Unmanned aerial vehicle rapid autonomous exploration method based on spatial layering

Abstract

The invention provides an unmanned aerial vehicle rapid autonomous exploration method based on spatial layering. The method comprises the steps that S1, point cloud information around an unmanned aerial vehicle is collected to construct a three-dimensional voxel map, and an area in a boundary is recognized; s2, dividing an area in the boundary into a plurality of sub-areas by adopting an online space decomposition algorithm, constructing a target cost matrix, combining path length cost and speed optimization cost, solving a non-closed traveling salesman problem with boundary constraint, and determining a traversal sequence of each sub-area; s3, generating a sphere safety space in each sub-region according to a traversal sequence, and generating a candidate viewpoint set by adopting polar coordinate Gaussian distribution sampling; and S4, screening an optimal viewpoint from the candidate viewpoint set based on the global loss function, optimizing the target function by adopting a B-spline trajectory to generate a local trajectory, and controlling the unmanned aerial vehicle to execute a flight exploration task along the local trajectory. The method has the beneficial effects that the application scene of the unmanned aerial vehicle can be expanded, optimal exploration path planning is realized, and the collision risk of unknown obstacles is avoided.

Description

Technical Field

[0001] This invention relates to the technical field of autonomous navigation for unmanned aerial vehicles (UAVs), and more specifically, to a method for rapid autonomous exploration of UAVs based on spatial layering. Background Technology

[0002] In recent years, unmanned aerial vehicle (UAV) technology, especially quadcopter UAV systems, has made groundbreaking progress. Their unique vertical takeoff and landing capabilities, high maneuverability, and relatively low manufacturing costs have enabled them to demonstrate significant application value in several key areas. In disaster relief, UAV systems have achieved multi-level applications, ranging from simple aerial reconnaissance to complex rescue missions. By carrying dedicated communication relay equipment, UAVs can quickly establish temporary communication networks in disaster areas where communication infrastructure has been damaged; real-world data shows they can establish stable communication coverage with a radius of up to 5 kilometers. Simultaneously, UAV systems integrating infrared thermal imaging technology can accurately detect vital signs, achieving a temperature resolution of ±0.5℃ in rubble search and rescue, significantly improving search and rescue efficiency. In the handling of hazardous chemical spills, UAV-mounted sampling systems can collect gas and liquid samples without contacting the pollution source, effectively ensuring the safety of rescue personnel.

[0003] In the field of industrial inspection, drone technology is driving innovation in traditional inspection methods. High-precision imaging systems combined with advanced computer vision algorithms enable drones to identify minute defects on the surface of cultural relics, down to 0.1mm, achieving nearly 10 times the accuracy of manual inspection. In power system maintenance, drones can perform millimeter-level damage inspections on insulators of high-voltage transmission lines, completing in a single flight what would traditionally take several days. For assessing the corrosion status of large storage tanks, the multispectral imaging systems on drones can penetrate surface coatings and accurately identify early corrosion characteristics, providing crucial data support for preventative maintenance.

[0004] Exploring special environments is one of the most challenging application areas for UAV technology. In 3D modeling of underground pipe networks, UAV-equipped lidar systems can achieve centimeter-level accuracy in spatial modeling even without GPS signals. In monitoring nuclear contamination areas, UAV radiation detection systems can generate real-time radiation intensity distribution maps, providing crucial information for emergency response decisions. For exploring the interior of earthquake ruins, micro-UAVs can traverse narrow gaps and accurately reconstruct the internal structure through real-time 3D reconstruction technology, significantly improving the safety and effectiveness of search and rescue operations.

[0005] However, existing unmanned aerial vehicle (UAV) systems still face numerous technical bottlenecks in practical applications. Regarding autonomous flight capabilities, most systems still rely on GPS navigation, performing poorly in scenarios where GPS signals are scarce, such as complex built-up environments or underground spaces. Path planning algorithms generally suffer from low computational efficiency, with planning delays reaching several seconds in large-scale scenarios, making it difficult to meet real-time obstacle avoidance requirements. Furthermore, existing 3D mapping algorithms have high computational complexity, making it difficult to achieve high-quality real-time reconstruction on resource-constrained airborne platforms. These technical limitations severely restrict the practical application of UAVs in mission-critical scenarios and urgently require breakthroughs through technological innovation. Summary of the Invention

[0006] The technical problem this invention aims to solve is how to expand the application scenarios of UAVs, achieve optimal exploration path planning, and avoid the risk of collisions with unknown obstacles. To overcome the shortcomings of the above-mentioned existing technologies (or related technologies), this invention provides a rapid autonomous exploration method for UAVs based on spatial layering.

[0007] This invention provides a rapid autonomous exploration method for unmanned aerial vehicles (UAVs) based on spatial layering. A UAV hardware platform is pre-built, with at least one UAV mounted on its takeoff platform. The UAV is equipped with a 3D lidar. The rapid autonomous exploration method includes the following steps: Step S1: After the UAV takes off and flies to the target area, control the three-dimensional lidar to continuously collect point cloud information around the UAV, construct a three-dimensional voxel map based on the point cloud information, and identify the region within the boundary of the three-dimensional voxel map. Step S2: The region within the boundary is divided into multiple sub-regions using an online spatial decomposition algorithm. A target cost matrix is ​​constructed and combined with path length cost and speed optimization cost to solve a non-closed traveling salesman problem with boundary constraints to determine the traversal order of each sub-region. Step S3: Generate a spherical safe space in each of the sub-regions according to the traversal order, and generate a candidate viewpoint set by sampling in the spherical safe space using a polar coordinate Gaussian distribution. Step S4: Based on the pre-constructed global loss function, the optimal viewpoint is selected from the candidate viewpoint set. Based on the optimal viewpoint, a local trajectory is generated using a B-spline trajectory optimization objective function with velocity constraints. The UAV is then controlled to perform a flight exploration mission along the local trajectory.

[0008] Compared with existing technologies, the rapid autonomous exploration method for unmanned aerial vehicles based on spatial layering proposed in this invention has the following advantages: This invention utilizes a UAV equipped with a 3D LiDAR to collect point cloud information and constructs a 3D voxel map and performs positioning in real time based on the point cloud information. This enables the UAV to achieve fully autonomous exploration in complex environments where GPS signals are lacking, greatly expanding the application scenarios of UAVs. An online spatial decomposition algorithm is used to divide the large area within the boundary into sub-regions, and the traversal order of the global sub-regions is optimized by combining the non-closed traveling salesman problem. This effectively solves the long-distance backtracking problem caused by random or short-sighted exploration in traditional methods, achieving near-optimal exploration path planning, thereby significantly reducing the total flight distance and time. By generating a spherical safe space within the sub-regions and sampling a set of candidate viewpoints within it, all observations and flight planning of the UAV are ensured to be carried out within a known safe range, fundamentally avoiding the risk of collision with unknown obstacles.

[0009] In one possible implementation, in step S1, the point cloud information is accumulated to build a point cloud map, and then the point cloud map is simplified into an Euclidean symbolic distance field map as the three-dimensional voxel map through an octree structure.

[0010] Compared with existing technologies, the above technical solution can simplify the original point cloud map into an Euclidean symbolic distance field map through an octree structure, which greatly compresses the amount of data and reduces the computation and storage overhead. Furthermore, the Euclidean symbolic distance field map can directly provide the distance information from any point in space to the nearest obstacle, enabling the path planning algorithm to perform collision detection and safe distance constraints quickly and efficiently, thereby supporting real-time decision-making and flight of UAVs in complex environments.

[0011] In one possible implementation, in step S2, the online spatial decomposition algorithm employs a longest-edge-first segmentation strategy, which includes strategy A and strategy B. Strategy A involves performing secondary segmentation along the longest axis when the aspect ratio of the sub-region exceeds 2:1, and strategy B involves segmenting the square sub-region along the x-axis.

[0012] Compared with existing technologies, the above technical solution can avoid the generation of excessively narrow sub-regions, which is conducive to UAVs conducting efficient viewpoint observation and path coverage within them.

[0013] In one possible implementation, the target cost matrix in step S2 is constructed using the following calculation formula: in, Represents the target cost matrix; This represents the first adjustable weighting coefficient; ( ) indicates that the A-star algorithm is used to calculate the shortest path length; Indicates the first The aforementioned sub-regions; Indicates the first The aforementioned sub-regions; This represents the second adjustable weighting coefficient; ( ) indicates the calculation of the path turning angle; ( ) indicates the calculation of minimum radius of curvature.

[0014] In one possible implementation, in step S2, MTZ constraints are used to eliminate sub-loops that occur during the solution process when solving the non-closed traveling salesman problem, and auxiliary variables are introduced. The access order of each sub-region is indicated by the following constraints: in, Indicates the first The access order of the sub-regions; Indicates the first The access order of the sub-regions; This represents the total number of all the sub-regions.

[0015] It is a binary variable, representing the node To the node Is there a path? =1 indicates that a path exists. =0 indicates that there is no path.

[0016] In one possible implementation, the conditions for generating the spherical safety space in step S3 are: in, This represents the radius of the safe space of the sphere; ( ) indicates the minimum value calculation; Indicates the first Segment boundary; ( ) indicates the calculation of boundary length; This indicates the sensor's maximum detection distance.

[0017] In one possible implementation, the global loss function used in step S4 includes a boundary coverage term and a velocity optimization term. The boundary coverage term is calculated using the truncated sensor ray method to determine the number of observable boundary voxels to quantify the coverage of unknown regions by candidate viewpoints. The velocity optimization term is obtained using the following formula: in, This represents the speed optimization term; This indicates the adjustable weighting coefficient for the direction angle; This indicates the path turning angle of the UAV from the current viewpoint to the candidate viewpoint; This represents the distance-adjustable weighting coefficient; This indicates the distance between the candidate viewpoint and the nearest obstacle.

[0018] In one possible implementation, the B-spline trajectory optimization objective function in step S4 is constructed using the following expression: in, This represents the objective function for optimizing the B-spline trajectory. This represents the fourth adjustable weighting coefficient; This indicates the maximum permissible speed of the drone; Indicates the first local trajectory The speed of each control point; This represents the fifth adjustable weighting coefficient; Indicates the first local trajectory Acceleration at each control point; This represents the sixth adjustable weighting coefficient; Indicates the first local trajectory The accelerator at each control point. Attached Figure Description

[0019] Figure 1 This is a flowchart of the steps of the present invention; Figure 2 This is a schematic diagram of a candidate viewpoint for the present invention; Figure 3 This is an example diagram of the gradient guidance loss for the UAV according to the present invention; Figure 4 This is a performance analysis diagram of candidate viewpoint generation in this invention; Figure 5 This is a schematic diagram of a real-world test scenario for the present invention; Figure 6 This is a schematic diagram of the remote interface of the present invention. Detailed Implementation

[0020] First, those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention. Those skilled in the art can make adjustments as needed to adapt to specific application scenarios.

[0021] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0022] See Figure 1 This invention discloses a rapid autonomous exploration method for unmanned aerial vehicles (UAVs) based on spatial layering. A UAV hardware platform is pre-built, with at least one UAV mounted on its takeoff platform. The UAV is equipped with a 3D LiDAR. The rapid autonomous exploration method includes the following steps: Step S1: After the UAV takes off and flies to the target area, control the 3D LiDAR to continuously collect point cloud information around the UAV, construct a 3D voxel map based on the point cloud information, and identify the region within the boundary of the 3D voxel map. Step S2: The region within the boundary is divided into multiple sub-regions using an online spatial decomposition algorithm. The target cost matrix is ​​constructed and combined with the path length cost and speed optimization cost to solve the non-closed traveling salesman problem with boundary constraints and determine the traversal order of each sub-region. Step S3: Generate a spherical safe space in each sub-region according to the traversal order, and generate a candidate viewpoint set by sampling in the spherical safe space using polar coordinate Gaussian distribution. Step S4: Based on the pre-constructed global loss function, the optimal viewpoint is selected from the candidate viewpoint set. Based on the optimal viewpoint, a local trajectory is generated by optimizing the objective function of B-spline trajectory with velocity constraints. The UAV is then controlled to perform flight exploration tasks along the local trajectory.

[0023] In this embodiment of the invention, a drone hardware platform is built. The drone is designed to include a flight controller, an onboard computer, a 3D LiDAR, and an inertial sensor. Specifically, a Pixhawk 4 is used as the flight controller, equipped with an NVIDIA Jetson NX onboard computer, and equipped with an Ouster OS1-64 3D LiDAR and a BMI088 inertial measurement unit, forming a complete perception and control system.

[0024] In this embodiment of the invention, in step S1, point cloud information is accumulated to establish a point cloud map. Then, the point cloud map is simplified into an Euclidean symbolic distance field map as a three-dimensional voxel map through an octree structure. In specific implementation, the UAV is placed on the takeoff platform, and a body coordinate system is established along the direction of the UAV body, with the X-axis pointing forward and the Z-axis pointing vertically upward. The size of the space to be explored is defined. The three-dimensional lidar collects the surrounding point cloud information at a speed of 10Hz, with about 130,000 points collected per frame. After the onboard computer receives the point cloud information, it triggers a callback function to perform point cloud registration. An improved ICP algorithm is used to achieve point cloud matching. At the same time, radar IMU information is fused. Kalman filtering is used to fuse the odometry calculated from the point cloud information and the odometry calculated from the radar IMU information to obtain more accurate pose information than a single sensor. The positioning accuracy reaches ±2cm. The UAV uses this position information to replace GPS and can fly in indoor, underground and other scenarios without GPS signals.

[0025] In this embodiment of the invention, in step S2, the online spatial decomposition algorithm employs a longest-edge-first segmentation strategy, which includes strategy A and strategy B. Strategy A involves secondary segmentation along the longest axis when the aspect ratio of a sub-region exceeds 2:1. Strategy B involves segmenting square sub-regions along the x-axis. Specifically, after the odometry processing described above, point cloud information is accumulated to establish a point cloud map, which is then simplified into an Euclidean Symbolic Distance Field (ESDF) map using an octree structure. The map resolution is set to 0.1m. The UAV will understand the obstacle distribution and exploration completion status based on the map status. The UAV divides the region within the boundary using its body coordinate system, taking the first quadrant of this body coordinate system as the first sub-region, the second quadrant as the second sub-region, and so on. When the volume of a sub-region is greater than 50m³, it will automatically be segmented along the longest axis, dividing the large region into smaller sub-regions. This process recursively divides a large region within the boundary into several exploration blocks until each sub-region meets the exploration conditions.

[0026] In this embodiment of the invention, when completing the exploration task in each exploration block, the UAV sets the boundaries between the mapped and unmapped areas as boundaries, and uses an improved DBSCAN clustering algorithm (neighborhood radius 0.5m) to segment the boundaries. For each boundary segment, polar coordinate Gaussian sampling is applied. =0.3, σ_φ=0.2) to generate a candidate viewpoint set. When sampling, the sphere safety space constraint is considered (radius R_s is calculated according to the actual environment). By arranging and combining the candidate viewpoint set, a dual objective optimization function including path length and turning cost is established to find the shortest target viewpoint set that can complete the exploration of each current boundary segment. During the exploration process, the new boundary is updated in real time, and new unknown areas are continuously explored until there are no boundaries left in this exploration block.

[0027] In this embodiment of the invention, the target cost matrix in step S2 is constructed using the following calculation formula: in, Represents the target cost matrix; This represents the first adjustable weighting coefficient; ( ) indicates that the A-star algorithm is used to calculate the shortest path length; Indicates the first Sub-regions; Indicates the first Sub-regions; This represents the second adjustable weighting coefficient; ( ) indicates the calculation of the path turning angle; ( ) indicates the calculation of minimum radius of curvature.

[0028] In this embodiment of the invention, in step S2, MTZ constraints are used to eliminate sub-loops that occur during the solution process when solving the non-closed traveling salesman problem, and auxiliary variables are introduced. The access order of each sub-region is indicated by the following constraints: in, Indicates the first The order of accessing each sub-region; Indicates the first The order of accessing each sub-region; This represents the total number of nodes, which is the total number of nodes in each sub-region. It is a binary variable indicating whether to start from node To the node There is a path. =1 indicates that a path exists. =0 indicates that there is no path.

[0029] In this embodiment of the invention, the conditions for generating the spherical safe space in step S3 are as follows: in, Represents the radius of the safe space of the sphere; ( ) indicates the minimum value calculation; ( ) indicates the minimum value calculation; Indicates the first Segment boundary; ( ) indicates the calculation of boundary length; This indicates the sensor's maximum detection distance.

[0030] In this embodiment of the invention, the global loss function used in step S4 includes a boundary coverage term and a velocity optimization term. The boundary coverage term is calculated by using the truncated sensor ray method to calculate the number of observable boundary voxels to quantify the coverage of the unknown region by the candidate viewpoint. The velocity optimization term is obtained by the following formula: in, Indicates speed optimization terms; This indicates the adjustable weighting coefficient for the direction angle; This indicates the turning angle of the drone's path from the current viewpoint to the candidate viewpoint; This represents the distance-adjustable weighting coefficient; This indicates the distance between the candidate viewpoint and the nearest obstacle.

[0031] In this embodiment of the invention, the objective function for optimizing the B-spline trajectory in step S4 is constructed using the following expression: in, This represents the objective function for optimizing B-spline trajectories. This represents the fourth adjustable weighting coefficient; Indicates the maximum permissible speed of the drone; Represents the first [item] on the local trajectory The speed of each control point; This represents the fifth adjustable weighting coefficient; Represents the first [item] on the local trajectory Acceleration at each control point; This represents the sixth adjustable weighting coefficient; Represents the first [item] on the local trajectory The accelerator at each control point.

[0032] In this embodiment of the invention, during actual implementation, an unfamiliar environment (such as an underground garage or forest area) is first selected, and the UAV prepares for takeoff. After the UAV takes off, the lidar collects point cloud information in the environment at a frequency of 10Hz. After the onboard computer receives the point cloud information, it triggers a callback function to perform point cloud information registration. Here, an improved ICP algorithm is used to achieve point cloud information matching. At the same time, radar IMU information is fused. The point cloud odometry and radar IMU information are fused through Kalman filtering to obtain accurate 6DOF pose information.

[0033] In this embodiment of the invention, after initial pose estimation is completed, point cloud information is accumulated and a point cloud map is constructed. First, an original point cloud map is established, and then simplified into an ESDF map using an octree structure. The resolution of the ESDF map is set to 0.1m. The ESDF map can accurately reflect the distribution of obstacles and spatial distance relationships, providing an important basis for path planning. During the environmental modeling process, according to... Figure 2 The viewpoint generation range shown employs an online spatial decomposition algorithm to divide the region within the boundary into several sub-regions. Specifically, using the body coordinate system as a reference, the region within the boundary is divided into four quadrants. First, the sub-regions in the first quadrant are explored. After completion, the second, third, and fourth quadrants are explored sequentially. When the area of ​​a sub-region in a quadrant exceeds 25 m², it is automatically recursively divided along the longest axis until the explorable condition is met. This hierarchical exploration strategy effectively avoids long-distance backtracking problems. Figure 3 The gradient-guided loss distribution shown is a mechanism that ensures the drone maintains a smooth flight trajectory when switching sub-regions.

[0034] In this embodiment of the invention, during the exploration task execution phase within a single sub-region, the UAV first detects the boundary between the mapped region and the unknown region, such as... Figure 4 The viewpoint generation performance analysis shown uses an improved DBSCAN clustering algorithm to segment the boundary, and generates a candidate viewpoint set by polar coordinate Gaussian sampling (σ_θ=0.3, σ_φ=0.2) for each boundary segment.

[0035] In this embodiment of the invention, an improved A* algorithm is used for global path search during the path planning stage. The heuristic weight is set to 1.2. Feasible paths that meet the minimum safe distance (0.5m) are searched in the ESDF map. Then, cubic spline curve interpolation is used to optimize the feasible paths, ensuring the continuity of the three dimensions of position, velocity, and acceleration. The flight controller performs trajectory tracking at a frequency of 200Hz, and the maximum acceleration is limited to within 2.0m / s². The entire flight process is as follows: Figure 6 The experiments shown included scenarios such as underground parking garages, forests, and warehouses. The drone maintained a stable average flight speed of 1.45 m / s in an environment without GPS.

[0036] In this embodiment of the invention, the ground workstation is as follows: Figure 6 As shown, a connection is established with the drone via Wi-Fi, with communication latency controlled within 150ms. The ground workstation displays the drone's flight status (including six-DOF pose), local path (highlighted in red), and 3D point cloud map in real time. All sensor data and processing results are automatically recorded and stored, supporting playback analysis after the task is completed, and providing data support for algorithm optimization.

[0037] In the description of this invention, the references to "one embodiment," "some embodiments," "in this embodiment," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0038] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A rapid autonomous exploration method for unmanned aerial vehicles based on spatial layering, characterized in that, A drone hardware platform is pre-built, with at least one drone mounted on its takeoff platform. The drone is equipped with a 3D lidar. The rapid autonomous exploration method for the drone includes the following steps: Step S1: After the UAV takes off and flies to the target area, control the three-dimensional lidar to continuously collect point cloud information around the UAV, construct a three-dimensional voxel map based on the point cloud information, and identify the region within the boundary of the three-dimensional voxel map. Step S2: The region within the boundary is divided into multiple sub-regions using an online spatial decomposition algorithm. A target cost matrix is ​​constructed and combined with path length cost and speed optimization cost to solve a non-closed traveling salesman problem with boundary constraints to determine the traversal order of each sub-region. Step S3: Generate a spherical safe space in each of the sub-regions according to the traversal order, and generate a candidate viewpoint set by sampling in the spherical safe space using a polar coordinate Gaussian distribution. Step S4: Based on the pre-constructed global loss function, the optimal viewpoint is selected from the candidate viewpoint set. Based on the optimal viewpoint, a local trajectory is generated using a B-spline trajectory optimization objective function with velocity constraints. The UAV is then controlled to perform a flight exploration mission along the local trajectory.

2. The rapid autonomous exploration method for unmanned aerial vehicles according to claim 1, characterized in that, In step S1, the point cloud information is accumulated to build a point cloud map, and then the point cloud map is simplified into an Euclidean symbolic distance field map as the three-dimensional voxel map through an octree structure.

3. The rapid autonomous exploration method for unmanned aerial vehicles according to claim 1, characterized in that, In step S2, the online spatial decomposition algorithm employs a longest-edge-first segmentation strategy, which includes strategy A and strategy B. Strategy A involves performing a secondary segmentation along the longest axis when the aspect ratio of the sub-region exceeds 2:1, and strategy B involves segmenting the square sub-region along the x-axis.

4. The rapid autonomous exploration method for unmanned aerial vehicles according to claim 1, characterized in that, The target cost matrix in step S2 is constructed using the following calculation formula: in, Represents the target cost matrix; This represents the first adjustable weighting coefficient; ( ) indicates that the A-star algorithm is used to calculate the shortest path length; Indicates the first The aforementioned sub-regions; Indicates the first The aforementioned sub-regions; This represents the second adjustable weighting coefficient; ( ) indicates the calculation of the path turning angle; ( ) indicates the calculation of minimum radius of curvature.

5. The rapid autonomous exploration method for unmanned aerial vehicles according to claim 1, characterized in that, In step S2, when solving the non-closed traveling salesman problem, MTZ constraints are used to eliminate sub-loops that occur during the solution process, and auxiliary variables are introduced. The access order of each sub-region is indicated by the following constraints: in, Indicates the first The access order of the sub-regions; Indicates the first The access order of the sub-regions; This represents the total number of all the aforementioned sub-regions; It is a binary variable, representing the node To the node Is there a path? =1 indicates that a path exists. =0 indicates that there is no path.

6. The rapid autonomous exploration method for unmanned aerial vehicles according to claim 1, characterized in that, In step S3, the conditions for generating the spherical safety space are as follows: in, This represents the radius of the safe space of the sphere; ( ) indicates the minimum value calculation; Indicates the first Segment boundary; ( ) indicates the calculation of boundary length; This indicates the sensor's maximum detection distance.

7. The rapid autonomous exploration method for unmanned aerial vehicles according to claim 1, characterized in that, The global loss function used in step S4 includes a boundary coverage term and a velocity optimization term. The boundary coverage term is calculated by using the truncated sensor ray method to calculate the number of observable boundary voxels to quantify the coverage of unknown regions by candidate viewpoints. The velocity optimization term is obtained by the following formula: in, This represents the speed optimization term; This indicates the adjustable weighting coefficient for the direction angle; This indicates the path turning angle of the UAV from the current viewpoint to the candidate viewpoint; This represents the distance-adjustable weighting coefficient; This indicates the distance between the candidate viewpoint and the nearest obstacle.

8. The rapid autonomous exploration method for unmanned aerial vehicles according to claim 1, characterized in that, The objective function for optimizing the B-spline trajectory in step S4 is constructed using the following expression: in, This represents the objective function for optimizing the B-spline trajectory. This represents the fourth adjustable weighting coefficient; This indicates the maximum permissible speed of the drone; Indicates the first local trajectory The speed of each control point; This represents the fifth adjustable weighting coefficient; Indicates the first local trajectory Acceleration at each control point; This represents the sixth adjustable weighting coefficient; Indicates the first local trajectory The accelerator at each control point.

Images