A precise docking control method for a delivery robot

Abstract

The application discloses a kind of distribution robot precision docking control method, comprising: obtaining the target pose of target logistics point and the current pose of the end of docking execution mechanism, with current pose as benchmark Left invariant error of target pose is constructed, and left invariant error is mapped to Lie algebra space by logarithmic mapping, obtain six-dimensional error vector;Gain tensor and differential gain tensor are constructed, and integral gain tensor is constructed according to inertia matrix, so that gain is adaptively adjusted with the change of end load and configuration;Comprehensive generation control force helix acting on the end of execution mechanism;According to control force helix drive docking execution mechanism, until six-dimensional error vector converges to within the preset threshold value.

Description

Technical Field

[0001] This invention relates to the field of control engineering technology, specifically to a precise docking control method for delivery robots. Background Technology

[0002] With the rapid development of logistics automation and intelligent warehousing technologies, delivery robots are playing an increasingly important role in last-mile delivery tasks in various indoor and outdoor scenarios. In these tasks, delivery robots are typically equipped with multi-degree-of-freedom robotic arms or other docking actuators, requiring their end effectors to precisely insert or mate with interfaces at target logistics points, such as shelf slots, exchange platform slots, or charging docks, at specified spatial positions and orientations. This precise docking control places extremely high demands on pose accuracy, motion compliance, and environmental adaptability, and is one of the core technological aspects of a delivery robot's autonomous operation capability.

[0003] Currently, a common control scheme for the last-mile delivery robot involves using vision or laser sensors to acquire the six-dimensional target pose of the logistics point in space. This pose is typically expressed as three-dimensional position coordinates and three attitude angles, such as roll, pitch, and yaw. The current last-mile pose is then subtracted from the target pose in corresponding dimensions to obtain position and attitude deviation vectors. Based on this, the controller configures independent proportional, integral, and derivative (PID) controllers for each dimension of the position and attitude deviations. The proportional, integral, and derivative coefficients of each controller are tuned to a set of fixed values ​​through trial and error or experience. From these, three-dimensional control forces and three-dimensional control torques are calculated and synthesized into a control force helical drive to propel the robot's last-mile motion until all deviations converge. Some improved schemes adjust the proportional coefficients in segments according to the magnitude of the deviation or introduce fuzzy rules to fine-tune the coefficients, but the basis for these adjustments always stems from the numerical value and trend of the deviation signal itself.

[0004] However, the above-mentioned technical solutions have limitations in the gain coefficient settings of the position and attitude control channels, which cannot reflect the dynamic changes in inertial characteristics caused by configuration changes and load changes during the movement of the robot end effector. Therefore, it is difficult to achieve high-precision, high-compliance, and high-reliability position and attitude alignment control at the same time. Summary of the Invention

[0005] The present invention aims to at least partially solve the technical problems in the above-mentioned technologies.

[0006] Therefore, this invention discloses a precise docking control method for delivery robots, comprising the following steps:

[0007] Obtain the target pose of the target logistics point in space, and obtain the current pose of the end effector of the docking mechanism;

[0008] On the special Euclidean group SE(3), the left invariant error of the target pose is constructed based on the current pose, and the left invariant error is mapped to the Lie algebra space through logarithmic mapping to obtain a six-dimensional error vector;

[0009] The six-dimensional error vector is used as the proportional term error;

[0010] Select a reference pose, transfer the six-dimensional error vector at different times to the reference coordinate system corresponding to the reference pose through the adjoint transformation and accumulate it, and then transfer the accumulated result back to the volume coordinate system corresponding to the current pose through the adjoint transformation to obtain the integral term error;

[0011] Based on the current volume velocity at the end of the actuator, combined with the left-invariant error, the Lie derivative is calculated using the right Jacobian inverse mapping on the special Euclidean group SE(3) to obtain the differential term error;

[0012] Based on the inertia matrix of the actuator end during motion and the change of the inertia matrix in the configuration space, a proportional gain tensor and a differential gain tensor are constructed, and an integral gain tensor is constructed based on the inertia matrix so that the gain is adaptively adjusted with the end load and configuration changes.

[0013] Based on the environmental perception information near the end of the actuator, an environmental deformation tensor reflecting the geometric constraints of the local free space is generated, and the differential gain tensor is corrected using the environmental deformation tensor.

[0014] The proportional term error is combined with the proportional gain tensor, the integral term error is combined with the integral gain tensor, and the differential term error is combined with the corrected differential gain tensor to generate a control force spiral acting on the end of the actuator.

[0015] The docking actuator is driven by the control force spiral until the six-dimensional error vector converges to a preset threshold.

[0016] In addition, the precise docking control method for delivery robots disclosed in this invention may also have the following additional technical features:

[0017] Furthermore, the left-invariant error is obtained by group operation of the inverse of the current pose and the target pose; the logarithmic mapping solves for the rotation vector on the rotation component, and obtains the generalized translation error by combining the left Jacobian inverse of the rotation vector on the translation component.

[0018] Furthermore, the acquisition of the integral term error specifically includes:

[0019] The initial end pose is recorded as the reference pose;

[0020] In each control cycle, the six-dimensional error vector in the current body coordinate system is mapped to the reference coordinate system using the adjoint matrix of the relative transformation between the current pose and the reference pose.

[0021] Integrate the mapped error vector over time within the reference coordinate system;

[0022] By using the adjoint matrix of the relative transformation from the reference pose to the current pose, the integration result is mapped back to the current body coordinate system, thus obtaining the integration term error.

[0023] Furthermore, the acquisition of the differential term error specifically includes:

[0024] The volume velocity at the end of the actuator is transformed by the SE(3) right Jacobian inverse mapping with the negative error vector as the variable. The result is the differential error that characterizes the rate of error change in the tangent space of the manifold.

[0025] Furthermore, the proportional gain tensor and the differential gain tensor are constructed as follows:

[0026] Based on the inertia matrix at the end of the actuator, it is scaled by the proportional coefficient and the damping coefficient respectively, and a curvature compensation term is constructed according to the Christofel symbol of the inertia matrix as the configuration changes. The curvature compensation term is superimposed on the proportional gain tensor to suppress overshoot in the direction of high inertia change; the integral gain tensor is obtained by scaling the inertia matrix by the integral coefficient.

[0027] Furthermore, the inertia matrix is ​​obtained in real time through robot joint configuration and rigid body dynamic parameters.

[0028] Furthermore, the environmental perception information is obtained from the point cloud surrounding the end point captured by the depth camera; the environmental deformation tensor is generated as follows:

[0029] Extract the free space boundary profile in the vertical plane along the docking direction;

[0030] Perform conformal mapping on the boundary profile to map the free space region to the canonical domain;

[0031] Based on the local Jacobian matrix of the conformal mapping at the end projection point, an anisotropic environmental deformation tensor is constructed, the components of which are amplified in the narrow direction of free space;

[0032] The environmental deformation tensor is added to the differential gain tensor with a preset weight to increase damping in the narrow direction.

[0033] Furthermore, the conformal mapping is a numerical approximation of the Schwarz-Christophe map.

[0034] Furthermore, the end effector is the end effector of a robotic arm; the control force helix is ​​a six-dimensional force helix containing three-dimensional force and three-dimensional torque; the control force helix is ​​mapped to joint torque through the Jacobian matrix of the robotic arm, and dynamic feedforward compensation is superimposed to drive the movement of each joint.

[0035] Furthermore, the left-invariant error, integral term error, and differential term error are all located in the Lie algebra space of the current end-body coordinate system, and are uniformly performed in this space when generating the control force spiral.

[0036] The precise docking control method for delivery robots disclosed in this invention has at least the following beneficial effects:

[0037] (1) By constructing the control gain as a gain tensor associated with the end inertia matrix, the proportional gain and differential gain can be adaptively adjusted in real time with the end configuration and load changes. When the end inertia increases, the control stiffness automatically increases, and when the inertia decreases, the control action automatically converges. Thus, when the end inertia characteristics undergo large-scale dynamic changes, the dynamic response quality remains consistent, effectively avoiding overshoot, oscillation or slow response that occurs in fixed gain schemes in this type of scenario.

[0038] (2) By introducing the environmental deformation tensor to anisotropically correct the differential gain tensor, the damping characteristics automatically increase in the narrow direction of the local free space and remain small in the open direction. Compliant contact can be achieved in geometrically confined space without manual segmentation or preset force control threshold. At the same time, it maintains rapid convergence in free space, thus improving the compliance and efficiency of the docking process from the mechanism level.

[0039] (3) By defining the pose error on a special Euclidean group and expressing it uniformly with Lie algebra vectors, the position error and attitude error are processed in a unified geometric framework, avoiding the coupling error and singularity problem introduced by establishing control channels for position and attitude separately, and further improving the accuracy of pose alignment and the global stability of control.

[0040] Additional features and advantages of this invention will be set forth in the description which follows, or may be learned by practicing the invention. Attached Figure Description

[0041] The technical solution and beneficial effects of the present invention will become apparent and readily understood from the following description in conjunction with the accompanying drawings, wherein:

[0042] Figure 1 This is a flowchart of the precise docking control method for the delivery robot of the present invention;

[0043] Figure 2 This is a flowchart of Embodiment 1 of the precise docking control method for delivery robots of the present invention;

[0044] Figure 3 This is a flowchart of Embodiment 2 of the precise docking control method for delivery robots of the present invention;

[0045] Figure 4 This is a flowchart of Embodiment 3 of the precise docking control method for delivery robots of the present invention. Detailed Implementation

[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0047] The precise docking control method for delivery robots disclosed in this invention will now be described with reference to the accompanying drawings.

[0048] like Figure 1 As shown, a method for precise docking control of a delivery robot includes the following steps:

[0049] Obtain the target pose of the target logistics point in space, and obtain the current pose of the docking execution mechanism's end effector;

[0050] On the special Euclidean group SE(3), the left invariant error of the target pose is constructed based on the current pose, and the left invariant error is mapped to the Lie algebra space through logarithmic mapping to obtain a six-dimensional error vector.

[0051] The six-dimensional error vector is used as the proportional term error;

[0052] Select a reference pose, transfer the six-dimensional error vector at different times to the reference coordinate system corresponding to the reference pose through the adjoint transformation and accumulate it, and then transfer the accumulated result back to the volume coordinate system corresponding to the current pose through the adjoint transformation to obtain the integral term error;

[0053] Based on the current volume velocity at the end of the actuator, combined with the left-invariant error, the Lie derivative is calculated using the right Jacobian inverse mapping on the SE(3) group to obtain the differential term error;

[0054] Based on the inertia matrix of the actuator end effector during motion and the change of the inertia matrix in the configuration space, a proportional gain tensor and a differential gain tensor are constructed, and an integral gain tensor is constructed based on the inertia matrix so that the gain is adaptively adjusted with the end effector load and configuration changes.

[0055] Based on environmental perception information near the end of the actuator, an environmental deformation tensor reflecting the geometric constraints of local free space is generated, and the differential gain tensor is corrected using the environmental deformation tensor.

[0056] The proportional term error is combined with the proportional gain tensor action, the integral term error is combined with the integral gain tensor action, and the differential term error is combined with the corrected differential gain tensor action to generate a control force spiral that acts on the end of the actuator.

[0057] The control force is driven by a spiral to dock the actuator until the six-dimensional error vector converges to a preset threshold.

[0058] The left-invariant error is obtained by grouping the inverse of the current pose with the target pose; the logarithmic map solves for the rotation vector on the rotation component, and the generalized translation error is obtained by combining the left Jacobian inverse of the rotation vector on the translation component.

[0059] The acquisition of the integral term error specifically includes:

[0060] Record the initial end pose as the reference pose;

[0061] In each control cycle, the six-dimensional error vector in the current body coordinate system is mapped to the reference coordinate system using the adjoint matrix of the relative transformation between the current pose and the reference pose.

[0062] Integrate the mapped error vector over time within the reference coordinate system;

[0063] By using the adjoint matrix of the relative transformation from the reference pose to the current pose, the integration result is mapped back to the current body coordinate system, thus obtaining the integration term error.

[0064] The acquisition of differential term error specifically includes:

[0065] The volume velocity at the end of the actuator is transformed by the SE(3) right Jacobian inverse mapping with the negative error vector as the variable. The result is the differential error that characterizes the rate of error change in the tangent space of the manifold.

[0066] The proportional gain tensor and the differential gain tensor are constructed as follows:

[0067] Based on the inertia matrix at the end of the actuator, it is scaled by the proportional coefficient and the damping coefficient respectively. A curvature compensation term is constructed according to the Christofel symbol of the inertia matrix as the configuration changes. The curvature compensation term is superimposed on the proportional gain tensor to suppress overshoot in the direction of high inertia change. The integral gain tensor is obtained by scaling the inertia matrix by the integral coefficient.

[0068] The inertia matrix is ​​obtained in real time through the robot's joint configuration and rigid body dynamic parameters.

[0069] Environmental perception information is obtained from the point cloud surrounding the end point captured by the depth camera; the environmental deformation tensor is generated as follows:

[0070] Extract the free space boundary profile in the vertical plane along the docking direction;

[0071] Perform conformal mapping on the boundary profile to map the free space region to the canonical domain;

[0072] Based on the local Jacobian matrix of the conformal mapping at the end projection point, an anisotropic environmental deformation tensor is constructed, and the components of this tensor in the narrow direction of free space are amplified.

[0073] The environmental deformation tensor is added to the differential gain tensor with preset weights to increase damping in the narrow direction.

[0074] Conformal mapping is a numerical approximation of the Schwarz-Christophe map.

[0075] The actuator end effector is the end effector of the robotic arm; the control force helix is ​​a six-dimensional force helix containing three-dimensional force and three-dimensional torque; the control force helix is ​​mapped to joint torque through the Jacobian matrix of the robotic arm, and dynamic feedforward compensation is superimposed to drive the movement of each joint.

[0076] The left-invariant error, integral term error, and differential term error all reside in the Lie algebra space of the current end-body coordinate system, and are uniformly performed in this space when generating the control force screw.

[0077] Example 1

[0078] This embodiment is designed for an indoor warehousing scenario. The delivery robot is equipped with a six-degree-of-freedom robotic arm as the docking execution mechanism. The robotic arm is equipped with a depth camera and grippers at its end. The target logistics point is a standardized docking slot on the shelf.

[0079] like Figure 2 As shown, in this embodiment, the depth camera collects point cloud data in front of the end effector in real time, and obtains the six-dimensional target pose of the target logistics point through point cloud segmentation and target recognition algorithms; simultaneously, the current pose of the end effector is calculated in real time through the forward kinematics of the robotic arm. The inertia matrix of the end effector during its movement is calculated in real time by the robot controller using a Lagrange dynamics model based on the current joint angles, the mass and inertia parameters of each link.

[0080] The technical focus of this embodiment is as follows: At the final stage of docking, when the end-effector approaches the grid but has not yet entered, the depth camera's field of view is severely obstructed by the surrounding shelf partitions, resulting in a narrow free space vertically and a relatively wide free space horizontally. The generation process of the environmental deformation tensor at this point is as follows: The free space boundary contour in the vertical plane along the docking direction at the current height of the end-effector is extracted. This contour is presented as a flattened polygon compressed vertically. A Schwarz-Christophe map is performed on this polygon to approximate it to the interior of a unit circle. The local Jacobian matrix of this mapping at the projection point of the end-effector is calculated. Due to the high compression in the vertical direction, the eigenvalues ​​of this Jacobian matrix in the vertical direction are significantly greater than those in the horizontal direction. Based on this, an environmental deformation tensor is constructed, with its vertical component amplified. After adding this environmental deformation tensor to the differential gain tensor with preset weights, the damping of the end-effector in the vertical direction is significantly increased, making the gripper highly compliant when approaching the upper and lower edges of the grid, avoiding rigid collisions with the shelf partitions. Meanwhile, the damping in the horizontal direction is relatively small, maintaining the ability to quickly track the center position of the grid. After the end smoothly slides into the slot, the error converges to within the threshold, and the gripper completes the handover of goods.

[0081] For the remaining technical details of this embodiment, please refer to the above technical details, which will not be repeated in this embodiment.

[0082] Example 2

[0083] This embodiment targets an outdoor last-mile delivery scenario. The delivery robot is equipped with a wheeled mobile chassis and a five-degree-of-freedom robotic arm, which grips the goods at the end. The target logistics point is a parcel exchange platform slot located on a community wall. In this embodiment, the target pose is obtained by fusing a LiDAR and a monocular vision system mounted on the end of the robotic arm. First, the initial target pose is estimated by visually recognizing the QR code tag on the slot, and then the LiDAR refines the target pose over a short distance.

[0084] like Figure 3As shown, in this embodiment, during the extension of the robotic arm towards the slot, the robotic arm configuration changes from a retracted state to a large-arm-length state, resulting in a several-fold change in the equivalent inertia matrix at the end effector. If a fixed gain is used, the low gain in the retracted state leads to a sluggish initial response, while the same gain in the extended state results in significant overshoot due to the substantial increase in inertia. In this embodiment, the construction of the proportional gain tensor and differential gain tensor adapts in real-time to the configuration change: in the retracted state, the components of the inertia matrix are relatively small, and the proportional gain tensor and differential gain tensor maintain correspondingly low levels, resulting in smooth motion; as the robotic arm extends segment by segment, the diagonal components of the inertia matrix gradually increase, and simultaneously, the Christoffel sign term caused by the configuration change is calculated in real-time and superimposed on the proportional gain tensor, forming curvature compensation. This ensures that the proportional gain is synchronously enhanced in the direction of increasing inertia, and the differential damping is synchronously matched, suppressing overshoot in the direction of high inertia change. The integral gain tensor is also obtained by scaling the inertia matrix, ensuring that the integral action has consistent error accumulation characteristics under different configurations. Thus, throughout the entire process from the retracted state to the fully extended docking position, the dynamic response of the end remains consistent, the positional error converges smoothly along the geodesic line, and finally the device is accurately inserted into the slot.

[0085] For the remaining technical details of this embodiment, please refer to the above technical details, which will not be repeated in this embodiment.

[0086] Example 3

[0087] This embodiment is designed for a cold chain warehousing scenario. The delivery robot is equipped with a four-degree-of-freedom Cartesian coordinate docking actuator. The end effector of the actuator is a retractable cargo pallet, and the target logistics point is a positioning pin hole assembly on a cold storage shelf. In this embodiment, the docking actuator has no rotational degree of freedom, and the end effector's posture remains constant during movement, requiring only the control of three translational degrees of freedom.

[0088] like Figure 4As shown, in this embodiment, although the degree of freedom is reduced, the weight of the goods carried by the end during docking is different in each task, and there is an assembly gap between the positioning pin and the pin hole of the shelf in the cold storage. When the existing technology uses fixed gain control, lightly loaded goods are easily jammed when they contact the guide slope of the pin hole due to excessive stiffness, while heavy-loaded goods cannot be pushed in smoothly due to insufficient stiffness. In this embodiment, the control method is carried out on the translation subgroup of the SE(3) group: the left invariant error degenerates into the position error vector, and the integral term error and the differential term error only act on the translation component. The proportional gain tensor is determined by the equivalent translational inertial mass of the current end. The heavier the load, the larger the diagonal component of the gain tensor, providing a stronger driving force to overcome friction and guide resistance; at the same time, the depth camera acquires the point cloud around the pin hole and extracts the free space boundary contour of the pin hole entrance area. Since the pin hole entrance has a conical guide chamfer, the free space is radially contracted. The environmental deformation tensor generated by conformal mapping makes the radial damping gain gradually increase as the end approaches the center of the pin hole, and the end automatically and smoothly slides in when it contacts the guide slope. When heavy-load and light-load cargo alternately perform docking tasks, the proportional gain and differential damping automatically match according to their respective loads, eliminating the need for manual readjustment. After the error converges, the pallet descends to place the cargo onto the positioning pin, completing the handover.

[0089] For the remaining technical details of this embodiment, please refer to the above technical details, which will not be repeated in this embodiment.

[0090] In summary, the precise docking control method for delivery robots disclosed in this invention has at least the following beneficial effects:

[0091] (1) By constructing the control gain as a gain tensor associated with the end inertia matrix, the proportional gain and differential gain can be adaptively adjusted in real time with the end configuration and load changes. When the end inertia increases, the control stiffness automatically increases, and when the inertia decreases, the control action automatically converges. Thus, when the end inertia characteristics undergo large-scale dynamic changes, the dynamic response quality remains consistent, effectively avoiding overshoot, oscillation or slow response that occurs in fixed gain schemes in this type of scenario.

[0092] (2) By introducing the environmental deformation tensor to anisotropically correct the differential gain tensor, the damping characteristics automatically increase in the narrow direction of the local free space and remain small in the open direction. Compliant contact can be achieved in geometrically confined space without manual segmentation or preset force control threshold. At the same time, it maintains rapid convergence in free space, thus improving the compliance and efficiency of the docking process from the mechanism level.

[0093] (3) By defining the pose error on a special Euclidean group and expressing it uniformly with Lie algebra vectors, the position error and attitude error are processed in a unified geometric framework, avoiding the coupling error and singularity problem introduced by establishing control channels for position and attitude separately, and further improving the accuracy of pose alignment and the global stability of control.

[0094] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for precise docking control of a delivery robot, characterized in that, Includes the following steps: Obtain the target pose of the target logistics point in space, and obtain the current pose of the end effector of the docking mechanism; On the special Euclidean group SE(3), the left invariant error of the target pose is constructed based on the current pose, and the left invariant error is mapped to the Lie algebra space through logarithmic mapping to obtain a six-dimensional error vector; The six-dimensional error vector is used as the proportional term error; Select a reference pose, transfer the six-dimensional error vector at different times to the reference coordinate system corresponding to the reference pose through the adjoint transformation and accumulate it, and then transfer the accumulated result back to the volume coordinate system corresponding to the current pose through the adjoint transformation to obtain the integral term error; Based on the current volume velocity at the end of the actuator, combined with the left-invariant error, the Lie derivative is calculated using the right Jacobian inverse mapping on the special Euclidean group SE(3) to obtain the differential term error; Based on the inertia matrix of the actuator end during motion and the change of the inertia matrix in the configuration space, a proportional gain tensor and a differential gain tensor are constructed, and an integral gain tensor is constructed based on the inertia matrix so that the gain is adaptively adjusted with the end load and configuration changes. Based on the environmental perception information near the end of the actuator, an environmental deformation tensor reflecting the geometric constraints of the local free space is generated, and the differential gain tensor is corrected using the environmental deformation tensor. The proportional term error is combined with the proportional gain tensor, the integral term error is combined with the integral gain tensor, and the differential term error is combined with the corrected differential gain tensor to generate a control force spiral acting on the end of the actuator. The docking actuator is driven by the control force spiral until the six-dimensional error vector converges to a preset threshold.

2. The precise docking control method for delivery robots according to claim 1, characterized in that, The left-invariant error is obtained by group operation of the inverse of the current pose and the target pose; the logarithmic mapping solves for the rotation vector on the rotation component, and obtains the generalized translation error by combining the left Jacobian inverse of the rotation vector on the translation component.

3. The precise docking control method for delivery robots according to claim 1, characterized in that, The acquisition of the integral term error specifically includes: The initial end pose is recorded as the reference pose; In each control cycle, the six-dimensional error vector in the current body coordinate system is mapped to the reference coordinate system using the adjoint matrix of the relative transformation between the current pose and the reference pose. Integrate the mapped error vector over time within the reference coordinate system; By using the adjoint matrix of the relative transformation from the reference pose to the current pose, the integration result is mapped back to the current body coordinate system, thus obtaining the integration term error.

4. The precise docking control method for delivery robots according to claim 1, characterized in that, The acquisition of the differential term error specifically includes: The volume velocity at the end of the actuator is transformed by the SE(3) right Jacobian inverse mapping with the negative error vector as the variable. The result is the differential error that characterizes the rate of error change in the tangent space of the manifold.

5. The precise docking control method for delivery robots according to claim 1, characterized in that, The proportional gain tensor and the differential gain tensor are constructed as follows: Based on the inertia matrix at the end of the actuator, it is scaled by the scaling factor and the damping factor respectively, and a curvature compensation term is constructed according to the Christofel symbol of the inertia matrix as the configuration changes. The curvature compensation term is superimposed on the scaling gain tensor to suppress overshoot in the direction of high inertia change. The integral gain tensor is obtained by scaling the inertia matrix by the integral coefficient.

6. The precise docking control method for delivery robots according to claim 5, characterized in that, The inertia matrix is ​​obtained in real time through robot joint configuration and rigid body dynamic parameters.

7. The precise docking control method for delivery robots according to claim 1, characterized in that, The environmental perception information is obtained from the point cloud surrounding the end point captured by the depth camera; the environmental deformation tensor is generated as follows: Extract the free space boundary profile in the vertical plane along the docking direction; Perform conformal mapping on the boundary profile to map the free space region to the canonical domain; Based on the local Jacobian matrix of the conformal mapping at the end projection point, an anisotropic environmental deformation tensor is constructed, the components of which are amplified in the narrow direction of free space; The environmental deformation tensor is added to the differential gain tensor with a preset weight to increase damping in the narrow direction.

8. The precise docking control method for delivery robots according to claim 7, characterized in that, The conformal mapping is a numerical approximation of the Schwarz-Christophe map.

9. The precise docking control method for delivery robots according to claim 1, characterized in that, The actuator end is the end of the robotic arm; the control force helix is ​​a six-dimensional force helix containing three-dimensional force and three-dimensional torque; the control force helix is ​​mapped to joint torque through the Jacobian matrix of the robotic arm, and dynamic feedforward compensation is superimposed to drive the movement of each joint.

10. The precise docking control method for a delivery robot according to any one of claims 1 to 9, characterized in that, The left-invariant error, integral term error, and differential term error are all located in the Lie algebra space of the current end-body coordinate system, and are uniformly performed in this space when generating the control force spiral.

Images