A quadruped robot and a mechanical arm cooperative unknown environment ultra-wideband autonomous station deployment and self-supervised positioning method, system, terminal and medium

Abstract

The application discloses a kind of four-legged robot and mechanical arm collaborative unknown environment under ultra-wideband autonomous station and self-supervision positioning method, system, terminal and medium, method includes: acquisition environment three-dimensional point cloud, generates candidate position set of base station layout, filters layout position, in accordance with the layout position screened out, executes autonomous grabbing and accurate placement operation, combines robot pose and kinematics and calculates base station initial coordinate, utilizes UWB ranging information between base station and carries out partition independent optimization, obtains the base station coordinate after optimization;With SLAM trajectory as pseudo true value, combined with environmental point cloud data, automatically generate LOS label or NLOS label, construct UWB positioning model;Real-time receive UWB ranging sequence, utilize the UWB positioning model constructed, realize high-precision positioning.The application can realize the closed-loop cooperation of environment perception, base station layout and positioning model training, can improve the positioning reliability in unknown environment and robot operation autonomy.

Description

Technical Field

[0001] This invention relates to the field of autonomous localization and environmental perception technology for mobile robots, and in particular to a method, system, terminal and medium for ultra-wideband autonomous deployment and self-supervised localization in unknown environments in collaboration between a quadruped robot and a robotic arm. Background Technology

[0002] As lunar exploration, deep-earth resource development, disaster relief, and industrial inspection missions continue to advance into unknown and extreme environments, autonomous operation technologies relying on mobile robots have been extensively studied. However, in typical unknown scenarios such as the lunar surface, deep-earth mines, and post-disaster ruins, there are neither global navigation satellite system signals nor pre-deployed positioning infrastructure. Traditional ultra-wideband positioning technologies that rely on fixed base stations are difficult to apply directly, urgently requiring robots to have the ability to autonomously build positioning base stations in unknown environments. In contrast, quadruped robots possess superior terrain adaptability and mobile exploration capabilities, while collaborative robotic arms have precise operational capabilities. The deep integration of these two provides a new technological path for the autonomous deployment of positioning base stations by quadruped robots and robotic arms in unknown environments.

[0003] Existing UWB positioning systems mostly assume that base station locations are known and the environmental structure is prior, using methods such as genetic algorithms and particle swarm optimization to search for the optimal deployment combination from a set of candidate locations. However, these methods all rely on manually preset base station candidate locations and a global prior map, essentially still falling under the category of human decision-making, and cannot enable robots to autonomously decide on base station deployment locations in unknown environments. Meanwhile, most existing research on robotic arm operations focuses on fixed workstations in known structured environments. The mobile robot's environmental exploration and the robotic arm's physical operations are independent at the algorithm level, lacking a collaborative mechanism for simultaneous execution of exploration and operation, making it difficult to support robots in autonomously completing the entire process of base station deployment.

[0004] Furthermore, in complex indoor environments, UWB positioning faces a severe decline in non-line-of-sight (NLOS) accuracy. While existing data-driven deep learning methods can effectively suppress NLOS errors, they generally rely on large-scale manually labeled data, and obtaining labeled ground truth data in unknown environments is itself a major challenge.

[0005] Therefore, existing technologies still have shortcomings. Summary of the Invention

[0006] To address the aforementioned shortcomings of existing technologies, this invention provides a method, system, terminal, and medium for ultra-wideband autonomous deployment and self-supervised localization in unknown environments, utilizing a quadruped robot and a robotic arm in collaboration. The technical solution adopted by this invention is as follows: In a first aspect, the present invention provides a method for ultra-wideband autonomous deployment and self-supervised localization in unknown environments using a quadruped robot and a robotic arm in collaboration, the method comprising: The system collects 3D point clouds of the environment, automatically identifies spatial partitions, boundary corners, and regional connectivity points, generates a set of candidate locations for base station deployment, and selects deployment locations according to the rule of prioritizing connected regions and then supplementing spatial corners. Based on the selected deployment locations, the robotic arm performs autonomous grasping and precise placement operations, and calculates the initial coordinates of the base station by combining the robot's pose and kinematics. It also uses UWB ranging information between base stations to perform independent optimization of each partition, and obtains the optimized base station coordinates. Using SLAM trajectories as pseudo-ground values, LOS or NLOS labels are automatically generated by combining environmental point cloud data, and a UWB localization model suitable for the current environment is constructed using a self-supervised training method. The system receives UWB ranging sequences in real time, uses the constructed UWB positioning model to output the weights and correction coefficients of each base station, uses the correction coefficients to correct the original ranging values, and substitutes the corrected ranging values ​​and corresponding weights into a weighted least squares framework to achieve high-precision positioning.

[0007] In one implementation, a 3D point cloud of the environment is acquired, and spatial partitions, boundary corners, and regional connectivity points are automatically identified to generate a set of candidate locations for base station deployment, including: The quadruped robot uses a lidar-equipped device to collect real-time 3D point clouds of the environment. Vertical obstacles are extracted using RANSAC-based planar segmentation, and the overall environment is divided into several sub-spatial regions by connected component analysis to obtain spatial partitioning results. Boundary corner points are extracted for each subspace region through convex hull operation and polygon approximation, and passable connecting areas between subspace regions are extracted from the ground point cloud as connected points to form a set of candidate locations for base station deployment.

[0008] In one implementation, the placement locations are selected by prioritizing connected regions and then supplementing spatial corner points. Based on the selected locations, the robotic arm performs autonomous grasping and precise placement operations. This includes: After obtaining the set of candidate locations for base station deployment, the deployment locations are selected according to the rule of prioritizing connected regions and then supplementing spatial corner points; The quadruped robot's robotic arm uses an end-effector camera for vision guidance to grab a base station from a container on the side of the quadruped robot. Based on an optimal planar search algorithm, the grabbed base station is then precisely placed at the selected deployment location.

[0009] In one implementation, the depth camera is rigidly connected to the end effector of the robotic arm, and the transformation relationship from the camera coordinate system to the end effector coordinate system of the robotic arm is pre-calibrated using hand-eye calibration, wherein the hand-eye calibration matrix is... ,satisfy:

[0010] in, This is the pose of the robotic arm's end effector. The pose of the calibration board as observed by the camera; The hand-eye calibration matrix is ​​used to describe the rigid transformation between the camera coordinate system and the end-effector coordinate system of the robotic arm.

[0011] In one implementation, the initial coordinates of the base station are calculated by combining the robot's pose and kinematics. Then, UWB ranging information between base stations is used for independent optimization in different zones to obtain the optimized base station coordinates, including: By combining the quadruped robot's pose and the robotic arm's kinematics, the initial estimated position of the base station in the world coordinate system is calculated through forward kinematics, thus obtaining the initial coordinates of the base station. Using the UWB ranging information between base stations collected during the base station deployment process, the base station coordinates are grouped and optimized independently according to the spatial partitioning results. The L-BFGS-B method is used to solve the problem and obtain the optimized base station coordinates.

[0012] In one implementation, SLAM trajectories are used as pseudo-ground values, and LOS or NLOS labels are automatically generated by combining environmental point cloud data. A UWB localization model suitable for the current environment is constructed using self-supervised training, including: Using SLAM trajectories as pseudo-true values, linear interpolation is used to record the associated spatial location for each UWB ranging information; Dynamic object points are removed by a dynamic point filtering method based on local density. Line segments between the tag location and the base station are sampled at equal intervals. LOS or NLOS tags are automatically generated based on the comparison between the maximum continuous occlusion length and a threshold. Using LOS or NLOS labels as supervision signals, a CNN-BiLSTM-MultiheadAttention feature extraction network is trained, and a random forest regressor is trained using the extracted joint features to construct a UWB localization model.

[0013] In one implementation, the method further includes: During the deployment of base stations, SLAM trajectory, LiDAR point cloud map, and real-time ranging data between the placed base stations and UWB tags on the quadruped robot are collected simultaneously during the movement of the quadruped robot. This results in a set of UWB ranging sequences with associated spatial location information, as well as corresponding environmental point cloud data.

[0014] Secondly, embodiments of the present invention also provide an ultra-wideband autonomous station deployment and self-supervised localization system for unknown environments in which a quadruped robot and a robotic arm collaborate, wherein the system is used to implement the steps of the ultra-wideband autonomous station deployment and self-supervised localization method for unknown environments in which a quadruped robot and a robotic arm collaborate as described in any of the above solutions, and the system includes: The base station deployment and precise placement module is used to collect 3D point clouds of the environment, automatically identify spatial partitions, boundary corners and regional connectivity points, generate a set of candidate locations for base station deployment, and select deployment locations according to the rule of prioritizing connected regions and then supplementing spatial corners. Based on the selected deployment locations, the robotic arm performs autonomous grasping and precise placement operations, and calculates the initial coordinates of the base station by combining the robot's pose and kinematics. It also uses UWB ranging information between base stations to perform independent partition optimization to obtain the optimized base station coordinates. The UWB localization model building module is used to automatically generate LOS or NLOS labels by combining SLAM trajectories as pseudo-ground values ​​with environmental point cloud data, and to build a UWB localization model suitable for the current environment using a self-supervised training method. The UWB positioning solution module is used to receive UWB ranging sequences in real time, use the constructed UWB positioning model to output the weights and correction coefficients of each base station, use the correction coefficients to correct the original ranging values, and substitute the corrected ranging values ​​and corresponding weights into the weighted least squares framework to achieve high-precision positioning.

[0015] Thirdly, embodiments of the present invention also provide a terminal, wherein the terminal includes a memory, a processor, and an ultra-wideband autonomous station deployment and self-supervised localization program for unknown environments in which a quadruped robot and a robotic arm collaborate, stored in the memory and capable of running on the processor, wherein when the processor executes the ultra-wideband autonomous station deployment and self-supervised localization program for unknown environments in which a quadruped robot and a robotic arm collaborate, the steps of the ultra-wideband autonomous station deployment and self-supervised localization method for unknown environments in which a quadruped robot and a robotic arm collaborate, as described in any of the above-mentioned schemes, are implemented.

[0016] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a program for ultra-wideband autonomous station deployment and self-supervised localization in an unknown environment in collaboration between a quadruped robot and a robotic arm, the program for ultra-wideband autonomous station deployment and self-supervised localization in an unknown environment in collaboration between a quadruped robot and a robotic arm implementing the steps of the method for ultra-wideband autonomous station deployment and self-supervised localization in an unknown environment in collaboration with a quadruped robot and a robotic arm as described in any of the above-mentioned schemes on the computer-readable storage medium.

[0017] Beneficial Effects: Compared with existing technologies, this invention provides a method for autonomous base station deployment and self-supervised localization in unknown environments using a quadruped robot and a robotic arm in collaboration. First, a 3D point cloud of the environment is collected, and spatial partitions, boundary corners, and regional connectivity points are automatically identified to generate a set of candidate locations for base station deployment. Deployment locations are selected based on the rule of prioritizing connected regions and then supplementing spatial corners. According to the selected locations, the robotic arm performs autonomous grasping and precise placement operations. The initial coordinates of the base stations are calculated by combining the robot's pose and kinematics. Independent optimization of each partition is performed using UWB ranging information between base stations to obtain optimized base station coordinates. Using SLAM trajectories as pseudo-ground values, LOS or NLOS labels are automatically generated based on environmental point cloud data. A UWB localization model suitable for the current environment is constructed using self-supervised training. UWB ranging sequences are received in real time. Using the constructed UWB localization model, the weights and correction coefficients of each base station are output. The original ranging values ​​are corrected using the correction coefficients. The corrected ranging values ​​and corresponding weights are substituted into a weighted least squares framework to achieve high-precision localization. This invention enables closed-loop coordination of environmental perception, base station deployment, and positioning model training, which can improve the positioning reliability and robot autonomy in unknown environments. Attached Figure Description

[0018] Figure 1 This is a flowchart of a preferred embodiment of the ultra-wideband autonomous deployment and self-supervised localization method for a quadruped robot and robotic arm working in an unknown environment, according to an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the structured perception results of environmental point clouds in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of hand-eye calibration in the ultra-wideband autonomous deployment and self-supervised localization method for a quadruped robot and robotic arm working in an unknown environment, according to an embodiment of the present invention.

[0021] Figure 4 This is a technical roadmap for an ultra-wideband autonomous deployment and self-supervised localization method in an unknown environment, which is a manifestation of the present invention, involving a quadruped robot and a robotic arm working together.

[0022] Figure 5 This is a schematic diagram of the workspace and near-vertical grasping in the ultra-wideband autonomous deployment and self-supervised localization method for a quadruped robot and robotic arm working in an unknown environment, according to an embodiment of the present invention.

[0023] Figure 6 This is a schematic diagram of base station location estimation in the ultra-wideband autonomous deployment and self-supervised localization method for unknown environments in collaboration between a quadruped robot and a robotic arm, according to an embodiment of the present invention.

[0024] Figure 7This is a technical framework diagram of an ultra-wideband autonomous deployment and self-supervised positioning system for unknown environments, which is a collaborative system of a quadruped robot and a robotic arm according to an embodiment of the present invention.

[0025] Figure 8 A schematic diagram of a terminal provided in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0027] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content, operations, or steps, nor does it require execution in the described order. For example, some operations or steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0028] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0029] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. For example, "first control information" and "second control information" are only used to distinguish different control information and do not limit their order.

[0030] Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or the order of execution, and that the words "first" and "second" do not necessarily imply that they are different.

[0031] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0032] The method in this embodiment can be applied to a terminal, which can be an intelligent electronic device such as a computer. Figure 1 As shown in the figure, the ultra-wideband autonomous deployment and self-supervised localization method for unknown environments in which a quadruped robot and a robotic arm cooperate in collaboration specifically includes the following steps: Step S100: Collect 3D point cloud data of the environment, automatically identify spatial partitions, boundary corners and regional connectivity points, generate a set of candidate locations for base station deployment, and select deployment locations according to the rule of prioritizing connected regions and then supplementing spatial corners. Based on the selected deployment locations, the robotic arm performs autonomous grasping and precise placement operations, and calculates the initial coordinates of the base station by combining the robot pose and kinematics. The UWB ranging information between base stations is used to perform independent optimization of the partitions to obtain the optimized base station coordinates.

[0033] To achieve autonomous deployment of base stations in unknown spaces using a quadruped robot and a robotic arm in a collaborative manner, this embodiment includes an integrated hardware and software system consisting of an embodied intelligent hardware integration platform and two core software modules: 1. Embodied Intelligent Hardware Integration Platform: This embodiment constructs an integrated mobile and operational hardware platform for autonomous deployment of base stations in unknown environments using a quadruped robot and a robotic arm in a collaborative manner. The hardware includes: a quadruped robot, a six-degree-of-freedom collaborative robotic arm, a depth camera, a LiDAR, and several UWB positioning tags and base station modules. The core innovation of this platform lies in the deep integration of the quadruped robot's all-terrain mobile exploration capabilities, the collaborative robotic arm's high-precision physical manipulation capabilities, and the environmental perception capabilities of multi-source sensors, forming a physical execution base with a closed loop of mobile, perception, and operation. In traditional methods, base station deployment relies on manual on-site work, with environmental exploration and base station operation performed separately by independent devices; however, this embodiment integrates mobile exploration, environmental perception, and physical manipulation into a single robot platform, enabling the robot to simultaneously perform environmental exploration, structural analysis, and base station deployment tasks in unknown environments.

[0034] Specifically, this embodiment uses a quadruped robot as the mobile platform. The quadruped robot possesses active joint actuation capabilities, enabling it to perform various gaits such as walking, trotting, and jumping. It exhibits unparalleled mobility in unstructured terrain such as stairs, gravel, and slopes compared to wheeled platforms. Unlike wheeled platforms that passively adapt to terrain, the quadruped robot, through active adjustment of its leg joints, can maintain horizontal stability on rough ground, providing a relatively stable operating base for the robotic arm mounted on it. The quadruped robot is responsible for carrying the UWB (Ultra Wideband) base station module and, based on its structured perception of the environment (such as...), ... Figure 2 As shown in the diagram, the system navigates sequentially to each of the deployed target locations (spatial corners or doorway connections). Its navigation and positioning are provided by laser SLAM (Simultaneous Localization and Mapping), enabling autonomous positioning and path tracking in unknown environments without GNSS (Global Navigation Satellite System) signals.

[0035] In this embodiment, a six-DOF collaborative robotic arm is fixedly mounted on the back of a quadruped robot. A depth camera is mounted at the end effector of the robotic arm, forming a vision-guided configuration based on the end effector camera. The core design of this configuration is that the depth camera is rigidly connected to the end effector of the robotic arm, and the transformation relationship from the camera coordinate system to the end effector coordinate system is pre-calibrated through hand-eye calibration. A schematic diagram of hand-eye calibration is shown below. Figure 3 As shown in the figure. Let the hand-eye calibration matrix be... ,satisfy:

[0036] in, The end effector pose is given by robot kinematics in the robot's base coordinate system. The pose of the calibration board is observed by the camera, and the pose of the calibration board is obtained through camera visual recognition. The hand-eye calibration matrix describes the rigid transformation from the camera coordinate system to the robotic arm's end effector coordinate system. This calibration process is performed once before the mission begins, and the calibration results are reused throughout the deployment mission. The robotic arm is responsible for the autonomous grasping and precise placement of the base station, and is the key actuator for completing the physical deployment loop.

[0037] In this embodiment, a 16-line LiDAR is mounted on the front of a quadruped robot, with a horizontal field of view of 360° and a vertical field of view of 30°, used for large-scale environmental mapping and structured perception. The LiDAR collects point cloud data at a frequency of 10Hz and constructs a global point cloud map in real time using a LiDAR SLAM algorithm. Figure 1 On the one hand, it is used for the localization and navigation of the quadruped robot itself, and on the other hand, it provides spatial geometric priors for deployment location decisions, including environmental structural information such as wall positions, room outlines, and doorway connections.

[0038] In this embodiment, the depth camera is mounted at the end of the robotic arm, with an RGB image resolution of 1920×1080 and a depth resolution of 1280×720, for fine-grained near-field perception. After the quadruped robot navigates to the roughly positioned location, the depth camera acquires RGB-D images at a frequency of 30Hz, providing visual data support for target detection and placement plane search.

[0039] The quadruped robot carries several UWB base station modules (such as LinkTrack PB modules), which are autonomously grasped and precisely placed at key locations in the environment by a robotic arm. Each UWB base station module has a built-in high-precision clock and wireless communication module, and is in a ready-to-deploy state after being powered on. After the base station module is grasped, transported, and placed at the planned location by the robotic arm, the system records the location coordinates and number of the base station, marking it as deployed, and only then can the base station be used for subsequent positioning. The base stations and the UWB tags carried by the quadruped robot together constitute a complete positioning system, providing absolute position reference for subsequent operations.

[0040] In practical applications, to address the two core issues of base station deployment location decision-making and placement execution in unknown environments, traditional UWB deployment methods rely on global prior maps and manually preset candidate locations, which cannot be directly applied in unknown environments. This embodiment proposes a progressive deployment strategy: during the exploration process of a quadruped robot, the environmental point cloud structure is analyzed in real time, automatically identifying spatial partitions, boundary corners, and regional connectivity points to generate a set of candidate locations for base station deployment. Deployment locations are then selected according to the rule of prioritizing connected regions and supplementing spatial corners. Subsequently, a robotic arm performs autonomous grasping and precise placement, and the initial coordinates of the base station are calculated by combining the robot's pose and kinematics. Finally, UWB ranging information between base stations is used for independent partition optimization.

[0041] To achieve a structured representation of the environment and standardized extraction of its locations, the overall space needs to be rationally divided into zones. Combined with... Figure 4 As shown in the diagram, this embodiment uses a LiDAR mounted on a quadruped robot to collect 3D point clouds of the environment in real time. Then, RANSAC (Random Sample Consensus)-based planar segmentation is used to extract vertical obstacles. Combined with connected component analysis, the overall environment is divided into several sub-space regions, resulting in spatial partitioning. Next, boundary corner points of each sub-space region are extracted using convex hull operations and polygon approximation. Passable connecting regions between sub-space regions are extracted from the ground point cloud as connected points, forming a set of candidate locations for base station deployment.

[0042] Specifically, the RANSAC method constructs a planar model through random sampling and determines interior points based on the distance error from a point to the plane. Its core judgment condition is:

[0043] in, The unit normal vector of the fitted plane. ; Let be any point in the point cloud, and be the coordinates of that point in 3D space. ; For the plane offset term, the plane equation The constant term in; For point The Euclidean distance to the fitted plane; This is the distance threshold used in RANSAC plane segmentation to determine the distance to an interior point. When the distance from a point to the plane is less than the threshold, it is determined to be an interior point in the current plane.

[0044] To further address the merging failure problem caused by normal vector error, an improved point-to-surface distance difference condition based on the centroid of the patch and the normal vector is introduced:

[0045] in, This is the difference between the point and the surface, used to determine whether two vertical surfaces belong to the same plane. It is a three-dimensional vector between the centroids of two vertical obstacle facets to be merged. , Let be the global normal vector of the two faces. When the difference is less than a set threshold, the two faces are determined to belong to the same plane and a merging operation is performed. Next, after projecting the vertical obstacle points onto a two-dimensional plane, density clustering is used to aggregate the discrete vertical obstacle fragments, forming effective partitions with spatial segmentation capabilities. Based on the effective partitions, a two-dimensional occupancy grid is constructed, and the overall environment is divided into [variables] using connected component analysis. Each is an independent subspace region.

[0046] In the process of spatial corner point identification and connected point detection, this embodiment extracts the boundary contour for each effective subspace and then fits the minimum convex polygon of the region using convex hull operation. (Convex hull) Defined as a set containing the coordinates of the center points of all subspace grids. Minimal convex set:

[0047] in, : Convex hull, containing the set of points The smallest convex polygon, The coefficients are convex combination coefficients, satisfying The weighting coefficients.

[0048] Furthermore, this embodiment employs the Douglas-Peucker polygon approximation algorithm to normalize the convex hull contour, by setting a fitting accuracy threshold. Redundant points in the contour are removed, retaining only corner points that characterize the spatial boundary. The fitting accuracy threshold is dynamically calculated proportionally to the contour perimeter.

[0049] in, The threshold for fitting accuracy is the point removal threshold in polygon approximation. This is the accuracy scaling factor, a scaling factor that controls the fitting accuracy. is the perimeter of the convex hull profile, and the total length of the convex hull boundary.

[0050] The core principle of polygon approximation is: traverse all points on the contour, and if the perpendicular distance from a point to the line connecting its two adjacent points is less than 1 / 3, then the approximation is considered complete. If the distance is greater than or equal to a certain value, the point is discarded; only points with a distance greater than or equal to a certain value are retained. The key points are used as corner points.

[0051] In base station deployment, relying solely on spatial corner points can easily lead to problems such as base stations serving only a single sub-area, severe signal attenuation when penetrating walls, and redundant deployment numbers. Connected areas, as accessible hubs shared by multiple sub-spaces, are the optimal locations for deploying shared base stations. Connectivity point identification combines the spatial distribution characteristics of ground point clouds and non-vertical obstacle points. This embodiment first extracts the ground reference height from the point cloud:

[0052] in, The ground reference height is the average height of the ground point cloud. This represents the number of ground reference points, the number of sampling points used to calculate ground height. For the height coordinates of a single point, the first... The height values ​​of each ground sampling point.

[0053] Subsequently, this embodiment removes structural points of vertical obstacles, retains valid points of non-vertical obstacles on the ground layer, and performs neighborhood spatial matching in a two-dimensional plane to locate connected regions between zones. The coordinates of the center point of the connected region are obtained by weighted averaging of the points within the region:

[0054] in, The coordinates of the center point of the connected region are the weighted average position of all points within the connected region. The number of points within a connected region is the total number of sampling points constituting the connected region. Let be the coordinates of a single point within a connected region, and let be the coordinates of the first point within the connected region. The three-dimensional coordinates of each point.

[0055] Furthermore, after obtaining the set of candidate locations for base station deployment, this implementation selects deployment locations according to the rule of prioritizing connected regions and then supplementing spatial corner points. Next, combined with... Figure 4 As shown, dynamic deployment decisions are made, that is, the robotic arm of the quadruped robot grasps the base station from the container on the side of the quadruped robot through vision guidance by the end-effector camera, and accurately places the grasped base station at the selected deployment position based on the optimal planar search algorithm.

[0056] Specifically, this embodiment follows the rule of prioritizing connected points and supplementing corner points to select deployment locations: when navigation reaches the vicinity of the target corner point and a valid connected area is detected, the base station is preferentially deployed to the center of the connected area; if there is no valid connected point in the current area, the optimal geometric deployment rule is continued to be used to deploy the base station at the spatial corner point.

[0057] Let the actual number of base stations deployed within the subspace area be . The number of corner points identified in this area is Then the dynamic quantity constraint can be expressed as:

[0058] in, The actual number of base stations deployed is the total number of UWB base stations already installed in the sub-region. The number of corner points represents the total number of spatial corner points identified within the sub-region. This constraint inherits the traditional method's basic requirement of deploying at least four base stations in small spaces to meet the 3D positioning calculation, while also allowing for adaptive adjustment of the deployment scale based on the number of corner points in large-scale spaces.

[0059] When performing base station grasping and placement, the robotic arm employs an end-effector camera configuration. A rigid transformation relationship needs to be established between the camera coordinate system and the end-effector coordinate system, i.e., the hand-eye calibration matrix. :

[0060] The meanings of the parameters are the same as in formula (1).

[0061] In this embodiment, when capturing base stations, YOLOv8-seg target detection is used to identify base station objects, and point cloud instance segmentation is performed by combining depth images to obtain a refined point cloud of the target. Candidate grasping poses are generated using the GPD framework. Select the optimal grasping pose:

[0062] in, For the set of candidate grasp poses, the GPD framework generates all feasible grasp poses. For a single candidate capture pose, set The first in Grab the pose. To achieve the optimal grasping pose, the quality function is used. The optimal grasping pose was determined through evaluation. The capture quality function is a scoring function that evaluates the quality of candidate capture poses.

[0063] After determining the optimal grasping pose, a near-vertical constraint is applied to the grasping pose (the angle between the gripper direction and gravity ≤ 30°), enabling the robotic arm to approach the base station using an approximately straight-line trajectory from top to bottom. This naturally avoids the risk of collision with the container wall, and the workspace and near-vertical grasping are optimized. Figure 5 As shown in the image.

[0064] The algorithm performs a layered traversal along the vertical direction near the approximate placement location, using DBSCAN clustering to identify candidate planar regions. First, clusters with projected areas smaller than a threshold are filtered out; then, the remaining candidate regions are... In the middle, prioritize distance. Minimal clustering; if distances are equal, choose the area cluster. Larger clustering. Based on the MoveIt! tool, it plans with the goal of optimal grasping pose and optimal placement location, allowing up to 3 grasping retries.

[0065] Furthermore, this embodiment performs base station location estimation and independent zone optimization. Specifically, combined with Figure 6 As shown in the figure. This embodiment combines the quadruped robot's pose and the robotic arm's kinematics to calculate the initial position of the base station:

[0066] in, The initial location of the base station is estimated, and the base station coordinates are derived from kinematics. The position of the quadruped robot is its position in the world coordinate system (provided by SLAM). This represents the rotation matrix for the quadruped robot's posture, specifically the rotation matrix from the robot's body frame to the world frame. . This refers to the offset of the robotic arm base, which is the fixed offset of the robotic arm base within the robot's body. The end effector offset is the kinematic offset from the robot arm base to the end effector (calculated from the joint angle). The gripper offset is the fixed offset from the end effector of the robotic arm to the gripping point at the base station.

[0067] This embodiment utilizes UWB ranging information between base stations to perform independent optimization based on spatial partitioning results. Let the... The area contains One base station, UWB ranging matrix satisfy The objective function is optimized as follows:

[0068] in: For UWB ranging values, the first Base stations in the area With base station The distance between them is measured by UWB. For the base station coordinates, the first The first in the region The three-dimensional coordinates of each base station. Ranging noise, random error in UWB measurements. Let be the set of base station coordinates, the first... The set of coordinate vectors of all base stations within the region. is the regularization coefficient, which controls the weight of the kinematic priors in the optimization. The kinematic coordinates are derived using the prior values ​​of the base station coordinates obtained through kinematic derivation using formula (11). The optimization problem in this embodiment is solved using the L-BFGS-B method to obtain the optimized base station coordinates. The L-BFGS-B algorithm is a numerical method for large-scale nonlinear optimization problems, particularly suitable for problems with boundary constraints, and is a variant of the L-BFGS algorithm. The key steps of the L-BFGS-B algorithm include: initialization, quasi-Newton update, line search, boundary handling, and termination conditions. This algorithm uses special techniques in each iteration to ensure that the updated parameters satisfy the boundary conditions, making it suitable for engineering and scientific problems with upper and lower bounds on the parameters.

[0069] During the deployment of base stations, SLAM trajectory (providing pseudo-true values ​​of position), LiDAR point cloud map (providing environmental geometric information), and real-time ranging data between the placed base stations and UWB tags on the quadruped robot are collected simultaneously during the movement of the quadruped robot. This results in a set of UWB ranging sequences with associated spatial location information, as well as the corresponding environmental point cloud data.

[0070] Step S200: Using SLAM trajectories as pseudo-true values, automatically generate LOS or NLOS labels by combining environmental point cloud data, and construct a UWB localization model suitable for the current environment using self-supervised training.

[0071] After base station deployment and coordinate optimization, a UWB positioning model suitable for the current environment is automatically constructed without the need for manual data annotation. Traditional supervised learning methods rely on large-scale manually labeled data, which is difficult to obtain in unknown environments. This invention uses SLAM trajectories as pseudo-ground values ​​and records the associated spatial location of each UWB ranging information record through linear interpolation. Then, a dynamic point filtering method based on local density is used to remove dynamic object points, and the line segments between the tag location and the base station are sampled at equal intervals. Based on the comparison between the maximum continuous occlusion length and a threshold, LOS or NLOS tags are automatically generated. LOS (Line of Sight) and NLOS (Non-Line of Sight) are two key concepts in wireless communication and positioning technology used to describe the signal propagation path. LOS refers to a direct line-of-sight path between the transmitter and receiver without any obstacles. NLOS indicates that there are obstacles between the transmitter and receiver, and the signal cannot reach the receiver directly, but can only reach the receiver through reflection, diffraction, or other propagation methods. Next, using LOS or NLOS labels as supervision signals, the CNN-BiLSTM-MultiheadAttention feature extraction network is trained, and the extracted joint features are used to train a random forest regressor to construct the UWB localization model. The CNN-BiLSTM-MultiheadAttention network is a deep learning model that combines a convolutional neural network (CNN), a bidirectional long short-term memory neural network (BiLSTM), and a multi-head attention mechanism.

[0072] Specifically, let the SLAM trajectory sequence be... The UWB ranging sequence is For any UWB sampling time The spatial location at that moment is obtained through linear interpolation:

[0073] in: This is a SLAM trajectory sequence, including timestamps. and location The time sequence. This is a UWB ranging sequence, including timestamps. and distance measurement value The time sequence. The interpolated position at the UWB time point is the pseudo-spatial position corresponding to the UWB sampling time point obtained through linear interpolation. This interpolation result associates a high-confidence pseudo-spatial position value with each UWB ranging record.

[0074] After obtaining the tag location, it is necessary to further determine whether the signal propagation path between that location and each base station is obstructed by environmental structures. The original point cloud contains non-static points introduced by dynamic objects, which can constitute false alarm interference in occlusion detection. This embodiment uses a dynamic point filtering method based on local density. For the downsampled point set... any point in Its density is defined as... Center, radius The number of sampling points contained in the XY plane neighborhood:

[0075] in, For point density, with points Center, radius The number of sampling points in the neighborhood. For the points whose density needs to be calculated, the point cloud Any point in it. For point The projected coordinates of the point Two-dimensional projection on the XY plane. The radius is used to calculate density, and the neighborhood radius is used to calculate local point density. A density threshold is set. Points with a density below the threshold are identified as dynamic points and removed to obtain a static structured point cloud. .

[0076] For label position With base station location The line segments between points are sampled at equal intervals, and the minimum Euclidean distance from each sampling point to the static point cloud is calculated. An occlusion indicator function is defined, and a binary occlusion state sequence is obtained after traversing all sampling points along the path. A discrimination criterion based on the length of consecutive occlusion segments is introduced, where the length of the consecutive occlusion segment is denoted as . The maximum value of the length of all consecutive occlusion segments is taken as the core metric for the degree of occlusion. Ultimately based on With threshold The comparison generates pseudo-tags:

[0077] in: For LOS / NLOS tags, Indicates NLOS state, This indicates a LOS (Lost in Serving) status. : Length of continuous occlusion segment, the first The number of sampling points or physical length of a segment with continuous occlusion. Maximum continuous occlusion length: the maximum length of all continuous occlusion segments on the path. Occlusion length threshold: The minimum continuous occlusion length for determining NLOS tags. The core idea of ​​this discrimination mechanism is that only a blockage of a certain physical length will have a substantial impact on the NLOS label of the UWB signal, while the momentary occlusion of isolated sampling points should not lead to NLOS misjudgment.

[0078] Next, this embodiment will continue The feature vectors of adjacent time points are stacked along the time axis to form the input sequence. In terms of spatial feature extraction, the ranging values ​​of each base station are organized into a two-dimensional grid image according to their projected coordinates on the XY plane. This continuous... The two-dimensional distance images at each time step are stacked into a three-dimensional tensor. Feature extraction is performed using a multi-layer two-dimensional convolutional neural network, and the feature vectors are flattened into spatial feature vectors by global average pooling. .

[0079] For temporal feature extraction, a bidirectional long short-term memory (BiLSTM) network is used to model the input sequence. BiLSTM processes the sequence in both forward and reverse temporal directions: the forward LSTM captures the trend of ranging values ​​from the past to the future, while the backward LSTM captures abrupt changes in the signal from the future to the past. After concatenating the hidden states from both directions, the output at each time step... It also encodes the context information before and after that moment.

[0080] A multi-head self-attention mechanism is introduced on top of BiLSTM. The output sequence of BiLSTM is... The input is a multi-head self-attention layer. Each attention head independently calculates the dependency weights within the sequence. The results from each head are then concatenated and linearly transformed to output the global context feature vector. This mechanism enables the model to directly correlate the current time with similar signal patterns from several times ago, and to assign higher weights to ranging abrupt changes containing key location correction information at the boundary moments of NLOS events.

[0081] Spatial features extracted by CNN Temporal characteristics of multi-head self-attention output The data is then stitched together, and a gated fusion mechanism is used to adaptively adjust the contribution ratio of each feature channel.

[0082] in, : Fusion feature vectors, gated joint features after fusion. : Gating vector, a vector that controls the degree of passage of each feature channel, with element values ​​in the range [0,1]. For element-wise multiplication, use the Hadamard product (element-wise multiplication). : Original feature vector, the original input features without CNN and LSTM processing. ,in, The sigmoid activation function maps the output of the gated network to the [0,1] interval. : Gated network weight matrix, learnable weight parameters. : Gated network bias term, a learnable bias parameter.

[0083] In this embodiment, the training data is divided into a training set, a validation set, and an internal test set in a 7:1.5:1.5 ratio. The UWB localization model training employs a phased strategy: the first phase uses self-supervised pseudo-labels as supervision signals to train the CNN-BiLSTM-MultiheadAttention feature extraction network; the second phase fixes the deep network parameters and uses the extracted joint features to train a random forest regressor, thus obtaining the UWB localization model. The loss function is defined as the weighted sum of the weight prediction error and the correction coefficient prediction error:

[0084] in, The total loss function is the weighted sum of the prediction errors of the weighted prediction errors and the prediction errors of the correction coefficients. This represents the total number of base stations, specifically the number of UWB base stations already deployed in the system. The weights are true values, and the base station weight pseudo-labels are generated through a self-supervised method (usually set according to LOS / NLOS status). , where is the predicted base station weights output by the random forest regressor. The pseudo-labels of the correction coefficients are generated through a self-supervised method to provide the true values ​​of the correction coefficients (1.0 in LOS state and >1 in NLOS state). The predicted correction coefficients are the output of the random forest regressor. The weighting coefficient is used to control the weight of the prediction error of the correction coefficient in the total loss.

[0085] Step S300: Receive UWB ranging sequences in real time, use the constructed UWB positioning model to output the weights and correction coefficients of each base station, use the correction coefficients to correct the original ranging values, and substitute the corrected ranging values ​​and corresponding weights into the weighted least squares framework to achieve high-precision positioning.

[0086] During the online localization phase after model deployment, the system receives UWB ranging sequences in real time. Since ranging data from future moments is unavailable during online localization, BiLSTM uses only historical data for one-way forward propagation: for the current moment... , the past The feature vector sequence at time n The input network uses only the forward LSTM to extract temporal features, while the backward LSTM is disabled or only uses observed historical data during online localization. This allows the model to... The computational complexity is reduced, and the weights and correction coefficients at the current moment are output in real time, meeting the real-time requirements of the robot platform for positioning delay.

[0087] Specifically, the original ranging value is corrected using the predicted correction coefficient: Then, the corrected ranging values ​​and their corresponding weights are substituted into the weighted least squares framework:

[0088] in, The original UWB ranging value is the corrected ranging value. Correction coefficient The corrected result. The original UWB ranging value, time Measured base stations The original distance between the robot tag and the robot tag. For correction coefficients, base station The ranging correction coefficients are output by the random forest regressor. For base station weights, base station The reliability weights are output by the random forest regressor. For the base station coordinates, the optimized first... The three-dimensional world coordinates of each base station. The timing is obtained by solving the weighted least squares method to determine the location result. The robot's position estimation is performed. This nonlinear optimization problem is solved iteratively using the Levenberg-Marquardt algorithm to obtain accurate positioning results.

[0089] Finally, this embodiment performs moving average filtering on the positioning results at multiple consecutive time points to suppress high-frequency jitter:

[0090] in, This is the filtered positioning result; the robot position after being smoothed by a moving average filter. The sliding window size is the window length for the moving average filter. To locate the result of a historical moment, the moment The original location estimate.

[0091] The ultra-wideband autonomous base station deployment and self-supervised localization method for unknown environments based on the quadruped robot and robotic arm collaboration in this embodiment can realize closed-loop collaboration of environmental perception, base station deployment and localization model training, so as to improve the localization reliability and robot operation autonomy in unknown environments.

[0092] Based on the above embodiments, the present invention also provides an ultra-wideband autonomous deployment and self-supervised localization system for unknown environments, where a quadruped robot and a robotic arm cooperate. The system in this embodiment can be used to implement the steps in the above method embodiments. Specifically, as... Figure 7 As shown, the system includes: a base station deployment and precise placement module 10, a UWB positioning model construction module 20, and a UWB positioning calculation module 30. Specifically, the base station deployment and precise placement module 10 is used to collect 3D point clouds of the environment, automatically identify spatial partitions, boundary corners, and regional connectivity points, generate a set of candidate locations for base station deployment, and select deployment locations according to the rule of prioritizing connected regions and then supplementing spatial corners. Based on the selected deployment locations, a robotic arm performs autonomous grasping and precise placement operations, and calculates the initial coordinates of the base station by combining the robot's pose and kinematics. It then uses UWB ranging information between base stations to perform independent partition optimization to obtain the optimized base station coordinates. The UWB positioning model construction module 20 is used to automatically generate LOS or NLOS labels using SLAM trajectories as pseudo-ground values ​​and combined with environmental point cloud data, and constructs a UWB positioning model suitable for the current environment using self-supervised training. The UWB positioning solution module 30 is used to receive UWB ranging sequences in real time, use the constructed UWB positioning model to output the weights and correction coefficients of each base station, use the correction coefficients to correct the original ranging values, and substitute the corrected ranging values ​​and corresponding weights into the weighted least squares framework to achieve high-precision positioning.

[0093] The principles of each module in the embodiment of the ultra-wideband autonomous deployment and self-supervised positioning system in an unknown environment, which is a collaboration between a quadruped robot and a robotic arm, are the same as the implementation process of each step in the above method embodiment, and will not be elaborated further here.

[0094] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 8 As shown. The terminal may include one or more processors 100 ( Figure 8(Only one is shown in the image), memory 101, and computer program 102 stored in memory 101 and executable on one or more processors 100. For example, a program for ultra-wideband autonomous deployment and self-supervised localization in an unknown environment with quadruped robot and robotic arm collaboration. When one or more processors 100 execute computer program 102, they can implement the various steps in the embodiment of the method for ultra-wideband autonomous deployment and self-supervised localization in an unknown environment with quadruped robot and robotic arm collaboration. Alternatively, when one or more processors 100 execute computer program 102, they can implement the functions of each module / unit in the embodiment of the device for ultra-wideband autonomous deployment and self-supervised localization in an unknown environment with quadruped robot and robotic arm collaboration, which is not limited here.

[0095] In one embodiment, the processor 100 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0096] In one embodiment, memory 101 can be an internal storage unit of the terminal, such as a hard disk or RAM. Memory 101 can also be an external storage terminal of the terminal, such as a plug-in hard disk, smart media card (SM), secure digital card (SD), flash card, etc., all equipped on the terminal. Furthermore, memory 101 can include both internal and external storage units. Memory 101 is used to store computer programs and other programs and data required by the terminal. Memory 101 can also be used to temporarily store data that has been output or will be output.

[0097] Those skilled in the art will understand that Figure 8 The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0098] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), direct memory bus RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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; and these 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 method for ultra-wideband autonomous deployment and self-supervised localization in unknown environments using a quadruped robot and a robotic arm in collaboration, characterized in that, The method includes: The system collects 3D point clouds of the environment, automatically identifies spatial partitions, boundary corners, and regional connectivity points, generates a set of candidate locations for base station deployment, and selects deployment locations according to the rule of prioritizing connected regions and then supplementing spatial corners. Based on the selected deployment locations, the robotic arm performs autonomous grasping and precise placement operations, and calculates the initial coordinates of the base station by combining the robot's pose and kinematics. It also uses UWB ranging information between base stations to perform independent optimization of each partition, and obtains the optimized base station coordinates. Using SLAM trajectories as pseudo-ground values, LOS or NLOS labels are automatically generated by combining environmental point cloud data, and a UWB localization model suitable for the current environment is constructed using a self-supervised training method. The system receives UWB ranging sequences in real time, uses the constructed UWB positioning model to output the weights and correction coefficients of each base station, uses the correction coefficients to correct the original ranging values, and substitutes the corrected ranging values ​​and corresponding weights into a weighted least squares framework to achieve high-precision positioning.

2. The method for ultra-wideband autonomous deployment and self-supervised localization in unknown environments using a quadruped robot and robotic arm in collaboration, as described in claim 1, is characterized in that... The system acquires 3D point clouds of the environment, automatically identifies spatial partitions, boundary corners, and regional connectivity points, and generates a set of candidate locations for base station deployment, including: The quadruped robot uses a lidar-equipped device to collect real-time 3D point clouds of the environment. Vertical obstacles are extracted using RANSAC-based planar segmentation, and the overall environment is divided into several sub-spatial regions by connected component analysis to obtain spatial partitioning results. Boundary corner points are extracted for each subspace region through convex hull operation and polygon approximation, and passable connecting areas between subspace regions are extracted from the ground point cloud as connected points to form a set of candidate locations for base station deployment.

3. The method for ultra-wideband autonomous deployment and self-supervised localization in unknown environments using a quadruped robot and robotic arm in collaboration, as described in claim 2, is characterized in that... The deployment locations are selected based on a rule that prioritizes connected regions and then supplements spatial corner points. Following the selected locations, the robotic arm performs autonomous grasping and precise placement operations. This includes: After obtaining the set of candidate locations for base station deployment, the deployment locations are selected according to the rule of prioritizing connected regions and then supplementing spatial corner points; The quadruped robot's robotic arm uses an end-effector camera for vision guidance to grab a base station from a container on the side of the quadruped robot. Based on an optimal planar search algorithm, the grabbed base station is then precisely placed at the selected deployment location.

4. The method for ultra-wideband autonomous deployment and self-supervised localization in unknown environments using a quadruped robot and robotic arm in collaboration, as described in claim 3, is characterized in that... The depth camera is rigidly connected to the end effector of the robotic arm. The transformation relationship from the camera coordinate system to the end effector coordinate system of the robotic arm is pre-calibrated through hand-eye calibration, and the hand-eye calibration matrix is ​​as follows: ,satisfy: in, This is the pose of the robotic arm's end effector. The pose of the calibration board as observed by the camera; The hand-eye calibration matrix is ​​used to describe the rigid transformation between the camera coordinate system and the end-effector coordinate system of the robotic arm.

5. The method for ultra-wideband autonomous deployment and self-supervised localization in unknown environments using a quadruped robot and robotic arm in collaboration, as described in claim 4, is characterized in that... By combining robot pose and kinematics to calculate the initial coordinates of the base station, and using UWB ranging information between base stations to perform independent optimization in different zones, the optimized base station coordinates are obtained, including: By combining the quadruped robot's pose and the robotic arm's kinematics, the initial estimated position of the base station in the world coordinate system is calculated through forward kinematics, thus obtaining the initial coordinates of the base station. Using the UWB ranging information between base stations collected during the base station deployment process, the base station coordinates are grouped and optimized independently according to the spatial partitioning results. The L-BFGS-B method is used to solve the problem and obtain the optimized base station coordinates.

6. The method for ultra-wideband autonomous deployment and self-supervised localization in unknown environments using a quadruped robot and robotic arm in collaboration, as described in claim 5, is characterized in that... Using SLAM trajectories as pseudo-ground values, and combining them with environmental point cloud data to automatically generate LOS or NLOS labels, a UWB localization model suitable for the current environment is constructed using a self-supervised training approach, including: Using SLAM trajectories as pseudo-true values, linear interpolation is used to record the associated spatial location for each UWB ranging information; Dynamic object points are removed by a dynamic point filtering method based on local density. Line segments between the tag location and the base station are sampled at equal intervals. LOS or NLOS tags are automatically generated based on the comparison between the maximum continuous occlusion length and a threshold. Using LOS or NLOS labels as supervision signals, a CNN-BiLSTM-MultiheadAttention feature extraction network is trained, and a random forest regressor is trained using the extracted joint features to construct a UWB localization model.

7. The method for ultra-wideband autonomous deployment and self-supervised localization in unknown environments using a quadruped robot and robotic arm in collaboration, as described in claim 6, is characterized in that... The method further includes: During the deployment of base stations, SLAM trajectory, LiDAR point cloud map, and real-time ranging data between the placed base stations and UWB tags on the quadruped robot are collected simultaneously during the movement of the quadruped robot. This results in a set of UWB ranging sequences with associated spatial location information, as well as corresponding environmental point cloud data.

8. A system for ultra-wideband autonomous deployment and self-supervised localization in unknown environments, characterized in that: The system is used to implement the steps of the ultra-wideband autonomous deployment and self-supervised localization method for quadruped robot and robotic arm collaboration in unknown environments as described in any one of claims 1-7. The system includes: The base station deployment and precise placement module is used to collect 3D point clouds of the environment, automatically identify spatial partitions, boundary corners and regional connectivity points, generate a set of candidate locations for base station deployment, and select deployment locations according to the rule of prioritizing connected regions and then supplementing spatial corners. Based on the selected deployment locations, the robotic arm performs autonomous grasping and precise placement operations, and calculates the initial coordinates of the base station by combining the robot's pose and kinematics. It also uses UWB ranging information between base stations to perform independent partition optimization to obtain the optimized base station coordinates. The UWB localization model building module is used to automatically generate LOS or NLOS labels by combining SLAM trajectories as pseudo-ground values ​​with environmental point cloud data, and to build a UWB localization model suitable for the current environment using a self-supervised training method. The UWB positioning solution module is used to receive UWB ranging sequences in real time, use the constructed UWB positioning model to output the weights and correction coefficients of each base station, use the correction coefficients to correct the original ranging values, and substitute the corrected ranging values ​​and corresponding weights into the weighted least squares framework to achieve high-precision positioning.

9. A terminal, characterized in that, The terminal includes a memory, a processor, and an ultra-wideband autonomous deployment and self-supervised localization program for an unknown environment in which a quadruped robot and a robotic arm collaborate. The processor executes the ultra-wideband autonomous deployment and self-supervised localization program for an unknown environment in which a quadruped robot and a robotic arm collaborate, and implements the steps of the ultra-wideband autonomous deployment and self-supervised localization method for an unknown environment in which a quadruped robot and a robotic arm collaborate as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an ultra-wideband autonomous station deployment and self-supervised localization program for an unknown environment in which a quadruped robot and a robotic arm cooperate. The ultra-wideband autonomous station deployment and self-supervised localization program for an unknown environment in which a quadruped robot and a robotic arm cooperate implements the steps of the ultra-wideband autonomous station deployment and self-supervised localization method for an unknown environment in which a quadruped robot and a robotic arm cooperate, as described in any one of claims 1-7, on the computer-readable storage medium.

Images