A Multi-UAV Online Collaborative Exploration Coverage Method in Unknown Environments

Abstract

This invention belongs to the field of autonomous control and path planning technology for unmanned aerial vehicle (UAV) swarms. Specifically, it discloses a method for online collaborative exploration and coverage of multiple UAVs in unknown environments. The method includes constructing a basic data structure for online exploration and coverage of multiple UAVs in unknown environments and extracting an effective frontier cluster set. Based on the concept of the vehicle path problem, task nodes in the effective frontier cluster set are uniquely assigned to each UAV, generating a low-frequency global task plan. According to the low-frequency global task plan, each UAV locks onto a local target and generates a local obstacle avoidance trajectory using a three-dimensional tangent guidance strategy. During multi-UAV communication, an incremental synchronization mechanism is used to exchange local map increments and task states, and a frontier cluster occupancy locking and mutual exclusion advancement mechanism is used to resolve multi-UAV target conflicts. This invention solves the problems of existing technologies in unifying online environment representation, frontier target construction, global task organization, local feasible trajectory generation, and low-overhead multi-UAV communication protection.

Description

Technical Field

[0001] This invention belongs to the field of autonomous control and path planning technology for unmanned aerial vehicle (UAV) swarms, specifically relating to a method for online collaborative exploration and coverage of multiple UAVs in unknown environments. Background Technology

[0002] Unmanned aerial vehicles (UAVs) are widely used in disaster search and rescue, mountain patrol, and environmental monitoring due to their mobility, ease of deployment, and low personnel safety risks. In these tasks, UAVs typically need to continuously search, detect, and cover designated areas to obtain information such as terrain, obstacles, targets, or environmental conditions. For large-scale, heavily obstructed, complex terrain, or communication-restricted areas, a single UAV is constrained by its endurance, payload, sensor field of view, and mission time limits, making it difficult to efficiently complete exploration and coverage tasks. Therefore, collaborative exploration and coverage by multiple UAVs has significant application value.

[0003] Under known environmental conditions, coverage planning can typically be optimized offline based on a complete map and fixed task points. However, in scenarios such as disaster sites, mountain valleys, or urban complexes, the terrain, obstacles, threats, and passable spaces within the task area often need to be gradually revealed through sensor observations. Consequently, coverage planning is transformed into an online closed-loop process of simultaneous exploration and mapping, coverage operations, and path replanning.

[0004] Existing methods for exploring unknown environments often employ frontier-driven approaches, next-best-viewpoint methods, random sampling trees, or information gain maximization to generate candidate targets, enabling online exploration to a certain extent. However, in multi-UAV collaborative coverage scenarios, these methods still suffer from several drawbacks, including frequent changes in frontier targets with map updates, the tendency for multiple UAVs to repeatedly approach the same high-yield area, unbalanced task loads, high computational overhead for 3D local obstacle avoidance, difficulty in uniformly representing unstructured obstacles, and challenges in maintaining consistent collaborative states under communication constraints.

[0005] Therefore, there is an urgent need to propose an online collaborative exploration and coverage method for multiple UAVs in unknown environments that can unify online environment representation, frontier target construction, global task organization, local feasible trajectory generation, and low-overhead communication protection for multiple UAVs, so as to improve coverage efficiency, path security, and collaborative robustness in unknown and complex environments. Summary of the Invention

[0006] The purpose of this invention is to address the problems in existing technologies, such as frequent changes in frontier targets with map updates, the tendency of multiple drones to repeatedly approach the same high-yield area, unbalanced task load, large computational overhead for 3D local obstacle avoidance, difficulty in uniformly representing unstructured obstacles, and difficulty in maintaining consistent collaborative states under communication constraints. This invention proposes a multi-UAV online collaborative exploration and coverage method for unknown environments.

[0007] The technical solution of this invention is: a method for online collaborative exploration and coverage by multiple unmanned aerial vehicles in an unknown environment, comprising the following steps: Constructing a basic data structure for online exploration and coverage of unknown environments by multiple UAVs: Based on online grid maps, establish online grid information status and coverage markers in unknown environments, integrate the local observation increments of each UAV into a global dynamic map, and use axis-aligned bounding boxes to structurally represent unstructured obstacles; Based on the basic data structure, the frontier point set is extracted and clustered into a frontier cluster set. Then, the frontier cluster set is subjected to utility evaluation, accessibility gating and visibility gating to obtain an effective frontier cluster set. In response to the low-frequency global reorganization trigger signal, based on the concept of multi-machine global dynamic map and vehicle routing problem, the task nodes in the effective frontier cluster set are uniquely assigned to each UAV, and the access sequence of each UAV is optimized to generate a low-frequency global task plan. Based on the low-frequency global mission plan, each UAV locks onto a local target and generates a local obstacle avoidance trajectory that meets safety constraints using a three-dimensional tangent guidance strategy based on axis-aligned bounding boxes. During multi-machine communication, local map increments and task status are exchanged through an incremental synchronization mechanism, and multi-machine target conflicts are resolved by using a leading cluster occupancy locking and mutual exclusion advancement mechanism to maintain collaborative consistency and achieve online collaborative exploration and coverage of multiple UAVs in unknown environments.

[0008] The beneficial effects of this invention are: 1. This invention integrates online grid mapping, overlay marking, local-global map updating, and axis-aligned bounding box obstacle representation, which can uniformly maintain the semantics of unknown environments, overlay status, and obstacle geometry information, thereby improving the efficiency of online mapping, collision detection, and obstacle avoidance planning.

[0009] 2. This invention transforms dynamic point-level frontiers into stable cluster-level objectives through incremental frontier updates, clustering, benefit evaluation, and feasibility gating, thereby reducing task size, ineffective objective selection, and repetitive replanning.

[0010] 3. This invention uses the concept of vehicle routing problem to perform low-frequency two-parameter global task allocation and access sorting, achieving unique allocation and load balancing of front clusters, and reducing multi-machine target contention, duplicate coverage and path crossing.

[0011] 4. This invention uses a three-dimensional tangent-guided local trajectory generation based on axis-aligned bounding boxes, and combines target locking, escape, and debris sweeping mechanisms to improve continuous propulsion capabilities in complex obstacle, leading-edge jitter, and terminal debris scenarios.

[0012] 5. This invention maintains collaborative consistency under conditions of limited communication or time-varying topology by using incremental communication, cluster occupancy mutual exclusion, target exclusion, and expiration release mechanisms, thereby reducing communication load and supporting task recovery. Attached Figure Description

[0013] Figure 1 The diagram shows a flowchart of a multi-UAV online collaborative exploration and coverage method in an unknown environment.

[0014] Figure 2 The diagram shows the flight paths of multiple drones covering an unknown environment.

[0015] Figure 3 The image shows a schematic diagram of the status and overlay markers of online raster map information.

[0016] Figure 4 The diagram shows a schematic of horizontal tangent obstacle avoidance trajectory generation based on AABB.

[0017] Figure 5 The diagram shows a schematic of vertical overpass obstacle-around trajectory generation based on AABB.

[0018] Figure 6 The diagram shows a comparison of coverage rates for different global layer methods.

[0019] Figure 7 The diagram shows a comparison of unit returns for different global layer methods.

[0020] Figure 8 The diagram shows a comparison of the overlap rates between different global layer methods.

[0021] Figure 9 The diagram shows a comparison of coverage rates for different local layer methods.

[0022] Figure 10 The diagram shows a comparison of unit returns for different local layer methods.

[0023] Figure 11 The diagram shows a comparison of the overlap rates between different local layer methods. Detailed Implementation

[0024] Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, and are not intended to limit the scope of the invention.

[0025] Example 1: like Figure 1 As shown, a multi-UAV online collaborative exploration and coverage method in an unknown environment includes the following steps: S1. Constructing the basic data structure for online exploration and coverage of multiple UAVs in unknown environments: Based on online grid maps, establish online grid information status and coverage markers in unknown environments, integrate the local observation increments of each UAV into a global dynamic map, and use axis-aligned bounding boxes to structurally represent unstructured obstacles. S2. Based on the basic data structure, extract the frontier point set and cluster it into a frontier cluster set. Then, perform utility evaluation, accessibility gating and visibility gating on the frontier cluster set to obtain an effective frontier cluster set. S3. In response to the low-frequency global reorganization trigger signal, based on the idea of ​​multi-machine global dynamic map and vehicle routing problem, the task nodes in the effective frontier cluster set are uniquely assigned to each UAV, and the access sequence of each UAV is optimized to generate a low-frequency global task plan. S4. Based on the low-frequency global mission plan, each UAV locks onto a local target and generates a local obstacle avoidance trajectory that meets safety constraints using a three-dimensional tangent guidance strategy based on the axis-aligned bounding box. S5. During multi-machine communication, local map increments and task status are exchanged through an incremental synchronization mechanism, and multi-machine target conflicts are resolved by using a leading cluster occupancy locking and mutual exclusion advancement mechanism to maintain collaborative consistency and achieve online collaborative exploration and coverage of multiple UAVs in unknown environments. In this embodiment, step S1 specifically includes: S11. Establish online raster information status and overlay markers.

[0026] Specifically, in unknown environments, multi-UAV systems do not rely on complete offline maps, but instead continuously update environmental information based on real-time sensor observations from each UAV. The task space is discretized into a set of three-dimensional grids. For any grid cell ,exist Constantly maintain the status of its online map information for:

[0027] in, This represents a set of three-dimensional raster cells after the task space has been discretized. Represents any grid cell, Indicates the current moment. Represents grid exist Real-time online map information status; This indicates unobserved or unobserved graticules, i.e., unknown graticules; A grid that has been observed and satisfies the passability or flightability constraints is a free grid. Indicates obstacles, terrain obstructions, or non-flying spaces, i.e., occupies a grid cell; An unreachable grid indicates an area that is not an obstacle but is temporarily unreachable under the current connectivity or local planning conditions. This is used to prevent drones from repeatedly triggering invalid attempts in narrow passages, dead ends, or areas where repeated failures occur.

[0028] Maintain overlay markers for any grid cell Its values ​​are as follows:

[0029] in, Represents grid exist Time-over markers, Represents grid It has been effectively observed by the drone's sensors and included in the coverage statistics under unobstructed conditions. Represents grid It has not yet been effectively covered. Based on the above coverage flags, it is possible to define... Global coverage at any time Specifically, it is expressed as follows:

[0030] in, Indicates the number of grid cells that have been covered. This indicates the total number of grid cells within the task space.

[0031] S12. Establish a mechanism for merging local map updates and global dynamic maps.

[0032] Specifically, in a multi-UAV system, let the first... A drone in Maintain local maps constantly It is used for real-time decision-making and local replanning; at the global level, it maintains the merged global dynamic map. It is used for frontier target organization, cluster-level task allocation, and multi-machine conflict resolution. Let the first... A drone in The set of sensor observations obtained at each time point is The observation set includes visible obstacle information, terrain height samples, potential threat information, and coverage observation information. The local map of the UAV is then generated through a state transition operator. An update will be performed, specifically as follows:

[0033] in, Indicates the drone's serial number. Indicates the first A drone in A constantly maintained local map Indicates the first A drone in The set of sensor observations obtained at any given time, This represents the local map state transition operator. During the aforementioned local map update process, for grid cells within the sensor's field of view that meet the unobstructed condition, their state is changed from... Updated to or For a raster that meets the coverage criterion, mark it as covered. Set to 1 and accumulate coverage statistics; for areas that have been selected as candidate targets multiple times but have failed in local planning or are currently not connected, mark them as... Alternatively, a temporary exclusion set can be added to reduce the frequency of invalid replanning.

[0034] When any UAV satisfies the communication constraints, its local incremental set relative to the global map is used. Upload or share to the global layer, and use the fusion operator The global dynamic map is updated as follows:

[0035] in, express A global dynamic map at any given moment. This represents the global dynamic map after incorporating local increments. Indicates the first The local map increment of the drone relative to the global map. This represents the map fusion operator; This includes at least the addition of free grids, occupied grids, covered grids, and unreachable markers. By synchronizing only incremental information, communication bandwidth consumption is reduced and duplicate writes are suppressed, enabling the global dynamic map to maintain progressive consistency under communication-constrained conditions.

[0036] S13. Establish axis-aligned bounding box representations of unstructured barriers.

[0037] Specifically, to enable subsequent local track generation to quickly integrate online-updated unstructured terrain and obstacle information, the obstacles, terrain protrusions, or threat boundaries obtained from online observation are enclosed in an axis-aligned bounding box set. Let... The set of axis-aligned bounding boxes at time is Specifically, it is expressed as follows:

[0038] in, express The set of axis-aligned bounding boxes at each moment. Indicates the first A surrounding box, This represents the number of bounding boxes currently being maintained online. They represent the first Each bounding box has its lower and upper boundaries along the 3D coordinate axes. Each axis-aligned bounding box satisfies the following six boundary constraints:

[0039] in, This represents the coordinate components of the point to be detected in the three-dimensional coordinate system. Rewriting the above boundary constraints in a unified half-space form yields the... The equivalent representation of the half-space of a bounding box:

[0040] in, Indicates the first The half-space equivalent expression of a bounding box This represents the three-dimensional spatial point to be detected. Representing three-dimensional real space, and They represent the first The first enclosed box The normal vectors and boundary constants of the half-space constraints, superscript This indicates transpose.

[0041] To reduce the impact of positioning errors, sensing errors, and control errors on flight safety, a safety expansion process is performed on the axis-aligned bounding box to obtain an expanded bounding box. Specifically, it is expressed as follows:

[0042] in, Indicates the first The enclosed box is an expanded enclosed box after safety expansion. For horizontal safety margin, For vertical safety margin, bounding boxes that are spatially adjacent and have similar upper height bounds are merged using intersection-union ratio (IUGR) to reduce the computational complexity of subsequent local planning, as shown below:

[0043] in, Indicates bounding box and The intersection and union ratio, This represents the volume of the intersection of two bounding boxes. This represents the volume of the union of two bounding boxes. The merging threshold, For the bounding box merging operator, This is the bounding box after the merger.

[0044] Thus, the established online raster information status, coverage markers, local maps, global dynamic maps, and axis-aligned bounding boxes (AABB) structured obstacle representations together constitute the basic data structure for online exploration and coverage of multi-UAVs in unknown environments, and provide a unified input for subsequent frontier target generation, global mission reorganization, local track generation, and communication consistency protection.

[0045] In this embodiment, step S2 specifically includes: S21. Extract the frontier set based on the online raster map and perform incremental updates.

[0046] Specifically, in unknown environments, the frontier is used to characterize the boundary between known free space and unknown space, serving as a source of candidate targets for multi-UAV exploration and coverage missions. Based on the online raster information state obtained from S1, unknown raster sets are defined respectively. and free grid sets as follows:

[0047]

[0048] in, express A set of raster cells whose state is unknown at any given time. express The set of grid cells is in a free state at any given time. If a free grid cell is adjacent to an unknown grid cell, then the free grid cell is determined to be a leading edge grid cell, thus obtaining the instantaneous leading edge set. Specifically, it is expressed as follows:

[0049] in, express The set of instantaneous frontiers at any given moment. Represents grid Adjacent grid cells, For grid The neighborhood operator is used. To avoid increasing computational overhead due to point-by-point enumeration of the neighborhood, morphological operations are employed to construct the leading edge mask. First, the unknown grid mask is... The neighborhood is expanded, and then the intersection with the free grid mask is calculated. Permanently unreachable regions are removed to obtain the leading edge mask. Specifically, it is expressed as follows:

[0050] in, express The leading edge mask of time, Indicates a free grid indicator mask. Indicates an unknown grid indicator mask. express Structural elements For the dilation operator, This is a mask for permanently unreachable areas, used to avoid misclassifying areas outside the boundary, permanent obstacles, or permanently unreachable spaces as exploration targets. Mapping the true raster indices in the front mask to the world coordinate system yields the front point set:

[0051] in, Indicates the first The position of the front edge point in the horizontal world coordinate system Represents a two-dimensional real number space. express Number of points ahead of time.

[0052] Since the frontier set in the unknown environment changes frequently with map updates, an incremental update method driven by a change window is adopted to reduce the overhead of full map recalculation and minimize task assignment oscillations. Let the state change mask be... Specifically, it is expressed as follows:

[0053] in, express A mask representing the state change at a given moment relative to the previous moment. Indicates all grid cells are in State distribution at time, This represents the state distribution of all grid cells at the previous time step. The structuring element is used to perform state change masking. The expansion yields a locally updated domain. :

[0054] in, Indicates a local update field. This represents the structural element used to expand the state change region. The leading edge global layer performs morphological cleanup according to a set cleanup cycle, and the failed leading edge is retained according to a set lifetime threshold to suppress leading edge jitter. Only after each sensing write is this data processed. Refresh the frontier increment within the covered area to obtain:

[0055] in, express The incremental mask for the frontier needs to be updated constantly. This represents the global frontier layer. And will... Incorporate into the global frontier layer For grids that no longer meet the frontier determination criteria, a survival time threshold is accumulated, and they are removed once the threshold is exceeded. At the same time, closing operations and small hole removal are performed periodically to improve the temporal stability of the frontier morphology and the applicability of clustering.

[0056] S2. Cluster based on the frontier point set and construct candidate target utility.

[0057] Because the number of individual frontier points increases rapidly with map resolution, directly using frontier points as task units would result in excessively large allocation sizes and sensitivity to noise. Therefore, the frontier point set... Aggregates into a frontier cluster set Specifically, construct the adjacency graph of the frontier points. An edge is established when any two front edge points satisfy a distance threshold:

[0058] in, and Represent any two frontier points, This represents the Euclidean distance between any two front edges. This represents the threshold for connected distances between leading edges. This represents the set of adjacent edges of the front edge. Indicates the connection of the first The and the first The edges of each leading edge point are used. The above formula calculates the connected components of the adjacency graph to obtain candidate leading edge clusters. Components with a cluster size smaller than a threshold are considered noise and discarded. When all components are smaller than the threshold, all leading edge points are merged into one cluster to ensure uninterrupted online closed-loop progression. For any leading edge cluster... Its cluster center location Take the average position of the leading edge points within the cluster:

[0059] in, Indicates the first A frontier cluster, Indicates the first The cluster center of a frontier cluster, Cluster The number of frontier points included. This represents any leading edge point within the cluster. Furthermore, using... Construct a structure with a radius of centered on . neighborhood grid set Specifically, it is expressed as follows:

[0060] in, Indicates Centered on, with radius The neighborhood raster set, Indicates the radius of the candidate target benefit assessment. For grid Mapping to the world coordinate system. Let Indicates from candidate target Observation grid Whether the line of sight is unobstructed or not, the observable information gain of the leading edge cluster. It is expressed as follows:

[0061] in, Indicates the first A cutting-edge cluster in The observable information gain at time step, Indicates from the cluster center To grid The value for determining unobstructed line of sight.

[0062] Net new coverage potential in the neighborhood It is expressed as follows:

[0063] in, Indicates the first A cutting-edge cluster in Net new coverage potential at any time Used to exclude already covered grid cells. To reduce duplicate coverage and repeated traversal, the coverage rate within the neighborhood of the leading cluster is... Statistical analysis:

[0064] in, Indicates the first Coverage rate within the neighborhood of each frontier cluster This indicates the number of graticles contained in the neighborhood. An exponential redundancy reduction factor is then constructed. :

[0065] in, Indicates the first Redundancy reduction factor of frontier clusters, This represents an exponential function with the natural constant as its base. This represents the redundancy reduction strength factor. From this, the effective new coverage potential after reduction is obtained. :

[0066] For the A drone in Location at any moment With frontier cluster center Define the arrival cost in the passable grid space. :

[0067] in, Indicates the first The drone to the The cost of reaching the center of a frontier cluster Indicates the first A drone in Location at any given moment This is the set of feasible paths from the current location of the drone to the center of the leading cluster. This represents any one of the feasible paths. Representing a path The first A discrete path point, This represents the number of path points. Considering both effective coverage potential and arrival cost, a utility function is constructed for cluster sorting and allocation. :

[0068] in, Indicates the first The drone was selected as the first The utility value of a frontier cluster, To cover potential weights, To achieve cost weighting, the point-level frontier is transformed into a cluster-level candidate objective with benefit, cost, and redundancy attributes through the above method, reducing the task dimension of subsequent global organization.

[0069] In this embodiment, step S3 specifically includes: S31. Establish a low-frequency global reorganization triggering mechanism driven by multiple constraints.

[0070] Specifically, in unknown environments, the frontier ensemble continuously evolves with UAV observations. If global task allocation and local online decision-making operate at the same high frequency, slight fluctuations in the frontier can easily lead to frequent target switching, plan oscillations, and increased communication load. Therefore, this invention employs a multi-timescale framework combining low-frequency global organization with high-frequency local execution, performing global task reorganization only when reorganization conditions are met.

[0071] Let the set of globally stable frontier points be... Its scale is Specifically, it is expressed as follows:

[0072] in, express The global stable frontier set at time t, This represents the number of globally stable frontier points. Operators are extracted for the globally stable frontier. This refers to the global dynamic map obtained through fusion in step S1. Let the time of the last successful generation of the global plan be... Define the relative rate of change of the frontier size as follows:

[0073] in, This represents the rate of change of the current global frontier size relative to the last global planning time. This indicates the number of frontier points recorded when the global plan was successfully generated last time.

[0074] Let the global reorganization cycle be... The threshold for the rate of change of the frontier scale is Then the global reorganization trigger signal It is expressed as follows:

[0075] in, This indicates a global reorganization trigger signal; Indicates periodic triggering. This indicates an event-triggered event. Periodic triggering is used to prevent global plans from becoming outdated over time, while event triggering is used to quickly respond to changes in the environmental structure when the current edge quantity suddenly increases or decreases.

[0076] To prevent global-level anomalies caused by null fronts or invalid plans, null front gating and plan validity gating are further introduced. Let the current global plan cache be... The function for determining the effectiveness of the plan is: Then the global reorganization execution criteria It is expressed as follows:

[0077] when Upon inception, a global plan is generated based on the vehicle routing problem concept, and the plan cache, frontier scale statistics, and reorganization time are updated, as detailed below:

[0078]

[0079] in, This indicates the criteria for executing the overall restructuring. Indicates the current global plan cache; This represents a global plan generation operator based on the concept of the vehicle routing problem. This represents the set of current locations of all drones. For the number of drones, For the first A drone in The position at that moment; and This indicates that the current frontier scale and current moment will be written into the global reorganization record.

[0080] S32. Construct cluster-level task nodes, revenue normalization, and edge cost functions.

[0081] Specifically, based on the effective frontier cluster set obtained in step S2, the centers of each frontier cluster are used to form a cluster-level task node set. Specifically, it is expressed as follows:

[0082] in, express The set of cluster-level task nodes at any given moment. The current number of effective frontier clusters, For the first A frontier cluster center. To adapt to changes in the revenue scale under different scenarios and task stages, the corrected revenue obtained in S2 is... Normalization is performed to obtain the expected return per unit. :

[0083] in, Indicates the first Normalized unit expected return of a frontier cluster, This represents the maximum correction gain among all currently effective frontier clusters.

[0084] To ensure that global task organization simultaneously considers distance cost, line-of-sight occlusion, and the risk of repeated traversal, a system is constructed starting from a point... Time edge cost function Let the distance between the two points be... Linear accessibility is The overlap coverage ratio caused by the path crossing an already covered area is: ,but:

[0085] in, Indicates from point Time The cost of the edge This represents the Euclidean distance between two points. Point and points Visual accessibility between them This represents the gaze detection function. This indicates the percentage of overlap caused by the path traversing an already covered area. The penalty coefficient for non-direct-view advancement. This is the penalty coefficient for repeated traversal.

[0086] This allows us to construct the cost matrix from UAV to cluster. Cluster-to-cluster cost matrix Specifically, it is expressed as follows:

[0087]

[0088] in, Indicates the first The drone's current location to the The cost of reaching a frontier cluster Indicates the first The first frontier cluster to the first The transfer cost of a frontier cluster Indicates the first A frontier cluster center.

[0089] S33. Cluster-level task package unique allocation and global plan generation.

[0090] Specifically, to achieve mutual exclusion and load balancing among multiple UAVs, this invention approximates the cluster-level vehicle path problem in an online greedy manner at the global layer. For the first... The cumulative cost of maintaining the current mission package using drones and the cluster index at the end of the task package For any unassigned front cluster Define its appended to the first The base incremental cost at the end of a drone mission package as follows:

[0091] in, This indicates that the unassigned frontier clusters will be... Added to the The basic incremental cost at the end of a drone mission package. Indicates the first The terminal cluster index of the current mission packet of the drone; when The condition indicates that the task package is empty. In this case, the cost from the UAV to the cluster is used; otherwise, the cost from the end cluster to the candidate cluster is used.

[0092] A comprehensive evaluation function is constructed by introducing a load balancing regularization term and a revenue reward term on top of the basic incremental cost. :

[0093] in, Indicates the frontier cluster Assigned to the The overall evaluation value when flying a drone This is a load balancing regularization coefficient used to suppress overload caused by a drone being continuously inserted into tasks; This is a value-reward strength used to prioritize the allocation of high-yield frontier clusters. Each selection makes... The smallest "drone-frontier cluster" combination Perform the allocation and update. and This continues until all valid frontier clusters have been allocated.

[0094] Therefore, we obtain the first... Cluster-level mission package collection for drones :

[0095] And it satisfies the cluster-level mutual exclusion coverage constraint:

[0096] in, Indicates the first A drone in Cluster-level task packages at specific times. Through this method, each effective front cluster is assigned to only one UAV within the same global planning cycle, avoiding repeated convergence of multiple UAVs to the same front cluster.

[0097] After obtaining the cluster-level mission packages for each drone Subsequently, the global layer further generates cluster-level access sequences for each UAV to enhance spatial continuity and reduce cross-regional interference. For the first... First, based on drones and Construct the initial access sequence Then, perform 2-opt swaps within a finite number of iterations to reduce the total cost of open-chaining. :

[0098] Ultimately, the low-frequency plan output by the global layer Represented as:

[0099] in, Indicates the first The total open-chain cost of a cluster-level access sequence for a drone. For the first Cluster-level access sequence of drones Indicates the first in the access sequence A frontier cluster index, Indicates the number of frontier clusters in the task package; This indicates a low-frequency plan output from the global layer. A cluster-level set of task nodes. For the first The mission package for deploying drones, Indicates all A collection of drones.

[0100] Thus, step S3 achieves the structured organization, unique allocation, and load-balanced propulsion of the unknown environment frontier clusters in the low-frequency global layer, providing a stable prior for the subsequent generation of high-frequency tracks in the local layer.

[0101] In this embodiment, step S4 specifically includes: S41. Local target locking based on global task package.

[0102] Specifically, the cluster-level task package output by the low-frequency global layer only provides the macroscopic propulsion direction and access order for each UAV. During actual execution, the local layer still needs to continuously generate local trajectories that satisfy 3D reachability, obstacle avoidance, line-of-sight occlusion, and platform motion constraints. For the... The drone is deployed, and the local layer follows the access sequence output by S3. Select the current primary target To avoid frequent target switching due to short-term frontal jitter or partial occlusion, a lock count is set for the current primary target. .when At that time, select a new primary target from the task package header and reset the lock count; within each local planning cycle, perform the following updates:

[0103] in, Indicates the first A drone in The target lock count is maintained at each moment. During the lock period, the local layer prioritizes generating tracks around the current primary target; when the target is reached, the target fails, or the lock count expires, the system switches to the next candidate target in the task package. This mechanism maintains planned inertia between the global task package and local execution, reducing track discontinuities caused by target jitter.

[0104] S42. Visible obstacle screening and direct segment collision detection.

[0105] Specifically, in each local planning cycle, the first The drone updates the local map based on current sensor observations, and obtains the current set of visible obstacles using the AABB structured representation in S1. Let the drone's current position be... The current local target point is For directly connected line segments With the dilated AABB set Perform collision detection:

[0106] in, Indicates the current location of the drone. Indicates the current local target point. Indicates from arrive The straight line segment, This represents the collision detection function. express The expanded AABB set at each moment. If the above conditions are met, it means that there is no collision on the direct connection from the current position to the target point, and the local layer advances directly along this direction; if the collision detection result is non-zero, the first collision AABB is identified and tangential guidance around the obstacle is triggered.

[0107] When a collision occurs in the direct connection segment, let the expansion AABB of the first collision be... To ensure that the candidate points for horizontal obstacle avoidance have consistent geometric constraints, a system is first established on the horizontal plane. Projected straight line:

[0108]

[0109] in, Represents the coordinates of any point on the horizontal plane. Represents the coefficients of the equation of the projected line. Indicates the current position Horizontal coordinates Indicates the target point The horizontal coordinates. Then take auxiliary points. Make it located at the point And in the direction perpendicular to the projected line, and let .Depend on , , Uniquely determine the auxiliary projection plane :

[0110] in, Indicates by , and auxiliary points Determined auxiliary projection plane, For plane normal vector components, For plane offset terms, This represents the height component of the three-dimensional coordinates. The auxiliary projection plane is used to constrain the horizontal obstacle-around candidate tangent points, ensuring that the generated track points lie within a geometrically consistent three-dimensional plane.

[0111] Let the four corners of the projection rectangle of the first collision AABB on the horizontal plane be... , , , According to the safe distance Expanding the corner points outwards yields two-dimensional candidate tangent points:

[0112] in, Indicates the first Two-dimensional candidate tangent points, The first of the projected rectangles Corner points, and This represents the horizontal coordinate of the corner point. This indicates the safe distance extending outward from the tangent point.

[0113] Project the two-dimensional candidate tangent point to The three-dimensional tangent point is obtained:

[0114] in, Indicates the first Three-dimensional candidate tangent points, These represent the three-dimensional coordinate components of the tangent point. The auxiliary projection plane can be represented by Unique solution for height This generates a set of candidate sub-paths for horizontal obstacle avoidance. To ensure the safety of candidate paths, a feasibility check is performed on each tangent point; that is, no candidate tangent point can fall inside any other expanded AABB.

[0115] When horizontal detours are too costly or horizontal tangency points are not feasible, vertical over-the-top candidate paths are generated. Assume a direct connection segment... The intersection points of the projection on the horizontal plane with the entrance and exit points of the first-collision AABB projection rectangle are respectively... and :

[0116] in, and These represent the entrance and exit intersection points of the horizontal projection of the direct connection segment and the first-collision AABB projection rectangle, respectively. These represent the horizontal coordinates of the corresponding intersection points.

[0117] Let the upper limit height of the first collision AABB be... The safe height for crossing the roof is:

[0118] in, Indicates the safe height for exceeding the top. This indicates the upper limit of the height of the first collision with AABB. This represents the overshoot safety margin. Then, it represents the set of intermediate waypoints for vertical overshoot candidate paths. Represented as:

[0119] in, This represents the set of intermediate waypoints for vertical overpass candidate paths, where the two 3D points in the set are the overpass entry waypoint and the overpass exit waypoint, respectively. In other words, a local obstacle bypass segment can be represented as:

[0120] For any candidate branch path Its broken line length Represented as:

[0121] in, Indicates candidate branch path The length of the broken line, Indicates intermediate waypoints in the candidate paths. This indicates the number of intermediate waypoints. Since vertical overpasses typically involve additional ascent, descent, and energy expenditure, a vertical path penalty factor is introduced. Construct the equivalent lengths of horizontal and vertical candidate paths:

[0122] in, This represents the equivalent length of the vertical top-crossing candidate path. This represents the equivalent length of the horizontal obstacle avoidance candidate path. Represents the vertical path penalty coefficient and .

[0123] The local layer prioritizes collision risk as the first priority and equivalent length as the second priority when making branch decisions to select the optimal candidate path.

[0124] in, This represents the optimal candidate path selected by the local layer. This indicates the collision risk or number of tangent conflicts of the candidate path. This represents the equivalent length of the candidate path. The output is a local obstacle avoidance trajectory that balances safety and efficiency.

[0125] S43. Local tail end stalling and escape.

[0126] Specifically, the local layer will obtain the optimal candidate path As the current local trajectory segment, position updates are performed according to a set step size, and the local map and overlay markers are updated synchronously. Let the first... The additional coverage gain of the drone within the current local window is The drone is considered to have experienced partial stagnation when the following conditions are met:

[0127] in, Indicates the first The increased coverage benefits of drones within the current local window. This is the threshold for local advancement. If local stagnation persists and meets the preset escape cooldown time, a temporary escape trajectory is generated on the current online map, and the drone escapes by following the temporary escape trajectory.

[0128] In the later stages of a mission, the remaining frontiers often exhibit characteristics of being small-scale, discrete, and fragmented, which can easily lead to long-distance, low-yield wanderings. To address this, the local layer combines the cluster-level mission package output by S3, the frontier utility obtained by S2, and the target locking mechanism to perform priority sorting and nearest-neighbor sweeping on the fragmented frontiers. This allows the UAV to prioritize the processing of high-yield fragmented frontiers within its own area of ​​responsibility, reducing cross-regional backtracking and repeated sweeping by multiple UAVs.

[0129] Thus, S4 achieves a closed-loop conversion from low-frequency global mission packets to high-frequency executable trajectories, and improves the efficiency of local obstacle avoidance, continuous propulsion, and terminal convergence in unknown and complex environments.

[0130] In this embodiment, step S5 specifically includes: S51. Establish a communication triggering and incremental synchronization mechanism.

[0131] Specifically, during the online exploration phase of an unknown environment, the local map, coverage status, leading-edge targets, and mission execution status of the multi-UAV system are all updated asynchronously over time. Requiring all UAVs to synchronize a complete map in real time would result in a high communication load; relying entirely on the local autonomous decision-making of each UAV could easily lead to repeated approaching of the same leading-edge cluster and local congestion. Therefore, this invention employs an incremental communication mechanism based on communication radius and communication cooldown period, exchanging only state increments that have a critical impact on collaborative coverage.

[0132] Let the first drones and the first A drone in The positions at the time are respectively and The communication radius is The communication cooling-off period is The last time the two machines communicated was when The two machines will trigger a communication when the following conditions are met:

[0133] in, and They represent the first frame and the first A drone in Location at any given moment Indicates the communication radius. Indicates the time of the last communication between the two machines. This indicates a communication cooldown period. After communication is triggered, drones do not exchange complete global maps, but instead exchange local incremental information. For the first For a drone, the local incremental information includes at least newly added free grids, newly added occupied grids, newly added covered grids, recent trajectory segments, current primary target, cluster occupancy records, and candidate target summaries, which can be specifically represented as:

[0134] in, Indicates the first A drone in Local incremental information that is ready to be exchanged at any time. For local map increments, To add a new set of overlay markers, This is a summary of recent trajectories. For the current primary objective, Increment the lock table for cluster occupancy.

[0135] The receiving party updates its local collaborative state and global dynamic map cache through idempotent fusion:

[0136]

[0137] in, For idempotent fusion operators, for The current time frame has covered the entire raster set. This is to merge the set of covered rasters after adding new cover markers. Through the incremental synchronization method described above, progressive consistency between the cover view and the status of critical tasks can be achieved under limited bandwidth.

[0138] S52. Establish a frontier cluster occupancy locking and mutual exclusion advancement mechanism.

[0139] Specifically, the leading cluster ensemble continuously evolves with observations, and strong competition can easily arise when multiple UAVs advance in parallel. Relying solely on target reallocation within each local cycle would result in frequent target switching and chasing oscillations. Therefore, this invention establishes a leading cluster occupancy locking mechanism between low-frequency global organization and high-frequency local execution. For any leading cluster... Maintain its occupancy record Specifically, it is expressed as follows:

[0140] in, Indicates frontier cluster Occupancy records Indicates the current occupying frontier cluster The drone's serial number, This indicates the expiration time of the occupancy record. To occupy the locked step size. When the first Select the leading edge cluster for the drone When the current primary target is used, write or renew the lease. During the lockout period, other drones reduce the priority of the frontier cluster or treat it as an already occupied target when selecting targets or updating mission packages, thus ensuring that the same frontier cluster is primarily advanced by a single drone within the same local execution window.

[0141] When communication is available, neighboring UAVs exchange cluster occupancy lock table increments and perform idempotent fusion. If different UAVs conflict to occupy the same leading edge cluster, a decision is made based on arrival cost, track continuity, and lock time. Let the... UAVs for front-line clusters The cost of the ruling is , can be represented as:

[0142] in, Indicates the first UAVs for front-line clusters The cost of conflict adjudication For the first A drone reaches the front cluster The cost, The continuous cost of switching tracks or deviating from the current course. The cost of locking the state, For the corresponding weights.

[0143] The drone with the lower adjudication cost is selected as the owner of the frontier cluster, and other drones release the target and reselect a mission. For frequent releases of occupancy caused by short-term occupancy or frontier jitter, the drone is only removed from the globally stable frontier layer after exceeding the hysteresis upper bound to maintain cluster occupancy consistency.

[0144] Under limited communication bandwidth, synchronizing the complete covered map and the complete frontier map can cause significant communication overhead. Therefore, this invention allows each UAV to synchronize only its own recent trajectory, local occupancy, and cluster target summary, imposing an exclusion penalty on neighboring UAVs in local target scoring. Let the... Candidate targets or mission packages for the drone The locations of other drones are set as follows The repulsion radius is The repulsion strength is Then the target rejects punishment It is expressed as follows:

[0145] in, Indicates the first Target exclusion penalty for drone candidate targets or mission packages. Indicates except the first The location collection of other drones besides the main drone. Indicates the repulsion radius. Indicates the strength of repulsion. This represents the minimum distance from a candidate target or mission package to other drone ensembles. Target rejection penalties enable drones to proactively avoid areas already advanced by neighboring drones and crowded frontal zones when selecting targets, resulting in a dispersed advancement trend and reducing the probability of multiple drones converging near the same frontal zone.

[0146] S53. Collaborative recovery of multiple UAVs under conditions of loss of connection, reconnection, and individual failure.

[0147] Specifically, during intermittent communication link interruptions, each UAV retains its local map, current task package, and cluster occupancy record, and continues to generate local tracks and advance coverage tasks according to S4. When the UAV re-enters communication range and meets the communication cooldown period, it exchanges the map increment, coverage increment, and cluster occupancy status during the disconnection period through an incremental synchronization mechanism, and corrects duplicate occupancy targets according to the conflict resolution mechanism.

[0148] If the occupancy record of a certain frontier cluster exceeds the expiration time If the lease is not renewed, the frontier cluster is determined to be released and re-enter the candidate task set; if a UAV is detected to have no communication, no trajectory update, or has not renewed its lease for its occupied target for an extended period, the incomplete frontier cluster in its corresponding task package is released and redistributed in the next S3 low-frequency global task reorganization. Thus, even under conditions of limited communication, topology time-varying, or individual failure, the multi-UAV system can still maintain coverage closure and task recovery capabilities.

[0149] Example 2: Based on Example 1, this embodiment of the invention presents an unknown and complex three-dimensional task scenario to verify the effectiveness of the proposed multi-UAV online collaborative exploration and coverage method driven by the frontier vehicle path problem in an unknown environment. The main parameter settings are shown in Table 1.

[0150] Table 1 Main Parameter Settings

[0151] like Figure 2 As shown, this unknown and complex 3D mission scenario only provides the initial boundary of the mission area, base coordinates, basic safety constraints, and the initial state of the UAV. The terrain undulations, obstacle distribution, and threat information within the area are gradually revealed as the UAV's sensors observe. Three drones depart from the base and operate according to the following complete process: Initialize the online grid map, coverage markers, and drone status; each drone writes real-time observations and updates the local map during flight; enclose newly added obstacles and terrain protrusions as axis-aligned bounding boxes and perform safe expansion; extract leading edge points from the boundaries of the known free space and unknown space and perform incremental maintenance; aggregate leading edge points into leading edge clusters and calculate candidate benefits, arrival costs, and feasibility; perform low-frequency global task reorganization when periodic or event triggering conditions are met, obtaining cluster-level task packages and access sequences for each drone; each drone selects local targets based on its own task package and generates executable tracks through direct detection, horizontal tangential obstacle avoidance, vertical over-the-top obstacle avoidance, and branch adjudication; continuously update the coverage status during advancement and perform extrication and cleanup in stalled or terminal debris scenarios; when the drones meet the communication conditions, exchange map increments, coverage increments, task status, and cluster occupancy records, complete conflict adjudication and task recovery, until the coverage completion condition or simulation step limit is reached.

[0152] Figure 3 The diagram shows an online raster map that can distinguish between unknown, free, occupied, unreachable, and covered states, and can continuously update the coverage markers based on sensor observations, providing a basis for coverage statistics and frontier extraction. Figure 4 and Figure 5 A schematic diagram of a tangent-guided local planning method in a 3D scene is presented, which realizes local obstacle avoidance in unknown environments and reduces the amount of local planning computation. Figure 6 , Figure 7 and Figure 8 The comparison results of different global layer methods in terms of coverage, unit benefit and overlap rate show that the low-frequency two-parameter global task planning of the present invention can reduce inter-machine overlap and improve unit benefit while maintaining high coverage. Figure 9 , Figure 10 and Figure 11 Further comparisons of different local layer methods in terms of coverage, unit benefit, and overlap rate demonstrate that the 3D tangent-guided local trajectory generation method of this invention can reduce multi-UAV convergence and redundant coverage while maintaining good terminal coverage propulsion capability. In summary, this embodiment shows that the multi-UAV online collaborative exploration coverage method driven by the leading-edge vehicle path problem in unknown environments proposed in this invention can complete online mapping, leading-edge target organization, global task allocation, local safe trajectory generation, and communication-constrained collaborative recovery even without a complete prior map, achieving good overall results in terms of coverage, inter-UAV overlap rate, unit benefit, and load balancing. To further quantitatively compare the planning effects of each method, relevant comparison results are shown in Tables 2 and 3.

[0153] Table 2 Comparison data of global layer experiments

[0154] In the global layer comparative experiment, the local planners all adopted the three-dimensional tangent guidance method of this invention. The example data in Table 2 shows that the global planning method of this invention, while maintaining high coverage, reduces the overlap rate, average mileage, and running time among multiple UAVs, and improves the unit benefits of multiple UAVs and the coordination among UAVs.

[0155] Table 3 Comparison data of local layer experiments

[0156] In the local layer comparative experiments, the global planning method adopted the two-parameter planning method of this invention. Example data in Table 3 shows that the three-dimensional tangent guidance method of this invention can reduce the overlap rate, average mileage, and running time among multiple unmanned aerial vehicles (UAVs), improve unit revenue and inter-UAV coordination, and reduce multi-UAV convergence behavior.

[0157] Example 3: Based on Embodiment 1, this embodiment of the invention provides a multi-UAV online collaborative exploration and coverage system for unknown environments, which can be used to implement the multi-UAV online collaborative exploration and coverage method for unknown environments as described in the foregoing embodiments. The system includes: The first module is used to construct the basic data structure for online exploration and coverage of multiple UAVs in unknown environments: based on online grid maps, online grid information status and coverage markers are established in unknown environments, the local observation increments of each UAV are merged into a global dynamic map, and axis-aligned bounding boxes are used to structure unstructured obstacles. The second module is used to extract the frontier point set based on the basic data structure and cluster them into a frontier cluster set. Then, the frontier cluster set is subjected to utility evaluation, accessibility gating and visibility gating to obtain an effective frontier cluster set. The third module is used to respond to the low-frequency global reorganization trigger signal, and based on the idea of ​​multi-machine global dynamic map and vehicle routing problem, uniquely assign the task nodes in the effective frontier cluster set to each UAV, optimize the access sequence of each UAV, and generate a low-frequency global task plan. The fourth module is used to enable each UAV to lock onto local targets based on the low-frequency global mission plan, and to generate local obstacle avoidance tracks that meet safety constraints based on axis-aligned bounding boxes and a three-dimensional tangent guidance strategy. The fifth module is used to exchange local map increments and task status through an incremental synchronization mechanism during multi-machine communication, and to resolve multi-machine target conflicts by using a leading cluster occupancy locking and mutual exclusion advancement mechanism, thereby maintaining collaborative consistency and enabling online collaborative exploration and coverage of multiple UAVs in unknown environments.

[0158] According to embodiments of the present invention, the present invention also provides an electronic device, a readable storage medium, and a computer program product.

[0159] In an exemplary embodiment, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described in Embodiment 1 above.

[0160] In an exemplary embodiment, the readable storage medium may be a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method described in Embodiment 1 above.

[0161] In an exemplary embodiment, the computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1 above.

[0162] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0163] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0164] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0165] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0166] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0167] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for multi-UAV online collaborative exploration and coverage in an unknown environment, characterized in that, Includes the following steps: Constructing a basic data structure for online exploration and coverage of unknown environments by multiple UAVs: Based on online grid maps, establish online grid information status and coverage markers in unknown environments, integrate the local observation increments of each UAV into a global dynamic map, and use axis-aligned bounding boxes to structurally represent unstructured obstacles; Based on the basic data structure, the frontier point set is extracted and clustered into a frontier cluster set. Then, the frontier cluster set is subjected to utility evaluation, accessibility gating and visibility gating to obtain an effective frontier cluster set. In response to the low-frequency global reorganization trigger signal, based on the concept of multi-machine global dynamic map and vehicle routing problem, the task nodes in the effective frontier cluster set are uniquely assigned to each UAV, and the access sequence of each UAV is optimized to generate a low-frequency global task plan. Based on the low-frequency global mission plan, each UAV locks onto a local target and generates a local obstacle avoidance trajectory that meets safety constraints using a three-dimensional tangent guidance strategy based on axis-aligned bounding boxes. During multi-machine communication, local map increments and task status are exchanged through an incremental synchronization mechanism, and multi-machine target conflicts are resolved by using a leading cluster occupancy locking and mutual exclusion advancement mechanism to maintain collaborative consistency and achieve online collaborative exploration and coverage of multiple UAVs in unknown environments.

2. The multi-UAV online collaborative exploration and coverage method in an unknown environment according to claim 1, characterized in that, The specific method for establishing online raster information status and overlay markers in unknown environments based on online raster maps is as follows: Discretize the task space in an unknown environment into a set of three-dimensional grids. For any grid cell ,exist Maintain arbitrary grid cells at all times Online raster information status : in, This indicates an unknown raster, i.e., a raster that has not been observed or has insufficient observation confidence; This indicates a free grid, which is a grid that has been observed and satisfies the passability or flightability constraints; This indicates that the grid is occupied, i.e., obstacles, terrain occupancy, or non-flying space; This indicates an unreachable grid, which is an area that is not an obstacle but is temporarily unreachable under the current connectivity state or local planning conditions. It is used to prevent drones from repeatedly triggering invalid attempts in narrow passages, dead ends, or areas where repeated failures occur. For any grid cell Maintain overlay markers: in, Represents grid exist Time-over markers, Represents grid It has been effectively observed by the drone's sensors and included in the coverage statistics under unobstructed conditions. Represents grid It has not yet been effectively covered.

3. The multi-UAV online collaborative exploration and coverage method in an unknown environment according to claim 2, characterized in that, The method for fusing the local observation increments of each UAV into a global dynamic map is as follows: The local map of each UAV is updated by the local map state transition operator; When any UAV meets the communication constraints, the local map increments are merged into a global dynamic map, and the global dynamic map is updated through the map fusion operator.

4. The multi-UAV online collaborative exploration and coverage method in an unknown environment according to claim 3, characterized in that, The method of using axis-aligned bounding boxes to structurally represent unstructured obstacles is as follows: Wrap obstacles, terrain protrusions, or threatening boundaries as an axis-aligned bounding box set: in, express The set of axis-aligned bounding boxes at each moment. Indicates the first A surrounding box, This represents the number of bounding boxes currently being maintained online. They represent the first The bounding box has a lower and upper boundary along the three-dimensional coordinate axes, and each axis-aligned bounding box satisfies six boundary constraints: in, This represents the coordinate components of the spatial point to be detected in the three-dimensional coordinate system; Performing a safe expansion process on the axis-aligned bounding box yields an expanded bounding box: in, Indicates the first The enclosed box is an expanded enclosed box after safety expansion. For horizontal safety margin, For vertical safety margin; For spatially adjacent bounding boxes with upper height limits below a preset threshold, merging is triggered based on the intersection-union ratio (IoU) threshold, thus completing the structured representation of unstructured barriers.

5. The multi-UAV online collaborative exploration and coverage method in an unknown environment according to claim 4, characterized in that, The method for extracting frontier point sets based on basic data structures and clustering them into frontier cluster sets, and then performing utility evaluation, accessibility gating, and visibility gating on the frontier cluster sets to obtain effective frontier cluster sets is as follows: Based on the online raster information state, we define the unknown raster set and the free raster set, and then construct the instantaneous front set. : in, express The set of instantaneous frontiers at any given moment. This represents a set of three-dimensional raster cells after the task space has been discretized. Represents grid Adjacent grid cells, For grid neighborhood operators, Represents grid exist Real-time online map information status; Constructing a leading-edge mask using morphological operations: in, express The leading edge mask of time, Indicates a free grid indicator mask. Indicates an unknown grid indicator mask. express Structural elements For the dilation operator, A mask for permanently inaccessible areas, used to avoid misidentifying areas outside the boundary, permanent obstacles, or permanently inaccessible spaces as exploration targets; Mapping the raster indices of true front masking points in the instantaneous front set to the world coordinate system yields the front point set. ; front set Aggregate into a set of front clusters, calculate the cluster center position of each front cluster, and construct a neighborhood grid set centered on the cluster center position. : in, Indicates the first The cluster center of a frontier cluster, Indicates Centered on, with radius The neighborhood raster set, Indicates the radius of the candidate target benefit assessment. This represents a set of three-dimensional raster cells after the task space has been discretized. For grid Mapping to the world coordinate system; Construct a utility function for cluster sorting and allocation based on neighborhood grid sets, and calculate the th... The drone was selected as the first The utility value of a frontier cluster; Applying morphological dilation consistent with the safety margin to the occupied area yields an inflated occupied mask. and define the free element mask. : in, Indicates a free-state grid indicator mask; in the free cell set Label the connected components in the neighborhood, and denote the first... The connected component where the drone is currently located is ; If any frontier cluster center The connected components of the grid neighborhood and If there is no intersection, it is determined that the frontier cluster is currently affecting the drone. If unreachable, remove it; otherwise, use reachability gating. Perform coarse collision detection on the direct connection segment from the current position of the UAV to the center of the leading cluster. If the direct connection segment intersects with the expanded bounding box set, the arrival priority is reduced. If the frontier cluster center after reachability gating If the line of sight to a key unknown grid in its neighborhood is blocked by occupied cells or unknown uncertain areas, the expected revenue is reduced by the visibility gating reduction factor to obtain the gated candidate revenue. Based on utility values ​​and gating candidate returns, frontier clusters are selected from the frontier cluster set to obtain the effective frontier cluster set.

6. The multi-UAV online collaborative exploration and coverage method in an unknown environment according to claim 5, characterized in that, The specific method for constructing the utility function is as follows: Calculate the net new coverage potential of the neighborhood of the frontier cluster: in, Indicates the first A cutting-edge cluster in Net new coverage potential at any time Represents grid exist Time-over markers, Used to exclude grid cells that are already covered. Indicates from the cluster center To grid The value for determining unobstructed line of sight is 1 if there is no obstruction, and 0 otherwise. Coverage rate within the neighborhood of the statistical frontier cluster: in, Indicates the first Coverage rate within the neighborhood of each frontier cluster Indicates the first The number of grid cells contained within the neighborhood of each front cluster; Construct an exponential redundancy reduction factor: in, Indicates the first Redundancy reduction factor of a frontier cluster, This represents the redundancy reduction strength factor. Represents an exponential function with the natural constant as its base; Calculate the effective new coverage potential after reduction : ； Define arrival cost: in, Indicates the first The drone to the The cost of reaching the center of a frontier cluster Indicates the first A drone in Location at any given moment This is the set of feasible paths from the current location of the drone to the center of the leading cluster. This represents any feasible path. Indicates a feasible path The first A discrete path point, Indicates a feasible path The first A discrete path point, Indicates the number of path points; Based on the reduced effective new coverage potential and arrival cost, a utility function for cluster sorting and allocation is constructed: in, Indicates the first The drone was selected as the first The utility value of a frontier cluster, To cover potential weights, To reach the cost weight.

7. The multi-UAV online collaborative exploration and coverage method in an unknown environment according to claim 5, characterized in that, The global reorganization trigger signal is: in, This indicates a global reorganization trigger signal. Indicates the overall restructuring cycle. This represents the threshold for the rate of change of the frontier size. This represents the rate of change of the current global frontier size relative to the last global planning time. express The number of globally stable frontier points at time t. This indicates the number of frontier points recorded when the global plan was successfully generated last time. Indicates periodic triggering. Indicates that an event has been triggered; Based on the concepts of multi-drone global dynamic maps and vehicle routing problems, the method of uniquely assigning task nodes in the effective frontier cluster set to each UAV and optimizing the access sequence of each UAV to generate a low-frequency global task plan is as follows: The cluster-level task node set is formed by the centers of each frontier cluster in the effective frontier cluster set; Constructing the frontier of the cluster-level task node set To the front line edge cost function : in, Indicates from point Time The cost of the edge This represents the Euclidean distance between two points. Point and points Visual accessibility between them This represents the gaze detection function. This indicates the percentage of overlap caused by the path traversing an already covered area. The penalty coefficient for non-direct-view advancement. The penalty coefficient for repeated traversal; Constructing the cost matrix from UAV to cluster Cluster-to-cluster cost matrix : in, Indicates the first The drone's current location to the The cost of reaching a frontier cluster Indicates the first The first frontier cluster to the first The transfer cost of a frontier cluster Indicates the first A frontier cluster center, Indicates the first A drone in The position at that moment; Define any unassigned front cluster Added to the The base incremental cost at the end of a drone mission package for: in, Indicates the first The terminal cluster index of the current mission packet of the drone; when If the task package is empty, the cost from the drone to the cluster is used; otherwise, the cost from the end cluster to the candidate cluster is used. By introducing a load balancing regularization term and a revenue reward term on the basic incremental cost, a comprehensive evaluation function is constructed: in, Indicates the frontier cluster Assigned to the The overall evaluation value when flying a drone Indicates the first The cumulative cost of maintaining the current mission package using drones. This is a load balancing regularization coefficient used to suppress overload caused by a drone being continuously inserted into tasks; The value reward intensity is used to prioritize the allocation of high-yield frontier clusters; To The expected return per unit obtained after normalization; Choose to make the comprehensive evaluation value The smallest "drone-frontier cluster" combination Perform the allocation and update the "Drone-Frontier Cluster" combination. corresponding and This continues until all effective frontier clusters have been assigned, resulting in the [number]th [cluster]. A drone in A set of cluster-level task packages that always satisfy cluster-level mutual exclusion coverage constraints; For the A drone, based on the cost matrix from drone to cluster. Cluster-to-cluster cost matrix Construct the initial access sequence This leads to the output of a low-frequency global task plan: in, This indicates a low-frequency plan output from the global layer. A cluster-level set of task nodes. Indicates the first A drone in Cluster-level task packages at specific times Indicates the number of drones. Indicates the number of cluster-level task packages. Indicates all A collection of drones.

8. The multi-UAV online collaborative exploration and coverage method in an unknown environment according to claim 4, characterized in that, The method for generating local obstacle avoidance trajectories that meet safety constraints by enabling each UAV to lock onto local targets based on the low-frequency global mission plan and using a three-dimensional tangent guidance strategy based on axis-aligned bounding boxes is as follows: For the A drone, according to the initial access sequence Select the current primary target And set a lock count for the current primary target. ;when When the target is locked, a new primary target is selected from the head of the mission package and the lock count is reset. During the lock period, a track is generated around the current primary target first. When the target is achieved, the target fails, or the lock count expires, switch to the next candidate target in the task package; Based on the drone's current location With the current local target point direct line segments between Establish on the horizontal plane The projected line, and take auxiliary points. make auxiliary points Located at the point And in the direction perpendicular to the projected line, by , , Uniquely determine the auxiliary projection plane ; For direct-connection line segments Collision detection is performed with the expanded bounding box set, extracting the four corners of the projected rectangle of the first-collision expanded bounding box on the horizontal plane, and expanding the safety distance outward based on the tangent point. Expanding the corner points outwards yields two-dimensional candidate tangent points; Project the two-dimensional candidate tangent points onto the auxiliary projection plane. The three-dimensional tangent points are obtained, and a set of horizontal obstacle avoidance candidate paths is generated. Feasibility tests are performed on each tangent point to ensure that no candidate tangent point falls inside other expanding bounding boxes; When the cost of horizontal detour exceeds a preset threshold or horizontal tangent point is not feasible, a vertical overpass candidate path is generated: the horizontal projection of the direct connection segment between the current position of the UAV and the current local target point is determined, and the entry and exit points of the projection rectangle of the first-collision expansion bounding box are determined; the overpass safety height is determined based on the upper bound height of the first-collision expansion bounding box and the overpass safety margin, and a vertical overpass candidate path is generated. Introducing a vertical path penalty coefficient The equivalent length of the vertical obstacle crossing candidate path is obtained by weighting the broken line length of the vertical obstacle crossing candidate path, and the equivalent length of the horizontal obstacle bypass candidate path is the broken line length of the horizontal obstacle bypass candidate path. The branch decision is made with collision risk as the first priority and equivalent length as the second priority, and the optimal candidate path is selected. Optimal candidate path As the current local trajectory segment obstacle avoidance track, the position is updated according to the set step size, and the local map and overlay markers are updated simultaneously.

9. The multi-UAV online collaborative exploration and coverage method in an unknown environment according to claim 1, characterized in that, The specific method for exchanging local map increments and task states through incremental synchronization is as follows: When the distance between two drones is less than or equal to the communication radius And the time since the last communication is greater than or equal to the communication cooldown period. When communication is triggered, the two UAVs exchange local incremental information; the local incremental information includes newly added free grids, newly added occupied grids, newly added covered grids, recent trajectory segments, current primary target, cluster occupancy records, and candidate target summaries; The receiving party updates its local collaborative state and global dynamic map cache through idempotent fusion.

10. The multi-UAV online collaborative exploration and coverage method in an unknown environment according to claim 1, characterized in that, The specific method for resolving multi-machine target conflicts and maintaining cooperative consistency by utilizing the front cluster occupancy locking and mutual exclusion propulsion mechanism is as follows: When the Select the leading edge cluster for the drone When the current primary target is to write to or renew the frontier cluster. The occupancy record includes the UAV number currently occupying the leading cluster and the expiration time of the occupancy record. During the lockout period, other UAVs can reduce the leading cluster size when selecting targets or updating mission packages. Prioritize or consider it as an already occupied target; When different UAVs conflict and occupy the same leading edge cluster, the cost of each UAV reaching the leading edge cluster, the continuity cost caused by track switching or deviation from the current course, and the cost of locking state are weighted and summed to obtain the decision cost of each UAV for the leading edge cluster. The drone with the lowest adjudication cost is selected as the owner, and the other drones are reassigned to new missions.

Images