Mobile robot autonomous exploration method based on topological graph

Abstract

The invention discloses a mobile robot autonomous exploration method based on a topological graph, and relates to the field of intelligent robots. According to the method, a framework topological structure of an environment is introduced, exploration behaviors are guided by utilizing topology of the environment, and the method comprises incremental framework extraction, framework topological graph updating and exploration strategy finite-state machine. According to the method, the BFMs phenomenon is relieved to a great extent, complex TSP problem solving is avoided, and efficient exploration incremental skeleton extraction is achieved with extremely low calculation consumption; the exploration behavior of the robot is effectively guided, invalid movement is remarkably reduced, information acquisition is maximized, and resource consumption is minimized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent robots, and in particular to an autonomous exploration method of a mobile robot based on a topological graph. Background Art

[0002] Intelligent robots have always been the goal pursued in many fields, but in many operational scenarios, they are still in a state of manual operation and the ability to operate autonomously is still lacking. In recent years, autonomous robotic exploration has attracted increasing attention thanks to its potential applications in multiple fields such as search and rescue, environmental monitoring, and planetary exploration. The main goal of autonomous exploration is to enable robots to efficiently navigate and map unknown environments without human intervention and without prior information to serve other subsequent tasks. A key challenge of autonomous exploration is the need for efficient decision-making in uncertain and dynamic environments. Robots must continuously make intelligent decisions about where to move next to maximize information acquisition while minimizing resource consumption such as time, energy, and computing power.

[0003] Early exploration strategies typically employed boundary-based or sampling-based methods, greedily moving toward unknown regions with the highest information gain. However, these strategies, lacking a global perspective and based on heuristic information gain estimation, often result in low-quality trajectories in complex environments, severely impacting exploration efficiency. For example, if the unknown boundary of a small region decreases with exploration, and the information gain estimate is lower than that of another small region, the exploration will greedily move toward the region with higher utility. As the exploration utility decreases, the exploration returns to the previous small region, repeating the process, leading to severe back-and-forth maneuvers (BFMs).

[0004] To solve this problem, many methods in recent years have taken global and local information into consideration at the same time, adopted the coarse-fine hierarchical idea, represented all currently known global and local information in different ways, constructed and solved the TSP problem, and thus obtained the globally optimal path.

[0005] However, due to the unpredictability of the unknown space, this global optimum based on partially observable information can be considered greedy in some ways. Therefore, while the aforementioned problems are alleviated, they still exist. Because information changes during the exploration process, the solution is usually only valid at the current moment, and solving high-frequency TSP optimization problems can impose a heavy computational burden.

[0006] Therefore, those skilled in the art have devoted themselves to developing a topological graph-based autonomous exploration method for mobile robots. This invention effectively guides the robot's exploration behavior by introducing the skeleton topology of the environment, significantly reducing ineffective movement, maximizing information acquisition, and minimizing resource consumption. Summary of the Invention

[0007] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is to improve the efficiency of autonomous exploration of robots.

[0008] To achieve the above objectives, the present invention provides a topology-based autonomous exploration method for a mobile robot, which utilizes the topology of the environment to guide the exploration behavior.

[0009] Furthermore, the skeleton topology graph is used to replace the roadmap to abstract the topological structure of the environment from the geometric space.

[0010] Furthermore, the skeleton topology map first obtains the environment structure skeleton map, and then generates the environment topology map from the skeleton map.

[0011] Furthermore, by recording the values ​​of wavefront propagation and the types of trigger waves, incremental updates of skeleton extraction are achieved.

[0012] Furthermore, a lightweight incremental update based on wavefront propagation is performed to integrate frontiers, viewpoints, information gain, and path planning.

[0013] Furthermore, the bifurcation point information is extracted through the skeleton to generate a skeleton topology map.

[0014] Furthermore, the skeleton topology is cropped based on the wavefront propagation information.

[0015] Furthermore, the exploration strategy is encapsulated as a finite state machine.

[0016] Further, the following steps are included:

[0017] Step 1: Input sensor data into SLAM and output the positioned and registered point cloud.

[0018] Step 2: UFOMap receives SLAM data and updates the status of the 3D occupancy grid;

[0019] Step 3: Skeleton extraction is performed incrementally based on the state of the mesh;

[0020] Step 4: Generate and prune the skeleton topology based on the current exploration branch and the updated skeleton;

[0021] Step 5: The exploration strategy finite state machine performs a turntable conversion and executes the exploration strategy according to the state of the skeleton topology graph;

[0022] Step 6: Repeat the above steps until the state machine reaches the terminal state.

[0023] Existing autonomous exploration technologies are limited by incomplete environmental information, making it difficult to reach globally optimal decisions. Under these circumstances, greedy "optimal solutions" based on predictions of information gain can easily lead to back-and-forth motion (BFMs), resulting in low-quality exploration trajectories and reduced exploration efficiency. Furthermore, to mitigate this problem, TSP-based methods require continuous optimization, imposing a high computational burden. However, due to incomplete information, they cannot fundamentally address the issue. This paper utilizes the topology of the environment to guide exploration behavior, proposing a novel autonomous exploration method based on a lightweight skeleton topology graph, which significantly mitigates the occurrence of BFMs. Furthermore, this guidance based on environmental topology avoids the complex TSP problem, enabling efficient exploration with extremely low computational overhead. This method replaces the traditional randomly sampled roadmap with a skeleton topology graph, accurately and concisely abstracting the environment's topological structure from geometric space. To address this representation, a lightweight incremental update mechanism based on wavefront propagation is designed, implicitly integrating key elements such as the frontier, viewpoint, information gain, and path planning, avoiding ray casting and redundant path searches. In this topological representation, we introduce high-level intent into the exploration process, prioritizing the completion of the current exploration branch to minimize inefficient round-trip operations (BFMs) while avoiding the complex Traveling Salesman Problem (TSP) solution. With this idea at its core, we design a comprehensive exploration strategy and encapsulate it as a finite state machine, enabling efficient exploration with low computational overhead.

[0024] The topological maps generated by existing topological map generation methods have high redundancy, which will bring great computational pressure to subsequent navigation path planning. The present invention considers the generation of topological maps from the perspective of environmental geometric characteristics and proposes the concept of skeleton topological maps, which is a lightweight representation of environmental geometric structures. First, a skeleton map of the environmental structure is obtained, and then a minimalist topological map is generated from the skeleton map. Through the skeleton, information such as bifurcation points is extracted to generate a skeleton topological map, which greatly reduces the number of redundant topological nodes; in addition, the skeleton topological map can be further trimmed with the help of the properties of wavefront propagation.

[0025] Existing skeleton generation methods mostly rely on offline extraction, making it difficult to implement incremental online updates as the robot explores and moves. This paper proposes an incremental skeleton extraction method based on wavefront propagation. Skeleton extraction is viewed as the process of wavefront propagation triggered by obstacles and unknown areas until collision. By recording the wavefront propagation values ​​and the types of triggering waves, incremental updates of the skeleton extraction are achieved.

[0026] The present invention relates to the field of autonomous exploration by mobile robots, and is particularly suitable for scenarios where robots are deployed in unknown environments or where the environment changes over time, rendering the original map invalid and requiring re-map reconstruction, such as warehouse automation, indoor service robots, rescue and detection missions, etc.

[0027] A key challenge in autonomous exploration is the need for efficient decision-making in uncertain and dynamic environments. The robot must continuously make intelligent decisions about where to move next to maximize information acquisition while minimizing resource consumption such as time, energy, and computing power.

[0028] At present, mainstream autonomous exploration methods have shortcomings in exploration efficiency and computational delay. In particular, when exploring unknown environments, they are prone to swinging back and forth, resulting in low exploration efficiency and waste of time.

[0029] The present invention effectively guides the robot's exploration behavior and significantly reduces invalid movements by introducing the skeleton topology of the environment.

[0030] like Figure 8 As shown in the figure, the proposed method was tested in four complex simulation environments. Compared with the existing advanced algorithms, the exploration time can be reduced by more than 19.8%, and the moving path length can be reduced by more than 15.6%. At the same time, the calculation delay of each round of decision-making is more than 4 times faster than the existing methods. The specific quantitative indicators are shown in the table below.

[0031]

[0032]

[0033] Compared with the prior art, the present invention has the following obvious substantial features and significant advantages:

[0034] 1. The present invention significantly reduces the occurrence of BFMs phenomenon, improves exploration efficiency, and significantly reduces the computational cost of exploration decision-making.

[0035] 2. The present invention generates a newly proposed skeleton topology graph with the help of the skeleton graph to achieve a minimalist representation of the geometric structure of the environment.

[0036] 3. During the autonomous exploration process of the present invention, as the robot moves and continuously collects new sensor data, the skeleton map of the environment can be updated incrementally and quickly to support high-frequency exploration decisions.

[0037] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is an overall system block diagram of a preferred embodiment of the present invention;

[0039] Figure 2 This is an incremental skeleton extraction algorithm process of a preferred embodiment of the present invention;

[0040] Figure 3 8 neighborhood diagram of a preferred embodiment of the present invention;

[0041] Figure 4 This is a schematic diagram of skeleton extraction according to a preferred embodiment of the present invention;

[0042] Figure 5 This is a skeleton topology graph updating algorithm flow of a preferred embodiment of the present invention;

[0043] Figure 6 It is a schematic diagram of a skeleton topology diagram of a preferred embodiment of the present invention;

[0044] Figure 7 It is an exploration strategy finite state machine of a preferred embodiment of the present invention;

[0045] Figure 8 It is a schematic diagram of a simulation environment of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following describes several preferred embodiments of the present invention with reference to the accompanying drawings to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0047] In the drawings, components with identical structures are denoted by the same reference numerals, and components with similar structures or functions are denoted by similar reference numerals. The size and thickness of each component shown in the drawings are arbitrary and are not limited by the present invention. For clarity, the thickness of components in some places in the drawings is appropriately exaggerated.

[0048] like Figure 1 As shown in the figure, the specific content of the implemented topology-based mobile robot autonomous exploration method is as follows: the overall system architecture includes three core parts: incremental skeleton extraction, skeleton topology map update, and exploration strategy finite state machine.

[0049] In addition to the core components, the system also includes a simultaneous localization and mapping (SLAM) module, which uses data from the onboard lidar and IMU as input to provide the system with a positioned and registered point cloud for constructing a three-dimensional occupancy grid map, UFOMap. UFOMap can efficiently record and manage changed grids for use by other modules; and a local planner module based on a trajectory library, which inputs a local destination and outputs control commands in the form of linear velocity and angular velocity, allowing the mobile robot to safely reach the local destination.

[0050] The process of the entire system is as follows: sensor data is input into SLAM, and the positioned and registered point cloud is output; UFOMap receives SLAM data to update the state of the three-dimensional occupied grid; the skeleton extraction module is executed incrementally according to the grid state change; the skeleton topology map is generated and pruned based on the current exploration branch and the updated skeleton; the exploration strategy finite state machine performs state transitions and executes the exploration strategy according to the state of the skeleton topology map; the above process is repeated until the state machine reaches the termination state.

[0051] The detailed process is as follows:

[0052] 1. Incremental Skeleton Extraction

[0053] We hope to use the environment topology to guide the robot's exploration behavior to reduce BFMs. Therefore, we must first quickly extract a concise topological structure. The generalized Voronoi Diagram (GVD) maximizes the distance from the skeleton point to the obstacle, thus providing a compact environment geometry representation. The obstacles are represented as n convex point sets, of which two convex point sets C i and C j Two-equidistant faces between Defined as:

[0054]

[0055] Among them, d i (x), d j (x), d k ((x) is the distance from point x to any convex point set C i , C j , C k The minimum distance, From point x to convex point set C i The direction of the closest point in the vector. GVD is the union of all two equidistant faces:

[0056]

[0057] However, in mobile robot exploration tasks, this strict definition will bring a large computational burden, that is, each GVD extraction requires traversing and calculating the distance from the point in the free space to the obstacle, and it is difficult to update incrementally. Therefore, an approximate method is adopted to iteratively confirm the candidate points of the skeleton and support incremental updates. The pseudo code is as follows Figure 2 The input of the algorithm is the skeleton and wavefront, with the grid as the unit, and each grid s has three states: unknown, occupied, and free. s ∈{unknown, occupied, free}, wavefront waveFronts is initialized to 0. The definition of the 8-neighborhood Adj8(s) of the current grid s is as follows Figure 3 As shown. Iterate all free grids in the ROI area until no grid meets the conditions, so that the skeleton can be obtained. The whole process is equivalent to the wavefront propagation starting from the free grid close to the obstacle. The position where the two waves collide is the skeleton, as shown Figure 4 (c). Each round of iteration is divided into two stages, corresponding to different judgment conditions as follows ( Figure 2 , lines 11-12):

[0058] (a)A(s)=1, (b)2≤B(s)≤6,

[0059] (c)n1*n3*n5=0, (d)n3*n5*n7=0,

[0060] (e)n1*n3*n7=0, (f)n1*n5*n7=0,

[0061] Where A(s) represents the number of transitions from 0 to 1 experienced when traversing its neighborhood in the order n1, n2, ..., n8, and B(s) represents the number of non-zero neighbors of s:

[0062] B(s)=n1+n2+n3+…+n7+n8,

[0063] If state(s ni )=free then n i =1, otherwise n i =0.

[0064] In order to achieve incremental extraction, the wavefront (that is, the number of iterations) is recorded. The unknown area is regarded as a special occupied grid, and the wavefront propagating from this type of grid is marked as a special wavefront ( Figure 2 , lines 13-18). During the iteration process, the wavefront of the grid currently in the free state is first determined. If the current iteration number is less than the wavefront at that point, the point is skipped ( Figure 2 , line 9). Finally, the wavefront values ​​of the unknown area grid will be reset to 0 ( Figure 2 , line 22). This avoids repeated computation of the wavefront propagation triggered by the occupied grid, accounts for the uncertainty of unknown regions, and ensures the consistency and stability of skeleton extraction. Furthermore, the wavefront reflects the width of the path, which also provides information for subsequent topology pruning and exploration decisions.

[0065] Figure 4 Describes the incremental skeleton extraction, where (a) and (b) represent the iThe wavefront propagation process at time T i Time to T i+1 The movement of the robot at time T, where the shaded part represents i The wavefront triggered by obstacles at any time is the retained part, so it is only necessary to perform the above iterative confirmation process on other newly added parts, thereby achieving incremental and fast skeleton extraction.

[0066] 2. Skeleton topology update

[0067] The proposed skeleton topology is defined as in and ε represent nodes and edges respectively. There are four types of nodes. The graph grows in a tree-like manner, so it has the concepts of parent nodes and child nodes. It is divided according to the number of parent nodes and child nodes:

[0068] Endpoint (N T ): a parent node, no child nodes

[0069] Connection point (N C ): a parent node and a child node

[0070] Bifurcation point (N B ): one parent node, multiple child nodes

[0071] ·Entry point (N I ): multiple parent nodes, one child node

[0072] On this basis, we define the first bifurcation point (N) encountered when a node is traced back to the parent node. B ) or the entry point (N I ) type of point is defined as the parent hub node of the node. The pseudo code of the entire update process is as follows Figure 5 shown.

[0073] 2.1 Graph Extension

[0074] In the initialization phase, the skeleton mesh point closest to the starting point is marked as s Home , call the function genNode(s Home ) Generate the corresponding skeleton topology node and join the frontier candidate queue

[0075] After initialization, during the normal exploration process, for the front candidate queue For each node in the graph, we will traverse its eight neighborhoods, generate corresponding nodes for the grids that belong to the skeleton points and are in the free state, and mark these points as the child nodes of the central node. Repeat this process to expand the graph in a breadth-first manner ( Figure 5, lines 15-19). If there is an unknown node in the neighborhood of the current center node, add the node to As the starting point for the next expansion ( Figure 5 , lines 5, 19). If there are already generated nodes in the neighborhood Then the candidate pair of points will be Join the sink candidate queue ( Figure 5 , line 8). Then, according to the previous definition, call the function setN0deType(s) to assign the type of the current node ( Figure 5 , line 12). Then all newly generated child nodes will be added to the expansion queue, and the above process will be repeated to perform breadth-first graph expansion ( Figure 5 , lines 13-14).

[0076] 2.2 Graph Simplification

[0077] Since skeleton generation is affected by the shape of the space occupied, it generates small edges, which is not conducive to reflecting the topological structure of the environment and affects the simplicity of the graph, as shown in the figure. Therefore, we reuse the previously recorded wavefronts. When two waves meet, if the propagation distance is very short, it is considered to be an unnecessary skeleton caused by the irregular shape distribution of the obstacle or the sensor error. Based on this assumption, we traverse all newly generated endpoint type nodes (N T ), if the wavefront here is less than the threshold, then backtrack to the parent node and delete the node until a bifurcation point (N B ) or the entry point (N I ) type of node is stopped.

[0078] Then update the edges according to the newly generated nodes; in order to facilitate subsequent path search and other navigation tasks, only B 、N I 、N T Establish an edge between nodes of type N, connecting the two nodes C The type of nodes and their number serve as path attributes and length attributes of topological edges.

[0079] An example of a skeleton topology diagram is Figure 6 shown.

[0080] 3. Exploration Strategy Finite State Machine Based on Skeleton Topology

[0081] According to the topological graph extracted in the first two sections, a lot of information such as frontier, viewpoint, gain, exploration trajectory, etc. is implied. On this basis, a computationally efficient exploration strategy is proposed, which can be used as follows: Figure 7 The finite state machine shown is described, which includes six states, which are described in detail as follows.

[0082] State 1: Exploration guided by frontier nodes

[0083] This is the most important state in the entire exploration strategy. In this state, if the frontier node queue If it is empty, it enters state 3; if it is not empty, the frontier node queue The nodes in are classified into different branches according to their parent hub nodes. Nodes with the same parent hub node are considered to be in the same branch. If there is a branch that is the same as the current exploration branch, the nearest frontier node in the branch is sent to state 5; if not, enter state 2;

[0084] This approach avoids the complex construction and solution of the TSP problem; uses scene topology information to prioritize the completion of the current exploration branch, thereby minimizing the occurrence of BFMs; and, due to the characteristics of the skeleton, the target point is located in the center of the road, which is conducive to safe movement and observation, thereby improving overall exploration efficiency.

[0085] State 2: Exploration branch update

[0086] This means that the current branch exploration is complete, and a branch must be selected from the other candidate branches obtained in state 1 to be updated as the new current exploration branch. The selection strategy can be flexible. Here, according to the principle of small area priority, the branch corresponding to the node with the smallest corresponding wavefront value is greedily selected to be updated as the current exploration branch. As mentioned above, the smaller the wavefront count, the narrower the area. Other unselected branches are treated as a group and stored in the stack missedBranches for later backtracking. The system then transitions to state 5 and selects the nearest frontier node in the updated exploration branch as the local destination.

[0087] State 3: Backtracking

[0088] Go back in time from missedBranches to check whether the current group has any valid unexplored points. If so, go to state 2. If not, pop the next group from missedBranches for inspection. If no valid group is found until missedBranches is empty, the entire area has been explored and go to state 6.

[0089] State 4: Boundary-guided exploration

[0090] When the scene is too wide, the topology graph may not be able to continue growing. In this case, it is necessary to switch to boundary-guided mode. Only the grids of the state changes recorded by UFOMap are searched for the intersection of known and unknown nodes. All boundary grids are found. Then, only the distance between boundary grids is used as the cost to construct and solve the TSP problem, thus obtaining the exploration path sequence. Then, the process goes to state 5 and the first grid of the obtained exploration path is used as the local destination.

[0091] State 5: Path Planning

[0092] The A* algorithm is used for path planning. By expanding the skeleton topology graph in a tree-like manner, it can efficiently backtrack to parent nodes and retrieve paths and information from topological edges. Therefore, the path planning algorithm only needs to plan the mobile robot's current position and the first given path point, significantly reducing the search space for the path planning algorithm.

[0093] After planning is completed, if the robot approaches the local destination and the skeleton state corresponding to the local destination grid is still unknown state(skeleton, s) = unknown, it means that this is an overly wide scene. At this time, it will enter state 4 to reselect the local destination; otherwise, it will be sent to the motion control module for execution, and at the same time enter state 1 to continue exploration and decision-making.

[0094] Status 6 e 6): Exploration completed.

[0095] The preferred embodiments of the present invention have been described in detail above. It should be understood that numerous modifications and variations based on the concepts of the present invention are possible without inventive effort by those skilled in the art. Therefore, any technical solution that can be derived by one skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A mobile robot autonomous exploration method based on a topological graph, characterized in that: Leveraging the topology of the environment to guide exploration behavior.

2. The mobile robot autonomous exploration method based on topological graph according to claim 1, characterized in that: The skeleton topology map is used instead of the roadmap to abstract the topological structure of the environment from the geometric space.

3. The mobile robot autonomous exploration method based on topological graph as claimed in claim 2, characterized in that: The skeleton topology map first obtains the environment structure skeleton map, and then generates the environment topology map from the skeleton map.

4. The mobile robot autonomous exploration method based on topological graph as claimed in claim 2, characterized in that: Through the skeleton, the bifurcation point information is extracted to generate a skeleton topology map.

5. The mobile robot autonomous exploration method based on topological graph according to claim 4, characterized in that: The skeleton is cropped based on wavefront propagation.

6. The mobile robot autonomous exploration method based on topological graph according to claim 4, characterized in that: Incremental skeleton extraction based on wavefront propagation.

7. The mobile robot autonomous exploration method based on topological graph according to claim 6, characterized in that: By recording the values ​​of wavefront propagation and the types of trigger waves, incremental updates of skeleton extraction are achieved.

8. The mobile robot autonomous exploration method based on topological graph according to claim 1, characterized in that: Lightweight incremental update based on wavefront propagation, integrating frontiers, viewpoints, information gain and path planning.

9. The mobile robot autonomous exploration method based on topological graph according to claim 1, characterized in that: Encapsulate the exploration strategy as a finite state machine.

10. The mobile robot autonomous exploration method based on topological graph according to claim 1, characterized in that: The following steps are involved: Step 1: Input sensor data into SLAM and output the point cloud after positioning and registration; Step 2: UFOMap receives SLAM data and updates the status of the three-dimensional occupied grid; Step 3: skeleton extraction, which is performed incrementally based on the state-changing mesh; Step 4: Generate and prune the skeleton topology map based on the current exploration branch and the updated skeleton; Step 5: The exploration strategy finite state machine performs a turntable conversion and executes the exploration strategy according to the state of the skeleton topology graph; Step 6: Repeat the above steps until the state machine reaches the terminal state.

Images