A UAV autonomous exploration method based on boundary guidance

Abstract

The invention discloses an autonomous exploration method for an unmanned aerial vehicle based on boundary guidance, the method comprising the following steps: step S1: constructing a three-dimensional grid map based on an octree, step S2: detecting boundary points based on voxel states, step S3: updating local and global boundary points, step S4: constructing a P_RRG based on a pruning strategy, step S5: sampling RRG bias based on a double-layer sliding window, step S6: local path planning based on path exploration gain, and step S7: global path planning based on the nearest observed boundary point. The autonomous exploration algorithm of the invention adopts an RRG hierarchical autonomous exploration algorithm based on boundary guidance, combined with boundary exploration and a fast exploration random graph algorithm. On the one hand, boundary points guide the fast exploration random graph bias growth, and on the other hand, it can ensure that the unmanned aerial vehicle can take into account both local and global planning, thereby improving the autonomous exploration efficiency of the unmanned aerial vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular to an autonomous exploration method of an unmanned aerial vehicle based on boundary guidance. Background Art

[0002] Autonomous exploration of drones is an important research direction in the field of drones. Drones with autonomous navigation capabilities can replace humans in tasks such as environmental exploration in unknown environments, thereby reducing the danger of human operation and search costs. The autonomy and intelligence of drones are the basis of autonomous exploration and have received widespread attention.

[0003] The current exploration methods are mainly divided into two methods: those based on boundary point detection and those based on viewpoint sampling. The former relies on the profit function model of independent boundary points, and often adopts a greedy strategy in the exploration process, selecting the boundary points with the highest profit value in each iteration for exploration. The latter selects the sampling path with the highest overall utility based on the RRT or PRM algorithm. Due to the limited viewing angle, the area with smaller exploration gain will be ignored, resulting in a high reciprocity of UAV exploration. The above exploration strategies all have their own advantages and limitations. It is difficult to achieve complete and rapid autonomous exploration using only a single method. Combining the advantages of various methods can complement each other and improve the efficiency of autonomous exploration.

[0004] The RRG algorithm randomly samples nodes in the task space based on RRT, and then reconnects each node in the random tree with other nodes whose distance is less than the set threshold to form a denser and rapidly expanding random graph, which can converge to the optimal solution faster than the RRT algorithm. However, on the one hand, due to the high randomness of the node expansion direction and position in RRG, it is difficult for the algorithm to explore efficiently when the environment changes. In this case, the number of leaf nodes and paths increases dramatically, and many invalid points and paths may be generated. On the other hand, when drones conduct autonomous exploration in unknown environments, they need to consider the coherence of global exploration. When there are discontinuous unknown areas, the exploration mission may terminate prematurely due to lack of global information.

[0005] Aiming at the problem of low exploration efficiency of UAVs in narrow and complex environments, boundary detection is combined with random sampling of viewpoints, and a boundary-guided RRG hierarchical autonomous exploration algorithm is proposed. By combining local and global exploration, the present invention aims to ensure efficient exploration of UAVs in complex environments. In the autonomous exploration process, boundary extraction and pruning-based RRG bias sampling are introduced to improve the sampling density of random graphs in unknown areas and ensure the integrity of autonomous exploration of UAVs. Summary of the invention

[0006] The purpose of the present invention is to provide a method for autonomous exploration of unmanned aerial vehicles based on boundary guidance. In view of the problem of low exploration efficiency of the RRG algorithm in narrow and complex environments, a RRG hierarchical autonomous exploration algorithm based on boundary guidance is proposed. First, boundary points are extracted in the sliding window to make up for the lack of random sampling of viewpoints, and at the same time, they are used as historical observation information to guide the unmanned aerial vehicle to unknown areas. Then, part of the feasible path is retained through the pruning strategy, and a local random graph is constructed based on biased sampling of target boundary points in a double-layer sliding window to improve the sampling density of unknown areas. Then, a lightweight global random graph is incrementally constructed through the local random graph, and the unmanned aerial vehicle is guided to explore unknown sub-areas in the map based on the strategy of the nearest observed boundary point to ensure the integrity of the unmanned aerial vehicle's autonomous exploration.

[0007] To achieve the above object, the present invention provides the following technical solution: a method for autonomous exploration of a drone based on boundary guidance, the method comprising the following steps:

[0008] Step S1: 3D grid map construction based on octree, using 3D lidar sensor to obtain information about the surrounding environment in real time for fusion and update, and constructing 3D grid map octomap based on octree. In the 3D grid map, small cubes of the same size - voxels are used to represent the passable area, prohibited area and unknown voxel area in the map environment, which are expressed as: ;

[0009] Step S2: Boundary point detection based on voxel state, based on the 3D grid map state, within the sliding window range Inside, centered on the body, boundary detection resolution Subdivide the detection area for the offset;

[0010] Step S3: Update local and global boundary points. When the UAV conducts autonomous exploration, local exploration and global exploration are performed alternately. During the exploration process, the UAV will extract boundary points of the local fixed sliding window in real time as the body moves until no new boundary is extracted, that is, when there are no observable boundary points around the UAV, the local planning ends and is converted to global planning. During global planning, the optimal boundary point in the global boundary will be selected as the next target boundary to continue exploring. During the exploration process, local and global boundary points will be updated in real time, and both local and global boundary points must meet the conditions of random graph viewpoints or UAV observations. In step S2, boundary point judgment is performed only based on the map state, but since the map state is updated with a certain probability, there may be some misidentification in the extraction of boundary points. The judgment conditions for boundary point update are as follows: ,

[0011] In the formula, represents the local boundary point, To extract the range of local boundaries, and Represents the viewpoint on the random graph and the field of view of the drone. When the detected boundary points meet the above judgment conditions, they will not only be stored in the local boundary point set and updated, but also be stored in the global boundary point set.

[0012] Step S4: P_RRG construction based on the pruning strategy does not completely abandon the random graph generated by the algorithm at the last moment, but retains part of the random exploration graph P_RRG within the local planning range when the drone moves to the next node. When the drone reaches the target point planned last time, within the sliding window of the local planning, all feasible paths from the drone to the random graph node are searched with the aircraft as the center, and then the pruned random graph P_RRG is constructed;

[0013] Step S5: Based on the RRG bias sampling of the double sliding window, a part of the random graph is retained in step S4, and a new local random graph is constructed based on the bias sampling of this graph. A real-time sliding window is determined according to the maximum range of the real-time point cloud around the body. , and according to the fixed sliding window Limit the range of the real-time sliding window and generate an adaptive sliding window ;

[0014] when When The target local boundary point is the center and the radius is Random sampling is performed within the range of to improve the sampling rate of random exploration graph nodes in unknown areas. The calculation formula is as follows: ,

[0015] In the formula, is the sampling point, The three target local boundary points with the largest exploration gain in the exploration direction;

[0016] when When the adaptive sliding window Random sampling in the probability interval retains the basic global search property, and the sampling threshold The larger it is, the more goal-oriented the sampling is, the better the global search and path planning effect is. The value of cannot be too large;

[0017] Step S6: Local path planning based on path exploration gain. When the UAV autonomously explores, multiple traversable paths can be generated with the body as the center on the local exploration random graph. The target point selection needs to be based on reasonable evaluation indicators, so that the UAV can explore unknown areas more quickly and comprehensively. The target point selection strategy based on the path exploration gain selects the path with the highest path exploration gain. The end point of the path is the local planning target point. The nodes on the random graph are regarded as viewpoints with detection function. The viewpoint gain is defined as the gain calculation volume. Unknown voxels within The selection of the quantity and gain calculation volume size can integrate the range of node sampling. The viewpoint gain function calculation in this algorithm is as follows: ,,

[0018] Define the path exploration gain as , the set of traversable paths in the random graph is recorded as , and the set of path points on each path , the path exploration gain is calculated as follows: ,

[0019] In the formula, Represents the shortest flight path in the set Path, Representative Viewpoint The gain, and Node Represents secondary path points At a radius of The number of local boundary points and viewpoints observed within the range; Represents the distance from the viewpoint to the drone, is a weight function used to exclude paths that are very different from the current exploration direction. is the weight coefficient;

[0020] Step S7: Based on the global path planning of the nearest observed boundary point, a lightweight global random graph is incrementally constructed on the basis of the local random graph to ensure that the explored area is covered. Then, the nearest observable global boundary point is searched on the global graph. Finally, the sampling radius is searched with the boundary point as the center. The viewpoint with the largest gain within the range is used as the target point of global planning. When the UAV covers a new unknown area during flight, the new boundary points are extracted and converted back into local planning. During the exploration process, the UAV alternates between local planning and global planning until there are no observable boundary points, and the exploration ends.

[0021] Furthermore, in step S2, each subdivision point is calculated by looping through the voxel states of each subdivision point and its surroundings. If the voxel where the subdivision point is located is in an unknown state and there is at least one unknown voxel and one free voxel around it, the point is judged to be a boundary point. When the boundary points are extracted, a large number of dense boundary points with similar positions will be released, and the boundary points will be downsampled using the voxel downsampling method.

[0022] Furthermore, in step S5, in the RRG bias sampling algorithm based on the double sliding window, a distribution mechanism based on uniform probability is introduced, and a sampling threshold is set in the random point generation process. , and generate random sample values , compare the random sample value with the sampling threshold.

[0023] The technical effects and advantages of the present invention are as follows: a lightweight global random graph is incrementally constructed through a local random graph, and the drone is guided to explore unknown sub-areas in the map based on the nearest observed boundary point strategy to ensure the integrity of the drone's autonomous exploration. The autonomous exploration algorithm of the present invention adopts the RRG hierarchical autonomous exploration algorithm based on boundary guidance, combined with boundary exploration and rapid exploration random graph algorithms. On the one hand, the boundary points guide the rapid exploration of the random graph bias growth, and on the other hand, it can ensure that the drone can take into account both local and global planning, thereby improving the autonomous exploration efficiency of the drone. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 It is the flow chart of the improved algorithm;

[0026] Figure 2 This is a schematic diagram of boundary point extraction based on sliding window;

[0027] Figure 3 Schematic diagram of some traversable paths within the local planning scope. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] The present invention provides a UAV autonomous exploration method based on boundary guidance, the process is as follows Figure 1 As shown, the method comprises the following steps:

[0030] Step S1: octree-based three-dimensional grid map construction, using a three-dimensional lidar sensor to obtain information about the surrounding environment in real time for fusion and update, and construct a three-dimensional grid map octomap based on the octree. In the three-dimensional grid map, small cubes of the same size, voxels, are used to represent the passable area, prohibited area and unknown voxel area in the map environment. The voxel status (idle, occupied, unknown) is represented as: ;

[0031] Step S2: Boundary point detection based on voxel state, based on the 3D grid map state, within the sliding window range Inside, centered on the body, boundary detection resolution Subdivide the detection area for the offset, calculate each subdivision point by looping, and judge the state of each subdivision point and the voxels around it. If the voxel where the subdivision point is located is in an unknown state and there is at least one unknown voxel and one free voxel around it, then the point is judged to be a boundary point, such as Figure 2 As shown, when extracting boundary points, a large number of dense boundary points with similar positions will be released, and the boundary points will be downsampled using the voxel downsampling method;

[0032] Step S3: Update local and global boundary points. When the UAV conducts autonomous exploration, local exploration and global exploration are performed alternately. During the exploration process, the UAV will extract boundary points of the local fixed sliding window in real time as the body moves until no new boundary is extracted, that is, when there are no observable boundary points around the UAV, the local planning ends and is converted to global planning. During global planning, the optimal boundary point in the global boundary will be selected as the next target boundary to continue exploring. During the exploration process, local and global boundary points will be updated in real time, and both local and global boundary points must meet the conditions of random graph viewpoints or UAV observations. In step S2, boundary point judgment is performed only based on the map state, but since the map state is updated with a certain probability, there may be some misidentification in the extraction of boundary points. The judgment conditions for boundary point update are as follows: ,

[0033] In the formula, represents the local boundary point, To extract the range of local boundaries, and Represents the viewpoint on the random graph and the field of view of the drone. When the detected boundary points meet the above judgment conditions, they will not only be stored in the local boundary point set and updated, but also be stored in the global boundary point set.

[0034] Step S4: P_RRG construction based on the pruning strategy does not completely abandon the random graph generated by the algorithm at the last moment, but retains part of the random exploration graph P_RRG within the local planning range when the drone moves to the next node. When the drone reaches the target point planned last time, within the sliding window of the local planning, all feasible paths from the drone to the random graph node are searched with the aircraft as the center, and then the pruned random graph P_RRG is constructed;

[0035] Step S5: Based on the RRG bias sampling of the double sliding window, a part of the random graph is retained in step S4, and a new local random graph is constructed based on the bias sampling of this graph. A real-time sliding window is determined according to the maximum range of the real-time point cloud around the body. , and according to the fixed sliding window Limit the range of the real-time sliding window and generate an adaptive sliding window ;

[0036] In step S5, in the RRG bias sampling algorithm based on the double sliding window, a distribution mechanism based on uniform probability is introduced, and a sampling threshold is set in the random point generation process. , and generate random sample values , compare the random sampling value with the sampling threshold;

[0037] when When The target local boundary point is the center and the radius is Random sampling is performed within the range of to improve the sampling rate of random exploration graph nodes in unknown areas. The calculation formula is as follows: ,

[0038] In the formula, is the sampling point, The three target local boundary points with the largest exploration gain in the exploration direction;

[0039] when When the adaptive sliding window Random sampling in the probability interval retains the basic global search property, and the sampling threshold The larger it is, the more goal-oriented the sampling is, the better the global search and path planning effect is. The value of cannot be too large;

[0040] Step S6: Local path planning based on path exploration gain. When the UAV autonomously explores, multiple traversable paths can be generated with the body as the center on the local exploration random graph. The target point selection needs to be based on reasonable evaluation indicators, so that the UAV can explore unknown areas more quickly and comprehensively. The target point selection strategy based on the path exploration gain selects the path with the highest path exploration gain. The end point of the path is the local planning target point. Figure 3 As shown in the figure, the nodes on the random graph are regarded as viewpoints with detection function, and the viewpoint gain is defined as the gain calculation volume Unknown voxels within The selection of the quantity and gain calculation volume size can integrate the range of node sampling. The viewpoint gain function calculation in this algorithm is as follows: ,

[0041] Define the path exploration gain as , the set of traversable paths in the random graph is recorded as , and the set of path points on each path , the path exploration gain is calculated as follows: ,

[0042] In the formula, Represents the shortest flight path in the set Path, Representative Viewpoint The gain, and Node Represents secondary path points At a radius of The number of local boundary points and viewpoints observed within the range; Represents the distance from the viewpoint to the drone, is a weight function used to exclude paths that are very different from the current exploration direction. is the weight coefficient;

[0043] Step S7: Based on the global path planning of the nearest observed boundary point, a lightweight global random graph is incrementally constructed on the basis of the local random graph to ensure that the explored area is covered. Then, the nearest observable global boundary point is searched on the global graph. Finally, the sampling radius is searched with the boundary point as the center. The viewpoint with the largest gain within the range is used as the target point of global planning. When the UAV covers a new unknown area during flight, the new boundary points are extracted and converted back into local planning. During the exploration process, the UAV alternates between local planning and global planning until there are no observable boundary points, and the exploration ends.

[0044] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A boundary-guided UAV autonomous exploration method, characterized in that: The method comprises the following steps: Step S1: constructing a three-dimensional grid map based on an octree; Step S2: Boundary point detection based on the voxel state of the three-dimensional grid map; Step S3: updating local and global boundary points based on observable conditions; Step S4: RRG construction based on pruning strategy; Step S5: RRG bias sampling based on double sliding windows; Step S6: local path planning based on path exploration gain; Step S7: global path planning based on the nearest observed boundary point; In step S4, a real-time sliding window W is determined according to the maximum range of the real-time point cloud around the aircraft body. lidar =(X min ,X max ,Y min ,Y max ), and according to the set fixed sliding window W f =(x min ,x max ,y min ,y max ) Limit the range of the real-time sliding window and generate an adaptive sliding window W a ={min(x min ,X min ),min(x max ,X max ),min(y min ,Y min ),min(y max ,Y max )}; In step S5, in the RRG bias sampling algorithm based on the double sliding window, a distribution mechanism based on uniform probability is introduced. In the process of random point generation, a sampling threshold 0<ε<1 is set, and a random sampling value ε is generated. rand =Rand()∈[0,1], compare the random sampling value with the sampling threshold; When ε<ε rand When <1, W f The target local boundary point is the center and the radius is R ε Random sampling is performed within the range of to improve the sampling rate of random exploration graph nodes in unknown areas. The calculation formula is as follows: q sample =q frontier +R ε *Rand[-1,1] In the formula, q sample is the sampling point, q frontier The three local boundary points of the target with the largest exploration gain in the exploration direction; When 0<ε rand <ε, in the adaptive sliding window W a Random sampling in .

2. The autonomous exploration method of a UAV based on boundary guidance according to claim 1, characterized in that: In step S2, each subdivision point in the local sliding window is calculated by looping and traversing, and the state of each subdivision point and the voxels around it is determined. If the voxel where the subdivision point is located is in an unknown state and there is at least one unknown voxel and one free voxel around it, the point is determined to be a boundary point. When the boundary points are extracted, a large number of dense boundary points with similar positions will be released, and the boundary points will be downsampled using the voxel downsampling method.

Images