An unmanned aerial vehicle autonomous exploration method for large-scale unknown scenes

Abstract

The application relates to a large-scale unknown scene-oriented unmanned aerial vehicle autonomous exploration method, which comprises the following steps: a voxel map is established by using binocular images, depth data in the binocular images and IMU data obtained in the flight process of an unmanned aerial vehicle; a large-scale unknown space is represented in a sparse manner to obtain clustering topological nodes of a plurality of divided spaces; a global coverage path is planned based on the clustering topological nodes of the plurality of divided spaces and the voxel map; local path planning is carried out according to an exploration viewpoint and the global coverage path, and trajectory optimization is carried out on the local path. The application can realize complete and efficient exploration of an unmanned aerial vehicle in a large-scale unknown scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle autonomous search, in particular to an unmanned aerial vehicle autonomous exploration method for large-scale unknown scenes. BACKGROUND

[0002] Unmanned aerial vehicles have broad application value in disaster relief, mine exploration and factory inspection due to their high flexibility, and how to improve the autonomous exploration efficiency in large-scale unknown scenes has attracted widespread attention from researchers. At present, the unmanned aerial vehicle autonomous exploration field mainly includes boundary-based, sampling-based and hybrid methods. The part between the explored free area and the unexplored area is usually established as an exploration boundary, and the unmanned aerial vehicle moves to the boundary and continuously updates the boundary during autonomous exploration.

[0003] Among them, the boundary-based method mostly adopts a greedy strategy, that is, the next exploration target is selected according to the minimum motion cost or the maximum information gain, which leads to local optimal exploration in large-scale unknown scenes, causing the unmanned aerial vehicle to miss some areas and have to revisit and explore, which greatly reduces the exploration efficiency. The sampling-based method mainly includes randomly sampling viewpoints and determining the best viewpoint for observing more unknown space through an utility function. Among them, the next best viewpoint planning is the earliest method widely used, which generates a rapidly expanding random tree through random sampling and calculates the branch with the maximum information gain to guide the robot movement. However, this method requires a large amount of calculation to evaluate the viewpoint gain, and the global information is also not fully considered in large-scale unknown scenes. In addition, the sampling-based method lacks directionality, is easy to produce a large number of redundant viewpoints, and may incorrectly estimate the exploration cost of some areas, affecting the exploration real-time performance and effectiveness.

[0004] As can be seen, both the boundary-based method and the sampling-based method are easily affected by local optimization, leading to incomplete exploration and back-and-forth movement and revisiting in the exploration area, which greatly affects the exploration efficiency. SUMMARY

[0005] The technical problem to be solved by the present application is to provide an unmanned aerial vehicle autonomous exploration method for large-scale unknown scenes, which can realize complete and efficient exploration of unmanned aerial vehicles in large-scale unknown scenes.

[0006] The technical solution adopted by the present application to solve the technical problem is: providing an unmanned aerial vehicle autonomous exploration method for large-scale unknown scenes, comprising the following steps:

[0007] S1, establishing a voxel map using the binocular map, depth data in the binocular map and IMU data obtained during the flight of the unmanned aerial vehicle;

[0008] S2, sparse representation of large unknown space to obtain a number of cluster topology nodes of divide and conquer space;

[0009] S3, planning global coverage path based on a number of cluster topology nodes of divide and conquer space and voxel map;

[0010] S4, local path planning according to the exploration viewpoint and the global coverage path, and trajectory optimization of the local path.

[0011] The step S2 of sparse representation of large unknown space to obtain a number of cluster topology nodes of divide and conquer space, specifically: through the decomposition and topological representation of the exploration space, the first sparse representation of the large unknown space is completed, a large number of topological nodes are obtained, the topological nodes are divided into corresponding divide and conquer spaces based on the divide and conquer idea, and the topological nodes in each divide and conquer space are clustered to realize the second sparse representation, thereby obtaining a number of cluster topology nodes of divide and conquer space.

[0012] The first sparse representation of the large unknown space through the decomposition and topological representation of the exploration space to obtain a large number of topological nodes, specifically includes:

[0013] (1) the large unknown space is hexagonally decomposed, during the decomposition, taking the coordinate (0, 0) as the origin, and the length E L and the width E W as the boundaries of the X-axis and Y-axis of the large unknown space, the hexagonal grids are arranged along the X-axis and Y-axis directions, and in the process of generating the grids, each column of hexagons is staggered and offset along the Y-axis direction.

[0014] (2) topological representation of the large unknown space after decomposition, the first sparse representation of the large unknown space is realized by using a topological graph, in which the topological nodes represent the areas occupied by the hexagonal grids, and the topological edges represent the passability of the unmanned aerial vehicle.

[0015] The topological nodes are divided into corresponding divide and conquer spaces based on the divide and conquer idea, and the topological nodes in each divide and conquer space are clustered to realize the second sparse representation, thereby obtaining a number of cluster topology nodes of divide and conquer space, specifically including:

[0016] (1) based on the divide and conquer idea, the topological nodes are divided into corresponding divide and conquer spaces according to their spatial distribution, thereby realizing the grouping of the topological nodes;

[0017] (2) the topological nodes in each divide and conquer space are clustered by using a k-mean clustering algorithm, thereby obtaining the cluster topology nodes of each divide and conquer space.

[0018] The step S3 plans a global coverage path based on the clustering topology nodes and the voxel map of the plurality of divide-and-conquer spaces, and specifically comprises:

[0019] (1) A method combining depth-first search and backtracking strategy is used to traverse and plan the clustering topology nodes to obtain a traversal path, and a midpoint of two clustering topology nodes connected by the traversal path is calculated, and the midpoint is taken as a connection point for sequentially exploring between the divide-and-conquer spaces;

[0020] (2) The grid cell where the unmanned aerial vehicle is located is decomposed, and unconnected regions are decomposed according to voxel occupancy information in the voxel map, and the center of each region is taken as a topology node, and the connectivity between regions is taken as an edge, and a connected graph is incrementally constructed to obtain free regions and unknown regions;

[0021] (3) The current position of the unmanned aerial vehicle is taken as a starting point, all topology nodes of the free regions and all topology nodes of the unknown regions are taken as exploration traversal points, and the connection point of the divide-and-conquer space adjacent to the position of the unmanned aerial vehicle is taken as a destination, and a global coverage path is planned.

[0022] The method combining depth-first search and backtracking strategy is used to traverse and plan the clustering topology nodes to obtain a traversal path, and specifically, the current position of the unmanned aerial vehicle is taken as a starting point, the Euclidean distance between the clustering topology nodes is taken as a traversal cost, and from the starting node, all unvisited adjacent clustering topology nodes are recursively traversed, and the path cost is recorded; when a certain clustering topology node is recursively searched, if it has been traversed, it is backtracked to the previous clustering topology node and other paths are tried; when all the clustering topology nodes are visited, the recursion ends and the traversal path is returned.

[0023] The center of each region is taken as a topology node, and the connectivity between regions is taken as an edge, and a connected graph is incrementally constructed, and specifically comprises:

[0024] For the edges in the free regions, an A* search algorithm is used to search for a path in two adjacent unit regions, if a path connecting the topology nodes in the two unit regions can be planned, the topology nodes between the two adjacent unit regions are connectable, and the path length is defined as the weight of the connected edge;

[0025] For a partially explored unit, the unit is divided, and the center of a known subunit or the center of a locally explored subunit is used to represent a free topology node, and the center of an unknown subunit is used to represent an unknown topology node; the free topology nodes are clustered to represent the center nodes of the known regions in the unit, and the unknown topology nodes in the unknown subunits are retained to guide exploration; an A* search algorithm is used to obtain a connection edge between the connectable free topology nodes and unknown topology nodes, and the path length is multiplied by a constant penalty factor to represent the weight of the connected edge;

[0026] For the edge in the unknown region, it is assumed that two adjacent unknown cells or sub-cells are connected, the path length between the topological nodes of the adjacent unknown cells or sub-cells is obtained by using A* search, and the path length is multiplied by a constant penalty factor to represent the weight of the edge.

[0027] The step S4 is according to the exploration viewpoint and the global coverage path to carry out local path planning, and the local path is optimized, and specifically includes:

[0028] (1) The exploration viewpoint of the unmanned aerial vehicle is determined by exploration boundary extraction and viewpoint sampling, and the unknown region is pushed along the global coverage path;

[0029] (2) Taking the current position of the unmanned aerial vehicle as the starting point and the exploration viewpoint of the unmanned aerial vehicle as the traversal point, the local exploration destination is determined according to the traversal order of the topological nodes in the global coverage path and the distance between the topological nodes in the path and the current position of the unmanned aerial vehicle, and the local exploration path is obtained by using the greedy asymmetric traveling salesman problem for solving;

[0030] (3) According to the motion constraint condition of the unmanned aerial vehicle, a trajectory optimization algorithm based on B-spline curve is used to optimize the local exploration path to obtain the local flight trajectory of the unmanned aerial vehicle.

[0031] The exploration viewpoint of the unmanned aerial vehicle is determined by exploration boundary extraction and viewpoint sampling, and specifically: the region between the explored free region and the unknown region is established as an exploration boundary, and then the representative viewpoint with coverage rate exceeding the threshold is obtained based on the exploration boundary by using a sampling method; for the same grid cell, the average pose of all representative viewpoints is calculated and taken as the exploration viewpoint of the unmanned aerial vehicle.

[0032] The technical scheme adopted by the present application to solve its technical problems is: a computer readable storage medium is provided, and a computer program is stored on the computer readable storage medium, the computer program is executed by a processor to realize the steps of the above-mentioned unmanned aerial vehicle autonomous exploration method for large-scale unknown scene.

[0033] Advantages

[0034] Compared with the prior art, the present application has the following advantages and positive effects: the present application decomposes the exploration space into uniform hexagonal grid in the sparse representation of large-scale unknown space, further divides the grid unit of local exploration and extracts the center of each region as a topological node, the space decomposition method can realize higher resolution representation of local exploration environment compared with the prior method, and is beneficial to guide the unmanned aerial vehicle to realize complete exploration. Meanwhile, for a large number of topological nodes generated by the sparse representation of large-scale unknown scene, the present application manages the nodes by using the divide-and-conquer idea, so as to convert the traversal planning problem of a large number of topological nodes into a plurality of identical sub-problems, compared with the prior method, the present application avoids the problems of storing and traversing a large number of topological nodes, and improves the path planning efficiency. Finally, the present application considers the guidance of global coverage path to the exploration viewpoint in local planning, solves the local optimal problem existing in autonomous exploration, avoids the redundant revisit in the exploration process, and is beneficial to improve the exploration efficiency. As can be seen from the above, the present application can realize complete and efficient exploration of the unmanned aerial vehicle in the large-scale unknown scene through the above process. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a system block diagram of the unmanned aerial vehicle autonomous exploration method for large-scale unknown scene of the first embodiment of the present application;

[0036] Figure 2 is a schematic diagram of the sparse representation of large-scale unknown space in the first embodiment of the present application;

[0037] Figure 3 is a connected graph constructed incrementally in the first embodiment of the present application;

[0038] Figure 4 is a schematic diagram of the global coverage path in the first embodiment of the present application;

[0039] Figure 5 is a schematic diagram of the local exploration path in the first embodiment of the present application. DETAILED DESCRIPTION

[0040] The present application will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present application and not used to limit the scope of the present application. In addition, it should be understood that those skilled in the art can make various modifications or changes to the present application after reading the content taught by the present application, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

[0041] The first embodiment of the present application relates to an unmanned aerial vehicle autonomous exploration method for large-scale unknown scene, and the system framework of the method is as follows Figure 1As shown, it mainly includes a UAV body system, a voxel mapping module, a sparse representation module of a large-scale unknown space, a global coverage path planning module, and a local path planning and trajectory optimization module. The UAV autonomous exploration method for a large-scale unknown scene in this embodiment specifically includes:

[0042] S1: The binocular image, depth data in the binocular image, and IMU data obtained during the flight of the UAV. The UAV body system in this embodiment adopts a quadrotor structure, carries a binocular camera and an IMU sensor, and can calculate the UAV odometer based on the obtained binocular image and IMU data in real time by using the VINS_Fusion algorithm, and then transmit the odometer together with the depth data in the binocular image to the voxel mapping module. In addition, the UAV body system adopts PX4 as a flight controller, converts the received optimized trajectory into a motor control signal, thereby realizing the access to each exploration viewpoint.

[0043] S2: The voxel mapping module takes the odometer and depth data as input to obtain a voxel map. In the voxel map, the distance of each voxel to the nearest surface is calculated by using the TSDF algorithm to represent the environment, and the distance value is stored in the voxel grid. A positive value indicates that the voxel is located outside the surface, and a negative value indicates that the voxel is located on the surface. In the application, only the voxels near the surface are stored, and the space utilization is optimized by using the sparse storage method.

[0044] S3: Sparse representation of a large-scale unknown space to obtain a plurality of clustered topological nodes of the divided space. In this step, the first sparse representation of the large-scale unknown space is completed by decomposing and topologically representing the exploration space, a large number of topological nodes are obtained, the topological nodes are divided into corresponding divided spaces based on the divide-and-conquer idea, and the second sparse representation is realized by clustering the topological nodes in each divided space to obtain a plurality of clustered topological nodes of the divided space.

[0045] In order to plan a global coverage guide path, the exploration space needs to be represented. The sparse representation of the large-scale unknown space in this embodiment includes the following steps as shown in Figure 2 .

[0046] (1) Hexagonal grid decomposition of the large-scale unknown space. When decomposing, the coordinate (0, 0) is taken as the origin, and the length E L and the width E W are taken as the boundaries of the X-axis and Y-axis of the large-scale unknown space, respectively, to determine the generation range of the hexagonal grid. The basic geometric parameters of the hexagonal grid include the side length S, the height H, and the width W, and the relationship between them is W = 2S. To form a honeycomb-like distribution, the hexagonal grid is arranged along the X-axis and Y-axis directions, with an X-axis coordinate interval of 0.75W and a Y-axis coordinate interval of H. In the process of generating the grid, each column of hexagons is staggered and offset in the Y-axis direction, with an offset y offset The current X-axis coordinate value x c is determined according to the following expression:

[0047]

[0048] where x0 is the horizontal coordinate value of the initial center point of the hexagonal grid, and mod represents the modulo operation.

[0049] (2) Topological representation of the large-scale unknown space after decomposition, first-layer sparse representation of the large-scale exploration space using a topological graph. In the first sparse representation, the topological nodes represent the areas occupied by the hexagonal grids, and the topological edges represent the passability of the UAV. Since the exploration space is an unknown scene, it is initially assumed that each node is passable. Assuming that the initial center point coordinates of the hexagonal grid are (x0, y0), the center coordinates of other hexagonal grids can be calculated according to the arrangement rules of the hexagonal grid, and the expression is:

[0050]

[0051] (3) To facilitate the conversion of the traversal planning of a large number of topological nodes into multiple identical small topological node planning problems, the topological nodes are divided into corresponding division spaces based on the divide-and-conquer idea according to their spatial distribution, achieving grouping of the topological nodes. In the division process, the average coordinates (x a ,y a ) of all topological nodes are first calculated, and the center lines passing through the coordinates and along the x-axis and y-axis are used as the division boundaries. Taking the division of topological nodes into four division spaces as an example, assuming that the coordinates of a topological node are (x i ,y i ), the division can be performed according to the following conditions: if x i ≥x a and y i ≥y a , the topological node is divided into the first division space; if x i ≤x a and y i ≥y a , the topological node is divided into the second division space; if x i ≥x a and y i ≤y a , the topological node is divided into the third division space; if x i ≥x aand y i ≤ y a If the number of topology nodes in the subspace is still large, recursive partitioning is performed on each subspace until the number of topology nodes meets the requirements of real-time path planning.

[0052] (4) In order to facilitate subsequent traversal planning of each subspace, the subspace needs to be sparsely represented. The k-mean clustering algorithm is used to cluster the topology nodes in each subspace to obtain the clustered topology nodes of each subspace. When clustering, the number of clustered nodes can be adjusted according to the size of the topology nodes in the subspace. In order to facilitate the representation of subsequent global coverage path planning, the number of clustered topology nodes of the subspace is set to 1 in this embodiment.

[0053] S4: Planning a global coverage path based on the clustered topology nodes of the subspaces and the voxel map.

[0054] In this step, in the global coverage path planning, first, a first layer global coverage path planning is performed, that is, the exploration order of each subspace is obtained by traversing the clustered topology nodes; then, the topology nodes in the subspace are planned in sequence in a second layer (the second layer planning is taken as an example in the figure with a solid line box dsp-1, after the exploration of the subspace is completed, the planning of dsp-2 to dsp-n is performed in sequence), and a connected graph is constructed according to the local area where the unmanned aerial vehicle is located during the exploration, and is aggregated with the second layer planning path to form the final global coverage guide path. Figure 1

[0055] This step of global coverage path planning includes:

[0056] (1) In order to obtain the optimal traversal order of each subspace, the depth-first search and backtracking strategy are combined to traverse the clustered topology nodes to obtain the traversal path, the midpoint of the two clustered topology nodes connected by the traversal path is calculated, and the midpoint is taken as the connection point of the exploration in sequence between the subspaces.

[0057] Specifically, the depth-first search and backtracking strategy are combined to traverse the clustered topology nodes to obtain the traversal path, that is, the current position of the unmanned aerial vehicle is taken as the starting point, the Euclidean distance between the topology nodes is taken as the traversal cost, from the starting node, all unvisited adjacent clustered nodes are recursively traversed, and the path cost is recorded; when a clustered node is recursively searched, if it has been traversed, it is backtracked to the previous clustered node and other paths are tried; when all clustered nodes are visited, the recursion ends and the shortest path is returned.

[0058] ​(2) In the sparse representation of large-scale unknown space, the obstacles are not considered in the ideal decomposition of the exploration space. In order to plan an effective global coverage path in actual exploration, the connectivity between the topological nodes needs to be updated in combination with the perception data. First, the grid cell where the UAV is located is decomposed, and the unconnected areas are decomposed according to the voxel occupancy information in the voxel map. The center of each region is taken as a topological node V, and the connectivity between regions is taken as an edge E. The connected graph is incrementally constructed, as shown in Figure 3 .

[0059] The incremental connected graph construction process realizes the construction of two connected graphs, namely the known free graph G f =(V f ,E f ) and the unknown graph G u =(V u ,E u ), which are used to represent free areas and unknown areas respectively. The connection between the two is realized using the bridging edge E p . Finally, the overall space can be represented by G=(V f ∪V u ,E f ∪E u ∪E p ).

[0060] In the incremental connected graph construction process, the local exploration unit is finely divided into sub-units, i.e. the hexagonal grid is evenly divided into six triangular grids, and the triangular grid is basically consistent with the UAV perception range, which helps to guide the UAV to realize fine exploration. For the explored unit or partially explored sub-unit, the average position of all voxels in the free area is calculated to construct the free topological node set V f , and the center of the region of the unexplored unit or the remaining sub-unit of the partially explored unit is extracted to construct the unknown topological node set V u . Only when the adjacent units / sub-units are both free or both unknown, they are considered to be connectable, and the connected edges are incrementally constructed using topological nodes, including the following processes:

[0061] For the edge E f in the free graph G f , the connectivity between the free topological node z m in the unit γ i and the free topological node z n in the adjacent unit γ j is evaluated in pairs. The A* search algorithm is used to search for a path in the two adjacent unit regions. If a path connecting the topological nodes in the two unit regions can be planned, the nodes z m and z n are connectable, and the path length is defined as the weight of the connected edge.

[0062] For the partially explored cell, first, the cell is finely divided, and the center of the known sub-cell or the center of the locally explored sub-cell is used to represent the free topology node, and the center of the unknown sub-cell is used to represent the unknown topology node. The free topology nodes are clustered to represent the center nodes of the known area in the cell, for example, z o , and the unknown topology nodes in the unknown sub-cell are used to guide the fine exploration. The A* search algorithm is used to obtain the connecting edge E p between the connected free topology nodes and unknown topology nodes. p Considering local uncertainty, the path length is multiplied by a constant penalty factor a u to represent the weight of the connecting edge.

[0063] For the edge E u in the unknown graph G u , since the scene structure in the space is unknown, it is assumed that two adjacent unknown cells (for example, topology nodes z r and z s ) or sub-cells (for example, topology nodes z p and z q ) are connected. The A* search algorithm is used to obtain the path length between the topology nodes in the adjacent unknown cells or sub-cells, and the path length is multiplied by a constant penalty factor b p to represent the weight of the edge.

[0064] After incrementally constructing the connected graph, if the connected graph G contains multiple disjoint components, it indicates that there is an inaccessible area in the environment, and in this case, the relevant topology nodes are directly deleted to be excluded from the exploration planning.

[0065] (3) According to the traversal order of the space exploration and the connected graph, the global coverage path is planned, specifically: taking the current position of the unmanned aerial vehicle as the starting point, taking all free area topology nodes and all unknown area topology nodes as exploration traversal points, and taking the adjacent points of the space of the unmanned aerial vehicle as the destination, the global coverage path is planned.

[0066] The traversal path of the topological nodes of the space without obstacles is shown in (a) of FIG. Figure 4 , but the actual exploration needs to combine the global coverage path planning with the connected graph, and the center of the explored area in the cell is used as the node for transition to the unknown area, and the local global coverage path considering connectivity is shown in (b) of FIG. Figure 4(b) in FIG. 6. In the above process, the traversal problem of each topology node needs to be solved, which can be summarized as an asymmetric traveling salesman problem. To minimize the exploration cost of the UAV, a cost matrix is constructed by evaluating the traversal time between each topology node, and the Lin-Hernighan heuristic algorithm is used to solve it, thereby obtaining the global coverage path. At the same time, when the UAV reaches the circular region with the topology node as the center and the radius of 2 meters, it is considered to reach the destination, at which time the UAV is guided into the next divide-and-conquer space according to the connected graph with the center node of the local exploration unit as the destination Figure 4 , and then the global path planning of the divide-and-conquer space is explored according to the above process. The above process is iterated until the path planning of all divide-and-conquer spaces is completed. In actual exploration, if the path to enter the next divide-and-conquer space cannot be searched according to the connected graph, the traversal path of the clustered topology node is re-planned.

[0067] In the asymmetric traveling salesman problem, the cost matrix C cp is calculated , first, a hybrid path planning method based on voxels and topology graphs is used to search for a path P c from the current position p n of the UAV to the topology node z cn . For short-distance search, i.e., when ||p c -z n ||<d thr , the A* algorithm based on voxels is used to search the path between the two points; for long-distance search, i.e., when ||p c -z n ||≥d thr , the A* algorithm is used to search the path on the connected graph closest to the two points. If there are unknown voxels / unknown topology nodes in the above path search, they are regarded as free voxels / free topology nodes, but the length of the searched path needs to be multiplied by the penalty term a p to represent the uncertainty of unknown voxels / unknown topology nodes. The above method is beneficial to realize efficient search of the path and speed up the calculation of the cost matrix.

[0068] Considering the kinematics of the UAV, higher costs should be assigned to longer and more tortuous paths. For the search path P c between the target positions p n and z cn ={p c ,z0,z1,...,z n}, its traversal time is estimated in a segmented manner. For a segment z i z i with a length of l i+1 , the traversal time t i is approximately calculated using a linear uniform acceleration motion model:

[0069]

[0070] where v m and a m are maximum speed and acceleration. is initial speed v i is the projection scalar of z i z i+1 along segment z i = 0 at the starting point of the UAV, and the expression of v i during the flight of the UAV is:

[0071]

[0072] The traversal time between p c and z n is calculated using the above formula as The traversal time between all pairs of target positions forms the cost matrix C cp .

[0073] S5: In local path planning and trajectory optimization, local path planning is performed according to the exploration viewpoint and the global coverage path, and trajectory optimization is performed on the local path to improve flight smoothness.

[0074] In local path planning and trajectory optimization, the global coverage path information is combined to explore the surrounding boundary in a flexible and globally optimized order, avoiding falling into local optimal exploration, which includes the following steps:

[0075] (1) The UAV exploration viewpoint is determined by exploration boundary extraction and viewpoint sampling, and the global coverage path is pushed forward to the unknown area to avoid revisiting exploration.

[0076] In exploration boundary extraction, the area between the explored free area and the unknown area is established as the exploration boundary, and then the representative viewpoints with coverage rate exceeding the threshold are obtained based on the exploration boundary using the sampling method. For the same grid cell, the average pose of all representative viewpoints is calculated and taken as the UAV exploration viewpoint. Since it comprehensively considers the distribution of the exploration boundary, it provides an effective position and orientation for guiding the UAV exploration, so the exploration viewpoint is taken as the traversal point of local exploration.

[0077] In the above exploration boundary extraction, the region growing algorithm is used to cluster the region between the explored free space and the unknown space adjacent to the explored free space to establish the exploration boundary, and the exploration boundary is also updated after the voxel map is updated using the latest sensor data, the exploration boundary whose occupancy state is changed, i.e., the unknown space is determined as free or occupied, is removed, and a new space boundary is searched, and the principal component analysis algorithm is used to divide the boundary cluster with a large size into two uniform boundaries.

[0078] (2) In local path planning, the current position of the unmanned aerial vehicle is taken as the starting point, the exploration viewpoint of the unmanned aerial vehicle is taken as the traversal point, and the local exploration destination is determined according to the traversal order of the topological nodes in the global coverage path and the distance between the topological nodes in the path and the current position of the unmanned aerial vehicle. Since the local region has fewer traversal nodes, the greedy asymmetric traveling salesman problem is used to obtain the local exploration path, as shown in (a) of FIG. 6. Figure 5

[0079] (3) In order to improve the smoothness of the flight of the unmanned aerial vehicle, the B-spline curve-based trajectory optimization algorithm is used to optimize the local exploration path according to the motion constraint condition of the unmanned aerial vehicle, so as to obtain the local flight trajectory of the unmanned aerial vehicle, as shown in (b) of FIG. 6. Figure 5

[0080] It can be found that the present application divides the exploration space into uniform hexagonal grids in the sparse representation of the large-scale unknown space, further divides the grid units of local exploration, and extracts the centers of each region as topological nodes. Compared with the existing method, the space decomposition method can realize a higher resolution representation of the local exploration environment, and is beneficial to guide the unmanned aerial vehicle to realize complete exploration. Meanwhile, for a large number of topological nodes generated by the sparse representation of the large-scale unknown scene, the present application uses the divide-and-conquer idea for management, so as to convert the traversal planning problem of a large number of topological nodes into a plurality of identical sub-problems, compared with the existing method, the problem of storing and traversing a large number of topological nodes is avoided, and the path planning efficiency is improved. Finally, in the local planning, the exploration viewpoint is traversed and explored considering the guidance of the global coverage path, the local optimal problem existing in the autonomous exploration is solved, the redundant revisit in the exploration process is avoided, and the exploration efficiency is improved. As can be seen from the above, the present application can realize complete and efficient exploration of the unmanned aerial vehicle in the large-scale unknown scene through the above process.

[0081] ​​The second embodiment of the present application relates to a computer readable storage medium, having stored thereon a computer program which, when executed by a processor, implements the steps of the above method for autonomous exploration of unmanned aerial vehicles facing large-scale unknown scenarios.

[0082] Those skilled in the art will appreciate that embodiments of the present application can be supplied as methods, systems, or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, magnetic disks and optical storage) embodying computer program code, computer readable program code, or code segments for use by or in connection with a computer or any instruction execution system.

[0083] The present application is described in reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowcharts and / or block diagrams block or blocks. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the function specified in the flowchart block or blocks.

[0084] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including an instruction means that implement the function specified in the flowcharts and / or block diagrams flow or flows and / or block or blocks. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the function specified in the flowchart block or blocks.

[0085] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowcharts and / or block diagrams flow or flows and / or block or blocks. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the function specified in the flowchart block or blocks.

[0086] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for autonomous exploration of unmanned aerial vehicles for large-scale unknown scenarios, characterized by, The method comprises the following steps: S1, a voxel map is established by using binocular images, depth data in the binocular images and IMU data obtained during the flight of the unmanned aerial vehicle; S2, sparse representation is performed on the large-scale unknown space to obtain clustered topological nodes of the divided spaces, specifically: first sparse representation is performed on the large-scale unknown space by exploring the decomposition and topological representation of the space to obtain a large number of topological nodes, the topological nodes are divided into corresponding divided spaces based on the divide-and-conquer idea, and the topological nodes in each divided space are clustered to achieve second sparse representation to obtain clustered topological nodes of the divided spaces; wherein the first sparse representation is performed on the large-scale unknown space by exploring the decomposition and topological representation of the space, a large number of topological nodes are obtained, specifically including: The large unknown space is hexagonally decomposed, and during the decomposition, the coordinate (0, 0) is taken as an origin, and the length E L and the width E W are respectively boundaries of X-axis and Y-axis of the large unknown space, the hexagonal grids are arranged along the X-axis and Y-axis directions, and during the generation of the grids, each column of hexagons is staggered and offset along the Y-axis direction; topological representation is performed on the decomposed large-scale unknown space, the topological graph is used to perform first sparse representation on the large-scale unknown space, and the topological nodes are used to represent the areas occupied by the hexagonal grids during the first sparse representation, the topological edges are used to represent the passability of the unmanned aerial vehicle; the topological nodes are divided into corresponding divided spaces based on the divide-and-conquer idea, the topological nodes are grouped, and the second sparse representation is achieved by clustering the topological nodes in each divided space to obtain clustered topological nodes of the divided spaces, specifically including: the topological nodes are divided into corresponding divided spaces according to their spatial distribution based on the divide-and-conquer idea to achieve grouping of the topological nodes; k-mean clustering algorithm is used to cluster the topological nodes in each divided space to obtain clustered topological nodes of each divided space; S3, global coverage path planning is performed based on the clustered topological nodes of the divided spaces and the voxel map; S4, local path planning is performed according to the exploration viewpoint and the global coverage path, and trajectory optimization is performed on the local path.

2. The method of claim 1, wherein, The global coverage path planning based on the clustered topological nodes of the divided spaces and the voxel map in step S3 specifically includes: (1) a method combining depth-first search and backtracking strategy is used to plan the traversal of the clustered topological nodes to obtain a traversal path, the midpoint of two clustered topological nodes connected by the traversal path is calculated, and the midpoint is taken as a connection point for sequentially exploring between the divided spaces; (2) the grid unit in which the unmanned aerial vehicle is located is decomposed, unconnected areas are decomposed according to the voxel occupancy information in the voxel map, the centers of the areas are taken as topological nodes, and the connectivity between the areas is taken as edges to incrementally build a connected graph to obtain free areas and unknown areas; (3) the topological nodes of all free areas and the topological nodes of all unknown areas are taken as exploration traversal points, and the connection points of the divided spaces adjacent to the location of the unmanned aerial vehicle are taken as destinations for planning the global coverage path.

3. The method of claim 2, wherein, The method of combining the depth-first search and the backtracking strategy is used to plan the traversal path of the clustering topological nodes, specifically: taking the current position of the unmanned aerial vehicle as the starting point, taking the Euclidean distance between the clustering topological nodes as the traversal cost, starting from the starting node, recursively traversing all unvisited adjacent clustering topological nodes, and recording the path cost; when the recursive search reaches a certain clustering topological node, if the clustering topological node has been traversed, the method backtracks to the last clustering topological node and tries other paths; when all the clustering topological nodes are visited, the recursion ends and the traversal path is returned.

4. The method of claim 2, wherein, The connected graph is incrementally constructed by taking the center of each region as a topological node and the connectivity between the regions as an edge, specifically including: For the edges in the free region, the A* search algorithm is used to search the path in two adjacent unit regions, if the path connecting the topological nodes in the two unit regions can be planned, the topological nodes between the two adjacent unit regions are connectable, and the path length is defined as the weight of the connected edge; For the partially explored unit, the unit is divided, and the center of the known sub-unit or the center of the locally explored sub-unit is used to represent the free topological node, and the center of the unknown sub-unit is used to represent the unknown topological node; the free topological nodes are clustered to represent the center nodes of the known regions in the unit, and the unknown topological nodes in the unknown sub-unit are retained to guide the exploration; the A* search algorithm is used to obtain the connecting edge between the connected adjacent free topological nodes and unknown topological nodes, and the path length is multiplied by a constant penalty factor to represent the weight of the connected edge; For the edges in the unknown region, it is assumed that the two adjacent unknown units or sub-units are connected, and the A* search algorithm is used to obtain the path length between the topological nodes in the adjacent unknown units or sub-units, and the path length is multiplied by a constant penalty factor to represent the weight of the edge.

5. The method of claim 1, wherein, The step S4 includes local path planning according to the exploration viewpoint and the global coverage path, and trajectory optimization of the local path, specifically including: (1) determining the unmanned aerial vehicle exploration viewpoint through boundary extraction and viewpoint sampling, and advancing along the global coverage path to the unknown region; (2) taking the current position of the unmanned aerial vehicle as the starting point, taking the unmanned aerial vehicle exploration viewpoint as the traversal point, determining the local exploration destination according to the traversal order of the topological nodes in the global coverage path and the distance between the topological nodes in the path and the current position of the unmanned aerial vehicle, using the greedy asymmetric traveling salesman problem to obtain the local exploration path; (3) using the B-spline curve-based trajectory optimization algorithm to optimize the local exploration path according to the motion constraint conditions of the unmanned aerial vehicle, to obtain the local flight trajectory of the unmanned aerial vehicle.

6. The method of claim 5, wherein, The unmanned aerial vehicle exploration viewpoint is determined through boundary extraction and viewpoint sampling, specifically: the region between the explored free region and the unknown region is established as an exploration boundary, and then the representative viewpoints with coverage rate exceeding a threshold are obtained based on the exploration boundary using a sampling method; for the same grid unit, the average pose of all representative viewpoints is calculated and taken as the unmanned aerial vehicle exploration viewpoint.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to realize the steps of the unmanned aerial vehicle autonomous exploration method for large-scale unknown scene in any one of claims 1-6.

Images