kinematic representation matrix, the target pose and the moving target pose are transformed by pose constraint to generate the initial foot landing point trajectory sequence, torso displacement trajectory sequence and arm end displacement trajectory sequence. A self-collision three-dimensional space mesh is established for the centaur robot. The trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement, and the trajectory sequence of the end displacement of the two arms are mapped to the self-collision three-dimensional space mesh for spatial interference verification to obtain the interference verification result. The state parameters in the unified kinematic representation matrix are adjusted according to the interference verification result to generate an interference-free cooperative motion trajectory. Based on the interference-free cooperative motion trajectory, foot control commands and dual-arm control commands are separated to control the centaur robot to perform cooperative tasks.

2. The method for collaborative trajectory planning of the end effector and foot of a centaur robot according to claim 1, characterized in that, The process involves acquiring the physical joint connection information of the centaur robot and, based on this information, constructing a unified kinematic representation matrix that includes quadrupedal state data, torso state data, and arm state data. Read the degree-of-freedom parameters of each joint position of the centaur robot and the link length data of adjacent joints from the physical joint connection information; Based on the degree of freedom parameters and the link length data, determine the independent motion parameters of the four legs corresponding to the four-legged state data, the single-axis rotation parameters of the waist corresponding to the trunk state data, and the coordinated motion parameters of the two arms corresponding to the two-arm state data. According to the joint connection order defined by the physical joint connection information, the quadrupedal independent motion parameters, the waist uniaxial rotation parameters, and the biarm coordinated motion parameters are configured into a data structure with a Jacobian matrix format to generate the unified kinematic representation matrix.

3. The method for collaborative trajectory planning of the end effector and foot of a centaur robot according to claim 1, characterized in that, The parsing of the global operation instructions includes the target pose and the moving target pose, and the pose constraint transformation of the target pose and the moving target pose is performed on the unified kinematic representation matrix to generate initial foot landing point trajectory sequences, torso displacement trajectory sequences and arm end-effector displacement trajectory sequences, including: Extract the ground projection coordinates of the moving target pose in the global coordinate system, and plan multiple sets of off-road dynamic support points based on the ground projection coordinates, so as to generate the trajectory sequence of the foot landing point based on the multiple sets of off-road dynamic support points; Based on the centroid offset constraints corresponding to the torso state data in the unified kinematic representation matrix, the centaur robot's centroid balance position at each of the off-road state support points is calculated, so as to generate the torso displacement trajectory sequence based on each of the centroid balance positions. Extract the spatial coordinates of the target pose in the local space, and combine them with the torso displacement trajectory sequence to perform inverse coordinate transformation on the spatial coordinates to generate the end-arm displacement trajectory sequence.

4. The method for collaborative trajectory planning of the end effector and foot effect of a centaur robot according to claim 1, characterized in that, The process of establishing the self-collision 3D space mesh of the centaur robot includes: Obtain the geometric bounding box size data of each physical mechanical component of the centaur robot; A three-dimensional voxel space is established with the centaur robot's center of mass as the origin; In the three-dimensional voxel space, the geometric bounding box size data is divided into multiple three-dimensional occupancy nodes according to a pre-configured voxel resolution corresponding to the geometric bounding box size data, so as to construct the self-collision three-dimensional space mesh representing the robot body's occupied area using the multiple three-dimensional occupancy nodes.

5. The method for collaborative trajectory planning of the end effector and foot of a centaur robot according to claim 4, characterized in that, The process of mapping the foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence onto the self-collision three-dimensional spatial mesh for spatial interference verification to obtain the interference verification result includes: The foot landing point trajectory sequence, the torso displacement trajectory sequence, and the arm end displacement trajectory sequence are respectively converted into space occupancy voxel blocks at the corresponding control time. The coordinates of the space-occupying voxel block are compared with the coordinates of the three-dimensional occupying node in the self-collision three-dimensional space mesh to obtain the coordinate overlap state between the space-occupying voxel block and the three-dimensional occupying node. When the coordinate coincidence state indicates that there are overlapping coordinate points, spatial interference is determined to have occurred, and the overlapping coordinate points are extracted as interference collision points to generate the interference verification result based on the interference collision points.

6. A trajectory collaborative planning device for the end effector and feet of a centaur robot, characterized in that, include: The representation matrix construction module is configured to acquire the physical joint connection information of the Centaur robot, and construct a unified kinematic representation matrix containing quadruped state data, trunk state data and arm state data based on the physical joint connection information. The initial trajectory generation module is configured to parse the target pose and the moving target pose contained in the global operation instructions, and perform pose constraint transformation on the target pose and the moving target pose based on the unified kinematic representation matrix to generate the initial foot landing point trajectory sequence, torso displacement trajectory sequence and arm end displacement trajectory sequence. The interference trajectory generation module is configured to establish a self-collision three-dimensional space mesh of the centaur robot, map the trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement and the trajectory sequence of the end displacement of the two arms to the self-collision three-dimensional space mesh for spatial interference verification, so as to obtain the interference verification result, and adjust the state parameters in the unified kinematic representation matrix according to the interference verification result, thereby generating an interference-free cooperative motion trajectory. The control command separation module is configured to separate foot control commands and dual-arm control commands based on the interference-free cooperative motion trajectory, so as to control the centaur robot to perform cooperative operations.

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 performs the method according to 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 performs the method described in any one of claims 1-5.

9. A centaur robot, characterized in that, Includes the robot body, quadruped actuators, torso actuators, dual-arm actuators, and controllers; The quadruped actuator, the torso actuator, and the dual-arm actuator are respectively connected to the robot body; The controller is communicatively connected to the quadruped actuator, the torso actuator, and the dual-arm actuator, respectively. The controller is configured to acquire the physical joint connection information of the centaur robot, and to construct a unified kinematic representation matrix containing quadrupedal state data, torso state data and arm state data based on the physical joint connection information. The controller is also configured to parse the target pose and the moving target pose contained in the global operation instructions, and perform pose constraint transformation on the target pose and the moving target pose based on the unified kinematic representation matrix to generate an initial foot landing point trajectory sequence, a torso displacement trajectory sequence and a double arm end displacement trajectory sequence. The controller is also configured to establish a self-collision three-dimensional space mesh of the centaur robot, map the trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement and the trajectory sequence of the end arms displacement to the self-collision three-dimensional space mesh for spatial interference verification, so as to obtain the interference verification result, and adjust the state parameters in the unified kinematic representation matrix according to the interference verification result, thereby generating an interference-free cooperative motion trajectory. The controller is also configured to separate foot control commands and dual-arm control commands based on the interference-free cooperative motion trajectory, and to control the quadruped actuator to perform foot movements based on the foot control commands, and to control the dual-arm actuator to perform end-effector operations based on the dual-arm control commands, so that the centaur robot can perform cooperative operations.

10. A trajectory collaborative planning system for the end effector and feet of a centaur robot, characterized in that, This includes global operation command generation equipment, trajectory collaborative planning equipment, and centaur robots; The global operation instruction generation device is communicatively connected to the trajectory collaborative planning device, and the trajectory collaborative planning device is communicatively connected to the centaur robot; The global operation instruction generation device is configured to generate global operation instructions that include the pose of the task target and the pose of the moving target, and send the global operation instructions to the trajectory collaborative planning device. The trajectory cooperative planning device is configured to acquire the physical joint connection information of the centaur robot, and to construct a unified kinematic representation matrix containing quadruped state data, trunk state data and arm state data based on the physical joint connection information. The trajectory collaborative planning device is also configured to parse the operation target pose and the moving target pose contained in the global operation instruction, and perform pose constraint transformation on the operation target pose and the moving target pose based on the unified kinematic representation matrix, so as to generate an initial foot landing point trajectory sequence, torso displacement trajectory sequence and arm end displacement trajectory sequence. The trajectory cooperative planning device is also configured to establish a self-collision three-dimensional space grid for the centaur robot, map the trajectory sequence of the foot landing point, the trajectory sequence of the torso displacement, and the trajectory sequence of the end displacement of the two arms to the self-collision three-dimensional space grid for spatial interference verification, so as to obtain the interference verification result, and adjust the state parameters in the unified kinematic representation matrix according to the interference verification result, thereby generating an interference-free cooperative motion trajectory. The trajectory cooperative planning device is also configured to separate foot control commands and dual-arm control commands based on the interference-free cooperative motion trajectory, and send the foot control commands and dual-arm control commands to the centaur robot. The centaur robot is configured to perform foot movements according to the foot control commands and to perform end-effector tasks according to the arm control commands, in order to complete collaborative work.

Images