A method and system for autonomous exploration of mobile robots in unknown environments

Abstract

The application discloses a kind of unknown environment under mobile robot autonomous exploration method and system, this method first real-time scanning environment obtains the positioning and point information of robot, updates 3D grid map of three-dimensional space representation;Then, boundary detection and candidate viewpoint sampling are carried out on the map, and the weight of viewpoint is calculated to screen expansion node;HybridA* algorithm is used for path planning, and the path is maintained into topological map;Pure path tracking method based on ray distance is used for path tracking, and PID algorithm is used to adjust speed, to improve the accuracy and real-time performance of path tracking.The application improves the completeness of the selection of expansion viewpoint, reduces the repeated exploration of the same area, and improves the path tracking accuracy, effectively solves the technical problems of robot autonomous exploration in large-scale unknown environment.

Description

Technical Field

[0001] This invention relates to a method and system for autonomous exploration by robots, specifically a method and system for autonomous exploration by mobile robots in unknown environments, belonging to the field of mobile robot technology. Background Technology

[0002] With the development of automation technology, ground robots are being used more and more widely in various industrial and exploratory fields. Especially in underground working environments such as coal mines, robots can perform dangerous tasks or tasks that are difficult for humans to reach. However, existing autonomous navigation systems for ground robots typically rely on pre-set maps and environmental information, which is impractical in unknown or dynamically changing scenarios. Therefore, developing an autonomous navigation method for mobile robots capable of effectively exploring large-scale unknown environments has become particularly important.

[0003] However, exploring large-scale environments fully and efficiently remains a challenging open problem, especially in the absence of prior maps, where navigation systems face significant uncertainty in path planning and decision-making. This uncertainty not only increases task complexity but can also lead to navigation failures or inefficiencies. Therefore, how to conduct effective path planning and decision-making in unknown or partially known environments has become a focus of research in recent years.

[0004] Traditional path planning methods typically rely on optimization techniques to generate feasible paths from a starting point to a destination by minimizing specific objective functions (such as path length, energy consumption, or time). These methods progressively build an environmental map during the exploration process and utilize incremental map information for real-time planning. However, this approach has significant limitations: first, incremental map building and maintenance require substantial computational resources, especially in large-scale environments where computational overhead increases significantly; second, the process of building and updating the global map can lead to planning delays, impacting the system's real-time performance.

[0005] In terms of decision-making, many methods perceive the environment by analyzing its geometric features (such as the shape, distance, and distribution of obstacles) and make optimal decisions based on these features. For example, some algorithms plan paths by detecting key points or boundaries in the environment or utilize local information for obstacle avoidance. However, these methods typically rely on maintaining a global map, leading to low computational efficiency. Furthermore, due to the uncertainty and dynamic changes in the environment, geometrically based decision-making methods may not be suitable for complex or unstructured terrain, especially when lateral errors accumulate, resulting in a significant decrease in navigation accuracy.

[0006] Conventional path-tracking methods (such as pure tracking algorithms) perform well in structured environments, but often struggle in unstructured terrain. Pure tracking algorithms guide the robot by following a predefined path, but when faced with complex terrain, lateral errors accumulate, leading to path deviations or even navigation failures. This limitation is particularly pronounced in environments lacking prior maps, as the robot cannot anticipate terrain changes and struggles to make timely adjustments.

[0007] In summary, although a large amount of research has been dedicated to path planning and decision-making in unknown or partially unknown environments, existing methods still have significant shortcomings in terms of computational efficiency, real-time performance, and adaptability. In particular, achieving efficient and accurate exploration without relying on a global map in large-scale, unstructured environments remains a pressing problem to be solved. Summary of the Invention

[0008] The purpose of this invention is to provide a method for autonomous exploration of mobile robots in unknown environments, which can improve the completeness of the selection of extended viewpoints, reduce repeated exploration of the same area, and improve path tracking accuracy. It effectively solves the technical problem of autonomous exploration of robots in large-scale unknown environments and can be applied to autonomous exploration of mobile robots in a wide range of unknown scenarios.

[0009] To achieve the above objectives, the present invention provides a method for autonomous exploration of a mobile robot in an unknown environment, comprising the following steps:

[0010] Step S1: Obtain the robot's real-time positioning information and point cloud information, register the point cloud information, and use the point cloud data information to update the 3D grid map representing the three-dimensional space.

[0011] Step S2: Perform incremental boundary detection in the updated 3D raster map, sample candidate viewpoints, evaluate the sampled viewpoints, and select the nodes to be expanded.

[0012] Step S3: Within the freely movable area detected by the robot, use the search-based HybridA* path planning algorithm to search for paths from the current robot position to all candidate viewpoints obtained based on sampling, and maintain the paths in the topology graph;

[0013] Step S4: Use a pure path tracing method based on dynamic ray distance to trace the path from the robot's current position to the extended node.

[0014] The autonomous exploration method for mobile robots in unknown environments according to embodiments of the present invention first acquires the robot's real-time positioning information and point cloud information, and converts the information into a grid map format to construct or update a 3D grid map. Next, boundary detection is performed on the updated 3D grid map, candidate viewpoints that meet the requirements are sampled, and all viewpoints are evaluated to filter out target nodes (i.e., extended nodes). Then, a search-based Hybrid A* path planning algorithm is used to search within the area where the robot can move freely, and the searched paths are maintained in the topology graph. The searched paths include paths from the current robot position to all candidate viewpoints. Finally, a pure path tracking method based on dynamic ray distance is used to track the path from the robot's current position to the target node, thus achieving path tracking. It should be noted that during the tracking process, the robot's real-time positioning information and point cloud information will change, and the 3D grid map will be updated in real time. Therefore, after tracking the target node, the robot will enter the next cycle of exploration and tracking, and so on, to achieve autonomous exploration in large-scale unknown environments.

[0015] In some embodiments of the present invention, step S1 specifically includes:

[0016] Step S11: Use a LiDAR-based SLAM module to scan the environment, obtain the robot's localization and point cloud information, process the point cloud information, and convert it into a grid map representation for real-time updating of the 3D grid map.

[0017] Step S12: Each node in the 3D raster map corresponds to a cubic region in the environment space. When new point cloud data arrives, the 3D raster map updates the state of the corresponding cubic region. Initially, all regions are marked as unknown space S. uk As the robot moves and the SLAM module scans, the unknown region is gradually explored and updated to a known state. The known state is divided into a known free space S. free and known occupied space S occ Given free space S free The area that the robot can move freely in, as detected by its sensors; the occupied space S is known. occ These are impassable areas detected by the robot using sensors.

[0018] In some embodiments of the present invention, step S2 specifically includes:

[0019] Step S21: Perform boundary detection based on the updated 3D grid map, and only detect the newly added grids and the robot's existing boundaries recorded in the 3D grid map;

[0020] Step S22: Uniformly and randomly sample around the leading edge boundary detected by the robot in the freely movable area detected by the sensor to obtain candidate viewpoints;

[0021] Step S23: Perform a multi-dimensional viewpoint comprehensive evaluation on the candidate viewpoints obtained from random sampling to select suitable candidate viewpoints.

[0022] In some embodiments of the present invention, the boundary in step S21 is defined as follows: in a known 3D space, if a free mesh has at least one unknown neighboring mesh of the same height, then the free mesh is defined as the boundary; let p i Let i be the i-th position of the robot. For robots in p i Global boundary at time, For from p i to p i+1 The boundary within the robot's range, For the robot from p i Move to p i+1 The updated sensor's sensing range of the free grid is then in Boundary detection is performed in the process.

[0023] In some embodiments of the present invention, the extended nodes selected in step S23 are suitable candidate viewpoints selected through multi-dimensional viewpoint comprehensive weight calculation. The weights include information gain, travel cost, and topological features, as detailed below:

[0024] Information gain is evaluated by comprehensively considering the number of unknown grid cells visible from the viewpoint. It is calculated using a simulated lidar ray projection method, where simulated laser rays are emitted from the viewpoint towards the detected leading edge boundary. The result is then calculated using the formula... Calculate the number of grid cells covering the unknown region, where δ i Let P(v) represent the contribution value of the i-th unknown grid, and let P(v) represent the information gain of viewpoint v.

[0025] Travel cost is based on the map's topology, including the actual shortest path length L(v0,v) from the current node to the candidate viewpoint and the number of nodes L traversed by that path. t (v0,v), where the shortest path length is obtained using the HybridA* algorithm;

[0026] Topological features are based on the viewpoint's position in the topological graph and explore the endpoints of the topological graph, where each endpoint represents a complete region to be explored, and nodes are the intersections of different regions.

[0027] Therefore, the formula for calculating viewpoint weight is: Among them, U(v) k () is the viewpoint vk The utility of G(v) k ) is the penalty function, P(v) k ) is the information gain, c is the penalty factor, and L(v0→v k () is the distance from the current viewpoint to the viewpoint v k The distance traveled.

[0028] In some embodiments of the present invention, the pure path tracking method based on dynamic ray distance in S4 tracks a predetermined path by controlling the robot's linear velocity and turning angle. The specific steps are as follows:

[0029] Step S41: Calculate the ray distance and the required turning angle based on the robot's current position, target position, and the next target point on the path;

[0030] Step S42: Calculate the ray distance. The ray distance is the straight-line distance from the robot's current position to the target point. The ray distance is dynamically adjusted based on the robot's speed and minimum turning radius. The calculation formula is: l d = k*V+C, where k is the velocity coefficient, V is the robot's velocity, and C is the robot's minimum turning radius;

[0031] Step S43: Calculate the steering angle. The steering angle θ is calculated using the Ackermann steering model and depends on the vehicle's wheelbase L and turning radius R. The calculation formula is: tanθ=L / R;

[0032] Step S44: Use the PID algorithm to adjust the robot's speed to ensure that the robot travels along the predetermined path.

[0033] In some embodiments of the present invention, during the path tracking process in step S44, the robot's position error is used to adjust the steering angle to reduce the robot's movement deviation and keep it on the path; wherein, the position error is the lateral error e(t).

[0034] This invention also discloses an autonomous exploration system for mobile robots in unknown environments, comprising:

[0035] The SLAM module is used to receive the robot's localization information and the real-time point cloud information around the robot body collected by the robot's body sensors. It registers the point cloud information to perform point cloud data modeling, builds and updates a 3D grid map.

[0036] The path analysis module receives 3D raster map information from the SLAM module, performs map front detection, generates candidate viewpoints based on sampling at the map front, and selects extended nodes through a multi-dimensional viewpoint comprehensive evaluation mechanism.

[0037] The path planning module generates paths from the robot's current position to all candidate viewpoints using a search-based HybridA* path planning algorithm, and maintains these paths in the topology graph.

[0038] The path tracking module works by the robot selecting extended nodes based on the path analysis module, obtaining the path through the extended nodes through the path planning module, and then using a pure path tracking method based on ray distance to track the path. Attached Figure Description

[0039] Figure 1 This is a flowchart of a method for autonomous exploration of a mobile robot in an unknown environment according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of the system of the present invention;

[0041] Figure 3 This is a schematic diagram of the overall process independently explored by the present invention;

[0042] Figure 4 This is a schematic diagram of the boundary detection process for finding target nodes in this invention;

[0043] Figure 5 This is a schematic diagram illustrating the principle of the pure path tracking method based on dynamic ray distance in this invention. Detailed Implementation

[0044] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0045] The following is for reference. Figures 1-5 A method for autonomous exploration of a mobile robot in an unknown environment according to an embodiment of the present invention is described, the method comprising:

[0046] Step S1: Obtain the robot's real-time positioning information and point cloud information, register the point cloud information, and use the point cloud data information to update the 3D grid map representing the three-dimensional space.

[0047] Step S2: Perform incremental boundary detection in the updated 3D grid map, sample candidate viewpoints, evaluate the sampled viewpoints, and select the nodes to be expanded.

[0048] Step S3: Within the freely movable area detected by the robot, use the search-based HybridA* path planning algorithm to search for paths from the current robot position to all candidate viewpoints obtained based on sampling, and maintain the paths in the topology graph;

[0049] Step S4: Use a pure path tracing method based on dynamic ray distance to trace the path from the robot's current position to the extended node.

[0050] Understandably, in the actual exploration and tracking process, the robot's real-time localization and point cloud information are first acquired and converted into a grid map format to construct or update a 3D grid map. Next, boundary detection is performed on the updated 3D map, candidate viewpoints that meet the requirements are sampled, and all viewpoints are evaluated to select the target node (i.e., the extended node). Then, a search-based Hybrid A* path planning algorithm is used to search within the area where the robot can move freely, and the searched paths are maintained in the topology graph. These searched paths include paths from the current robot position to all candidate viewpoints. Finally, a pure path tracking method based on dynamic ray distance is used to track the path from the robot's current position to the target node, thus achieving path tracking. It should be noted that during the tracking process, the robot's real-time localization and point cloud information will change, and the 3D grid map will be updated in real time. Therefore, after tracking the target node, the robot will enter the next cycle of exploration and tracking, and so on, to achieve autonomous exploration in large-scale unknown environments.

[0051] Therefore, the robot movement process is divided into path search and path tracking. The path is obtained by searching the HybridA* path planning algorithm, while the path tracking uses a pure path tracking method based on ray distance proposed in this invention. This is a path tracking technology for mobile robots, which is particularly suitable for car-like mobile robots with Ackermann steering model. It aims to improve the accuracy and real-time performance of path tracking, reduce tracking errors, and meet the dynamic constraints of the mobile robot.

[0052] In some embodiments of the present invention, step S1 specifically includes:

[0053] Step S11: Use a LiDAR-based SLAM module to scan the environment, obtain the robot's localization and point cloud information, and process the point cloud information to convert it into a grid map representation for real-time updating of the 3D grid map. The 3D grid map used is OctoMap, which is used to represent the three-dimensional environment. Its nodes store the indicators of the space represented by its child nodes, allowing certain branches to be ignored when traversing the octree. Each node corresponds to a Morton code for fast node lookup, thereby accelerating traversal.

[0054] Step S12: Each node in the 3D raster map corresponds to a cubic region in the environment space. When new point cloud data arrives, the 3D raster map updates the state of the corresponding cubic region. Specifically, each node in OctoMap corresponds to a cubic region (voxel) in the environment space. When new point cloud data arrives, OctoMap updates the state of the corresponding voxel, including the following steps:

[0055] Initially, all regions are labeled as unknown space S. uk As the robot moves and the SLAM module scans, the unknown region is gradually explored and updated to a known state. The known state is divided into a known free space S. free and known occupied space S occ Given free space S free The area that the robot can move freely in, as detected by its sensors; the occupied space S is known. occ These are impassable areas detected by the robot using sensors.

[0056] In some embodiments of the present invention, step S2 specifically includes:

[0057] Step S21: Perform boundary detection based on the updated 3D grid map. In order to improve the efficiency of boundary detection and avoid searching the entire map every time the map is updated, only detect the newly added grids and the robot's existing boundaries recorded in the 3D grid map.

[0058] Step S22: In the area where the robot can move freely as detected by the sensors (i.e., in the known free space S) free Candidate viewpoints are obtained by uniformly and randomly sampling around the detected leading edge boundary.

[0059] Step S23: Perform a multi-dimensional viewpoint comprehensive evaluation on the candidate viewpoints obtained from random sampling to select suitable candidate viewpoints.

[0060] In some embodiments of the present invention, the boundary in step S21 is defined as follows: in a known 3D space, if a free mesh has at least one unknown neighboring mesh of the same height, then the free mesh is defined as the boundary; let p i Let i be the i-th position of the robot. For robots in p i Global boundary at time, For from p i to p i+1 The boundary within the robot's range, For the robot from p i Move to p i+1 The updated sensor's sensing range of the free grid is then in Boundary detection is performed during this process. Because boundaries must appear in the updated mesh, some existing boundaries may cease to be boundaries due to the updating of neighboring meshes.

[0061] In some embodiments of the present invention, the extended nodes selected in step S23 are suitable candidate viewpoints selected by multi-dimensional viewpoint comprehensive weight calculation, the weights of which include information gain, travel cost and topological features, as detailed below:

[0062] Information gain is evaluated by comprehensively considering the number of unknown grid cells visible from the viewpoint. It is calculated using a simulated lidar ray projection method, where simulated laser rays are emitted from the viewpoint towards the detected leading edge boundary. The result is then calculated using the formula... Calculate the number of grid cells covering the unknown region, where δ i Let P(v) represent the contribution value of the i-th unknown grid, and let P(v) represent the information gain of viewpoint v.

[0063] Travel cost is based on the map's topology, including the actual shortest path length L(v0,v) from the current node to the candidate viewpoint and the number of nodes L traversed by that path. t (v0,v), where the shortest path length is obtained using the HybridA* algorithm;

[0064] Topological features are based on the viewpoint's position in the topological graph and explore the endpoints of the topological graph, where each endpoint represents a complete region to be explored, and nodes are the intersections of different regions. Prioritizing the exploration of the endpoints of the topological graph ensures that different regions are explored sequentially, avoiding unnecessary duplicate explorations.

[0065] Therefore, the formula for calculating the multidimensional viewpoint weight is: Among them, U(v) k () is the viewpoint v k The utility of G(v) k ) is the penalty function, P(v) k ) is the information gain, c is the penalty factor, and L(v0→v k () is the distance from the current viewpoint to the viewpoint v k The distance traveled.

[0066] At candidate viewpoint v k In step S2, the candidate viewpoint with the largest weight value is selected as the extended node v'0, and the robot moves from the current node to the extended node (target node) using the viewpoint weights calculated in step S2.

[0067] It should be noted that the multi-dimensional viewpoint comprehensive evaluation mechanism is adopted to evaluate the selection of expansion nodes. It fully considers the information gain, travel cost and topological characteristics of candidate nodes. Compared with the existing technology that only considers information gain, it improves the completeness of the selection of expansion viewpoints and effectively avoids the shortcomings of the existing technology that are prone to getting trapped in local optima when selecting viewpoints.

[0068] In some embodiments of the present invention, the pure path tracking method based on dynamic ray distance in S4 tracks a predetermined path by controlling the robot's linear velocity and turning angle. The specific steps are as follows:

[0069] Step S41: Calculate the ray distance and the required turning angle based on the robot's current position, target position, and the next target point (i.e., the extended node, hereinafter the same) on the path;

[0070] Step S42: Calculate the ray distance. The ray distance is the straight-line distance from the robot's current position to the target point. The ray distance is dynamically adjusted based on the robot's speed and minimum turning radius, and is used to determine the foresight during path tracking. Its calculation formula is: l d = k*V+C, where k is the velocity coefficient, V is the robot's velocity, and C is the robot's minimum turning radius;

[0071] Step S43: Calculate the steering angle. The steering angle θ is calculated using the Ackermann steering model and depends on the vehicle's wheelbase L and turning radius R. The calculation formula is: tanθ=L / R;

[0072] Step S44: Use the PID algorithm to adjust the robot's speed to ensure that the robot travels along the predetermined path.

[0073] In some embodiments of the present invention, during the path tracking process in step S44, the robot's position error is used to adjust the steering angle to reduce the robot's movement deviation and keep it on the path; wherein, the position error is the lateral error e(t).

[0074] This invention also discloses an autonomous exploration system for mobile robots in unknown environments, comprising:

[0075] The SLAM module is used to receive the robot's positioning information and the real-time point cloud information around the robot body collected by the robot's body sensors. It registers the point cloud information to perform point cloud data modeling, builds and updates a 3D grid map.

[0076] The path analysis module receives 3D raster map information from the SLAM module, performs map front detection, generates candidate viewpoints based on sampling at the map front, and selects extended nodes through a multi-dimensional viewpoint comprehensive evaluation mechanism.

[0077] The path planning module generates paths from the robot's current position to all candidate viewpoints using a search-based HybridA* path planning algorithm, and maintains these paths in the topology graph.

[0078] The path tracking module works by the robot selecting extended nodes based on the path analysis module, obtaining the path through the extended nodes through the path planning module, and then using a pure path tracking method based on ray distance to track the path.

[0079] It should be noted that the system of the present invention includes a SLAM module, a path analysis module, a path planning module, and a path tracking module; the output information of the SLAM module is transmitted to the path analysis module, the output information of the path analysis module is transmitted to the path planning module, the output information of the path planning module is transmitted to the path tracking module, and finally the path tracking module is connected to the SLAM module for further exploration and tracking.

[0080] Specifically, the functions and roles of each module are as follows:

[0081] I. SLAM Module:

[0082] In this implementation example, the SLAM module, as the upstream module of path planning and control, scans the robot's surrounding environment in real time to obtain the robot's localization and perception information and the registered point cloud data information, and uses the point cloud data information to update OctoMap in real time.

[0083] Each node in OctoMap corresponds to a cubic region (voxel) in the environment space. When new point cloud data arrives, OctoMap updates the state of the corresponding voxel, specifically including:

[0084] Initially, all regions are labeled as unknown space S. uk ;

[0085] As the robot moves and the SLAM module scans, the unknown area is gradually explored and updated to a known state, which is divided into the following two parts:

[0086] Given free space S free The area that the robot can move freely in, detected by its sensors.

[0087] Known occupied space S occ Impassable areas detected by the robot using sensors.

[0088] II. Path Analysis Module:

[0089] In this implementation example, the path analysis module processes the map information provided by the front-end SLAM module. This module is a key module for the robot to complete autonomous exploration tasks in unknown environments. Appropriate candidate viewpoint sampling and evaluation methods can greatly improve the efficiency of the robot's autonomous exploration tasks.

[0090] like Figure 4 As shown, in this implementation example, the path analysis module specifically includes the following steps:

[0091] First, boundary detection is performed quickly in the updated 3D grid map. A boundary is defined as a free grid in the known 3D space that has at least one unknown neighboring grid at the same height. To improve boundary detection efficiency, only newly added grids and existing boundaries of the mobile robot recorded in the 3D grid map need to be detected.

[0092] Let p i Let i be the i-th position of the robot. For robots in p i Global boundary at time, For from p i to p i+1 The boundary within the robot's range, For the robot from p i Move to p i+1 The updated sensor sensing range is a free grid; then in Boundary detection is performed because boundaries must appear in the updated mesh, and some existing boundaries may no longer be boundaries due to the updating of neighboring meshes.

[0093] Secondly, in the known free space S free Candidate viewpoints are obtained by uniformly and randomly sampling around the detected leading edge boundary.

[0094] Next, the candidate viewpoints obtained by random sampling are subjected to a comprehensive viewpoint weight calculation, which takes into account information gain, travel cost, and topological features. The specific steps include:

[0095] Information gain is evaluated, taking into account the number of unknown grid cells visible from the viewpoint, and is calculated using a ray projection method simulating LiDAR.

[0096] A simulated laser beam is emitted from the viewpoint toward the detected leading edge boundary, using the formula... Calculate the number of grid cells covering the unknown region, where δ i Let P(v) represent the contribution value of the i-th unknown grid, and let P(v) represent the information gain of viewpoint v.

[0097] Travel costs are based on the map's topology, including the actual shortest path length L(v0,v) from the current node to the candidate viewpoint and the number of nodes L traversed by that path. t (v0,v), where the shortest path length is obtained using the HybridA* algorithm.

[0098] Topological features are based on the viewpoint's position in the topological graph. The endpoints of the topological graph usually represent a complete region to be explored, while nodes are the intersections of different regions. Prioritizing the exploration of the endpoints of the topological graph can ensure that different regions are explored in sequence, avoiding unnecessary repeated exploration.

[0099] In summary, the viewpoint weight is calculated as follows: Among them, U(v) k () is the viewpoint v k The utility of G(v) k ) is the penalty function, P(v) k ) is the information gain, c is the penalty factor, and L(v0→v k () is the distance from the current viewpoint to the viewpoint v k The distance traveled.

[0100] III. Path Planning Module:

[0101] In this implementation example, the path planning module uses the search-based HybridA* path planning algorithm to search for the path from the current robot position to all candidate viewpoints. The path searched by the path analysis module to move the robot from the current position to the optimal viewpoint (i.e., the extended node) selected by the path analysis module is handed over to the path tracking module for tracking, and the remaining paths are maintained in the topology graph.

[0102] IV. Path Tracking Module:

[0103] In this implementation example, the path tracking module tracks the path from the robot's current position to the extended node transmitted by the path planning module, and based on the requirements of path smoothness and tracking accuracy, a pure path tracking method based on ray distance is used to track the path.

[0104] In this implementation example, the pure path tracking method based on ray distance tracks a predetermined path by controlling the robot's linear velocity and turning angle. During path tracking, the robot's current position, target position, and the next target point (i.e., the extended node) on the path are used to calculate the ray distance and the required turning angle. The specific calculation methods for the ray distance and turning angle are as follows:

[0105] Line-of-sight distance (LAD) refers to the straight-line distance from the robot's current position to the target point. It is dynamically adjusted based on the robot's speed and minimum turning radius, and is used for foresight when determining path tracking. The formula for calculating LAD is: l d = k*V+C, where k is the velocity coefficient, V is the robot's velocity, and C is the robot's minimum turning radius.

[0106] The steering angle θ is calculated using the Ackermann steering model, with the formula: tanθ=L / R, where L is the vehicle's wheelbase and R is the turning radius.

[0107] Path tracking control uses a PID (proportional-integral-derivative) algorithm to adjust the robot's speed, ensuring that the robot travels along the predetermined path. During path tracking, the robot's position error (lateral error e(t)) is used to adjust the steering angle to reduce deviation and stay on the path, enabling the robot to track the path more smoothly and accurately.

[0108] In summary, this invention addresses the significant challenges of autonomous exploration by mobile robots in large-scale unknown environments. By utilizing a path analysis module—specifically, the selection of extended nodes—it provides a multi-dimensional viewpoint comprehensive evaluation mechanism. This effectively improves the efficiency of path planning during exploration and reduces repetitive exploration of the same area, significantly enhancing the efficiency of autonomous exploration. The path tracking module proposes a pure path tracking method based on ray distance. It adjusts the robot's speed and minimum turning radius using the straight-line distance from the robot's current position to the target point, effectively improving the smoothness and accuracy of path tracking and reducing lateral tracking errors. Ultimately, through the cooperation of these modules, the invention solves the technical problems of low efficiency and high resource consumption inherent in existing technologies for autonomous exploration of ground-based mobile robots in unknown environments.

[0109] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for autonomous exploration of a mobile robot in an unknown environment, characterized in that, The method includes: Step S1: Obtain the robot's real-time positioning information and point cloud information, register the point cloud information, and use the point cloud data information to update the 3D grid map representing the three-dimensional space. Step S2: Perform incremental boundary detection in the updated 3D raster map, sample candidate viewpoints, evaluate the sampled viewpoints, and select the nodes to be expanded. Step S3: Within the freely movable area detected by the robot, use the search-based Hybrid A* path planning algorithm to search for paths from the current robot position to all candidate viewpoints obtained based on sampling, and maintain the paths in the topology graph; Step S4: Use a pure path tracing method based on dynamic ray distance to trace the path from the robot's current position to the extended node; Step S1 specifically includes: Step S11: Use a LiDAR-based SLAM module to scan the environment, obtain the robot's localization and point cloud information, process the point cloud information, and convert it into a grid map representation for real-time updating of the 3D grid map. Step S12: Each node in the 3D raster map corresponds to a cubic region in the environment space. When new point cloud data arrives, the 3D raster map updates the state of the corresponding cubic region. Initially, all regions are marked as unknown space. As the robot moves and the SLAM module scans, the unknown area is gradually explored and updated to a known state. The known state is divided into known free space. and known space occupied Known free space The area that the robot can move freely in, as detected by its sensors; the occupied space is known. These are impassable areas detected by the robot using sensors. Step S2 specifically includes: Step S21: Perform boundary detection based on the updated 3D grid map, and only detect the newly added grids and the robot's existing boundaries recorded in the 3D grid map; Step S22: Perform uniform random sampling around the leading edge boundary detected by the robot in the freely movable area detected by the sensor to obtain candidate viewpoints; Step S23: Perform a multi-dimensional viewpoint comprehensive evaluation on the candidate viewpoints obtained from random sampling in order to select suitable candidate viewpoints; The boundary in step S21 is defined as follows: in a known 3D space, if a free mesh has at least one unknown neighboring mesh of the same height, then the free mesh is defined as a boundary; let For the robot's first One location, For robots in Global boundary at time, From arrive The boundary within the robot's range, For robots from Move to The updated sensor's sensing range of the free grid is then in Boundary detection is performed in the process.

2. The method for autonomous exploration of a mobile robot in an unknown environment according to claim 1, characterized in that, The expanded nodes selected in step S23 are suitable candidate viewpoints selected through multi-dimensional viewpoint comprehensive weight calculation. The weights include information gain, travel cost, and topological features, as detailed below: Information gain is evaluated by comprehensively considering the number of unknown grid cells visible from the viewpoint. It is calculated using a simulated lidar ray projection method, where simulated laser rays are emitted from the viewpoint towards the detected leading edge boundary. The result is then calculated using the formula... Calculate the number of grid cells covering the unknown region, where This represents the contribution value of the i-th unknown grid. This represents the information gain of viewpoint v; Travel costs are based on the map's topology, including the actual shortest path length from the current node to candidate viewpoints. and the number of nodes traversed by the path The shortest path length is obtained using the Hybrid A* algorithm. Topological features are based on the viewpoint's position in the topological graph and explore the endpoints of the topological graph, where each endpoint represents a complete region to be explored, and nodes are the intersections of different regions. Therefore, the formula for calculating viewpoint weight is: ,in, It is a penalty function. It is information gain, and c is the penalty factor. It is from the current viewpoint to the viewpoint. The distance traveled.

3. The method for autonomous exploration of a mobile robot in an unknown environment according to claim 2, characterized in that, The pure path tracking method based on dynamic ray distance in S4 tracks a predetermined path by controlling the robot's linear velocity and turning angle. The specific steps are as follows: Step S41: Calculate the ray distance and the required turning angle based on the robot's current position, target position, and the next target point on the path; Step S42: Calculate the ray distance. The ray distance is the straight-line distance from the robot's current position to the target point. The ray distance is dynamically adjusted based on the robot's speed and minimum turning radius. The calculation formula is as follows: ,in V is the speed coefficient, C is the robot's speed, and C is the robot's minimum turning radius. Step S43: Calculate the steering angle. It is calculated using the Ackermann steering model, depending on the vehicle's wheelbase L and turning radius R. The calculation formula is as follows: ; Step S44: Use the PID algorithm to adjust the robot's speed to ensure that the robot travels along the predetermined path.

4. The method for autonomous exploration of a mobile robot in an unknown environment according to claim 3, characterized in that, During the path tracking process in step S44, the robot's position error is used to adjust the steering angle to reduce the robot's movement deviation and keep it on the path; wherein, the position error is the lateral error. .

5. A mobile robot autonomous exploration system for unknown environments, used to implement the mobile robot autonomous exploration method for unknown environments as described in any one of claims 1 to 4, characterized in that, include: The SLAM module is used to receive the robot's positioning information and the real-time point cloud information around the robot body collected by the robot's body sensors. It registers the point cloud information to perform point cloud data modeling, builds and updates a 3D grid map. The path analysis module receives 3D raster map information from the SLAM module, performs map front detection, generates candidate viewpoints based on sampling at the map front, and selects extended nodes through a multi-dimensional viewpoint comprehensive evaluation mechanism. The path planning module generates paths from the robot's current position to all candidate viewpoints using a search-based Hybrid A* path planning algorithm, and maintains these paths in the topology graph. The path tracking module works by the robot selecting extended nodes based on the path analysis module, obtaining the path through the extended nodes through the path planning module, and then using a pure path tracking method based on ray distance to track the path.

Images