Wheel-legged robot autonomous navigation and three-dimensional mapping method, device, equipment and medium

Abstract

The application discloses a kind of wheel-foot robot autonomous navigation and three-dimensional mapping method, device, equipment and medium, the method includes based on several key frames that wheel-foot robot is collected and constructs topological graph, in each planning period, according to node utility and travel cost from the topological graph selects several access nodes to construct access sequence;The node in the topological graph is the key pose extracted from key frame, the edge in the topological graph is the executable connection between node, generate candidate view point set in the local field of view corresponding to the access node, and according to node utility and travel cost from the candidate view point set selects access view point to construct view point access sequence;Control the wheel-foot robot moves according to the view point access sequence, and updates the topological graph and constructs three-dimensional environment graph.The application is configured with the node utility based on unknown space gain and degradation risk determination, realizes the path planning of real-time planning using laser radar point cloud in unknown space, considers mapping degradation risk simultaneously, avoids state estimation unstable and accumulative drift, and improves the precision of three-dimensional mapping.

Description

Technical Field

[0001] This application relates to the field of mobile robot technology, and in particular to a method, apparatus, equipment and medium for autonomous navigation and 3D mapping of a wheeled robot. Background Technology

[0002] With the continuous expansion of mobile robot applications in scenarios such as underground space inspection, complex building mapping, disaster relief, and industrial operation and maintenance, 3D mapping technology based on vehicle-mounted or airborne platforms has been extensively studied. However, in structurally constrained environments such as multi-story buildings, narrow corridors, stairwells, and narrow passages, traditional wheeled or tracked platforms are limited by their mobility and terrain adaptability, making it difficult to reliably complete high-quality mapping tasks. In contrast, wheeled-legged robots combine efficient wheeled movement with legged traversal capabilities, exhibiting stronger passability and environmental adaptability in complex terrains, and are widely used in 3D mapping of complex environments.

[0003] Existing 3D mapping systems utilizing wheeled robots generally employ a tightly coupled fusion framework based on LiDAR and inertial measurement units to achieve position estimation and map construction. However, in long corridors, large-area planar walls, or repetitive structures, the geometric constraints of point clouds exhibit significant orientation degradation characteristics, leading to unstable state estimation and cumulative drift, which in turn affects the accuracy of 3D mapping.

[0004] Therefore, existing technologies still need to be improved and enhanced. Summary of the Invention

[0005] The technical problem to be solved by this application is to provide a method, device, equipment and medium for autonomous navigation and 3D mapping of wheeled and legged robots, addressing the shortcomings of existing technologies.

[0006] To address the aforementioned technical problems, the first aspect of this application provides a method for autonomous navigation and 3D mapping of a wheeled robot, wherein the method specifically includes: A topology graph is constructed based on several key frames collected by the wheeled robot. The nodes in the topology graph are key poses extracted from the key frames, and the edges in the topology graph are executable connections between nodes. Each node carries at least a node utility determined based on unknown spatial gain and degradation risk. Within each planning cycle, several access nodes are selected from the topology graph based on node utility and travel cost to construct an access sequence. A candidate viewpoint set is generated within the local field of view corresponding to the access node, and an access viewpoint is selected from the candidate viewpoint set based on node utility and travel cost to construct a viewpoint access sequence. The wheeled robot is controlled to move according to the viewpoint access sequence, and the topology map is updated to construct a three-dimensional environment map.

[0007] The autonomous navigation and 3D mapping method for wheeled robots, wherein the key pose extraction process specifically includes: The wheel-to-ground contact point of each leg of the wheel-legged robot is determined based on the motion data of the wheel-legged robot, and the vertical fusion residual is determined based on the wheel-to-ground contact point; The odometry residual of the wheeled robot is determined based on the IMU data collected by the wheeled robot and the motion data. The registration residuals of the point cloud data collected by the wheeled robot are obtained, and the degradation risk is constructed by utilizing the uncertainty of point positioning and the uncertainty of voxel plane fitting. Based on the degradation risk, a weighting coefficient is constructed, and the weighting coefficient is used to solve the optimal error state for the vertical fusion residual, the odometry residual of the wheeled robot, and the registration residual, so as to obtain the attitude error. Based on the attitude error and the key attitude of the previous frame, the key attitude is determined.

[0008] The aforementioned method for autonomous navigation and 3D mapping of a wheeled robot, wherein determining the wheel-to-ground contact point of each leg of the wheeled robot based on its motion data specifically includes: The initial contact point is determined based on the joint angle vector and joint length of each leg of the wheel-legged robot; The initial contact point is geometrically corrected using the local terrain normal and wheel radius of the pre-stored probabilistic voxel map to obtain the wheel-ground contact point for each leg.

[0009] The aforementioned method for autonomous navigation and 3D mapping of wheeled robots, wherein determining the vertical fusion residual based on the wheel-ground contact point specifically includes: Obtain the projected height of the wheel-ground contact point on the local voxel plane; The height error of the wheel-ground contact point is determined based on the projected height, and a vertical fusion residual is constructed based on the height error.

[0010] The aforementioned method for autonomous navigation and 3D mapping of wheeled robots, wherein the construction of degradation risk using point positioning uncertainty and voxel plane fitting uncertainty specifically includes: The adaptive variance is determined by utilizing the uncertainty of point positioning and the uncertainty of voxel plane fitting. A degradation intensity index is constructed based on the adaptive variance, and the degradation risk is determined based on the degradation intensity index.

[0011] The aforementioned autonomous navigation and 3D mapping method for a wheeled robot, wherein the wheeled robot is equipped with a reinforcement learning-based control strategy, and controlling the wheeled robot to move according to the viewpoint access sequence specifically includes: For each accessed viewpoint in the viewpoint access sequence, the eigenvalues ​​corresponding to the covariance matrix of the local point cloud data of the accessed viewpoint are obtained, and the degradation factor of the accessed viewpoint is determined based on the eigenvalues. Based on the degradation factor, the current posture of the wheeled robot, and the current observation state, the current action of the wheeled robot is determined by the control strategy, and the wheeled robot is controlled according to the current action.

[0012] The aforementioned method for autonomous navigation and 3D mapping of a wheeled robot, wherein updating the topology map to construct a 3D environment map specifically includes: Candidate key nodes are generated at preset intervals of frames, and the node distance between the candidate key nodes and existing nodes in the topology graph is obtained. Candidate key nodes whose distance to other nodes is greater than a preset distance threshold are inserted into the topology graph, and edges are constructed for the inserted candidate key nodes with a preset vertical gating constraint to update the topology graph.

[0013] The second aspect of this application provides an autonomous navigation and 3D mapping device for a wheeled robot, wherein the autonomous navigation and 3D mapping device for a wheeled robot specifically includes: A construction module is used to construct a topology graph based on several key frames collected by a wheeled robot. The nodes in the topology graph are key poses extracted from the key frames, and the edges in the topology graph are executable connections between nodes. Each node carries at least node utility, which is determined based on unknown spatial gain and degradation risk. The global processing module is used to select several access nodes from the topology graph to construct an access sequence in each planning cycle, based on node utility and travel cost. The local processing module is used to generate a set of candidate viewpoints within the local field of view corresponding to the access node, and select access viewpoints from the candidate viewpoints based on node utility and travel cost to construct a viewpoint access sequence. The control module is used to control the wheeled robot to move according to the viewpoint access sequence and update the topology map to construct a three-dimensional environment map.

[0014] A third aspect of this application provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the autonomous navigation and 3D mapping method for wheeled robots as described above.

[0015] A fourth aspect of this application provides a terminal device, which includes: a processor and a memory; The memory stores a computer-readable program that can be executed by the processor; When the processor executes the computer-readable program, it implements the steps in the autonomous navigation and 3D mapping method for wheeled and legged robots as described above.

[0016] Beneficial effects: 1. This application configures node utility based on unknown spatial gain and degradation risk, realizing real-time path planning in unknown space using lidar point cloud while considering mapping degradation risk, avoiding unstable state estimation and cumulative drift, and improving the accuracy of 3D mapping.

[0017] 2. This application utilizes the motion data of a wheeled robot to determine the vertical fusion residual, and performs uncertainty-weighted voxel plane matching on the point cloud data collected by the wheeled robot, and introduces a degradation risk quantity to adaptively reweight the multi-source constraints. Even when the observability of the lidar decreases or the wheel slip disturbance increases, it still maintains a stable and continuous estimated attitude, avoiding the positioning divergence problem caused by the low inter-frame overlap of the lidar and the reduced inter-frame registration accuracy, thus improving the accuracy of 3D mapping.

[0018] 3. This application utilizes a lightweight scene degradation factor directly calculated from LiDAR frames, and uses a control strategy obtained through reinforcement learning to control the wheeled robot based on this degradation factor, thereby reducing the Z-axis drift of the odometry and autonomously reducing the intensity of motion when the registration difficulty between point cloud frames increases, thus achieving robust autonomous mapping for challenging terrains. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating the autonomous navigation and 3D mapping method for a wheeled robot provided in this application embodiment.

[0021] Figure 2 This is a schematic diagram illustrating the principle framework of the autonomous navigation and 3D mapping method for wheeled robots provided in the embodiments of this application.

[0022] Figure 3 Example diagram of the construction process for access sequences and viewpoint access sequences.

[0023] Figure 4 This is a schematic diagram of a wheeled robot.

[0024] Figure 5 This is a schematic diagram illustrating the technology of lidar environmental perception and planar detection for wheeled robots.

[0025] Figure 6 This is a timing diagram for motion compensation technology.

[0026] Figure 7 This is a scene illustration.

[0027] Figure 8 This is a schematic diagram of the autonomous navigation and 3D mapping device for a wheeled robot provided in an embodiment of this application.

[0028] Figure 9 A schematic block diagram of the terminal device provided in the embodiments of this application. Detailed Implementation

[0029] This application provides a method, apparatus, device, and medium for autonomous navigation and 3D mapping of a wheeled robot. To make the objectives, technical solutions, and effects of this application clearer and more explicit, the following detailed description is provided with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application.

[0030] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0031] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0032] It should be understood that the sequence number and size of each step in this embodiment do not imply the order of execution. The execution order of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of this application embodiment.

[0033] Research has revealed that as the application of mobile robots continues to expand in scenarios such as underground space inspection, complex building surveying, disaster relief, and industrial operation and maintenance, 3D mapping technology based on vehicle-mounted or airborne platforms has been extensively studied. However, in structurally constrained environments such as multi-story buildings, narrow corridors, stairwells, and narrow passages, traditional wheeled or tracked platforms are limited by their mobility and terrain adaptability, making it difficult to reliably complete high-quality mapping tasks. In contrast, wheeled-legged robots combine the efficient movement of wheels with the traversing capabilities of legs, exhibiting stronger passability and environmental adaptability in complex terrains, and are widely used in 3D mapping of complex environments.

[0034] Existing 3D mapping systems utilizing wheeled robots generally employ a tightly coupled fusion framework based on LiDAR and inertial measurement units to achieve position estimation and map construction. However, in environments such as long corridors, large-area planar walls, or repetitive structures, the geometric constraints of point clouds exhibit significant orientation degradation characteristics, leading to unstable state estimation and cumulative drift, which in turn affects the accuracy of 3D mapping.

[0035] To address the aforementioned issues, in this embodiment, a topology map is constructed based on several keyframes collected by a wheeled robot. Within each planning cycle, several visiting nodes are selected from the topology map based on node utility and travel cost to construct a visiting sequence. The nodes in the topology map are key poses extracted from the keyframes, and the edges are executable connections between nodes. A candidate viewpoint set is generated within the local field of view corresponding to each visiting node, and visiting viewpoints are selected from the candidate viewpoint set based on node utility and travel cost to construct a viewpoint visiting sequence. The wheeled robot is controlled to move according to the viewpoint visiting sequence, and the topology map is updated to construct a 3D environment map. This application configures node utility based on unknown spatial gain and degradation risk, enabling real-time path planning in unknown space using LiDAR point clouds while considering mapping degradation risk. This avoids unstable state estimation and cumulative drift, improving the accuracy of 3D mapping.

[0036] The application content will be further explained below with reference to the accompanying drawings and the description of the embodiments.

[0037] This embodiment provides a method for autonomous navigation and 3D mapping of a wheeled robot, such as... Figure 1 and Figure 2 As shown, the method includes: S10. Construct a topology map based on several key frames collected by the wheeled robot.

[0038] Specifically, the topology graph is an online topology graph maintained at the global layer, which is incrementally updated during the movement of the wheeled robot. Nodes in the topology graph represent key poses of the wheeled robot extracted from keyframes, and edges represent executable connections between nodes. These executable connections are based on obstacle avoidance constraints, wheeled leg kinematic constraints, and terrain traversability constraints. Furthermore, through... The shortest feasible cost is calculated once on the accessible voxel projection map and cached. During local incremental updates of the map (i.e., topology updates), only the affected edges are lazily updated to control online computation overhead and stabilize planning latency.

[0039] The topological graph can be represented as: , in, Representing a topological graph, Represents a set of nodes. Represents the set of edges.

[0040] Furthermore, each node in the topology map carries node attributes, which include not only node utility but also mapping gain and terrain description. The node utility, determined based on unknown space gain and degradation risk, reflects the node's comprehensive value in improving the overall quality of the 3D environment map within the current planning cycle. Unknown space gain is the proportion of unknown voxels within the node's field of view to the total number of visible voxels at that node, reflecting the amount of new environmental information the node brings to the map. Degradation risk is determined based on the degree of degradation of the geometric constraints of the LiDAR point cloud at the node, reflecting the uncertainty risk that may arise when estimating the state of the node's location.

[0041] For example, node utility can be represented as: , in, Indicates the first The node's utility and Indicates the weighting coefficient. Indicates the first Unknown spatial gain of nodes, Indicates the first Risk of node degradation.

[0042] It should be noted that when the update cycle of a node in the topology graph is every frame, the node utility is determined by the degradation risk corresponding to the point cloud data of that frame. When the update cycle of a node in the topology graph is every preset number of frames (such as 5 frames), the node utility is determined by the average degradation risk of the point cloud data of each frame in the preset number of frames. This average degradation risk is used to reflect the degree of continuous degradation within the time period corresponding to the preset data frame.

[0043] In one embodiment, the key pose extraction process specifically includes: H10. Determine the wheel-to-ground contact point of each leg of the wheel-legged robot based on the motion data of the wheel-legged robot, and determine the vertical fusion residual based on the wheel-to-ground contact point; H20. Determine the wheel-foot odometry residual based on the IMU data collected by the wheel-foot robot and the motion data; H30. Obtain the registration residuals of the point cloud data collected by the wheeled robot, and construct the degradation risk using the uncertainty of point positioning and the uncertainty of voxel plane fitting; H40. Based on the degradation risk, construct weighting coefficients, and use the weighting coefficients to solve for the optimal error state of the vertical fusion residual, wheel foot odometer residual, and registration residual to obtain the attitude error. H50. Based on the attitude error and the key attitude of the previous frame, determine the key attitude.

[0044] In step H10, the wheel-to-ground contact point is the actual contact point between the wheel axle of the wheel-legged robot and the ground, which can be obtained by geometrically correcting the center of the wheel axle. That is, when determining the wheel-to-ground contact point of each leg of the wheel-legged robot based on its motion data, the motion data of the wheel-legged robot at the moment of point cloud data acquisition can be obtained. Then, the center of the wheel axle can be determined based on this motion data, and then geometrically corrected to obtain the wheel-to-ground contact point.

[0045] Based on this, determining the wheel-to-ground contact point of each leg of the wheel-legged robot based on the motion data of the wheel-legged robot specifically includes: The initial contact point is determined based on the joint angle vector and joint length of each leg of the wheel-legged robot; The initial contact point is geometrically corrected using the local terrain normal and wheel radius of the pre-stored probabilistic voxel map to obtain the wheel-ground contact point for each leg.

[0046] Specifically, the joint angle vector and joint length are the motion data of the wheel-legged robot at the moment of point cloud data acquisition, and the initial contact point is the center position of the wheel axle at the moment of point cloud data acquisition. For example, Figure 4The wheel-legged robot shown can have its axle center position obtained by a cascade of homogeneous transformations of the hip, thigh, and lower leg segments. In other words, the expression for the axle center position can be: , , in, Indicates the center position of the wheel axle. Represents the hip joint transformation matrix. Represents the thigh joint transformation matrix. This represents the transformation matrix of the lower leg joint. Indicates the length of the lower leg joint. This indicates the position of the leg reference point.

[0047] Furthermore, the complex triangular expansion can be abstracted into a nonlinear positive kinematic mapping, which simplifies the expression for the wheel-axle center position to: , , in, Represents the joint angle vector. Indicates joint length. This represents a nonlinear positive kinematic mapping.

[0048] Therefore, when obtaining the initial contact point, it is only necessary to obtain the joint angle vector and joint length of each leg of the wheel-legged robot to obtain the initial contact point of each leg. Among them, the joint angle vector of each leg includes the hip joint, thigh joint, and calf joint, and the joint length includes the calf joint length.

[0049] However, in practical applications, directly using the wheel axle center as the contact point may introduce systematic geometric deviations. Therefore, in this embodiment, the initial contact point is geometrically corrected for slope consistency using the local terrain normal from the probabilistic voxel map and the wheel radius to obtain the wheel-ground contact point. The wheel-ground contact point is represented as: , in, Indicates the wheel-to-ground contact point. Indicates the wheel radius. This represents the local terrain normal of the probabilistic voxel map. The transpose of the rotation matrix from the world coordinate system to the body coordinate system.

[0050] This application uses the local terrain normal and wheel radius from the probabilistic voxel map to perform geometric correction for the initial contact point to ensure slope consistency. This allows the contact geometry to maintain physical consistency under different slopes, suppressing height drift and velocity pseudo-observation bias caused by slope changes from the source.

[0051] It should be noted that in practical applications, the initial contact point can be directly selected as the wheel-ground contact point, etc., according to the actual usage situation.

[0052] Furthermore, in step H10 above, the vertical fusion residual is used to reflect the degree of deviation between the wheel-to-ground contact point and the vertical position of the wheel-footed robot's body coordinate system. The vertical fusion residual is the estimation error of the globally consistent base height determined using the wheel-to-ground contact point. Specifically, determining the vertical fusion residual based on the wheel-to-ground contact point includes: Obtain the projected height of the wheel-ground contact point on the local voxel plane; The height error of the wheel-ground contact point is determined based on the projected height, and a vertical fusion residual is constructed based on the height error.

[0053] Specifically, to address the Z-axis drift problem in long-distance walking and degenerative structures, the height of the body base is constructed from each effective supporting leg, where, for example... Figure 5 As shown, the height of the fuselage base corresponding to each leg can be expressed as: , in, Indicates the height of the fuselage base. Represents a unit vector. Indicates the projection height. This represents the rotation matrix from the world coordinate system to the body coordinate system.

[0054] After obtaining the height of the fuselage base, a maximum likelihood method is used to perform variance-weighted fusion of the fuselage base heights corresponding to all legs to obtain a globally consistent height error (i.e., base height estimation error). Then, the vertical fusion residual is constructed using this height error, and this vertical fusion residual is expressed as: , in, Indicates the vertical fusion residual. This represents the pose transformation matrix from the world coordinate system to the body coordinate system.

[0055] Furthermore, in step H20 above, the wheel-foot odometry residual is used to reflect the deviation between the wheel-foot robot's motion model and its actual motion state. The wheel-foot odometry residual is constructed by fusing the wheel-foot robot's kinematic model, IMU data, and motion data. The IMU data may include the gyroscope bias vector, the accelerometer bias vector, the gravity vector in the world coordinate system, and the linear velocity vector in the world coordinate system, etc. The wheel-foot odometry residual can be obtained using existing methods, and no specific restrictions are imposed here.

[0056] Furthermore, in step H30 above, the registration residual is the covariance matrix obtained by modeling the spatial distribution characteristics and noise level of the matched point pairs during the uncertainty-weighted voxel plane matching process based on point cloud data. The process of obtaining the registration residual can be as follows: First, each distortion-free point in the point cloud data Transform to the world coordinate system to obtain the distortion-free points in the world coordinate system. The distortion-free points in the world coordinate system are represented as follows: , in, Represents the distortion-free point in the world coordinate system. This represents the rotation matrix from the body coordinate system to the lidar coordinate system. This represents the translation vector from the body coordinate system to the lidar coordinate system. This represents the translation vector from the world coordinate system to the body coordinate system.

[0057] Then, the registration residual from point to plane is constructed using the voxel plane parameters. This registration residual can be expressed as:

[0058] in, Indicates the registration residual. This represents the transpose of the normal vector. It represents the distance from the plane to the origin along the normal direction.

[0059] In one embodiment, the construction of degradation risk using point positioning uncertainty and voxel plane fitting uncertainty specifically includes: The adaptive variance is determined by utilizing the uncertainty of point positioning and the uncertainty of voxel plane fitting. A degradation intensity index is constructed based on the adaptive variance, and the degradation risk is determined based on the degradation intensity index.

[0060] Specifically, the adaptive variance is determined by projecting both point location uncertainty and map plane fitting uncertainty onto the normal direction. Point location uncertainty can be approximated using filtered covariance, and map plane fitting uncertainty can be obtained using the output of a plane fitting algorithm, such as the residual covariance after RANSAC fitting. This adaptive variance can be expressed as: , in, Indicates adaptive variance. This indicates uncertainty in point positioning. This indicates the uncertainty in map plane fitting. Represents the normal vector.

[0061] Then, an approximate Hessian is constructed using the residual Jacobian, and the minimum eigenvalue is extracted to form a degradation intensity index. This degradation intensity index is then truncated, normalized, and exponentially smoothed to obtain the degradation risk. The degradation risk can be expressed as: , in, express The risk of degradation over time, express time, This represents the total accommodation coefficient. express The degradation intensity index at any given time. This represents the truncation function.

[0062] Furthermore, in step H40, in order to process the multi-rate characteristics of the IMU (high frequency), the wheel-leg body sensing (mid frequency), and the lidar (low frequency), such as... Figure 6 As shown, this embodiment employs a buffering and time alignment strategy. The IMU is responsible for state forward propagation and is used for intra-frame point cloud distortion correction. The wheel speed and height observations of the wheeled robot are continuously buffered within one frame period, and iterative error state filtering and updating are performed after the point cloud data frame acquired by the LiDAR is aligned with the point-to-plane residual at the end of the frame. Specifically, this iterative error state filtering and updating process is as follows: First, a weighting coefficient is determined based on the degradation risk, which triggers a reliability reweighting. Specifically, when the degradation risk is high, the weight of the wheel-leg slope sensing kinematic constraints (speed / height, etc.) is increased, reducing the dependence on the weakly constrained LiDAR direction; conversely, when the degradation risk is low, the weight of the LiDAR is increased to fully utilize the effective information from LiDAR observations. This avoids positioning instability in feature-scarce regions, ensuring the overall stability and accuracy of state estimation.

[0063] Secondly, the vertical fusion residual, the wheel foot odometer residual, and the registration residual are stacked into a unified residual and a composite Jacobian, wherein the unified residual and the composite Jacobian can be expressed as: , in, Indicates uniform residuals, Indicates the compound Jacobi. Indicates the registration residual. Jacobi representing the registration residual. Indicates the residual of the wheel foot odometer. Jacobian, representing the residual of the wheel foot odometer. Indicates the vertical fusion residual. Jacobi represents the vertical fusion residual.

[0064] Finally, the Mahalanobis norm containing prior and multi-source observations is minimized through iterative relinearization; in each iteration, the optimal error state solution is: , in, Indicates attitude error. Indicates the weighting coefficient. The information matrix representing the prior state.

[0065] Furthermore, in step H50, after obtaining the attitude error, the attitude error is applied to the key attitude of the previous frame to obtain the key attitude of the current frame, thus realizing robust continuous state output across terrain and degraded structures.

[0066] S20. In each planning cycle, select several access nodes from the topology graph based on node utility and travel cost to construct an access sequence.

[0067] Specifically, such as Figure 3 As shown, the access sequence is a global layer sequence that allows backtracking without returning to the starting point. This sequence includes several access nodes, and these nodes have an access order. The access sequence is obtained by solving an objective function constructed based on node utility and travel cost. The objective function corresponding to the access sequence is: , in, This represents the control coefficient. This represents the node utility of candidate nodes in the global layer. Indicates the distance between candidate nodes. This indicates the number of candidate nodes, which are selected from the topology graph (e.g., based on the distance to the current position of the wheeled robot).

[0068] Furthermore, the objective function can be solved using a real-time heuristic strategy. Specifically, an initial solution is first constructed using a nearest neighbor greedy approach, followed by 2-opt local improvements under a finite iterative budget. In addition, when the graph undergoes incremental changes due to node insertion, risk reduction, or local edge cost updates, incremental refinement is only performed on the affected road segments, avoiding the combined overhead of global recalculation every cycle, thereby maintaining online consistency.

[0069] This application constructs an objective function based on node utility and travel cost when determining the access sequence. While ensuring travel cost, it controls the access order of candidate nodes through node utility. For example, the current distance between candidate node A and the wheeled robot may be less than the current distance between candidate node B and the wheeled robot. However, due to numerous obstacles and narrow space near node A, the node utility of candidate node A is less than that of candidate node B. Therefore, candidate node B is accessed earlier than candidate node A. This reduces the impact of degradation on mapping and ensures the integrity of the mapping for the entire region.

[0070] S30. Generate a set of candidate viewpoints within the local field of view corresponding to the access node, and select access viewpoints from the set of candidate viewpoints based on node utility and travel cost to construct a viewpoint access sequence.

[0071] Specifically, after determining the access sequence, the local optimization layer treats each access node in the access sequence as the current sub-target, and then generates an executable waypoint sequence (i.e., a viewpoint access sequence) for that current sub-target (i.e., the access node) within a scrolling local window. For example... Figure 3 As shown, the local optimization layer employs viewpoint sampling and candidate viewpoint graph search when constructing the viewpoint access sequence for each access node. Specifically, this viewpoint sampling and candidate viewpoint graph search method is as follows: First, viewpoint sampling is performed within the local field of view corresponding to the access node to generate a candidate viewpoint set. The candidate viewpoints in the candidate viewpoint set are maintained in tuple form, and the tuple corresponding to each candidate viewpoint is represented as follows: , in, Indicates the position of the three-dimensional viewpoint. The viewpoint height represents the safety margin used to maintain an effective field of view and accommodate terrain undulations. This represents the set of uncovered points that can be covered from this candidate viewpoint. This represents the set of frontier points that can be covered from the candidate viewpoint. Includes collision and access flags.

[0072] Secondly, a joint optimization function is constructed based on node utility and travel cost, with the objectives of minimizing cost and maximizing coverage utility. This joint optimization is then solved to obtain the viewpoint access sequence. The joint optimization function can be expressed as: , in, Denotes the joint optimization function, This represents the number of candidate viewpoints in the candidate viewpoint set. This represents the cost of the shortest feasible path calculated on the local walkable graph. This represents the control parameters of the local optimization layer. The node utility of the candidate viewpoint.

[0073] Finally, the joint optimization function is solved to obtain the viewpoint access sequence, in which there are feasible connections between the accessing viewpoints in the viewpoint access sequence. These feasible connections incorporate wheel foot drivability constraints, obstacle safety constraints, and degradation risk feedback, thereby avoiding actions that are detrimental to mapping but geometrically achievable, such as frequent sharp turns and short-range reciprocating oscillations, in highly anisotropic degenerate structures such as corridors.

[0074] S40. Control the wheeled robot to move according to the viewpoint access sequence, and update the topology map to construct a three-dimensional environment map.

[0075] Specifically, after obtaining the viewpoint access sequence corresponding to each access point in the access sequence, the wheeled robot is controlled to move according to the viewpoint access sequence. Furthermore, when the wheeled robot moves from one access viewpoint to its next access viewpoint, it is controlled to move according to the viewpoint access sequence corresponding to that access viewpoint. In other words, the viewpoint access sequence corresponding to an access viewpoint is the access path from that access viewpoint to the next access viewpoint.

[0076] Furthermore, due to structurally constrained environments (e.g., such as...) Figure 7 In the long corridors, stairwells, and narrow passages shown, local geometry exhibits strong orientation anisotropy, affecting the observability of lidar-based SLAM and increasing the risk of state estimation degradation. Simultaneously, under these conditions, drastic motions (such as high-speed cruise or large-angle yaw maneuvers) can lead to wheel slip and transient attitude disturbances, further exacerbating odometry drift and disrupting global state estimation. Figure 1 To address the coupling problem between perception and motion, embodiments of this application deploy a reinforcement learning-based control strategy on a wheeled robot. This control strategy introduces a degradation factor, which is combined with policy observation and reward. This enables the controller equipped with the control strategy to autonomously switch to a more robust and conservative wheel-leg cooperative motion mode when the positioning risk increases, thereby stabilizing the front-end estimation during long-term mapping.

[0077] Based on this, in one embodiment, the wheeled robot is equipped with a reinforcement learning-based control strategy, and controlling the wheeled robot to move according to the viewpoint access sequence specifically includes: For each accessed viewpoint in the viewpoint access sequence, the eigenvalues ​​corresponding to the covariance matrix of the local point cloud data of the accessed viewpoint are obtained, and the degradation factor of the accessed viewpoint is determined based on the eigenvalues. Based on the degradation factor, the current posture of the wheeled robot, and the current observation state, the current action of the wheeled robot is determined by the control strategy, and the wheeled robot is controlled according to the current action.

[0078] Specifically, the input data for the control strategy includes the degradation factor, the current posture of the wheeled robot, and the current observation state, while the output data is the current action of the wheeled robot. The control strategy is derived from reinforcement learning, enabling the wheeled robot to automatically adjust its motion pattern in complex environments. Particularly when facing geometrically degraded and dynamically changing environments, this strategy guides the robot to select a smoother, more stable motion pattern in highly degraded regions, while allowing the robot to recover higher motion efficiency in regions with sufficient geometric constraints.

[0079] The degradation factor is a lightweight yet information-rich degradation descriptor constructed from local point cloud data. In each control step, the local point cloud is projected onto the body coordinate system. The plane yields the set of projection points as follows Then, calculate the covariance matrix of the projected point set and the corresponding eigenvalues ​​of the covariance matrix. The discrete parameters and direction-specific parameters are constructed using two eigenvalues, where the discrete parameters and geometric parameters are expressed as follows: , , in, This represents a discrete parameter used to reflect the dispersion of the spatial distribution. It represents geometry and is used to reflect the anisotropy of plane geometry.

[0080] After obtaining the discrete and geometric parameters, in order to suppress noise and instantaneous fluctuations, the discrete and geometric parameters are smoothed using an exponential moving average (EMA) to obtain the initial degradation factor. Then, the initial degradation factor is normalized to obtain the degradation factor, where the degradation factor is expressed as: , in, Indicates the degradation factor. Indicates the initial degradation factor. This represents the truncation function. This represents the lower limit of the initial degradation factor. This represents the upper limit of the initial degradation factor.

[0081] Furthermore, during the training of the control strategy, based on the degradation factor, the motion control problem of the wheeled robot is formulated as a partially observable Markov decision process (POMDP). The decision data of the partially observable Markov decision process is represented as follows: , Representing decision-making data, Representing the state space, Represents the action space. Represents the observation space. Represents the state transition probability. Represents the reward function, This represents the discount factor. Then, a control policy is trained using Proximal Policy Optimization (PPO). The cumulative discounted return that maximizes the expected return of this control strategy is: , in, This indicates the expected long-term return when choosing an action.

[0082] Then, by using the cumulative reward of discounts that maximize the expected value, the wheeled robot is trained to adaptively adjust its gait patterns in complex environments, especially in geometrically degenerate regions. By adjusting the control strategy, unstable behavior and motion loss are reduced, thereby improving motion stability and mapping quality.

[0083] In one embodiment, to achieve robust gait control in geometrically complex environments, the observation space is a multimodal observation space. The observation states in this space, including the proprioceptive perception state of the wheeled robot and the external degradation cue observation state, can be represented as follows: , in, The base angular velocity and projected gravity vector of the wheeled robot are used to characterize instantaneous stability and disturbance level; This includes the positional displacement and velocity of the leg joints and wheel joints; This includes plane speed commands and yaw rate commands; This includes actions from the previous moment, used to enhance time smoothness and suppress high-frequency control oscillations; Including degradation factors.

[0084] The motion in the motion space is represented by the wheel-leg motion vector, which is expressed as: , in, Indicates the movement of the leg joint. This indicates the movement of the wheel joint.

[0085] In one embodiment, to incentivize the agent to balance accurate command tracking and motion stability, particularly in scenarios involving geometrically degraded threat state estimation consistency, the reward function is constructed as a weighted sum of task-specific terms and regularization terms, which can be expressed as: , in, express The reward function at time step, This represents the weighting coefficient of the reward item. This indicates a reward item.

[0086] At the task level, command tracking is encouraged through exponential penalties, and an uprightness factor is obtained by projecting the vertical component of the gravity vector for gating. Height-reward tracking and planar position-reward tracking are respectively represented as: , , in, This indicates planar position reward tracking. Indicates high reward tracking, This represents the desired linear velocity in the horizontal direction. This represents the actual linear velocity in the horizontal direction. Represents the gravitational component. Represents the normalization constant. This represents the desired angular velocity in the vertical direction. Indicates the actual vertical angular velocity. The scaling factor represents the speed error.

[0087] Furthermore, the reward function includes a degradation penalty term, which is negligible in open and geometrically rich environments, but can suppress high-speed translation in narrow, directional structures. This degradation penalty term is expressed as: , in, Indicates a degradation penalty item. Indicates the degradation factor. This represents the actual linear velocity in the horizontal direction.

[0088] When using a control strategy for control, after receiving a speed command, the control strategy will calculate the degradation factor in real time. When the degradation factor is high, the control strategy will automatically suppress aggressive behaviors (such as reducing wheel speed, smoothing yaw, and increasing leg support) to minimize slip and disturbances, thereby improving the estimation conditions.

[0089] Furthermore, in step S40, when controlling the wheeled robot using the control strategy, to avoid the need for auxiliary motion mapping during deployment, the leg joints can be controlled by offset commands, and the wheel joints can be controlled by speed commands. The desired position of the leg joints and the desired speed of the wheel joints are expressed as follows: , , in, Indicates the desired position of the leg joint. Represents the scaling vector at the joint level. Indicates the reference leg joint position. This represents the wheel speed scaling factor. This indicates the desired speed of the wheel joint.

[0090] After obtaining the control actions of the leg joint and wheel joint, the PD (Power Distribution System) generates control commands for the leg joint and wheel joint. These control commands can be represented as follows: , in, Control commands, Indicates the assembly of leg joints. Represents a set of wheel joints. Indicates proportional gain. Represents differential gain. Indicates the desired speed. Indicates the desired position. Indicates the actual location. Indicates the actual speed.

[0091] Furthermore, in practical applications, after receiving control commands, the control commands are passed through a lightweight safety filter to limit joint speed, ensuring robust operation in real-world environments.

[0092] Furthermore, in step H40 above, during the autonomous navigation process of the wheeled robot, the topology map is updated based on the point cloud data collected by the wheeled robot to construct a 3D environment map. Specifically, at the global layer, the topology map can be updated once every preset number of frames, while at the local optimization layer, the local topology map is updated every frame.

[0093] For example, the topology graph update process is explained from a global layer perspective. At the global layer, to constrain the size of the topology graph and suppress combinatorial explosion caused by redundant nodes, a spatiotemporal hybrid gating system is used to constrain nodes inserted into the topology graph. Based on this, updating the topology graph to construct a 3D environment graph specifically includes: Candidate key nodes are generated at preset intervals of frames, and the node distance between the candidate key nodes and existing nodes in the topology graph is obtained. Candidate key nodes whose distance to other nodes is greater than a preset distance threshold are inserted into the topology graph, and edges are constructed for the inserted candidate key nodes with a preset vertical gating constraint to update the topology graph.

[0094] Specifically, candidate key nodes are key poses extracted based on the acquired keyframes, and node distance is the minimum distance between the candidate key node and all existing nodes in the topology graph. The preset distance threshold is a pre-set condition used as a distance constraint for node insertion; that is, a candidate key node is considered to satisfy the distance constraint only when the distance to the node corresponding to the candidate key node is greater than the preset distance threshold.

[0095] Furthermore, after the candidate key node satisfies the distance constraint, the vertical gating is used as the height change constraint to determine whether the candidate key node can be inserted into the topology graph. The vertical gating is pre-set to avoid cross-layer misconnection of nodes in the topology graph. It can be selected according to the traversability of the wheeled robot to ensure executability.

[0096] Furthermore, in practical applications, the autonomous navigation and 3D mapping method for wheeled robots provided in this application can be applied to multi-robot systems, which may include multiple heterogeneous robots (such as wheeled, tracked, and wheeled robots) and drones. Specifically, when the autonomous navigation and 3D mapping method for wheeled robots is applied to a multi-robot system, the 3D mapping process through the multi-robot system can be as follows: Step 1: Initialization of Environmental Mapping and Degradation Sensing Planning Specifically, the system collects 3D environmental data using the onboard lidar and inertial measurement unit, performs data preprocessing and time synchronization on the 3D data to construct a preliminary environmental model based on a global coordinate system. This preliminary environmental model includes a topology graph, in which nodes are key poses extracted from keyframes and carry node attributes.

[0097] Step 2: Calculation of degradation risk and regional division While establishing a global map, traversable spatial regions are extracted by analyzing point cloud data, and these regions are then divided into several sub-regions. Each sub-region is assigned regional attributes, which may include center coordinates, operational difficulty, and degradation risk. Simultaneously, a task graph is constructed based on the reachability relationships between these sub-regions, providing a unified task model for multi-robot collaborative operations.

[0098] Step 3: Multi-robot task allocation and scheduling Based on the mobility, task difficulty, and endurance constraints of multiple heterogeneous robots and drones, a subset of tasks that each robot can execute is configured. In configuring this subset of tasks for each robot, a centralized or distributed scheduling strategy can be employed, taking into account factors such as task priority, execution time, and energy consumption, to ensure that task allocation across multiple robots minimizes overall operation time or energy consumption.

[0099] Step 4: Global Path Planning and Local Path Optimization After configuring a subset of tasks that each robot can execute, a corresponding access sequence is generated for each robot, and global path planning is performed on the topology graph. During path planning, constraints such as floor switching, narrow passages, and detours around restricted areas can be considered to ensure the safety and feasibility of the path. Global path planning and local path optimization can utilize the processing methods described above for global and local optimization layers.

[0100] Step 5: Real-time environmental perception and path selection When performing tasks, the robot uses LiDAR to collect point cloud data in real time and construct a local map. Multiple candidate trajectories are generated and evaluated within the local map, and the trajectory that meets the requirements of safe distance, smoothness, and stability is selected for navigation and obstacle avoidance. All control commands are adjusted through a closed-loop system based on degradation perception, ensuring that the robot can adapt to changes in complex environments and avoid motion instability or perception failure due to environmental geometric degradation.

[0101] Step 6: Heterogeneous point cloud fusion and local map sharing During multi-robot collaboration, each robot periodically outputs local keyframe point cloud data. The local keyframe point cloud data is used to determine the key nodes that need to be inserted into the topology map, so as to unify the local maps of each robot under the same global coordinate system. This enables continuous updating and sharing of local point cloud maps, ensuring real-time updating and consistency of the global map. It also optimizes the accuracy of the map and improves the environmental perception capability when multi-robots work together.

[0102] Step 7: Local Replanning and Task Rescheduling During task execution, the system monitors the task progress, energy consumption, and map quality of each robot in real time. When a path is blocked, energy consumption exceeds the limit, or map quality degrades, it triggers local replanning or task rescheduling, and only adjusts the affected tasks and paths, reducing interference with the global task plan. This ensures the system's adaptability in dynamic environments and enables it to respond quickly to emergencies.

[0103] The embodiments of this application realize global task organization, local intelligent navigation, and consistent mapping of multi-source point clouds in complex and constrained environments. This enables multimodal heterogeneous unmanned devices to achieve efficient collaborative operation and stable environmental perception under a unified spatial reference. It not only improves the efficiency of task allocation and path planning, but also enhances the accuracy and reliability of autonomous navigation and 3D mapping through multi-robot collaborative operation and real-time map updates.

[0104] Based on the above-described autonomous navigation and 3D mapping method for wheeled robots, this embodiment provides a device for autonomous navigation and 3D mapping of wheeled robots, such as... Figure 8 As shown, the autonomous navigation and 3D mapping device for the wheeled robot specifically includes: The construction module 100 is used to construct a topology graph based on several key frames collected by the wheeled robot. The nodes in the topology graph are key poses extracted from the key frames, and the edges in the topology graph are executable connections between nodes. The nodes carry at least node utility, which is determined based on unknown spatial gain and degradation risk. The global processing module 200 is used to select several access nodes from the topology graph to construct an access sequence in each planning cycle based on node utility and travel cost. The local processing module 300 is used to generate a set of candidate viewpoints within the local field of view corresponding to the access node, and select access viewpoints from the candidate viewpoints based on node utility and travel cost to construct a viewpoint access sequence. The control module 400 is used to control the wheeled robot to move according to the viewpoint access sequence and update the topology map to construct a three-dimensional environment map.

[0105] Based on the above-described autonomous navigation and 3D mapping method for wheeled robots, this embodiment provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the autonomous navigation and 3D mapping method for wheeled robots as described in the above embodiment.

[0106] Based on the aforementioned autonomous navigation and 3D mapping method for wheeled robots, this application also provides a terminal device, such as... Figure 9As shown, it includes at least one processor 20; a display screen 21; and a memory 22, and may also include a communications interface 23 and a bus 24. The processor 20, display screen 21, memory 22, and communications interface 23 can communicate with each other via the bus 24. The display screen 21 is configured to display a preset user guide interface in the initial setup mode. The communications interface 23 can transmit information. The processor 20 can invoke logical instructions in the memory 22 to execute the methods described in the above embodiments.

[0107] Furthermore, the logical instructions in the aforementioned memory 22 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0108] The memory 22, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, such as program instructions or modules corresponding to the methods in the embodiments of this disclosure. The processor 20 executes functional applications and data processing by running the software programs, instructions, or modules stored in the memory 22, thereby implementing the methods in the above embodiments.

[0109] The memory 22 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 22 may include high-speed random access memory (RAM) and non-volatile memory. Examples include 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, as well as transient storage media.

[0110] Furthermore, the specific process of loading and executing multiple instruction processors in the aforementioned storage medium and terminal device has been described in detail in the above method, and will not be repeated here.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 this application.

Claims

1. A method for autonomous navigation and 3D mapping of a wheeled-legged robot, characterized in that, The aforementioned method for autonomous navigation and 3D mapping of wheeled robots specifically includes: A topology graph is constructed based on several key frames collected by the wheeled robot. The nodes in the topology graph are key poses extracted from the key frames, and the edges in the topology graph are executable connections between nodes. Each node carries at least a node utility determined based on unknown spatial gain and degradation risk. Within each planning cycle, several access nodes are selected from the topology graph based on node utility and travel cost to construct an access sequence. A candidate viewpoint set is generated within the local field of view corresponding to the access node, and an access viewpoint is selected from the candidate viewpoint set based on node utility and travel cost to construct a viewpoint access sequence. The wheeled robot is controlled to move according to the viewpoint access sequence, and the topology map is updated to construct a three-dimensional environment map.

2. The autonomous navigation and 3D mapping method for a wheeled-legged robot according to claim 1, wherein, The extraction process of the key pose specifically includes: The wheel-to-ground contact point of each leg of the wheel-legged robot is determined based on the motion data of the wheel-legged robot, and the vertical fusion residual is determined based on the wheel-to-ground contact point; The odometry residual of the wheeled robot is determined based on the IMU data collected by the wheeled robot and the motion data. The registration residuals of the point cloud data collected by the wheeled robot are obtained, and the degradation risk is constructed by utilizing the uncertainty of point positioning and the uncertainty of voxel plane fitting. Based on the degradation risk, a weighting coefficient is constructed, and the weighting coefficient is used to solve the optimal error state for the vertical fusion residual, the odometry residual of the wheeled robot, and the registration residual, so as to obtain the attitude error. Based on the attitude error and the key attitude of the previous frame, the key attitude is determined.

3. The autonomous navigation and 3D mapping method for a wheeled-legged robot according to claim 2, wherein, The determination of the wheel-to-ground contact point of each leg of the wheel-legged robot based on its motion data specifically includes: The initial contact point is determined based on the joint angle vector and joint length of each leg of the wheel-legged robot; The initial contact point is geometrically corrected using the local terrain normal and wheel radius of the pre-stored probabilistic voxel map to obtain the wheel-ground contact point for each leg.

4. The autonomous navigation and 3D mapping method for a wheeled-legged robot according to claim 2, wherein, The determination of the vertical fusion residual based on the wheel-ground contact point specifically includes: Obtain the projected height of the wheel-ground contact point on the local voxel plane; The height error of the wheel-ground contact point is determined based on the projected height, and a vertical fusion residual is constructed based on the height error.

5. The method for autonomous navigation and 3D mapping of a wheeled robot according to claim 1 or 2, characterized in that, The specific methods for constructing degradation risk using point positioning uncertainty and voxel plane fitting uncertainty include: The adaptive variance is determined by utilizing the uncertainty of point positioning and the uncertainty of voxel plane fitting. A degradation intensity index is constructed based on the adaptive variance, and the degradation risk is determined based on the degradation intensity index.

6. The method for autonomous navigation and 3D mapping of a wheeled robot according to claim 1, characterized in that, The wheeled robot is equipped with a reinforcement learning-based control strategy, and controlling the wheeled robot to move according to the viewpoint access sequence specifically includes: For each accessed viewpoint in the viewpoint access sequence, the eigenvalues ​​corresponding to the covariance matrix of the local point cloud data of the accessed viewpoint are obtained, and the degradation factor of the accessed viewpoint is determined based on the eigenvalues. Based on the degradation factor, the current posture of the wheeled robot, and the current observation state, the current action of the wheeled robot is determined by the control strategy, and the wheeled robot is controlled according to the current action.

7. The method for autonomous navigation and 3D mapping of a wheeled robot according to claim 1, characterized in that, The process of updating the topology graph to construct a 3D environment graph specifically includes: Candidate key nodes are generated at preset intervals of frames, and the node distance between the candidate key nodes and existing nodes in the topology graph is obtained. Candidate key nodes whose distance to other nodes is greater than a preset distance threshold are inserted into the topology graph, and edges are constructed for the inserted candidate key nodes with a preset vertical gating constraint to update the topology graph.

8. A wheeled-legged robot autonomous navigation and three-dimensional mapping device, characterized in that, The aforementioned autonomous navigation and 3D mapping device for wheeled robots specifically includes: A construction module is used to construct a topology graph based on several key frames collected by a wheeled robot. The nodes in the topology graph are key poses extracted from the key frames, and the edges in the topology graph are executable connections between nodes. Each node carries at least node utility, which is determined based on unknown spatial gain and degradation risk. The global processing module is used to select several access nodes from the topology graph to construct an access sequence in each planning cycle, based on node utility and travel cost. The local processing module is used to generate a set of candidate viewpoints within the local field of view corresponding to the access node, and select access viewpoints from the candidate viewpoints based on node utility and travel cost to construct a viewpoint access sequence. The control module is used to control the wheeled robot to move according to the viewpoint access sequence and update the topology map to construct a three-dimensional environment map.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the steps in the autonomous navigation and 3D mapping method for wheeled robots as described in any one of claims 1-7.

10. A terminal device, comprising: include: Processor and memory; The memory stores a computer-readable program that can be executed by the processor; When the processor executes the computer-readable program, it implements the steps in the autonomous navigation and 3D mapping method for wheeled and legged robots as described in any one of claims 1-7.

Images