Method and apparatus for controlling mobile robot for server transportation

Abstract

The application provides a mobile robot control method and device for server carrying, the method comprising: realizing accurate positioning and mapping through a multi-sensor fusion SLAM navigation technology, accurately identifying a target cabinet through a 3D vision system and obtaining installation position information, calculating deviation parameters of a current pose of the server and a target installation position through a feature matching algorithm, and coordinating pose adjustment of a cargo platform and a lifter body based on deviation compensation. Through an integrated multi-dimensional pose compensation control mechanism, the application realizes full-process automatic control of the mobile robot from navigation positioning to accurate alignment, and effectively improves the positioning accuracy and success rate of automatic server installation.

Description

Technical Field

[0001] This invention relates to the field of equipment control technology, and in particular to a mobile robot control method and apparatus for server handling. Background Technology

[0002] In the automated server installation system of a supercomputing center, mobile robots need to accurately navigate to the target rack and complete the precise positioning and installation of the servers. Existing mobile robot control methods typically employ a pre-planned path navigation approach, where the robot moves to the target location along a pre-planned fixed route and then directly performs the server installation work.

[0003] However, in implementing this traditional control method, the robot cannot adaptively adjust according to the actual environment and the specific location of the target rack, resulting in frequent large deviations between the server pose and the rack installation position, severely affecting installation accuracy and success rate. Especially in the complex environment of a supercomputing center, due to environmental changes and the accumulation of positioning errors, the robot struggles to achieve precise alignment between the server and the target installation position. Summary of the Invention

[0004] The main objective of this invention is to solve the technical problem in the prior art where mobile robots cannot adaptively adjust their pose according to the actual environment and target position, resulting in the server and installation position not being accurately aligned.

[0005] The first aspect of this invention provides a control method for a mobile robot used for server handling. The mobile robot includes a mobile chassis, a lifting body, a body head, a cargo-carrying and assembly tool unit, and an electrical control cabinet. The control method for the mobile robot used for server handling includes:

[0006] When a server installation instruction is received from the supercomputing center, the mobile chassis and the head of the machine are controlled to perform environmental perception and to perform positioning and mapping based on the environmental perception.

[0007] During the positioning and mapping process, the mobile chassis is controlled to move to the target cabinet corresponding to the server installation command. When it reaches the target cabinet, the loading and assembly tool unit is controlled to identify the target cabinet position and obtain the installation position information of the target cabinet.

[0008] The deviation between the current pose of the server on the cargo and assembly tool unit and the installation position information is calculated, and the pose compensation parameters of the cargo and assembly tool unit are calculated based on the calculated deviation information.

[0009] The posture adjustment process is performed on the cargo and assembly tool unit according to the posture compensation parameters, and the height adjustment process is performed on the lifting body to obtain the server assembly posture aligned with the target installation position, thereby realizing the installation operation between the server and the target cabinet.

[0010] Optionally, in a first implementation of the first aspect of the present invention, the step of controlling the mobile chassis and the head of the machine to perform environmental perception and to perform positioning and mapping based on the environmental perception when receiving the server installation instruction from the supercomputing center includes:

[0011] The server installation instructions are parsed and processed to extract the location identifier of the target cabinet and the server specification parameters to obtain the navigation task parameters;

[0012] According to the navigation task parameters, the laser obstacle avoidance component and safety edge component in the mobile chassis are controlled to perform environmental data acquisition and processing, and the high-position safety laser component in the fuselage head is controlled to perform high-position environmental data acquisition and processing to obtain multi-level environmental perception data.

[0013] Feature extraction and map construction are performed on the multi-level environmental perception data, and robot pose estimation calculation is performed to obtain a real-time environmental map and current position coordinates;

[0014] Based on the real-time environment map and current location coordinates, combined with the navigation task parameters, a global path planning algorithm is performed to calculate the optimal path in a static environment, thereby obtaining the global navigation path and basic motion control commands.

[0015] Optionally, in a second implementation of the first aspect of the present invention, the step of controlling the mobile chassis to move to the target rack corresponding to the server installation command during the positioning and mapping process, and controlling the loading and assembly tool unit to identify the target rack position and obtain the installation position information of the target rack when it reaches the target rack, includes:

[0016] The electric casters in the mobile chassis are controlled to roll and turn according to the global navigation path and basic motion control commands, thereby enabling the mobile chassis to move.

[0017] During the movement of the mobile chassis, the 2D depth camera system in the cargo and assembly tool unit is controlled to perform dynamic environmental monitoring and processing to obtain real-time position feedback and dynamic obstacle avoidance correction commands. The real-time position feedback and dynamic obstacle avoidance correction commands are used to make real-time adjustments to the movement process.

[0018] During the movement, the image data acquired by the 2D depth camera system is processed to identify the target cabinet, and it is determined whether the mobile robot has reached the target cabinet.

[0019] If so, the 3D vision system in the loading and assembly tool unit is controlled to perform a three-dimensional scan of the target cabinet to obtain the three-dimensional geometric features of the target cabinet.

[0020] The installation location information of the target cabinet is obtained by calculating the installation location based on the three-dimensional geometric features using geometric feature recognition and spatial positioning algorithms.

[0021] Optionally, in a third implementation of the first aspect of the present invention, the step of calculating the installation location information of the target cabinet based on the three-dimensional geometric features using geometric feature recognition and spatial positioning algorithms includes:

[0022] The three-dimensional geometric features of the target cabinet are processed by dual 3D camera data fusion, and the geometric features of the positioning reference object of the target installation position are generated by stereo vision algorithm.

[0023] Based on the geometric features of the reference object, the target installation location feature recognition process is performed, and the spatial coordinates and geometric parameters of the target installation location are extracted and calculated by a geometric pattern recognition algorithm.

[0024] The installation attitude is calculated by determining the spatial coordinates and geometric parameters of the target installation location, and the standard attitude parameters for server insertion are calculated by combining the internal guide rail direction and depth constraints of the target cabinet, thus obtaining the installation location information of the target cabinet.

[0025] Optionally, in a fourth implementation of the first aspect of the present invention, the step of calculating the deviation between the current pose of the server on the cargo and assembly tool unit and the installation position information, and calculating the pose compensation parameters of the cargo and assembly tool unit based on the calculated deviation information, includes:

[0026] The geometric feature vector of the server surface is extracted from the server point cloud data obtained by the 3D camera in the fuselage head to obtain a three-dimensional feature description of the server's current pose.

[0027] Based on the three-dimensional feature description, feature matching processing is performed on the CAD model corresponding to the server to obtain the server's current precise pose.

[0028] The coordinate system transformation and deviation calculation are performed on the current precise pose of the server and the installation position information to obtain the six-degree-of-freedom deviation parameters;

[0029] Based on the six-degree-of-freedom deviation parameters, the kinematic model of the cargo platform and shaping and positioning components in the cargo and assembly tool unit is inversely solved to calculate the control quantities of lateral movement distance and axial rotation angle, thereby obtaining the pose compensation parameters of the cargo and assembly tool unit.

[0030] Optionally, in a fifth implementation of the first aspect of the present invention, the step of performing feature matching processing based on the three-dimensional feature description and the CAD model corresponding to the server to obtain the current precise pose of the server includes:

[0031] The multi-scale feature descriptor matrix is ​​obtained by performing multi-scale feature descriptor calculation on the three-dimensional feature description of the current pose of the server using the FPFH algorithm or SHOT algorithm.

[0032] Based on the multi-level feature descriptor matrix and the pre-built feature library of the CAD model corresponding to the server, a hierarchical matching process is performed to obtain a set of candidate matching point pairs;

[0033] The candidate matching point pair set is geometrically consistent with the RANSAC algorithm, and abnormal matching point pairs are removed to obtain reliable feature matching point pairs.

[0034] The optimal rigid body transformation matrix of the reliable feature matching point pair is calculated by Singular Value Decomposition (SVD), and the pose parameters are solved to obtain the current precise pose of the server.

[0035] Optionally, in a sixth implementation of the first aspect of the present invention, the step of adjusting the pose of the cargo and assembly tool unit according to the pose compensation parameters and adjusting the height of the lifting body to obtain a server assembly pose aligned with the target installation position, thereby realizing the installation operation between the server and the target rack, includes:

[0036] Based on the pose compensation parameters, motion control processing is performed on the cargo platform and shaping and positioning components in the cargo and assembly tool unit to obtain the adjusted pose of the cargo platform.

[0037] Based on the installation position information, the heavy-duty precision linear guide assembly and the heavy-duty silent ball screw assembly in the lifting body are subjected to lifting control processing to obtain the adjusted position of the lifting body.

[0038] Multi-axis coordinated motion control processing is performed on the adjusted posture of the cargo platform and the adjusted posture of the elevator body to obtain the coordinated motion state of the cargo and assembly tool unit and the elevator body.

[0039] Based on the coordinated motion state, the six-axis collaborative robotic arm and the robotic arm dexterous hand are controlled to perform server pushing and fixing operations, thereby obtaining the server assembly pose aligned with the target installation position and realizing the installation operation between the server and the target rack.

[0040] A second aspect of the present invention provides a mobile robot control device for server handling, the mobile robot comprising a mobile chassis, a lifting body, a body head, a cargo-carrying and assembly tool unit, and an electrical control cabinet, the mobile robot control device for server handling comprising:

[0041] The navigation and mapping module is used to control the mobile chassis and the head of the machine to perform environmental perception when it receives the server installation instruction from the supercomputing center, and to perform positioning and mapping based on the environmental perception.

[0042] The location recognition module is used to control the mobile chassis to move to the target cabinet corresponding to the server installation command during the positioning and mapping process. When it reaches the target cabinet, it controls the loading and assembly tool unit to perform target cabinet location recognition and obtain the installation location information of the target cabinet.

[0043] The deviation calculation module is used to calculate the deviation between the current pose of the server on the cargo and assembly tool unit and the installation position information, and to calculate the pose compensation parameters of the cargo and assembly tool unit based on the calculated deviation information.

[0044] The pose adjustment module is used to adjust the pose of the cargo and assembly tool unit according to the pose compensation parameters, and to adjust the height of the lifting body to obtain the server assembly pose aligned with the target installation position, thereby realizing the installation operation between the server and the target cabinet.

[0045] The aforementioned mobile robot control method and device for server handling achieves precise positioning and mapping through multi-sensor fusion SLAM navigation technology. It combines a 3D vision system to accurately identify the target server rack and acquire its installation position information. A feature matching algorithm calculates the deviation parameters between the server's current pose and the target installation position, and based on deviation compensation, coordinates the pose adjustments of the loading platform and the lifting body. This invention, through an integrated multi-dimensional pose compensation control mechanism, realizes fully automated control of the mobile robot from navigation and positioning to precise alignment, effectively improving the positioning accuracy and success rate of automatic server installation.

[0046] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the first embodiment of the mobile robot control method for server handling in this invention.

[0049] Figure 2 This is a schematic diagram of one embodiment of the mobile robot control device for server handling in this invention.

[0050] Figure 3 This is a schematic diagram of the structure of a mobile robot used for server handling in an embodiment of the present invention;

[0051] Figure 4 This is a schematic diagram of the mobile chassis of the mobile robot in an embodiment of the present invention;

[0052] Figure 5 This is a schematic diagram of the structure of the head of the mobile robot in an embodiment of the present invention;

[0053] Figure 6 This is a schematic diagram of the cargo-carrying and assembly tool unit of the mobile robot in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] The terms "comprising" and "having," and any variations thereof, used in the embodiments of this invention are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0056] To facilitate understanding of this embodiment, a mobile robot control method for server handling disclosed in this embodiment of the invention will first be described in detail. The mobile robot includes a mobile chassis 1, a lifting body 2, a body head 3, a cargo carrying and assembly tool unit 4, and an electrical control cabinet 5. Figure 1 As shown, this method includes the following steps:

[0057] S101. When receiving the server installation instruction from the supercomputing center, control the mobile chassis and the head of the machine to perform environmental perception, and perform positioning and mapping based on the environmental perception.

[0058] In one embodiment of the present invention, when a server installation instruction from a supercomputing center is received, controlling the mobile chassis and the head of the robot to perform environmental perception, and performing positioning and mapping based on the environmental perception includes: parsing the server installation instruction, extracting the location identifier of the target rack and server specification parameters to obtain navigation task parameters; controlling the laser obstacle avoidance component and safety edge component in the mobile chassis to perform environmental data acquisition and processing based on the navigation task parameters, and controlling the high-position safety laser component in the head of the robot to perform high-position environmental data acquisition and processing to obtain multi-level environmental perception data; performing feature extraction and map construction on the multi-level environmental perception data, and performing robot pose estimation calculation to obtain a real-time environmental map and current position coordinates; and performing global path planning algorithm processing based on the real-time environmental map and current position coordinates combined with the navigation task parameters to calculate the optimal path in a static environment, and obtain a global navigation path and basic motion control instructions.

[0059] Specifically, when the mobile robot receives the server installation instruction from the supercomputing center, electrical control cabinet 5 first parses the instruction. The control unit in electrical control cabinet 5 uses a lexical analyzer to segment the instruction text, dividing the continuous string according to predefined delimiters and keywords to identify key information segments such as cabinet number, location coordinates, and server model. Subsequently, a semantic parser extracts the target cabinet's location identifier field from the segmentation results using a combination of regular expression matching and dictionary lookup. This field contains spatial positioning data such as floor code, area number, and specific coordinates. It also extracts server specification parameter fields, including the server's length, width, height, and weight. An error detection mechanism is used during parsing to complete missing fields and standardize data with abnormal formats, ensuring the completeness and accuracy of the extracted information. Finally, the parsed discrete information is reorganized into structured navigation task parameters, stored in JSON format, containing the target's X, Y, and Z three-dimensional coordinates, the server's digitized specifications, and the task priority identifier.

[0060] Specifically, based on navigation mission parameters, electrical control cabinet 5 begins to control the mobile chassis 1 and the head of the fuselage 3 to perform coordinated environmental data acquisition and processing. For example... Figure 4 The mobile chassis 1 consists of a chassis body 101, a laser obstacle avoidance component 102, a status and warning light strip component 103, a safety contact edge component 104, and electric casters 105. For example... Figure 5The head section 3 of the fuselage consists of a high-position safety laser assembly 301, a top decorative shell 302, a status light strip assembly 303, and a 3D camera 304. During environmental data acquisition and processing, the electrical control cabinet 5 first sends an acquisition command to the laser obstacle avoidance assembly 102. The laser obstacle avoidance assembly 102 activates its laser emitter to emit laser pulses at a set scanning frequency. The laser receiver receives the reflected laser signals and calculates the round-trip time using the time-of-flight method, combining this with the speed of light to calculate the obstacle distance data. The scanning process involves rotating the camera in 0.36-degree angular increments, covering a complete 360-degree horizontal field of view, acquiring 1000 distance measurement points per scan. Simultaneously, the pressure sensor array in the safety contact edge assembly 104 remains active, immediately transmitting the pressure value and contact position information to the electrical control cabinet 5 when external pressure is detected. In parallel, the high-position safety laser assembly 301 performs the same laser scanning process, with its scanning plane positioned 500 mm above the laser obstacle avoidance assembly 102, acquiring distance data for high-position obstacles. After receiving raw data from different laser components, the electrical control cabinet 5 matches the multi-layer laser data at the same time using a timestamp alignment algorithm. Then, it applies a coordinate system transformation matrix to uniformly transform the data in each sensor coordinate system to the robot body coordinate system, forming multi-layer environmental perception data containing three-dimensional information such as distance, angle, and height.

[0061] Specifically, when the electrical control cabinet 5 performs feature extraction and map building processing on multi-level environmental perception data, the cargo loading and assembly tool unit 4 participates in visual data processing. For example... Figure 6The cargo loading and assembly tool unit 4 consists of a cargo loading platform and shaping and positioning component 401, a weight monitoring component 402, a 3D vision system 403, a 2D depth camera system 404, a six-axis collaborative robotic arm 405, and a robotic arm dexterous hand 406. During feature extraction, the electrical control cabinet 5 calls the ORB feature extraction algorithm to process the image frames acquired by the 3D camera 304. The algorithm first applies a FAST corner detector to calculate the response function value at each pixel position in the image, and uses a non-maximum suppression method to select the pixel with the largest local response value as candidate corner points. Then, the Harris corner response function is applied to the candidate corner points for secondary filtering, retaining feature points whose corner response values ​​exceed a set threshold. For each retained feature point, the algorithm extracts the grayscale distribution pattern of its surrounding 31×31 pixel area, and encodes the grayscale pattern into a 256-bit binary descriptor using the BRIEF descriptor algorithm. The map construction process uses a sliding window mechanism, maintaining a sliding window containing the 30 most recent keyframes. Whenever a new keyframe is added, the oldest keyframe is moved out of the window. The algorithm classifies feature points less than 5 meters from the current robot position as local map points and those greater than 5 meters as non-local map points based on their spatial distance. For non-local map points, the algorithm uses mixed-integer linear programming to establish an optimization model. The objective function is to minimize the total number of retained map points, with the constraint that at least 100 map points are retained for pose calculation in each keyframe. The optimization process employs a branch-and-bound algorithm, recursively searching to find the optimal subset of map points that satisfies the constraints, forming a sparse map point set.

[0062] Specifically, during the pose estimation calculation, the electrical control cabinet 5 applies an iterative nearest-neighbor algorithm to process the sparse map point set. The algorithm first establishes the correspondence between the currently observed feature point and the stored feature points in the map, finding the nearest neighbor map point for each observed feature point through descriptor distance calculation. Then, a reprojection error function is constructed, which calculates the pixel distance difference between the projected position of the observed feature point on the image plane and the position of the map point after projection through the current estimated pose. The algorithm uses the Levenberg-Marquardt optimization method to iteratively solve for the robot pose parameters that minimize the reprojection error, including 3D position coordinates and 3D rotation angles. During each iteration, the Jacobian matrix of the error function is calculated, and the increment of the pose parameters is updated until the error converges to a preset threshold. After pose estimation is completed, the algorithm identifies and releases redundant information such as unstable map points observed less than 3 times and abnormal correlations with reprojection errors greater than 2 pixels, obtaining optimized pose data. Based on optimized pose data, the incremental map building algorithm calculates the 3D spatial coordinates of newly observed feature points using triangulation, updates the observation statistics of existing map points in the global map, and generates a real-time environmental map with a grid resolution of 0.05 meters and the current position coordinates. During the global path planning algorithm, the electrical control cabinet 5 converts the real-time environmental map into an occupied grid map, with each grid marked as free, occupied, or unknown. The algorithm uses the current position coordinates as the starting point and the target cabinet position as the ending point, applying the A* search algorithm to find the optimal path on the grid map. During the search, the cost function of each grid is calculated, including the actual cost from the starting point to the current grid and the heuristic cost from the current grid to the ending point, prioritizing the expansion of grid nodes with the smallest cost function value. The algorithm considers the robot's kinematic constraints, limiting the change in steering angle between adjacent path points to ensure that the generated path conforms to the steering capability of the electric caster 105. The final output includes a global navigation path containing a sequence of path point coordinates, speed commands for each path segment, steering angle commands, and basic motion control commands.

[0063] Furthermore, the step of performing feature extraction and map construction on the multi-layered environmental perception data, as well as robot pose estimation calculation to obtain a real-time environmental map and current position coordinates, includes: extracting corner features from keyframes in the multi-layered environmental perception data using the ORB feature extraction algorithm and matching and associating them to obtain a set of keyframe feature points; constructing a sliding window map from the set of keyframe feature points using a fixed number of keyframes and dividing the map points into local map points and non-local map points to obtain hierarchical map point data; establishing a K-cover problem model for the hierarchical map point data using a mixed integer linear programming method and optimizing the selection of non-local map points to obtain a sparse map point set; performing pose estimation calculation on the sparse map point set using an iterative nearest point algorithm and releasing redundant map information to obtain optimized pose data; and updating the optimized pose data in real time using an incremental map construction algorithm and calculating precise coordinate positions to obtain a real-time environmental map and current position coordinates.

[0064] Specifically, when the electrical control cabinet 5 processes multi-level environmental perception data for feature extraction and map construction, it first uses the ORB feature extraction algorithm to extract corner features from key frames and perform matching and association. The electrical control cabinet 5 selects key frames with high image quality and reasonable spatial distribution from the multi-level environmental perception data. Key frame selection criteria include an image sharpness score greater than a set threshold, a displacement distance exceeding 0.5 meters from the previous key frame, or a rotation angle exceeding 10 degrees. The ORB feature extraction algorithm first applies a FAST corner detector to the selected key frames. This detector constructs a 16-pixel circular neighborhood at each pixel location in the image and determines whether a pixel is a corner by comparing the grayscale difference between the center pixel and its neighboring pixels. The algorithm sets a grayscale difference threshold of 20. When the grayscale difference between 9 or more consecutive pixels in the neighborhood and the center pixel exceeds the threshold, the center pixel is marked as a candidate corner. Subsequently, the algorithm applies a non-maximum suppression method, comparing the response intensity of candidate corners within a 3×3 pixel window and retaining the pixel with the highest response intensity as the final corner. For each extracted corner point, the algorithm extracts a 31×31 pixel image patch around it. The BRIEF descriptor algorithm converts the grayscale distribution pattern of this image patch into a 256-bit binary descriptor. Each bit of the descriptor is determined by comparing the grayscale values ​​of predefined pixel pairs within the image patch. During feature matching, the algorithm calculates the Hamming distance between the feature points extracted in the current frame and the feature points stored in previous keyframes. The Hamming distance measures the similarity between two binary descriptors; a smaller distance indicates greater similarity. The algorithm sets a Hamming distance threshold of 50. When the Hamming distance between two feature points is less than this threshold, a matching relationship is established, forming a keyframe feature point set containing feature point coordinates, descriptor information, and the matching relationship.

[0065] Specifically, based on the keyframe feature point set, the electrical control cabinet 5 constructs a sliding window map using a fixed number of keyframes and classifies map points into local map points and non-local map points. During the sliding window map construction process, the algorithm maintains a fixed-size window containing the 30 most recent keyframes. The window size is chosen based on a balance between computational efficiency and map coverage. When a new keyframe is added to the window, the oldest keyframe and its associated feature points are removed from the window to ensure the window size remains constant. The algorithm triangulates the feature points within the window, calculating their coordinates in 3D space through observations from at least two different perspectives. The triangulation process uses the least squares method, minimizing the reprojection error to solve for the optimal 3D coordinates of the feature points. The reprojection error refers to the difference between the actual pixel position of the feature point in each observation frame and the theoretical projection position calculated from the 3D coordinates. During map point classification, the algorithm classifies each 3D feature point based on its Euclidean distance to the current robot position. Feature points less than 5 meters from the current position are classified as local map points, which participate in real-time pose tracking and local optimization calculations. Feature points located more than 5 meters from the current position are classified as non-local map points, which are mainly used for global localization and loop closure detection. Simultaneously, the algorithm statistically analyzes the observation count, observation angle distribution, and descriptor stability of each map point, forming hierarchical map point data that includes the map point's 3D coordinates, classification labels, and attribute statistics.

[0066] Specifically, electrical control cabinet 5 uses mixed-integer linear programming to establish a K-cover problem model for hierarchical map point data and optimize the selection of non-local map points. In establishing the K-cover problem model, the algorithm transforms the map sparsity problem into a combinatorial optimization problem, aiming to minimize the total number of retained map points while ensuring that each keyframe contains at least K map points. The algorithm sets K to 100, ensuring that each keyframe has enough feature points for robust pose estimation calculations. In the mixed-integer linear programming model, each non-local map point corresponds to a binary decision variable; a value of 1 indicates retention, and a value of 0 indicates deletion. The constraint matrix describes the relationship between map points and keyframes; each element in the matrix indicates whether a map point was observed in a given keyframe. The objective function is a weighted sum of all decision variables, with weights determined based on the observation quality of the map points. Map points with more observations and smaller reprojection errors are assigned smaller weights, making them more likely to be retained. The optimization process employs a branch-and-bound algorithm, using recursive search and pruning strategies to find the optimal solution in the feasible solution space. The algorithm first calculates the optimal solution of the linear relaxation problem as the lower bound, and then gradually introduces integer constraints through branching operations. During the search process, it uses the bounding information to prune branches that cannot produce a better solution, and finally obtains the optimal map point selection scheme that satisfies all constraints, forming a sparse map point set.

[0067] Specifically, electrical control cabinet 5 uses an iterative nearest-point algorithm to perform pose estimation calculations on the sparse map point set and releases redundant map information. Subsequently, an incremental map construction algorithm is used for real-time map updates. During the pose estimation calculation, the iterative nearest-point algorithm first establishes a correspondence between the currently observed feature point and the map points already stored in the sparse map point set. This correspondence is established through a combination of feature descriptor matching and spatial distance constraints. The algorithm calculates the descriptor similarity and reprojection distance between each observed feature point and candidate map points, selecting the map point with the highest comprehensive score as the matching object. Based on the established point correspondence, the algorithm constructs a reprojection error function, which measures the pixel distance difference between the actual position of the observed feature point on the image plane and the theoretical position of the map point after projection of the current estimated pose. Pose estimation employs a nonlinear least squares optimization method, using the Levenberg-Marquardt algorithm to iteratively solve for the robot's six-degree-of-freedom pose parameters, including three-dimensional position coordinates and three-dimensional rotation angles, that minimize the reprojection error function. During each iteration, the algorithm calculates the Jacobian matrix of the error function with respect to the pose parameters and updates the increment of the pose parameters based on gradient information. Iteration stops when the pose change between two consecutive iterations is less than a set threshold. After pose estimation, the algorithm identifies and releases redundant map information such as abnormal matching relationships with reprojection errors greater than 2 pixels, unstable map points with fewer than 3 observations, and historical feature descriptors exceeding the sliding window range. It releases corresponding storage space through a memory reclamation mechanism, obtaining memory-optimized pose data. The incremental map construction algorithm uses the optimized pose data to update the global map in real time. Newly observed feature points are added to the global map after their 3D spatial coordinates are calculated using a multi-view triangulation method. The observation statistics of existing map points are updated, including observation count accumulation, descriptor information fusion, and co-view relationship maintenance. Based on the updated map information, the algorithm constructs an occupied grid map, discretizing the 3D space into grid cells with a resolution of 0.05 meters. Each grid cell is marked as free space, occupied space, or unknown space based on the distribution of map points within it. Based on the current optimized pose data, the algorithm calculates the robot's precise position coordinates in the global coordinate system. The accuracy of this coordinate information reaches the centimeter level. The final output includes a real-time environmental map containing the distribution of environmental obstacles, the division of passable areas, the distribution of feature points, and the current position coordinates.

[0068] S102. During the positioning and mapping process, control the mobile chassis to move to the target cabinet corresponding to the server installation command. When it reaches the target cabinet, control the loading and assembly tool unit to identify the target cabinet position and obtain the installation position information of the target cabinet.

[0069] In one embodiment of the present invention, during the positioning and mapping process, controlling the mobile chassis to move to the target cabinet corresponding to the server installation command, and upon reaching the target cabinet, controlling the loading and assembly tool unit to identify the target cabinet position and obtain the installation position information of the target cabinet includes: controlling the electric casters in the mobile chassis to roll and turn according to the global navigation path and basic motion control commands to realize the movement of the mobile chassis; during the movement of the mobile chassis, controlling the 2D depth camera system in the loading and assembly tool unit to perform dynamic environmental monitoring processing to obtain real-time position feedback and dynamic obstacle avoidance correction commands, which are used to adjust the movement process in real time; during the movement, performing target cabinet identification processing on the image data collected by the 2D depth camera system to determine whether the mobile robot has reached the target cabinet; if so, controlling the 3D vision system in the loading and assembly tool unit to perform a three-dimensional scan of the target cabinet to obtain the three-dimensional geometric features of the target cabinet; and calculating the installation position based on the three-dimensional geometric features using geometric feature recognition and spatial positioning algorithms to obtain the installation position information of the target cabinet.

[0070] Furthermore, the step of calculating the installation position information of the target cabinet based on the three-dimensional geometric features using geometric feature recognition and spatial positioning algorithms includes: performing dual 3D camera data fusion processing on the three-dimensional geometric features of the target cabinet, generating geometric features of a positioning reference object for the target installation position using a stereo vision algorithm; performing target installation position feature recognition processing based on the geometric features of the reference object, extracting and calculating the spatial coordinates and geometric parameters of the target installation position using a geometric pattern recognition algorithm; calculating the installation attitude based on the spatial coordinates and geometric parameters of the target installation position, and calculating the standard attitude parameters for server insertion in conjunction with the internal guide rail direction and depth constraints of the target cabinet to obtain the installation position information of the target cabinet.

[0071] Specifically, when the electrical control cabinet 5 calculates its installation position based on three-dimensional geometric features using geometric feature recognition and spatial positioning algorithms, it first performs dual 3D camera data fusion processing on the three-dimensional geometric features of the target cabinet. The 3D vision system 403 consists of two 3D cameras, which are installed at the bottom of the moving parts of the loading platform and can change position as the loading platform is adjusted. During the dual 3D camera data fusion processing, the electrical control cabinet 5 synchronously controls the two 3D cameras to scan the target cabinet, with each camera independently acquiring three-dimensional point cloud data and color image information of the cabinet surface. The stereo vision algorithm processes the data collected by the two cameras using the principle of parallax calculation. The algorithm first performs stereo correction on the images of the left and right cameras to eliminate camera installation errors and lens distortion. Subsequently, the algorithm establishes the correspondence between the same physical point in the left and right images using feature matching methods, and calculates the three-dimensional spatial coordinates of each matching point based on the principle of triangulation. Point cloud data acquired by two 3D cameras are unified to the coordinate system of the loading and assembly tool unit 4 through coordinate system transformation. The algorithm filters the fused point cloud data to remove noise points and data holes, forming complete three-dimensional point cloud data of the target cabinet surface. This point cloud data includes the geometry and spatial relationships of the cabinet's front panel, side panels, and internal structure, constituting the geometric features of the reference object.

[0072] Specifically, based on the geometric features of the reference object, electrical control cabinet 5 performs target installation location feature recognition processing, extracting and calculating the spatial coordinates and geometric parameters of the target installation location through a geometric pattern recognition algorithm. The geometric pattern recognition algorithm first performs structural analysis on the geometric features of the reference object, identifying the main geometric components of the cabinet, including the outer shell structure and the internal installation area. The algorithm identifies the position and orientation of the cabinet's front panel using a planar detection method, determining the cabinet's reference coordinate system. Within the front panel area, the algorithm searches for feature patterns in the mounting slots, which exhibit regular rectangular opening structures for the insertion and installation of server equipment. During feature recognition, the algorithm uses edge detection technology to identify the boundary contours of the slots, analyzing the distribution patterns of the horizontal and vertical boundary lines. Based on the size requirements in the server specifications, the algorithm filters target areas from the identified mounting slots that meet the size matching conditions. The filtering process considers the compatibility of the server's width and height parameters with the cabinet slot dimensions, ensuring that the selected installation location can accommodate the server equipment to be installed. The algorithm calculates the geometric center point of the target installation area as the installation reference position, and simultaneously extracts geometric parameters such as the boundary dimensions and opening depth of this area. These parameters define the spatial constraints during server installation, including the maximum allowable device size and insertion depth limits, ultimately forming the spatial coordinates and geometric parameters of the target installation location.

[0073] Specifically, electrical control cabinet 5 performs installation attitude calculations on the spatial coordinates and geometric parameters of the target installation location, and calculates the standard attitude parameters for server insertion based on the direction of the guide rails and depth constraints inside the target cabinet. The installation attitude calculation process first analyzes the structural characteristics of the guide rails inside the cabinet. The guide rails are metal strips inside the cabinet used to guide the sliding insertion of the server, distributed on the left and right sides of the installation slot. The algorithm detects and identifies the spatial position and extension direction of the guide rails through linear features in the point cloud data. The extension direction of the guide rails indicates the standard movement trajectory when the server is inserted into the cabinet. Based on the guide rail direction information, the algorithm determines the standard orientation for server insertion, i.e., the server device should be pushed into the cabinet along the guide rail direction. During the calculation of depth constraints, the algorithm measures the available depth space inside the cabinet from the front panel to the rear panel. This depth determines the maximum distance the server can be inserted. The algorithm identifies the position of the rear wall or internal obstruction structures of the cabinet through point cloud data and calculates the distance from the installation location to these constraint boundaries as depth limit parameters. Simultaneously, the algorithm detects whether there are existing devices, cables, or other obstacles inside the cabinet and assesses the impact of these factors on the server insertion path. Based on information such as spatial coordinates, guide rail direction, and depth constraints, the algorithm calculates the standard attitude parameters of the server at the target installation location. This includes three-dimensional position coordinates representing the precise position the server should reach, and three-dimensional rotation angles representing the correct orientation of the server relative to the rack coordinate system. These attitude parameters ensure that the server is aligned with the rack guide rails, the insertion depth meets safety requirements, and interference with the rack's internal structure is avoided. Ultimately, this results in complete target rack installation position information, including installation position coordinates, insertion direction vector, depth limit value, and attitude angles. This information provides a precise spatial positioning reference for subsequent pose adjustments of the loading and assembly tool units and for robotic arm operations.

[0074] S103. Calculate the deviation between the current pose and installation position information of the server on the cargo and assembly tool unit, and calculate the pose compensation parameters of the cargo and assembly tool unit based on the calculated deviation information.

[0075] In one embodiment of the present invention, the step of calculating the deviation between the current pose of the server on the cargo and assembly tool unit and the installation position information, and calculating the pose compensation parameters of the cargo and assembly tool unit based on the calculated deviation information, includes: extracting the geometric feature vector of the server surface from the server point cloud data acquired by the 3D camera in the fuselage head to obtain a three-dimensional feature description of the server's current pose; performing feature matching processing between the three-dimensional feature description and the CAD model corresponding to the server to obtain the server's current precise pose; performing coordinate system transformation and deviation calculation processing between the server's current precise pose and the installation position information to obtain six-degree-of-freedom deviation parameters; and performing inverse kinematics calculation processing on the kinematic model of the cargo platform and the shaping and positioning component in the cargo and assembly tool unit based on the six-degree-of-freedom deviation parameters to calculate the control quantities of lateral movement distance and axial rotation angle to obtain the pose compensation parameters of the cargo and assembly tool unit.

[0076] Specifically, when the electrical control cabinet 5 calculates the deviation between the current pose and installation position information of the server on the loading and assembly tool unit 4, it first extracts the geometric feature vector of the server surface from the server point cloud data acquired by the 3D camera 304 in the fuselage head 3. The fuselage head 3 consists of a high-position safety laser assembly 301, a top decorative shell 302, a status light strip assembly 303, and a 3D camera 304. The 3D camera 304 monitors the working status of the loading platform of the loading and assembly tool unit 4 from below. The 3D camera 304 acquires image data of the server surface and converts the two-dimensional image information into three-dimensional point cloud data through a depth estimation algorithm. The point cloud data contains the three-dimensional coordinate information of the server shell surface. The electrical control cabinet 5 preprocesses the acquired server point cloud data and uses a normal vector calculation method to analyze the geometric features of the server surface. The normal vector is perpendicular to the surface and is used to describe the surface orientation. The algorithm identifies the main geometric structure of the server, including planar areas such as the front panel, side panel, and top panel, and extracts the edge features and corner features of the server. Edges are the boundaries between different surfaces, and corners are the locations where multiple edge lines intersect, exhibiting good spatial distinguishability. The algorithm combines the extracted normal vectors, edge lines, corners, and other geometric features to form a feature vector. This vector, in numerical form, describes the server's current geometric state and spatial pose, constituting a three-dimensional feature description of the server's current pose.

[0077] Specifically, based on the 3D feature description of the server's current pose, electrical control cabinet 5 performs feature matching processing between this feature and the corresponding CAD model of the server. The CAD model is the computer-aided design model of the server, containing information such as the server's standard geometry, precise dimensions, and spatial relationships of its components. During the feature matching process, the algorithm extracts standard feature descriptions from the server's CAD model, using the same feature extraction method as the actual point cloud data. The result of the CAD model feature extraction serves as the matching benchmark, representing the geometric feature distribution of the server in its standard pose. The algorithm establishes the correspondence between actual features and standard features through feature similarity calculation, which is based on metrics such as Euclidean distance between feature vectors. The matching process employs an iterative optimization strategy. The algorithm initializes an estimated server pose, transforms the CAD model according to this pose, and then calculates the matching error between the transformed model features and the actual observed features. By continuously adjusting the pose parameters and recalculating the matching error, the algorithm finds the optimal pose solution that minimizes the error. The algorithm stops iterating when the pose change in continuous iterations is less than a preset threshold. After the matching process is completed, the server's current precise pose is obtained. This pose is represented by six degrees of freedom parameters, including three-dimensional position coordinates and three-dimensional rotation angles, which accurately describe the actual spatial state of the server on the cargo platform.

[0078] Specifically, electrical control cabinet 5 performs coordinate system transformation and deviation calculation on the server's current precise pose and installation position information. The coordinate system transformation process unifies the pose information represented in different coordinate systems to the same reference coordinate system. The server's current precise pose is based on the body coordinate system of the loading and assembly tool unit 4, while the installation position information is based on the target cabinet's coordinate system. The algorithm transforms the server pose from the loading platform coordinate system to the cabinet coordinate system using a coordinate transformation matrix, which includes a rotation matrix and a translation vector. After the coordinate system transformation, the algorithm calculates the deviation between the transformed server pose and the target installation position. The deviation calculation includes position deviation and attitude deviation. The position deviation is obtained by calculating the vector difference between the server's current position and the target installation position, including the position offset in the X, Y, and Z directions. The attitude deviation is obtained by calculating the angular difference between the server's current orientation and the target installation orientation, including the rotational deviation angle around the X, Y, and Z axes, forming a six-degree-of-freedom deviation parameter.

[0079] Specifically, the electrical control cabinet 5 performs inverse kinematics calculation on the kinematic model of the cargo platform and shaping / positioning component 401 based on the six degrees of freedom deviation parameters. The cargo platform and shaping / positioning component 401 support 100mm lateral movement to the left and right of the cargo platform and 3° axial rotation around the cargo entry / exit direction. The kinematic model describes the motion relationship of the cargo platform's mechanical structure, and the inverse kinematics calculation calculates the motion parameters required for each drive joint based on the desired end position and attitude of the cargo platform. The algorithm calculates the lateral movement distance based on the X-direction position deviation in the six degrees of freedom deviation parameters and the axial rotation angle based on the rotation deviation around the Z-axis. The inverse kinematics process considers the physical constraints of the cargo platform, and a saturation processing method is used when the calculated control quantity exceeds the physical limits. After the inverse kinematics calculation is completed, the pose compensation parameters of the cargo and assembly tool unit 4 are obtained, including the specific distance value of lateral movement and the specific angle value of axial rotation, which are used to control the cargo platform and shaping / positioning component 401 to achieve precise adjustment of the server's pose.

[0080] Furthermore, the step of performing feature matching processing based on the 3D feature description and the corresponding CAD model of the server to obtain the server's current precise pose includes: calculating multi-scale feature descriptors for the 3D feature description of the server's current pose using the FPFH algorithm or the SHOT algorithm to obtain a multi-level feature descriptor matrix; performing hierarchical matching processing based on the multi-level feature descriptor matrix and a pre-built feature library of the server's corresponding CAD model to obtain a set of candidate matching point pairs; performing geometric consistency verification on the set of candidate matching point pairs using the RANSAC algorithm and removing abnormal matching point pairs to obtain reliable feature matching point pairs; and calculating the optimal rigid body transformation matrix and solving the pose parameters for the reliable feature matching point pairs using SVD singular value decomposition to obtain the server's current precise pose.

[0081] Specifically, when electrical control cabinet 5 performs feature matching processing based on the 3D feature description of the server's current pose and the corresponding CAD model, it first performs multi-scale feature descriptor calculation on the 3D feature description of the server's current pose using either the FPFH algorithm or the SHOT algorithm. The FPFH algorithm, or Fast Point Feature Histogram algorithm, is a point cloud description method based on geometric features. This algorithm generates feature descriptors by analyzing the spatial distribution relationships of other points in the neighborhood of each feature point. The SHOT algorithm, or Unique Shape Context algorithm, constructs a spherical neighborhood centered on the feature point, divides the neighborhood into multiple spatial regions, and statistically analyzes the distribution characteristics of points within each region. Electrical control cabinet 5 selects an appropriate algorithm based on the density and geometric complexity of the server's point cloud data. The SHOT algorithm is used when the point cloud density is high and the geometric details are rich; the FPFH algorithm is used when the point cloud density is moderate and computational efficiency is high. During the multi-scale feature descriptor calculation process, the algorithm sets several different neighborhood radius parameters: 0.01 meters, 0.02 meters, and 0.05 meters, corresponding to different spatial scale ranges. Small-scale neighborhoods capture detailed features of the server surface, such as screw holes and interface locations; medium-scale neighborhoods describe the edges and corners of the server panel; and large-scale neighborhoods reflect the overall geometric contours and main structure of the server. The algorithm computes feature descriptors at each scale: the FPFH algorithm generates a 33-dimensional feature vector, and the SHOT algorithm generates a 352-dimensional feature vector. Feature descriptors from multiple scales are arranged and combined in scale order to form a multi-level feature descriptor matrix containing different levels of geometric information. The number of rows in this matrix equals the number of feature points, and the number of columns equals the product of the dimension of a single feature descriptor and the number of scales.

[0082] Specifically, based on a multi-level feature descriptor matrix, the electrical control cabinet 5 performs hierarchical matching processing according to this matrix and the pre-built feature library of the corresponding CAD model on the server. The pre-built feature library is a feature database generated by offline processing of the server's CAD model. This database contains feature descriptor information of the server's standard geometric model under different views and poses. During the feature library construction process, the algorithm performs dense sampling on the CAD model, uniformly distributing sampling points on the model surface, and then uses the same FPFH or SHOT algorithm as the actual point cloud processing to calculate the multi-scale feature descriptor for each sampling point. The feature library is indexed and organized according to the dimension and type of the feature descriptors, supporting fast similarity retrieval and matching operations. The hierarchical matching processing adopts a coarse-to-fine matching strategy. The algorithm first performs coarse matching at the large-scale feature level, searching for potential matching candidates by comparing the Euclidean distance or cosine similarity between large-scale feature descriptors. The purpose of large-scale matching is to quickly filter out possible correspondences and reduce the computational load of subsequent fine matching. Subsequently, the algorithm performs fine matching at the medium-scale and small-scale feature levels, verifying and optimizing the results of coarse matching through more detailed feature comparisons. During the fine matching process, the algorithm employs a nearest neighbor search method. For each feature point in the actual point cloud, it searches for the corresponding point with the smallest distance between its feature descriptors in a pre-built feature library. Simultaneously, the algorithm sets a distance threshold to filter unreliable matches; when the distance between feature descriptors exceeds the threshold, the match is rejected. After hierarchical matching processing, the algorithm statistically analyzes all matches that pass the threshold test, forming a candidate matching point pair set. This set contains the correspondence between actual feature points and CAD model feature points, along with matching confidence scores.

[0083] Specifically, electrical control cabinet 5 uses the RANSAC algorithm to verify the geometric consistency of the candidate matching point pair set and eliminates abnormal matching point pairs. RANSAC, or Random Sample Consensus Algorithm, is a robust parameter estimation method specifically designed for processing datasets containing outliers. The principle of geometric consistency verification is based on the constraint of rigid body transformation, namely, that correct matching point pairs should maintain geometric consistency after undergoing the same rigid body transformation. The algorithm randomly selects a minimum number of point pairs from the candidate matching point pair set to estimate the rigid body transformation parameters; for rigid body transformations in three-dimensional space, at least three non-collinear point pairs are required. The algorithm calculates rotation matrices and translation vectors based on the selected point pairs. The rotation matrix describes the angular transformation relationship between coordinate systems, and the translation vector describes the positional offset between the origins of the coordinate systems. The calculated rigid body transformation parameters are applied to all candidate matching point pairs, and the algorithm checks the geometric consistency of each point pair under the transformation, i.e., the distance error between the actual feature point after transformation and the corresponding CAD model feature point. When the distance error is less than a preset threshold, the matching point pair is considered consistent with the current transformation model and is called an interior point. The algorithm counts the number of interior points supporting the current transformation model and records the corresponding transformation parameters. The RANSAC algorithm repeatedly performs random sampling and consistency checks, generating a candidate transformation model and its corresponding set of interior points in each iteration. The algorithm retains the transformation model with the most interior points as the optimal solution, and the corresponding set of interior points represents the matching point pair with the best geometric consistency. Abnormal matching point pairs include erroneous matches caused by misjudgment of feature similarity, occlusion effects, noise interference, etc. These abnormal point pairs do not meet the geometric constraints of rigid body transformation and are effectively identified and eliminated through the iterative verification process of the RANSAC algorithm, ultimately yielding reliable feature matching point pairs.

[0084] Specifically, electrical control cabinet 5 uses Singular Value Decomposition (SVD) to calculate the optimal rigid body transformation matrix and solve for pose parameters for reliable feature matching point pairs. SVD is a matrix decomposition method that can decompose any matrix into the product of three special matrices, used in rigid body transformation estimation to solve for the optimal rotation matrix. The optimal rigid body transformation matrix calculation process first performs coordinate centering on the reliable feature matching point pairs, calculating the geometric centers of the actual feature point set and the CAD model feature point set respectively. Then, the coordinates of all points are subtracted from the corresponding geometric center coordinates to eliminate the influence of translation components. The centered point pairs only contain rotational transformation relationships, and the algorithm constructs a covariance matrix to describe the correlation between the two point sets. The covariance matrix is ​​obtained by calculating the sum of the coordinate products of the actual feature points and the corresponding CAD model feature points. This matrix contains all the information about the rotational relationship between the two point sets. SVD is applied to the covariance matrix, decomposing it into the product of a left singular vector matrix, a singular value diagonal matrix, and a right singular vector matrix. The optimal rotation matrix is ​​calculated through a specific combination of the left and right singular vector matrices, ensuring that the sum of squares of the distance errors between transformed point pairs is minimized. The translation vector is calculated using the difference between the geometric centers and the effect of the rotation transformation, describing the positional relationship between the origins of the two coordinate systems. The rigid body transformation matrix, composed of the rotation matrix and translation vectors, fully describes the spatial transformation relationship from the CAD model coordinate system to the actual server coordinate system. The pose parameter calculation process converts the rigid body transformation matrix into a standard six-DOF pose representation, including three-dimensional position coordinates and three-dimensional rotation angles. The rotation angles are calculated using Euler angle decomposition or axis-angle representation of the rotation matrix, and the position coordinates directly correspond to the component values ​​of the translation vectors, ultimately yielding the server's current precise pose in the cargo platform coordinate system.

[0085] S104. Adjust the position of the cargo and assembly tool unit according to the position compensation parameters, and adjust the height of the lifting body to obtain the server assembly position aligned with the target installation position, thereby realizing the installation operation between the server and the target rack.

[0086] In one embodiment of the present invention, the step of adjusting the posture of the loading and assembly tool unit according to the posture compensation parameters and adjusting the height of the lifting body to obtain a server assembly posture aligned with the target installation position, thereby realizing the installation operation between the server and the target rack, includes: performing motion control processing on the loading platform and shaping and positioning components in the loading and assembly tool unit according to the posture compensation parameters to obtain the adjusted posture of the loading platform; performing lifting control processing on the heavy-duty precision linear guide assembly and heavy-duty silent ball screw assembly in the lifting body according to the installation position information to obtain the adjusted posture of the lifting body; performing multi-axis coordinated motion control processing on the adjusted posture of the loading platform and the adjusted posture of the lifting body to obtain a coordinated motion state between the loading and assembly tool unit and the lifting body; and controlling a six-axis collaborative robotic arm and a robotic arm dexterous hand to perform server pushing and fixing operations according to the coordinated motion state to obtain a server assembly posture aligned with the target installation position, thereby realizing the installation operation between the server and the target rack.

[0087] Specifically, when the electrical control cabinet 5 performs posture adjustment processing on the cargo loading and assembly tool unit 4 according to the posture compensation parameters, it first performs motion control processing on the cargo platform and shaping and positioning component 401 in the cargo loading and assembly tool unit 4. The cargo loading and assembly tool unit 4 consists of the cargo platform and shaping and positioning component 401, a weight monitoring component 402, a 3D vision system 403, a 2D depth camera system 404, a six-axis collaborative robotic arm 405, and a robotic arm dexterous hand 406. The cargo platform and shaping and positioning component 401 provides a platform for carrying goods and a basis for adjusting the position and posture of goods on the platform. It supports 100mm of lateral movement of the entire cargo platform to the left and right, and supports 3° of axial rotation around the direction of goods entry and exit. According to the lateral movement distance value in the posture compensation parameters, the electrical control cabinet 5 sends control commands to the lateral drive motor of the cargo platform and shaping and positioning component 401. The motor drives the cargo platform to move precisely left and right along the guide rail according to the commands. The lateral movement mechanism employs a precision ball screw drive. The screw pitch determines the conversion relationship between the motor rotation angle and the displacement of the cargo platform. The electrical control cabinet 5 achieves precise displacement control by calculating the required motor rotation angle. Simultaneously, based on the axial rotation angle value in the pose compensation parameters, the electrical control cabinet 5 controls the rotation drive mechanism of the cargo platform to adjust its angle. The rotation mechanism uses a worm gear reducer, which has self-locking characteristics and high-precision transmission capability. During rotation, the cargo platform makes fine adjustments around its geometric center axis, with the adjustment range limited to ±3° to ensure the stability of the server on the platform. Motion control processing adopts a closed-loop feedback control method. The cargo platform and the shaping and positioning component 401 are equipped with position encoders to monitor the actual position and angle of the platform in real time. The electrical control cabinet 5 compares the difference between the actual value and the target value and adjusts the control output until the position error converges within the set accuracy range. After adjustment, the cargo platform reaches a new spatial position and posture, forming the adjusted posture of the cargo platform. This posture enables the server on the cargo platform to be precisely aligned with the target installation position in the horizontal plane.

[0088] Specifically, based on the installation location information, the electrical control cabinet 5 performs lifting control processing on the heavy-duty precision linear guide assembly 201 and the heavy-duty silent ball screw assembly 202 in the lifting body 2. The lifting body 2 consists of the heavy-duty precision linear guide assembly 201, the heavy-duty silent ball screw assembly 202, the metal welded square tube frame 203, the high-power high-torque servo motor system 204, the body decorative cover 205, and the monitoring display terminal 206. The installation location information includes the vertical height coordinates of the mounting slot in the target cabinet, which defines the target height position that the server needs to reach. The electrical control cabinet 5 calculates the lifting distance that the cargo and assembly tool unit 4 needs to lift based on the height parameters in the installation location information. The lifting distance is equal to the difference between the target installation height and the current cargo platform height. The heavy-duty precision linear guide assembly 201 provides guidance and support for the lifting movement, ensuring that the cargo and assembly tool unit 4 maintains its vertical movement trajectory during the lifting process and avoids horizontal deviation and swaying. The heavy-duty silent ball screw assembly 202, as the core component of the lifting drive, converts the rotary motion of the high-power, high-torque servo motor system 204 into linear motion through threaded transmission. The electrical control cabinet 5 sends lifting control commands to the high-power, high-torque servo motor system 204, and the motor drives the ball screw to rotate at a set speed and acceleration according to the commands. The lead parameter of the ball screw determines the distance the loading and assembly tool unit 4 rises or falls per revolution of the motor. The electrical control cabinet 5 precisely controls the lifting height by controlling the number of motor revolutions. During the lifting control process, the electrical control cabinet 5 monitors the data changes of the weight monitoring component 402. When abnormal fluctuations in load occur, the lifting movement is immediately stopped and a safety check is performed. The lifting body 2 supports linear sliding within a stroke of 0-2800mm and supports lifting and moving a maximum load of 650kg with an accuracy within ±0.1mm. After the lifting control is completed, the loading and assembly tool unit 4 reaches the vertical height corresponding to the target installation position, forming the adjusted posture of the lifting body.

[0089] Specifically, the electrical control cabinet 5 performs multi-axis coordinated motion control processing on the adjusted posture of the cargo platform and the lifting body to ensure the motion coordination between the cargo and assembly tool unit 4 and the lifting body 2. The purpose of multi-axis coordinated motion control processing is to eliminate the mutual influence generated during the adjustment of the cargo platform posture and the lifting body height, ensuring that the server reaches the precise target position and posture in space. The coordinated control algorithm considers the impact of the lateral movement and rotation adjustment of the cargo platform on the center of gravity distribution of the lifting body. When the cargo platform deviates from the center position, it will change the center of gravity position of the entire cargo and assembly tool unit 4, affecting the stability of the lifting motion. The electrical control cabinet 5 calculates the center of gravity offset based on the current position and load distribution of the cargo platform and adjusts the lifting control strategy to compensate for the impact of the center of gravity offset. At the same time, the coordinated control algorithm considers the impact of the lifting motion on the horizontal adjustment accuracy of the cargo platform. Mechanical vibration and inertial forces during the lifting process will interfere with the positional accuracy of the cargo platform. The electrical control cabinet 5 adopts a motion trajectory optimization method to reasonably arrange the cargo platform adjustment and lifting motion in a time sequence, avoiding the simultaneous execution of rapid movements in multiple axes. Coordinated motion control also includes synchronous control of the movement speed of each axis, ensuring that the movement speeds of the cargo platform and the elevator body are matched, avoiding mechanical stress and positioning errors caused by asynchronous movement. The electrical control cabinet 5 dynamically adjusts the motion parameters of each axis by monitoring the position feedback signals of each axis in real time, achieving a coordinated motion state between the cargo and assembly tool unit 4 and the elevator body 2. Coordinated motion means that the cargo platform and the elevator body simultaneously reach their respective target positions, and the server's position and orientation in three-dimensional space precisely match the target installation position, creating ideal initial conditions for subsequent robotic arm operations.

[0090] Specifically, the electrical control cabinet 5 controls the six-axis collaborative robotic arm 405 and the robotic arm dexterous hand 406 to perform server pushing and fixing operations based on the coordinated motion state. The six-axis collaborative robotic arm 405 has six rotary joints, enabling precise control of any position and posture in three-dimensional space. The robotic arm dexterous hand 406 is installed at the end of the robotic arm as an actuator that directly contacts the server. During the server pushing operation, the electrical control cabinet 5 first calculates the motion trajectory of the robotic arm from its current position to the server pushing position. The pushing position is located near the geometric center of the server's rear panel, which provides stable thrust transmission and uniform force distribution. The robotic arm dexterous hand 406 is equipped with a force sensor that can monitor the magnitude and direction of the contact force in real time during the pushing process. When the contact force exceeds the safety threshold, the robotic arm immediately stops pushing and sounds an alarm. The pushing operation adopts a constant speed control method. The robotic arm pushes the server into the installation position along the rack rails at a preset slow speed. The pushing speed setting takes into account factors such as server weight, rail friction resistance, and safety requirements. During the pushing process, the electrical control cabinet 5 continuously monitors the alignment between the server and the rack rails, using the 3D vision system 403 and robotic arm force feedback to determine if the server is sliding in along the correct trajectory. Once the server is fully inserted into the target installation position, the robotic arm's dexterous hand 406 performs a fixing operation, including manipulating the locking mechanism inside the rack, adjusting the support feet, and connecting cable interfaces. This fixing operation ensures the server maintains a stable installation position within the rack, preventing positional shifts caused by vibration or external forces. After the pushing and fixing operations are completed, the server reaches a final state of precise alignment with the target installation position, forming the server assembly posture. This posture meets the rack's mechanical constraints and electrical connection requirements, achieving a complete installation operation between the server and the target rack.

[0091] In this embodiment, precise positioning and mapping are achieved through multi-sensor fusion SLAM navigation technology. A 3D vision system is used to accurately identify the target cabinet and obtain its installation location information. A feature matching algorithm calculates the deviation parameters between the server's current pose and the target installation position, and based on deviation compensation, the posture of the cargo platform and the elevator body is coordinated and adjusted. This invention, through an integrated multi-dimensional pose compensation control mechanism, achieves fully automated control of the mobile robot from navigation and positioning to precise alignment, effectively improving the positioning accuracy and success rate of automatic server installation.

[0092] The above describes the control method of a mobile robot for server handling in an embodiment of the present invention. The following describes the control device of a mobile robot for server handling in an embodiment of the present invention. The mobile robot includes a mobile chassis, a lifting body, a body head, a cargo-carrying and assembly tool unit, and an electrical control cabinet. Please refer to [link to relevant documentation]. Figure 2 One embodiment of the mobile robot control device for server handling in this invention includes:

[0093] The navigation and mapping module 201 is used to control the mobile chassis and the head of the machine to perform environmental perception when it receives the server installation instruction from the supercomputing center, and to perform positioning and mapping based on the environmental perception.

[0094] The location recognition module 202 is used to control the mobile chassis to move to the target cabinet corresponding to the server installation command during the positioning and mapping process. When it reaches the target cabinet, it controls the loading and assembly tool unit to perform target cabinet location recognition and obtain the installation location information of the target cabinet.

[0095] The deviation calculation module 203 is used to calculate the deviation between the current pose of the server on the cargo and assembly tool unit and the installation position information, and to calculate the pose compensation parameters of the cargo and assembly tool unit based on the calculated deviation information.

[0096] The pose adjustment module 204 is used to perform pose adjustment processing on the cargo and assembly tool unit according to the pose compensation parameters, and to perform height adjustment processing on the lifting body to obtain the server assembly pose aligned with the target installation position, thereby realizing the installation operation between the server and the target cabinet.

[0097] In this embodiment of the invention, the mobile robot control device for server handling operates the aforementioned mobile robot control method for server handling. The device achieves precise positioning and mapping through multi-sensor fusion SLAM navigation technology, accurately identifies the target server rack and obtains its installation position information using a 3D vision system, calculates the deviation parameters between the server's current pose and the target installation position using a feature matching algorithm, and coordinates the pose adjustment of the loading platform and the lifting body based on deviation compensation. This invention, through an integrated multi-dimensional pose compensation control mechanism, achieves fully automated control of the mobile robot from navigation and positioning to precise alignment, effectively improving the positioning accuracy and success rate of automatic server installation.

[0098] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0099] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0100] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A mobile robot control method for server handling, characterized in that, The mobile robot includes a mobile chassis, a lifting body, a body head, a cargo-carrying and assembly tool unit, and an electrical control cabinet. The control method for the mobile robot used for server handling includes: When a server installation instruction is received from the supercomputing center, the mobile chassis and the head of the machine are controlled to perform environmental perception and to perform positioning and mapping based on the environmental perception. During the positioning and mapping process, the mobile chassis is controlled to move to the target cabinet corresponding to the server installation command. When it reaches the target cabinet, the loading and assembly tool unit is controlled to identify the target cabinet position and obtain the installation position information of the target cabinet. Geometric feature vectors of the server surface are extracted from the server point cloud data acquired by the 3D camera in the fuselage head to obtain a 3D feature description of the server's current pose. Multi-scale feature descriptor calculations are performed on the 3D feature description of the server's current pose using either the FPFH algorithm or the SHOT algorithm to obtain a multi-level feature descriptor matrix. Layered matching processing is performed between the multi-level feature descriptor matrix and a pre-built feature library of the server's corresponding CAD model to obtain a set of candidate matching point pairs. The RANSAC algorithm is used to verify the geometric consistency of the candidate matching point pairs and eliminate abnormal matches. Point pairs are matched to obtain reliable feature matching point pairs; the optimal rigid body transformation matrix is ​​calculated and pose parameters are solved by SVD singular value decomposition on the reliable feature matching point pairs to obtain the current precise pose of the server; coordinate system transformation and deviation calculation are performed on the current precise pose of the server and the installation position information to obtain six-degree-of-freedom deviation parameters; the kinematic model of the cargo platform and shaping and positioning components in the cargo and assembly tool unit is inversely solved according to the six-degree-of-freedom deviation parameters to calculate the control quantities of lateral movement distance and axial rotation angle to obtain the pose compensation parameters of the cargo and assembly tool unit; Based on the posture compensation parameters, motion control processing is performed on the cargo platform and shaping and positioning components in the cargo and assembly tool unit to obtain the adjusted posture of the cargo platform; based on the installation position information, lifting control processing is performed on the heavy-duty precision linear guide assembly and heavy-duty silent ball screw assembly in the elevator body to obtain the adjusted posture of the elevator body; multi-axis coordinated motion control processing is performed on the adjusted postures of the cargo platform and the elevator body to obtain the coordinated motion state of the cargo and assembly tool unit and the elevator body; based on the coordinated motion state, the six-axis collaborative robotic arm and the robotic arm dexterous hand are controlled to perform server pushing and fixing operations to obtain the server assembly posture aligned with the target installation position, thereby realizing the installation operation between the server and the target rack.

2. The mobile robot control method for server handling according to claim 1, characterized in that, When a server installation instruction is received from the supercomputing center, controlling the mobile chassis and the head of the machine to perform environmental perception, and performing positioning and mapping based on the environmental perception includes: The server installation instructions are parsed and processed to extract the location identifier of the target cabinet and the server specification parameters to obtain the navigation task parameters; According to the navigation task parameters, the laser obstacle avoidance component and safety edge component in the mobile chassis are controlled to perform environmental data acquisition and processing, and the high-position safety laser component in the fuselage head is controlled to perform high-position environmental data acquisition and processing to obtain multi-level environmental perception data. Feature extraction and map construction are performed on the multi-level environmental perception data, and robot pose estimation calculation is performed to obtain a real-time environmental map and current position coordinates; Based on the real-time environment map and current location coordinates, combined with the navigation task parameters, a global path planning algorithm is performed to calculate the optimal path in a static environment, thereby obtaining the global navigation path and basic motion control commands.

3. The mobile robot control method for server handling according to claim 2, characterized in that, During the positioning and mapping process, the mobile chassis is controlled to move to the target rack corresponding to the server installation command. Upon reaching the target rack, the loading and assembly tool unit is controlled to identify the target rack's location and obtain the target rack's installation location information, including: The electric casters in the mobile chassis are controlled to roll and turn according to the global navigation path and basic motion control commands, thereby enabling the mobile chassis to move. During the movement of the mobile chassis, the 2D depth camera system in the cargo and assembly tool unit is controlled to perform dynamic environmental monitoring and processing to obtain real-time position feedback and dynamic obstacle avoidance correction commands. The real-time position feedback and dynamic obstacle avoidance correction commands are used to make real-time adjustments to the movement process. During the movement, the image data acquired by the 2D depth camera system is processed to identify the target cabinet, and it is determined whether the mobile robot has reached the target cabinet. If so, the 3D vision system in the loading and assembly tool unit is controlled to perform a three-dimensional scan of the target cabinet to obtain the three-dimensional geometric features of the target cabinet. The installation location information of the target cabinet is obtained by calculating the installation location based on the three-dimensional geometric features using geometric feature recognition and spatial positioning algorithms.

4. The mobile robot control method for server handling according to claim 3, characterized in that, The step of calculating the installation location of the target cabinet based on the three-dimensional geometric features using geometric feature recognition and spatial positioning algorithms includes: The three-dimensional geometric features of the target cabinet are processed by dual 3D camera data fusion, and the reference geometric features of the positioning reference object of the target installation position are generated by stereo vision algorithm. Based on the geometric features of the reference object, the target installation location feature recognition process is performed, and the spatial coordinates and geometric parameters of the target installation location are extracted and calculated by a geometric pattern recognition algorithm. The installation attitude is calculated by determining the spatial coordinates and geometric parameters of the target installation location, and the standard attitude parameters for server insertion are calculated by combining the internal guide rail direction and depth constraints of the target cabinet, thus obtaining the installation location information of the target cabinet.

5. A mobile robot control device for server handling, characterized in that, For implementing the mobile robot control method for server handling according to any one of claims 1-4, the mobile robot includes a mobile chassis, a lifting body, a body head, a cargo-carrying and assembly tool unit, and an electrical control cabinet; the mobile robot control device for server handling includes: The navigation and mapping module is used to control the mobile chassis and the head of the machine to perform environmental perception when it receives the server installation instruction from the supercomputing center, and to perform positioning and mapping based on the environmental perception. The location recognition module is used to control the mobile chassis to move to the target cabinet corresponding to the server installation command during the positioning and mapping process. When it reaches the target cabinet, it controls the loading and assembly tool unit to perform target cabinet location recognition and obtain the installation location information of the target cabinet. The deviation calculation module is used to calculate the deviation between the current pose of the server on the cargo and assembly tool unit and the installation position information, and to calculate the pose compensation parameters of the cargo and assembly tool unit based on the calculated deviation information. The pose adjustment module is used to adjust the pose of the cargo and assembly tool unit according to the pose compensation parameters, and to adjust the height of the lifting body to obtain the server assembly pose aligned with the target installation position, thereby realizing the installation operation between the server and the target cabinet.

Images