Unmanned aerial vehicle autonomous exploration method based on historical boundary exploration path guidance

Abstract

The invention provides an unmanned aerial vehicle autonomous exploration method based on historical boundary exploration path guidance. The method comprises the following steps: constructing a grid map; generating an optimal viewpoint and a high-quality viewpoint set for local exploration path refinement by using a viewpoint quality judgment strategy based on the shielding rate; obtaining an optimal global exploration path traversing all boundaries through a global exploration path cost matrix optimization strategy and a historical guide global exploration path optimization strategy; constructing a local topological graph through a high-quality viewpoint set to refine a local exploration path, and searching an optimal exploration path with the lowest total cost by using a Dijkstra algorithm; and a yaw angle track capable of covering more boundaries is obtained through a multi-stage yaw angle planning strategy. According to the method, boundary cluster information around the target viewpoint is comprehensively considered, the yaw angle adjustment sequence is optimized, the influence of unmanned aerial vehicle yaw angle planning on the speed is comprehensively considered, and the exploration quality and efficiency of the unmanned aerial vehicle in various unknown environments are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle positioning and exploration, and particularly to an unmanned aerial vehicle autonomous exploration method based on historical boundary exploration path guidance. BACKGROUND

[0002] Multi-rotor unmanned aerial vehicles are particularly suitable as carriers for autonomous exploration tasks due to their high mobility, flexibility, and small size. The core task of autonomous exploration for unmanned aerial vehicles is to achieve real-time environmental perception, autonomous decision-making, and path planning in unknown environments to complete the exploration of unknown environments. The application scenarios are very wide, such as in post-disaster search and rescue tasks, unmanned aerial vehicles need to quickly traverse the debris area to find survivors; in urban planning, unmanned aerial vehicles can autonomously construct three-dimensional maps to support infrastructure design; in scientific research, unmanned aerial vehicles can penetrate unknown areas such as the polar region, forest, or ocean to collect valuable data.

[0003] Compared with traditional modes that rely on artificial or pre-set paths, autonomous exploration gives unmanned aerial vehicles higher flexibility and independence, enabling them to cope with complex and variable environments without external intervention. This ability not only improves task efficiency, but also significantly reduces the burden on operators, and even can replace humans to perform tasks in some high-risk scenarios.

[0004] Currently, although the related technology of autonomous exploration has initially realized the rapid autonomous exploration of unmanned aerial vehicles, there are still many technical bottlenecks:

[0005] Firstly, the traditional viewpoint sampling method does not consider the quality difference of viewpoints, resulting in a large number of low-quality viewpoints being executed for ray casting, which greatly wastes computing resources;

[0006] Secondly, there are defects in the design of the global exploration path sequence. Due to insufficient consideration of historical exploration information, problems such as inconsistent paths and redundancy often occur. Firstly, there is a lack of clear target guidance. Traditional methods are based on factors such as Euclidean distance, yaw angle change, and speed direction, which cannot provide global exploration paths with guiding properties for unmanned aerial vehicles, resulting in a lack of directionality in the exploration process. Secondly, the neglect of independent narrow areas, such as narrow areas near obstacles or environmental boundaries, is prone to be missed in exploration, and eventually these areas have to be revisited, resulting in a large amount of unnecessary return motion, significantly increasing the path length and exploration time, and reducing overall efficiency.

[0007] Thirdly, the yaw angle planning strategy is too simple to efficiently cover the environmental information. Due to the limited camera view, the traditional yaw angle planning method is usually only guided by the target viewpoint, and the environmental information of the surrounding boundary cluster is not fully considered. This limitation leads to the fact that the UAV is easy to miss the coverage and detection of the surrounding boundary cluster during the exploration process, forcing the UAV to move back and forth to scan the missed boundary information in the subsequent stage, thereby significantly increasing the path length and prolonging the exploration time.

[0008] These deficiencies seriously restrict the quality and efficiency of the autonomous exploration of the UAV, and therefore, there is an urgent need to provide a UAV autonomous exploration method based on historical boundary exploration path guidance. SUMMARY

[0009] In view of the deficiencies of the prior art, the present application provides a UAV autonomous exploration method based on historical boundary exploration path guidance, which can improve the exploration quality and efficiency of the UAV in various unknown environments.

[0010] The technical scheme of the present application is as follows: a UAV autonomous exploration method based on historical boundary exploration path guidance, comprising the following steps:

[0011] S1), constructing a grid map based on the depth image information of the depth camera sensor carried by the UAV;

[0012] S2), updating and clustering the boundaries in the grid map, and uniformly sampling the viewpoints with the cluster centers of each boundary as the cylinder centers; and generating the optimal viewpoints and a high-quality viewpoint set for local exploration path refinement using a viewpoint quality determination strategy based on the occlusion rate;

[0013] S3), obtaining the optimal global exploration path that traverses all boundaries through a global exploration path cost matrix optimization strategy and a historical guided global exploration path optimization strategy;

[0014] S4), refining the local exploration path by constructing a local topological map using the high-quality viewpoint set, and searching for the optimal exploration path with the lowest total cost using the Dijkstra algorithm;

[0015] S5), obtaining the yaw angle trajectory that can cover more boundaries through a multi-stage yaw angle planning strategy.

[0016] Preferably, in step S1), the voxels of the grid map include known free voxels, known collision voxels and unknown voxels; wherein the known free voxels represent the area that the UAV can safely pass through, corresponding to the space without obstacles; the known collision voxels represent the area where there are obstacles, marked as impassable; the unknown voxels represent the area that has not been perceived or explored, and the state information thereof is to be updated, and the state information of each voxel is represented by a discrete value.

[0017] As preferred, in step S2), the optimal viewpoint and the high-quality viewpoint set for local exploration path refinement are generated using the occlusion rate-based viewpoint quality determination strategy, specifically including the following steps:

[0018] S21), taking the cluster center of each boundary as the center of the cylinder, and then uniformly generating viewpoints V, i.e.:

[0019] V={v1,v2,...,v n};

[0020] Wherein, the position of the i-th viewpoint v i (p i ,ψ i ) is p i , and the yaw angle pointing to the boundary cluster center is ψ i ;

[0021] And the direction vector of the position p i of the viewpoint v i to the center p avg,i of the boundary cluster F i is taken as the main viewing direction of the camera;

[0022] S22), assuming that the viewing angle of the camera is FOV, and K sampling direction vectors {FOV1, FOV2,.., FOV K} are uniformly selected within the viewing angle range of the main viewing direction of the camera;

[0023] S23), ray casting detection is performed along each direction FOV k , if the ray preferentially contacts the obstacle, then the direction FOV k is occluded; if it contacts the boundary, it is considered as an effective coverage ray;

[0024] S24), the occlusion rate ray i of each viewpoint v i is calculated, i.e.:

[0025]

[0026] In the formula, K i is the occluded ray; K is the total number of sampling directions;

[0027] S25), the viewpoints are sorted from low to high according to the occlusion rate, the viewpoints with an occlusion rate lower than a preset occlusion threshold are screened, and the number of voxel coverage is calculated; then, the viewpoints are sorted from large to small according to the number of voxel coverage, and the first N vp viewpoints are selected as high-quality viewpoints to construct a high-quality viewpoint set VP.

[0028] As preferred, in step S3), the global exploration path cost matrix optimization strategy optimizes the ATSP cost matrix by introducing edge cost of exploration map and probability cost of independent narrow area in the cost matrix, i.e.:

[0029] M tsp = M tsp (0, k) + M tsp (i, j) + M tsp (j, i) + M tsp (k, 0);

[0030] wherein, m tsp is the cost matrix; M tsp (0, k) is the cost of the UAV from the current viewpoint vpo to the optimal viewpoint of each boundary cluster; M tsp (i, j) and M tsp (j, i) are the connection cost between the optimal viewpoints; M tsp (k, 0) is the cost from the optimal viewpoint of each boundary cluster to the current viewpoint vpo;

[0031] wherein, the cost M tsp (0, k) of the UAV from the current viewpoint vpo to the optimal viewpoint of each boundary cluster is expressed as:

[0032]

[0033] wherein, represents the time lower bound from the current viewpoint vpo to the optimal viewpoint of each boundary cluster; ω c is the weight coefficient of motion consistency; represents the motion consistency cost; ω d represents the weight coefficient of edge cost; is the edge cost of the optimal viewpoint exploring the map; is the probability cost of independent narrow area of the optimal viewpoint ; ω s is the weight of the probability cost. Other costs are expressed as:

[0034] m tsp (k, 0) = 0, k ∈ {1, 2,..., N cls};

[0035]

[0036]

[0037] wherein, is the cost of the UAV from the​​​​ to the time lower bound of the current viewpoint vp0; N cls is the number of the current boundary cluster.

[0038] As preferred, in step S3), the time lower bound from the current viewpoint vp0 to the optimal viewpoint of each boundary cluster is expressed as:

[0039]

[0040] where, is the total length of the path; is the safe path from the current viewpoint vp0 to the optimal viewpoint of each boundary cluster; p0, are the positions of the current viewpoint vp0 and the optimal viewpoint respectively; v max , ω max are the maximum speed and the maximum yaw rate respectively; ψ0, are the current viewpoint vp0, the optimal viewpoint and the yaw angle pointing to the boundary cluster center respectively.

[0041] As preferred, in step S3), the motion consistency cost is expressed as:

[0042]

[0043] where, α is an adjustment parameter; θ is the included angle between the current advancing direction of the UAV and the direction pointing to the next optimal observation point, used to represent the motion consistency; v0 is the speed direction of the UAV; p0, are the positions of the current viewpoint vp0 and the optimal viewpoint respectively.

[0044] As preferred, in step S3), the optimal viewpoint explores the edge cost of the map is expressed as:

[0045]

[0046] where, is expressed as the boundary cluster F where the optimal viewpoint k , is the cluster point position of the boundary cluster F where the optimal viewpoint k , is the shortest distance from the cluster point position to the map boundary along the x, y, z axes; D k is the optimal viewpoint distance from the current viewpoint vp0 of the UAV; r sensor is the maximum perception range of the camera; wherein, is the boundary cluster F k cluster center point is the shortest distance to each component of the exploration map edge.

[0047] As preferred, in step S3), the optimal viewpoint probability cost of the independent narrow area is expressed as:

[0048]

[0049] wherein, wherein is the optimal viewpoint is the boundary cluster F k , and is the cluster point position of the boundary cluster F k wherein the optimal viewpoint is located is the distance obtained by expanding from the cluster center point along the principal component direction and the reverse direction until the obstacle is contacted, and if there is no obstacle, the maximum expansion distance is l mid , l mid is half of the boundary cutting threshold.

[0050] As preferred, in step S3), the optimal global exploration path traversing all boundaries is obtained by using the historical guided global exploration path optimization strategy, and specifically includes the following steps:

[0051] S31), initialize the historical exploration path, and generate a temporary path as the initial historical perception path by traversing all boundary points in the current map i.e.

[0052]

[0053] wherein, p i is the position of the optimal viewpoint; N cls is the number of the current boundary cluster;

[0054] S32), define the position of the old boundary, and find the serial number of the old boundary sequence V old in the historical exploration path of the last time, wherein, and wherein the position p j of part of the optimal viewpoints is ∈V old ;

[0055] S33) Find the last old boundary point p k ; to explore the historical path of the previous moment Split into sequence A and sequence B, and remove the point p from the starting point to the last old boundary point in sequence A. k Given all the old boundary points between them, we get A = [p1, p2, ..., p k-1 The remaining sequence B is from p. k+1 The portion leading to the endpoint, i.e., B = [p] k+1 ,p k+2 ,...,p n ];

[0056] S34) Taking the current drone position p0 as the starting point, and the first point p of sequence B as the starting point... k+1 As the endpoint, combine sequence A and the newly added boundary sequence V. new The ATSP problem is constructed, and a new boundary sequence C is obtained. This new boundary sequence C represents the first point p of sequence B after traversing all newly added boundary points and sequence A from the current position of the UAV. k+1 The shortest path; then sequence C and sequence B are merged to obtain a new historical perception path.

[0057] S35), based on the current UAV position p0 and the newly added boundary sequence V new Generate an independent temporary shortest path P temp Then compare the temporary shortest path P. temp Cost C temp New paths to historical perception Cost C h With set threshold ∈ c The relationship between C; h -C temp <∈ c Then a new historical perception path will be used. If C h >C temp +∈ c Then use the temporary shortest path P. temp As a new global exploration path, and the temporary shortest path P is used... temp Update to the historical guidance path, i.e.:

[0058]

[0059] Preferably, in step S4), a local topology map is constructed using a high-quality set of viewpoints to refine the local exploration path, and the Dijkstra algorithm is used to search for the optimal exploration path with the lowest total cost; specifically, the following steps are included:

[0060] S41) Set the search threshold R cir And determine the search threshold R cir The number of boundary clusters N contained in the search area within the specified range. r ; Create graph nodes for the current viewpoint and the set of optimal viewpoints for each boundary cluster, and construct a directed acyclic graph; that is: connect the current position of the UAV to all the optimal viewpoints of the first boundary cluster, then connect all the optimal viewpoints of the first boundary cluster to all the optimal viewpoints of the second boundary cluster, and so on, layer by layer, until all the optimal viewpoints of the last boundary cluster are covered.

[0061] S42) Use Dijkstra's algorithm to search for the cost C. t Minimum optimal local exploration path VP opt Among them, the optimal local exploration path VP opt Represented as:

[0062]

[0063] In the formula, The search threshold R is respectively cir Optimal local exploration path VP within opt From middle boundary cluster 1 to boundary cluster N r The optimal viewpoint after local refinement.

[0064] Optimal Local Exploration Path (VP) opt Cost C t Represented as:

[0065]

[0066] In the formula, This indicates the distance from the UAV's position p0 to the optimal viewpoint. Lower bound of time; C c Represents the cost of motion consistency; λ c These are the weighting coefficients; Represents the optimal local exploration path VP opt The optimal viewpoint of the middle boundary cluster k after local refinement.

[0067] Preferably, in step S5), a multi-stage yaw angle planning strategy is used to obtain a yaw angle trajectory that can cover more boundaries, specifically including the following steps:

[0068] S51) Construct a coverage area S with the current UAV position p0 as the center and radius D; within the coverage area S, search for the center points of all boundary clusters, calculate the yaw angle from the current position to the center point of each boundary cluster, and select the maximum yaw angle ψ. max If the maximum yaw angle ψ max Yaw angle ψ to the optimal viewpoint of the targetg If the included angle does not exceed 90 degrees, then proceed to step S52; otherwise, use the ordinary yaw planning method for planning.

[0069] S52), calculate the yaw angle from the current yaw angle ψ0 to the maximum yaw angle ψ0 respectively. max , and from ψ max Yaw angle ψ to the optimal viewpoint of the target g The shortest times t1 and t2, and the shortest adjustment time t min ;Right now:

[0070]

[0071] t min =t1+t2;

[0072] In the formula, ω max Maximum yaw rate;

[0073] S53) Calculate the distance from the current position of the UAV to the optimal viewpoint of the target. The straight-line distance L at the set minimum speed v min Trajectory time T cur ,Right now:

[0074]

[0075] S54), compare the shortest adjustment time t min and trajectory time T cur If t min ≤T cur Then t min This serves as the lower bound for trajectory planning in time; and generates a trajectory from the UAV's current position to the target's optimal viewpoint position, obtaining the actual trajectory duration t. real Based on the actual trajectory duration t real The yaw angle adjustment time is as follows:

[0076]

[0077] In the formula, tψ1 and tψ2 are the yaw angle planning times for the first and second stages, respectively;

[0078] If t min >T cur Then for the maximum yaw angle ψ max Adjustments are made to make the new adjustment time T min ≤T cur ,Right now:

[0079] Δψ1=β·|ψ max -ψ0|;

[0080] Δψ2=β·|ψg -ψ nax |;

[0081] In the formula, Δψ1 and Δψ2 are the adjusted angle differences, respectively; β is the scaling factor.

[0082] S55), based on the adjusted angle difference, the maximum yaw angle ψ′ is recalculated. nax And the new shortest time T min , with T min As a lower bound for trajectory optimization, the actual trajectory duration T is obtained. real According to T real The yaw angle adjustment time is stretched to obtain the yaw angle planning time Tψ1 for the first stage and the yaw angle planning time Tψ2 for the second stage, ensuring that the yaw angle adjustment time matches the actual trajectory time.

[0083] The beneficial effects of this invention are as follows:

[0084] 1. This invention further improves the quality and efficiency of UAV exploration in various unknown environments;

[0085] 2. This invention filters high-quality viewpoints by evaluating occlusion rate, and performs ray projection calculations on the number of covered voxels only for high-quality viewpoints, thereby reducing computational complexity and improving exploration efficiency;

[0086] 2. This invention redesigns the ATSP cost matrix, taking into account environmental information, and introduces the edge cost of the exploration map and the probability cost of independent small areas into the cost matrix to generate a more reasonable global exploration path;

[0087] 3. This invention generates a more stable and consistent global exploration path by making full use of historical exploration path information. By preserving the access order of areas unaffected by map updates, it avoids making significant adjustments to the global path due to small-scale boundary changes. At the same time, it rationally integrates newly added boundary points into the historical exploration path to form a smooth transition path sequence. In this way, the consistency of the exploration direction can be maintained during the exploration process, reducing unnecessary backtracking.

[0088] 4. This invention refines the local exploration path and weighs other high-gain viewpoint combinations to obtain a path with higher exploration efficiency;

[0089] 5. This invention comprehensively considers the boundary cluster information around the target viewpoint, optimizes the yaw angle adjustment sequence, thereby improving boundary coverage efficiency, reducing unnecessary path backtracking, and comprehensively considers the impact of UAV yaw angle planning on speed, ensuring high-speed flight of the UAV without affecting coverage efficiency. Attached Figure Description

[0090] Figure 1 Flowchart of the method of the present application;

[0091] Figure 2 Schematic diagram of the view quality determination strategy based on the occlusion rate of the present application;

[0092] Figure 3 Schematic diagram of the probability cost and edge cost calculation of the present application;

[0093] Figure 4 Schematic diagram of the global exploration path generation guided by history of the present application;

[0094] Figure 5 Schematic diagram of the local exploration path refinement of the present application;

[0095] Figure 6 Schematic diagram of the multi-stage yaw angle planning strategy of the present application. DETAILED DESCRIPTION

[0096] The specific embodiments of the present application will be further described below in conjunction with the accompanying drawings:

[0097] As shown in the figure, the present embodiment provides a UAV autonomous exploration method based on historical boundary exploration path guidance, which includes the following steps: Figure 1 S1), constructing a grid map based on the depth image information of the depth camera sensor carried by the UAV;

[0098] In the present embodiment, the voxels of the grid map include known free voxels, known collision voxels and unknown voxels; wherein the known free voxels represent the area that the UAV can safely pass through, corresponding to the space without obstacles; the known collision voxels represent the area where there are obstacles, marked as impassable; the unknown voxels represent the area that has not been perceived or explored, and its state information is to be updated, the state information of each voxel is represented by a discrete value, for example, using 0 to represent unknown voxels, 1 to represent known free voxels, and 2 to represent known collision voxels, so as to efficiently store the structure information of the entire environment by a three-dimensional array.

[0099] S2), updating and clustering the boundaries in the grid map, and uniformly sampling the viewpoints with the cluster center of each boundary as the center of the cylinder; then using the view quality determination strategy based on the occlusion rate to generate the optimal viewpoint and the high-quality viewpoint set for local exploration path refinement; as shown in the figure, specifically including the following steps:

[0100] Figure 2 S21), taking the cluster center of each boundary as the center of the cylinder, and then uniformly generating the viewpoints V, i.e.:

[0101] V={v1,v2,...,v

[0102] V={v1,v2,...,v​n};

[0103] Wherein, the i-th viewpoint v i (p i ,ψ i p) i For position; ψ i The yaw angle pointing towards the boundary cluster center;

[0104] S22), viewpoint v i Position p i To the boundary cluster F i Center position p avg,i The direction vector is taken as the main view direction of the camera; and the camera's view is set as FOV. K sampling direction vectors {FOV1, FOV2, ..., FOV} are uniformly selected within the view range of the main view direction of the camera. K};

[0105] S23), FOV in all directions k Perform ray projection detection; if the ray preferentially contacts the obstacle, then the directional field of view (FOV) is determined. k If the ray is blocked, it is considered to be an effective coverage ray if it touches the boundary.

[0106] S24), calculate v for each viewpoint i occlusion rate ray i ,Right now:

[0107]

[0108] In the formula, K i The ray is blocked; K is the total number of sampling directions;

[0109] S25) Sort the viewpoints from low to high occlusion rate, filter viewpoints with occlusion rates below a preset occlusion threshold, and calculate their voxel coverage count; then sort them from largest to smallest voxel coverage count, and select the top N. vp Each viewpoint is used as a high-quality viewpoint to construct a high-quality viewpoint set VP. Viewpoints with high occlusion rates indicate limited observation of unknown space and offer little help for efficient exploration.

[0110] S3) The optimal global exploration path that traverses all boundaries is obtained through the global exploration path cost matrix optimization strategy and the history-guided global exploration path optimization strategy;

[0111] In this embodiment, the global exploration path cost matrix optimization strategy optimizes the ATSP cost matrix by introducing the edge cost of the exploration map and the probability cost of independent small areas into the cost matrix, that is:

[0112] M tsp =Mtsp (0,k)+M tsp (i,j)+M tsp (j,i)+M tsp (k,0);

[0113] In the formula, M tsp For the cost matrix; M tsp (0,k) represents the optimal viewpoint for the UAV from its current viewpoint vp0 to each boundary cluster. Cost; M tsp (i,j) and M tsp (j,i) represents the connection cost between optimal viewpoints; M tsp (k,0) represents the optimal viewpoint from each boundary cluster. Cost to the current viewpoint vp0.

[0114] like Figure 3 The diagram illustrates the calculation of probabilistic and edge costs. By comprehensively considering multiple factors such as flight distance, yaw angle variation, velocity direction consistency, cost of exploring map edges, and probabilistic cost of isolated small areas, a redesigned cost matrix is ​​obtained, leading to a more reasonable global exploration path for the ATSP problem. The solved path not only provides clear directional guidance, leading the UAV from the map edge inwards to cover unknown areas, but also prioritizes exploring boundary clusters located in isolated small areas, significantly reducing back-visits caused by missing corner regions. Furthermore, the path optimizes motion consistency, reducing drastic changes in the UAV's velocity direction and yaw angle, thereby lowering energy consumption and turning time. Overall, this global exploration path outperforms traditional methods in path length, exploration time, and coverage efficiency, providing a more efficient and robust solution for autonomous exploration tasks of UAVs in unknown environments.

[0115] In this embodiment, the UAV moves from the current viewpoint vp0 to the optimal viewpoint of each boundary cluster. Cost M tsp (0,k) is represented as:

[0116]

[0117] In the formula, This represents the distance from the current viewpoint vp0 to the optimal viewpoint of each boundary cluster. Lower bound of time; ω c The weighting coefficient for motion consistency; Represents the cost of motion consistency; ω d The weighting coefficient representing edge cost; For the optimal viewpoint Explore the edge costs of the map; For the optimal viewpoint the probability cost of the independent narrow region; ω s is the probability cost weight;

[0118] where the time lower bound from the current viewpoint vpo to the optimal viewpoint of each boundary cluster is represented as:

[0119]

[0120] where, is the total length of the path; represents the safe path from the current viewpoint vpo to the optimal viewpoint of each boundary cluster ; p0, v are the positions of the current viewpoint vpo and the optimal viewpoint , respectively; max , ω max are the maximum speed and the maximum yaw rate, respectively; ψ0, are the current viewpoint vpo, the optimal viewpoint , and the yaw angle pointing to the boundary cluster center, respectively.

[0121] the motion consistency cost is represented as:

[0122]

[0123] where, α is an adjustment parameter; θ is the included angle between the current advancing direction of the UAV and the direction pointing to the next optimal observation point, used to represent the motion consistency; v0 is the speed direction of the UAV; p0, are the positions of the current viewpoint vpo and the optimal viewpoint , respectively.

[0124] the optimal viewpoint explores the edge cost of the map is represented as:

[0125]

[0126] where, is represented as the boundary cluster F where the optimal viewpoint k , is the cluster point position of the boundary cluster F where the optimal viewpoint k is the shortest distance along the x, y, z axes from the cluster point position of the boundary cluster F k where the optimal viewpoint to the current viewpoint vpo of the UAV; r sensor ​​This represents the maximum sensing range of the camera; where, For the boundary cluster F k Cluster center Find the shortest distance to each component of the map's edge.

[0127] Optimal viewpoint The probability cost of an independent, small area The expression is:

[0128]

[0129] In the formula, where Represented as the optimal viewpoint The boundary cluster F k , and To achieve the best viewpoint The boundary cluster F k Cluster point location Starting from the boundary cluster center point, the maximum extension distance is the distance obtained after reaching the obstacle along the principal component direction and the opposite direction. If there is no obstacle, the maximum extension distance is l. mid , l mid It is half of the boundary cutting threshold.

[0130] Other costs are expressed as follows:

[0131] M tsp (k,0)=0,k∈{1,2,...,N cls};

[0132]

[0133] In the formula, From arrive Lower bound of time; N cls This represents the current number of boundary clusters.

[0134] like Figure 4 As shown, the history-guided global exploration path optimization strategy specifically includes the following steps:

[0135] S31) Initialize the historical exploration path by generating a temporary path as the initial historical perception path by traversing all boundary points in the current map. Right now:

[0136]

[0137] In the formula, p i The position of the optimal viewpoint; N cls This represents the current number of boundary clusters;

[0138] S32) Define the location of the old boundary and the historical exploration path in the previous moment. Searching for the old boundary sequence V old The serial number, where And the position p of some of the optimal viewpoints j ∈V old ;like Figure 4 As shown in (a);

[0139] S33) Find the last old boundary point p k ; to explore the historical path of the previous moment Split into sequence A and sequence B, and remove the point p from the starting point to the last old boundary point in sequence A. k Given all the old boundary points between them, we get A = [p1, p2, ..., p k-1 The remaining sequence B is from p. k+1 The portion leading to the endpoint, i.e., B = [p] k+1 ,p k+2 ,...,p n ];like Figure 4 As shown in (b);

[0140] S34) Taking the current drone position p0 as the starting point, and the first point p of sequence B as the starting point... k+1 As the endpoint, combine sequence A and the newly added boundary sequence V. new The ATSP problem is constructed, and a new boundary sequence C is obtained. This new boundary sequence C represents the first point p of sequence B after traversing all newly added boundary points and sequence A from the current position of the UAV. k+1 The shortest path; then sequence C and sequence B are merged to obtain a new historical perception path. like Figure 4 As shown in (c);

[0141] S35), based on the current UAV position p0 and the newly added boundary sequence V new Generate an independent temporary shortest path P temp ,like Figure 4 As shown in (d); then compare the temporary shortest path P. temp Cost C temp New paths to historical perception Cost C h With set threshold ∈ c The relationship between C; h -C temp <∈ c Then a new historical perception path will be used. If C h >C temp+∈ c then use the temporary shortest path P temp as the new global exploration path, and update the temporary shortest path P temp to the historical guidance path, that is:

[0142]

[0143] S4), refine the local exploration path by constructing a local topological graph through a high-quality viewpoint set, and search the optimal exploration path with the lowest total cost by using the Dijkstra algorithm; as shown in FIG. 5, the method specifically comprises the following steps: Figure 5

[0144] S41), set a search threshold R cir , and determine the number N cir of boundary clusters contained in the search area within the search threshold R r ; create graph nodes for the current viewpoint and the optimal viewpoint set of each boundary cluster, and construct a directed acyclic graph; that is, connect the current position of the unmanned aerial vehicle to all the optimal viewpoints connected to the first boundary cluster, then connect all the optimal viewpoints of the first boundary cluster to all the optimal viewpoints of the second boundary cluster, and so on, layer by layer, until all the optimal viewpoints of the last boundary cluster are covered;

[0145] S42), search the optimal local exploration path VP t with the minimum cost C opt by using the Dijkstra algorithm, wherein the optimal local exploration path VP opt is represented as:

[0146]

[0147] In the formula, respectively, are the optimal local exploration paths VP cir within the search threshold R opt , and the boundary cluster 1 to the boundary cluster N r are the optimal viewpoints after local refinement;

[0148] The cost C t of the optimal local exploration path VP opt is represented as:

[0149]

[0150] In the formula, represents the time lower bound from the position p0 of the unmanned aerial vehicle to the optimal viewpoint ; C c represents the motion consistency cost; λ c is a weight coefficient; represents the optimal local exploration path VP​opt Optimal viewpoint after local refinement of the middle boundary cluster k.

[0151] S5), obtain yaw angle trajectory covering more boundaries by multi-stage yaw angle planning strategy; as shown in the figure, specifically comprising the following steps: Figure 6

[0152] S51), taking the current position p0 of the unmanned aerial vehicle as the center and radius D to construct the coverage area S; wherein the radius D can be adjusted by the coefficient a, that is:

[0153] D = a r sensor ;

[0154] In the formula, r sensor is the maximum perception range of the camera;

[0155] In the coverage area S, search for the center points of all boundary clusters, calculate the yaw angle from the current position to the center point of each boundary cluster, and select the maximum yaw angle ψ max ; if the included angle between the maximum yaw angle ψ max and the yaw angle ψ g of the target optimal viewpoint is not more than 90 degrees, step S62) is executed, otherwise the ordinary yaw planning method is used for planning;

[0156] S52), respectively calculate the shortest time t1 and t2 from the current yaw angle ψ0 to the maximum yaw angle ψ max , and from ψ max to the yaw angle ψ g of the target optimal viewpoint, and the shortest adjustment time t min ; that is:

[0157]

[0158] t min = t1 + t2;

[0159] In the formula, ω max is the maximum yaw rate;

[0160] S53), calculate the straight line distance L from the current position of the unmanned aerial vehicle to the target optimal viewpoint The trajectory time T cur under the set minimum speed v min , that is:

[0161]

[0162] S54), compare the shortest adjustment time t min and the trajectory time T cur , if t min ≤ T cur , then t​min as the time lower bound of trajectory planning; and generate a trajectory from the current position of the UAV to the target optimal viewpoint position, and obtain the actual trajectory time t real , according to the actual trajectory time t real stretch the yaw angle adjustment time, that is,

[0163]

[0164] wherein tψ1 and tψ2 are the yaw angle planning time of the first stage and the second stage respectively;

[0165] If t min > T cur , the maximum yaw angle ψ max is adjusted, so that the new adjustment time T min ≤ T cur , that is,

[0166] Δψ1 = β·|ψ max - ψ0|; and

[0167] Δψ2 = β·|ψ g - ψ max |; and

[0168] wherein Δψ1 and Δψ2 are the adjusted angle difference; and β is a scaling factor;

[0169] S55), according to the adjusted angle difference, recalculate the maximum yaw angle ψ′ max and the new shortest time T min , take T min as the time lower bound of trajectory optimization, obtain the actual trajectory time T real , according to T real stretch the yaw angle adjustment time to obtain the yaw angle planning time of the first stage Tψ1 and the yaw angle planning time of the second stage Tψ2, so as to ensure that the yaw angle adjustment time matches the actual trajectory time.

[0170] S6), use Kinodynamic A* to perform path planning to obtain a safe path Path; and generate a smooth and safe MINCO trajectory based on the safe path Path; and send the generated yaw angle trajectory and the MINCO trajectory to the controller of the UAV. Wherein the MINCO trajectory is expressed as:

[0171]

[0172] wherein p(t) represents an m-dimensional M-segment polynomial trajectory, are the polynomial coefficients; the MINCO trajectory is defined by a piecewise linear complexity of a smooth mapping from the path points q and the time profile T of each piece of trajectory to the polynomial coefficients c which ensures high order continuity of the trajectory at the junction points.

[0173] The foregoing embodiments and descriptions are only illustrative of the principles and the best mode of the application, and the scope of the application should be determined not by the embodiments, but by the appended claims as construed according to the full breadth permitted by the patent laws.

Claims

1. A method for autonomous exploration of unmanned aerial vehicles (UAVs) based on historical boundary exploration path guidance, characterized in that, Includes the following steps: S1) Construct a raster map based on depth image information from the depth camera sensor mounted on the drone; S2) Update and cluster the boundaries in the raster map, and perform uniform viewpoint sampling with the cluster center of each boundary as the cylinder center; And an optimal viewpoint and a high-quality viewpoint set for refining local exploration paths are generated using a viewpoint quality judgment strategy based on occlusion rate. S3) The optimal global exploration path that traverses all boundaries is obtained through the global exploration path cost matrix optimization strategy and the history-guided global exploration path optimization strategy; S4) Construct a local topology map using a high-quality set of viewpoints to refine the local exploration path, and use Dijkstra's algorithm to search for the optimal exploration path with the lowest total cost; S5) A yaw angle trajectory that can cover more boundaries is obtained through a multi-stage yaw angle planning strategy.

2. The UAV autonomous exploration method based on historical boundary exploration path guidance according to claim 1, characterized in that: In step S2), an optimal viewpoint and a high-quality set of viewpoints for refining local exploration paths are generated using a viewpoint quality determination strategy based on occlusion rate. This specifically includes the following steps: S21) Using the cluster center of each boundary as the center of the cylinder, viewpoints V are then generated uniformly, i.e.: V={v1,v2,...,v n }; Wherein, the i-th viewpoint v i (p i ,ψ i p) i For position; ψ i The yaw angle pointing towards the boundary cluster center; and viewpoint v i Position p i To the boundary cluster F i Center p avg,i The direction vector is used as the camera's main viewpoint direction; S22) Let the camera's field of view be FOV. Within the field of view along the main view direction of the camera, uniformly select K sampling direction vectors {FOV1, FOV2, ..., FOV...} K }; S23), FOV in all directions k Perform ray projection detection; if the ray preferentially contacts the obstacle, then the directional field of view (FOV) is determined. k If the ray is blocked, it is considered to be an effective coverage ray if it touches the boundary. S24), calculate v for each viewpoint i occlusion rate ray i ,Right now: In the formula, K i The ray is blocked; K is the total number of sampling directions; S25) Sort the viewpoints from low to high occlusion rate, filter viewpoints with occlusion rates below a preset occlusion threshold, and calculate their voxel coverage count; then sort them from largest to smallest voxel coverage count, and select the top N. vp Each viewpoint is used as a high-quality viewpoint to construct a high-quality viewpoint set VP.

3. The UAV autonomous exploration method based on historical boundary exploration path guidance according to claim 2, characterized in that: In step S3), the global exploration path cost matrix optimization strategy optimizes the ATSP cost matrix by introducing the edge cost of the exploration map and the probability cost of independent small areas into the cost matrix, that is: M tsp =M tsp (0,k)+M tsp (i,j)+M tsp (j,i)+M tsp (k,0); In the formula, m tsp For the cost matrix; M tsp (0,k) represents the optimal viewpoint for the UAV from its current viewpoint vp0 to each boundary cluster. Cost; M tsp (i,j) and M tsp (j,i) represents the connection cost between optimal viewpoints; M tsp (k,0) represents the optimal viewpoint from each boundary cluster. Cost to the current viewpoint vp0.

4. The UAV autonomous exploration method based on historical boundary exploration path guidance according to claim 3, characterized in that: The drone moves from its current viewpoint vp0 to the optimal viewpoint of each boundary cluster. Cost M tsp (0,k) is represented as: In the formula, This represents the distance from the current viewpoint vp0 to the optimal viewpoint of each boundary cluster. Lower bound of time; ω c The weighting coefficient for motion consistency; Represents the cost of motion consistency; ω d The weighting coefficient representing edge cost; For the optimal viewpoint Explore the edge costs of the map; For the optimal viewpoint The probability cost of an isolated, small region; ω s For probability cost weights.

5. The UAV autonomous exploration method based on historical boundary exploration path guidance according to claim 4, characterized in that: In step S3), the optimal viewpoint Exploring the edge cost of the map Represented as: In the formula, Represented as the optimal viewpoint The boundary cluster F k , For the optimal viewpoint The boundary cluster F k Cluster point location The shortest distance to the map boundary along the x, y, z axes; D k For the optimal viewpoint Distance from the drone's current viewpoint vp0; r sensor This represents the maximum sensing range of the camera; where, For the boundary cluster F k Cluster center Find the shortest distance to each component of the map's edge.

6. The UAV autonomous exploration method based on historical boundary exploration path guidance according to claim 4, characterized in that: In step S3), the optimal viewpoint The probability cost of an independent, small area The expression is: In the formula, where Represented as the optimal viewpoint The boundary cluster F k , and To achieve the best viewpoint The boundary cluster F k Cluster point location Starting from the boundary cluster center point, the maximum extension distance is the distance obtained after reaching the obstacle along the principal component direction and the opposite direction. If there is no obstacle, the maximum extension distance is l. mid , l mid It is half of the boundary cutting threshold.

7. The UAV autonomous exploration method based on historical boundary exploration path guidance according to claim 4, characterized in that: In step S3), the optimal global exploration path that traverses all boundaries is obtained using a history-guided global exploration path optimization strategy. This includes the following steps: S31) Initialize the historical exploration path by generating a temporary path as the initial historical perception path by traversing all boundary points in the current map. Right now: In the formula, p i The position of the optimal viewpoint; N cls This represents the current number of boundary clusters; S32) Define the location of the old boundary and the historical exploration path in the previous moment. Searching for the old boundary sequence V old The serial number, where And the position p of some of the optimal viewpoints j ∈V old ; S33) Find the last old boundary point p k ; to explore the historical path of the previous moment Split into sequence A and sequence B, and remove the point p from the starting point to the last old boundary point in sequence A. k Given all the old boundary points between them, we get A = [p1, p2, ..., p k-1 The remaining sequence B is from p. k+1 The portion leading to the endpoint, i.e., B = [p] k+1 ,p k+2 ,...,p n ]; S34) Taking the current drone position p0 as the starting point, and the first point p of sequence B as the starting point... k+1 As the endpoint, combine sequence A and the newly added boundary sequence V. new The ATSP problem is constructed, and a new boundary sequence C is obtained. This new boundary sequence C represents the first point p of sequence B after traversing all newly added boundary points and sequence A from the current position of the UAV. k+1 The shortest path; then sequence C and sequence B are merged to obtain a new historical perception path. S35), based on the current UAV position p0 and the newly added boundary sequence V new Generate an independent temporary shortest path P temp Then compare the temporary shortest path P. temp Cost C temp New paths to historical perception Cost C h With set threshold ∈ c The relationship between C; h -C temp <∈ c Then a new historical perception path will be used. If C h >C temp +∈ c Then use the temporary shortest path P. temp As a new global exploration path, and the temporary shortest path P is used... temp Update to the historical guidance path, i.e.:

8. The UAV autonomous exploration method based on historical boundary exploration path guidance according to claim 7, characterized in that: In step S4), a local topology map is constructed using a high-quality set of viewpoints to refine the local exploration path, and Dijkstra's algorithm is used to search for the optimal exploration path with the lowest total cost; specifically, the following steps are included: S41) Set the search threshold R cir And determine the search threshold R cir The number of boundary clusters N contained in the search area within the specified range. r ; This will create graph nodes for the current viewpoint and the optimal set of viewpoints for each boundary cluster, constructing a directed acyclic graph; That is: connect the current position of the UAV to the high-gain viewpoint of the first boundary cluster, then connect all the best viewpoints of the first boundary cluster to the high-gain viewpoint of the second boundary cluster, and so on, layer by layer, until the high-gain viewpoint of the last boundary cluster is covered; S42) Use Dijkstra's algorithm to search for the cost C. t Minimum optimal local exploration path VP opt Among them, the optimal local exploration path VP opt Represented as: In the formula, The search threshold R is respectively cir Optimal local exploration path VP within opt From middle boundary cluster 1 to boundary cluster N r The optimal viewpoint after local refinement; Optimal Local Exploration Path (VP) opt Cost C t Represented as: In the formula, This indicates the distance from the UAV's position p0 to the optimal viewpoint. Lower bound of time; C c Represents the cost of motion consistency; λ c These are the weighting coefficients; Represents the optimal local exploration path VP opt The optimal viewpoint of the middle boundary cluster k after local refinement.

9. The UAV autonomous exploration method based on historical boundary exploration path guidance according to claim 8, characterized in that: In step S5), a yaw angle trajectory that can cover more boundaries is obtained through a multi-stage yaw angle planning strategy, specifically including the following steps: S51) Construct a coverage area S with the current UAV position p0 as the center and radius D; within the coverage area S, search for the center points of all boundary clusters, calculate the yaw angle from the current position to the center point of each boundary cluster, and select the maximum yaw angle ψ. max If the maximum yaw angle ψ max Yaw angle ψ to the optimal viewpoint of the target g If the included angle does not exceed 90 degrees, proceed to step S52; otherwise, use the ordinary yaw planning method for planning. S52), calculate the yaw angle from the current yaw angle ψ0 to the maximum yaw angle ψ0 respectively. max , and from ψ max Yaw angle ψ to the optimal viewpoint of the target g The shortest times t1 and t2, and the shortest adjustment time t min ;Right now: t min =t1+t2; In the formula, ω max Maximum yaw rate; S53) Calculate the distance from the current position of the UAV to the optimal viewpoint of the target. The straight-line distance L at the set minimum speed v min Trajectory time T cur ,Right now: S54), compare the shortest adjustment time t min and trajectory time T cur If t min ≤T cur Then t min As a lower bound for trajectory planning in time; It generates a trajectory from the UAV's current position to the target's optimal viewpoint position and obtains the actual trajectory duration t. real Based on the actual trajectory duration t real The yaw angle adjustment time is as follows: In the formula, tψ1 and tψ2 are the yaw angle planning times for the first and second stages, respectively; If t min >T cur Then for the maximum yaw angle ψ max Adjustments are made to make the new adjustment time T min ≤T cur ,Right now: Δψ1=β·|ψ max -ψ0|; Δψ2=β·|ψ g -ψ max |; In the formula, Δψ1 and Δψ2 are the adjusted angle differences, respectively; β is the scaling factor. S55), based on the adjusted angle difference, the maximum yaw angle ψ′ is recalculated. max And the new shortest time T min , with T min As a lower bound for trajectory optimization, the actual trajectory duration T is obtained. real According to T real The yaw angle adjustment time is stretched to obtain the yaw angle planning time Tψ1 for the first stage and the yaw angle planning time Tψ2 for the second stage, ensuring that the yaw angle adjustment time matches the actual trajectory time.

Images