MUTE-RRT*-based multi-unmanned aerial vehicle dynamic threat environment path planning method and device

Abstract

The invention discloses a multi-unmanned aerial vehicle dynamic threat environment path planning and device based on MUTE-RRT *, and the method comprises the following steps: carrying out the environment modeling: building a model of the influence of a threat region, a dynamic obstacle and a building on the path of an unmanned aerial vehicle; constructing a multi-unmanned aerial vehicle three-dimensional path planning problem according to the constraint condition and the target function; for the multi-unmanned aerial vehicle three-dimensional path planning problem, flight paths of multiple unmanned aerial vehicles are generated at the same time by using an improved RRT * algorithm; and trajectory optimization is carried out on the flight paths of the multiple unmanned aerial vehicles by adopting a Dubins method. According to the invention, the time dimension is introduced, and a space-time security detection mechanism is provided to ensure the safe flight of the unmanned aerial vehicle; an environment area segmentation strategy is provided, and the safety detection efficiency is improved; the RRT * algorithm is improved, the planning time is shortened, the path quality is improved, and therefore the path planning performance of the multi-unmanned-aerial-vehicle system is enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle (UAV) path planning, and in particular relates to a method and device for path planning in a dynamic threat environment for multiple UAVs based on MUTE-RRT*. Background Art

[0002] Drones (UAVs) have gained widespread application in commercial, agricultural, and military fields due to their significant advantages in flexibility, cost-effectiveness, and reduced casualties. Compared to individual drones, swarms offer greater efficiency, robustness, and scalability, enabling them to perform complex tasks such as reconnaissance, surveillance, and strike, which are difficult for individual drones to accomplish independently. However, as UAV swarm missions become increasingly complex, planning a set of optimal flight trajectories while satisfying flight constraints remains a challenging problem.

[0003] Urban environments have always been a key area for drone flight. They are often more complex, not only including tall and dense buildings, but also some threat areas, such as signal interference areas, and dynamic flying objects. This has a significant impact on the safe flight of drones. In complex urban environments with dynamic obstacles and threat areas, planning a set of feasible and conflict-free flight paths is crucial for drone flight missions. Currently, facing complex urban environments, few studies have considered dynamic obstacles and threat areas. Multi-drone path planning in complex urban environments with dynamic obstacles and threat areas remains a relatively difficult problem.

[0004] As mentioned in the review, existing technologies have the following problems: it is difficult for drone swarms to plan paths in urban environments that take into account dynamic obstacles and threat areas, and the methods used are imperfect. Summary of the Invention

[0005] The purpose of this invention is to solve the path planning problem of multiple UAVs in a complex urban environment with threat areas and dynamic obstacles.

[0006] To this end, the first aspect of the present invention proposes a multi-UAV dynamic threat environment path planning method based on MUTE-RRT*, comprising the following steps:

[0007] Modeling the environment, including the impact of threat zones, dynamic obstacles, and buildings on the drone’s path;

[0008] Construct a three-dimensional path planning problem for multiple UAVs based on constraints and objective functions;

[0009] For the three-dimensional path planning problem of multiple UAVs, an improved RRT* algorithm is used to simultaneously generate flight paths for multiple UAVs. The improved RRT* algorithm includes: dividing the entire environment into regions according to the x and y axes and assigning obstacles to corresponding sub-regions; first performing collision detection when generating new path segments to ensure that UAVs avoid each other during flight; adaptively adjusting the step size based on the region where the parent node of the sampling point is located and the distribution of obstacles; generating multiple candidate nodes using a node offset strategy, and selecting the optimal node using a candidate node evaluation strategy; setting a gravitational potential field for the end point to interact with the sampling point to guide the generation of candidate nodes; using a candidate node evaluation strategy to evaluate the quality of candidate nodes; and adopting a pruning strategy to remove redundant nodes in the path.

[0010] The Dubins method is used to optimize the flight paths of multiple UAVs.

[0011] Preferably, environmental modeling includes:

[0012] The building is constructed as a cube, and a safety distance is set for the building so that the drone cannot pass through; a safety radius is set for the drone; the dynamic obstacle is set as a sphere so that the drone always keeps a safe distance from it; the threat point is modeled as a cone-like object, which has the threat direction, threat range angle and threat distance; for is the working space, W o ,W t ,W f ∈W represents the impassable area, the threatening but passable area, and the safe passable area; the impassable area includes static buildings and dynamic obstacles W O =W b ∪W d (t); W b represents a static building, W d Represents a dynamic obstacle; there are m dynamic obstacles in the environment, and the position of the kth dynamic obstacle at time t is There are n drones A={1,2,...,n} in the environment. For drone a i Its path is represented as

[0013] P i (t) = {p i,0 (t),p i,1 (t),p i,2 (t),…p i,k (t),…,p i,n (t),p i,n+1 (t)}t∈[0,T i ]

[0014] where p i,0 (t) represents drone ai Starting point, p i,n+1 (t) represents the end point, n+1 represents the number of trajectory points, p i,k (t) represents drone a i At time t, the intermediate path point, T i Indicates drone a i flight time.

[0015] Preferably, the Dubins algorithm is used to construct a three-dimensional path between two obstacle-free points;

[0016] The optimal flight path between the start and end points is determined based on the heading and pitch angles of the start and end points, the minimum turning R of the drone, and the three-dimensional path planning of multiple drones, including:

[0017] The drone's heading angle γ represents its steering ability in the horizontal plane, that is, the lateral deviation angle of the drone relative to the forward direction; the drone's pitch angle ψ describes its attitude adjustment in the vertical plane, that is, the upward or downward tilt angle relative to the horizontal plane;

[0018] For continuous path points P on a path k , P k+1 , the UAV is at the path point P k The flight direction is calculated as follows:

[0019]

[0020] In the above formula, V k Indicates that the drone is in P k The flight direction on the drone is at the end point P n+1 The flight direction is consistent with the flight direction of the previous waypoint;

[0021] V k =(v xk ,v yk ,v zk ), according to the three-dimensional dynamic model of the UAV, the vector relationship between the direction angle and the pitch angle is as follows:

[0022] v xk =||V k ||cosγcosψ

[0023] v yk =||V k ||cosγsinψ

[0024] v zk =||V k ||sinγ

[0025] The calculation formula for the direction angle and pitch angle is obtained:

[0026]

[0027] Preferably, the multi-UAV collaborative path planning is modeled as a minimization optimization problem, with the goal of minimizing flight path costs and UAV flight risks while meeting complex mission-related requirements and constraints;

[0028] The objective function is as follows:

[0029]

[0030] Among them, N u Indicates the number of drones, N i Indicates drone a i The number of path segments, C m represents the constraint condition, x max Indicates the maximum coordinate of the drone in the x-axis direction, y max Indicates the maximum coordinate of the drone in the y-axis direction, zmax indicates the maximum coordinate of the drone in the z-axis direction, and x i (t) represents drone a i The x-axis coordinate, y-axis coordinate of the j-th path i (t) represents drone a i The y-axis coordinate of the j-th path, z i (t) represents drone a i The z-axis coordinate of the j-th path, d ij Indicates drone a i The length of the j-th path; w ij Indicates drone a i The degree to which the j-th path is threatened, the closer it is to 1, the greater the degree of threat it is threatened. The specific calculation is as follows:

[0031]

[0032] In the above formula, w k (p) represents the threat level of the kth threat source to p, d kp represents the distance between the kth threat source and point p, los = 1 means there is no building blocking the distance between the threat source and point p, θ kp represents the threat angle of the kth threat source to p, θ max Indicates the threat angle range, r f represents the safe traversable radius, r k represents the threat distance of the kth threat source, N t represents the number of navigable paths with threats, p i,j Indicates drone a i At the path point of the jth path;

[0033] The constraints are expressed as follows:

[0034] The first constraint is to ensure that the path does not exceed the maximum turning angle γ of the drone during traversal max , which is the key to ensure the safe and stable flight of the drone:

[0035] C1=γ―γ max ≤0

[0036] The second constraint is that the climb angle cannot exceed a predetermined maximum angle ψ max , which is expressed as follows:

[0037] C2=ψ―ψ max ≤0

[0038] The third constraint is that the drone cannot collide with static buildings during flight. For static buildings, a safe distance is set around them, and the drone avoids collision with them in space. It can be expressed as:

[0039]

[0040] B represents the static building set W b At any point, r i,safe Indicates drone a i The safety distance, P i (t) indicates;

[0041] The fourth constraint is that the drone cannot collide with dynamic obstacles, which is expressed as follows:

[0042]

[0043] r k,safe represents the threat radius of the kth dynamic obstacle, is the position of the kth dynamic obstacle at time t;

[0044] The fifth constraint is that collisions between drones are prohibited, which can be expressed as follows:

[0045]

[0046] j represents the jth drone.

[0047] Preferably, when generating a new path segment, collision detection is first performed to ensure that UAVs avoid each other during flight. During the construction of the expansion tree, each time a new path segment is added, not only is it checked for collisions with obstacles in the environment, but also spatiotemporal collision detection is performed on other expansion tree edges within the same time period to ensure safety between UAVs. Specifically, this includes:

[0048] First, the time range of the current path segment is determined, and then the path edges within this time period are filtered out from all expansion trees. Based on the divided areas, the path edges in the same area as the current path segment are further filtered to reduce the amount of calculation. Next, the shortest distance between these path edges and the current path segment is calculated. If this distance is less than the safe distance, further spatiotemporal detection is required, that is, the shortest distance between the two is recalculated at the same time point. If it is still less than the safe distance, it is determined that the connection cannot be made safely. If all tests pass, the path segment is added to the expansion tree.

[0049] Preferably, the step size is adaptively adjusted based on the area where the parent node of the sampling point is located and in combination with the obstacle distribution, a node offset strategy is used to generate multiple candidate nodes, and the optimal node is selected through a candidate node evaluation strategy, including:

[0050] First, multiple sampling points are generated, and multiple candidate nodes are generated by combining the artificial potential field method;

[0051] Adaptively adjust the step size based on the obstacle density in the area where the nearest parent node is located to optimize the distribution of candidate nodes;

[0052] All candidate nodes are ranked using the candidate node evaluation strategy, and the optimal candidate nodes are screened out through the spatiotemporal security detection mechanism to ensure the safety and feasibility of the path.

[0053] Preferably, the adaptive step size dynamically adjusts the step size according to the obstacle distribution in the environment, thereby optimizing the path extension process; specifically, the steps include:

[0054] Determine the sub-region where the nearest parent node of the current expansion tree is located;

[0055] Calculate the generation direction of the new node based on the direction of the nearest parent node pointing to the sampling point and the direction of the nearest parent node pointing to the target point;

[0056] Based on the above direction, three adjacent sub-regions of the sub-region along the direction are further determined to form a detection range including four sub-regions;

[0057] The building densities of these four sub-areas are comprehensively considered, and the step size is dynamically adjusted so that the path shrinks in areas with dense obstacles and expands in open areas, thereby improving the adaptability and efficiency of path search.

[0058] Preferably, the step of setting a gravitational potential field for the endpoint to interact with the sampling point and guide the generation of the candidate node specifically includes:

[0059] The artificial potential field function is set as follows, where ζ is the gravitational potential field coefficient:

[0060]

[0061] x rand is the sampling point location, x goal is the end position, is the boundary distance between the two potential fields;

[0062] The above formula combines conical and quadratic potential fields. At long distances, the conical potential field provides a fixed, large attraction, allowing the agent to move quickly toward the target. At close range, the quadratic potential field provides an attraction that decreases as the distance decreases, helping the object approach the target more accurately. Combining these two potential fields creates a potential field that provides sufficient attraction when the object is far away from the target, while providing more precise control when the agent approaches the target.

[0063] The gradient corresponding to the potential field is shown as:

[0064]

[0065] Among them, x nearest is the position of the point closest to the end point, is the boundary distance between the two potential fields, when When , the gradients of the quadratic potential field and the conical potential field are equal;

[0066] According to the random sampling point and the end point position, the gravitational force of the nearest parent node is calculated. The calculation formula is as follows:

[0067]

[0068] F t (x nearest ) is the gravitational force exerted on the nearest parent node, x rand is the sampling point location, F rand is the gravitational force on the sampling point.

[0069] Preferably, a candidate node evaluation strategy is used to evaluate the quality of candidate nodes, including:

[0070] The definition of the heuristic function is as follows:

[0071] f(x)=h(x)+g(x)

[0072] Where f(x) represents the value information of the current node x, h(x) represents the estimated cost from the current node to the end point, and g(x) represents the actual cost from the starting point to the current node;

[0073] The fitness value of x_new is quantified by the following formula:

[0074] f(x new )=g(x parent )+βh(xnew )+dis(x parent ,x new )

[0075] g(x parent ) is the starting point to the parent node x parent The path length consumed; h(x new ) is from candidate node x new Euclidean distance to the end point; dis(x parent ,x new ) is from x parent to x new The Euclidean distance of ; β is the balancer of the function, coordinating the dependence on the historical path and the response to the target proximity;

[0076] The pruning strategy specifically includes:

[0077] Starting from the first generated path, with the starting point as the starting node, check whether it can be safely connected to the subsequent nodes in sequence; if the test passes, directly cross the intermediate nodes until it can no longer cross; then, start from the node that cannot be connected and perform the same operation again until the end of the path; during the test, ensure that the path does not collide with other drones or obstacles at the time and space level to ensure flight safety

[0078] The second aspect of this application proposes a multi-UAV dynamic threat environment path planning device based on MUTE-RRT*, including:

[0079] Environmental Modeling Module: used to build models of the impact of threat areas, dynamic obstacles, and buildings on the drone's path;

[0080] Building module: Constructing a multi-UAV 3D path planning problem based on constraints and objective functions;

[0081] Path generation module: For the multi-UAV three-dimensional path planning problem, the improved RRT* algorithm is used to simultaneously generate flight paths for multiple UAVs. The improved RRT* algorithm includes: dividing the entire environment into regions according to the x and y axes and assigning obstacles to corresponding sub-regions; when generating new path segments, collision detection is first performed to ensure that UAVs avoid each other during flight; the step size is adaptively adjusted based on the region where the parent node of the sampling point is located and the distribution of obstacles; a node offset strategy is used to generate multiple candidate nodes, and the optimal node is selected through a candidate node evaluation strategy; a gravitational potential field is set for the end point to interact with the sampling point to guide the generation of candidate nodes; a candidate node evaluation strategy is used to evaluate the quality of candidate nodes; and a pruning strategy is adopted to remove redundant nodes in the path;

[0082] Optimization module: Use the Dubins method to optimize the flight paths of multiple UAVs.

[0083] This application has the following beneficial effects:

[0084] First, the multi-UAV path planning environment is expanded to complex urban environments containing threat areas and dynamic obstacles, making it closer to real-life application scenarios and providing important support for the development of multi-UAV path planning. In response to the challenges of multi-UAV path planning, the present invention introduces the time dimension based on the Rapidly-exploring Random Tree star (RRT*) algorithm and proposes a spatiotemporal safety detection mechanism to ensure the safe flight of UAVs. In addition, an environmental area segmentation strategy is proposed to improve the efficiency of safety detection. At the same time, the present invention improves the RRT* algorithm and proposes an adaptive step size strategy, a candidate node generation and evaluation strategy, and a path cutting strategy to optimize the path search process. These improvements effectively reduce planning time and improve path quality, thereby enhancing the path planning performance of the multi-UAV system. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 The method processing flow of this application. DETAILED DESCRIPTION

[0086] The present invention will be further described below with reference to the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention fall within the scope of protection of the present invention.

[0087] The core problem to be solved by this application is how to enable multiple drones to successfully traverse the environment from a starting position to a target location in an urban environment with dynamic obstacles and threat zones. The environment is densely populated with buildings, dynamic obstacles, and multiple threat points. Drones cannot fly through buildings and dynamic obstacles, but they can pass within the coverage area of ​​the threat points, but there is a risk of being destroyed by interference. The closer to the threat point, the higher the risk of being destroyed by interference. This invention models the urban environment with dynamic obstacles and threat zones, abstracting it into a problem space for drone path planning.

[0088] The present invention provides a multi-UAV path planning method based on the MUTE-RRT* algorithm, which is suitable for complex urban environments with threats of dynamic obstacles. In order to effectively realize the flight path planning of multiple UAVs in such environments, the present invention first models the entire environment, abstracts dynamic obstacles, threat areas and dense buildings into a problem space for UAV planning. On this basis, mathematical modeling is performed for multi-UAV path planning, and a Multi-Unmanned Aerial Vehicle Time Rapidly-exploring Random Tree star (MUTE-RRT*) algorithm is proposed. The MUTE-RRT* algorithm introduces the time dimension on the basis of the RRT* algorithm, comprehensively considering the multi-UAV collision avoidance constraints, UAV motion characteristics and the collaborative planning of multi-UAV flight paths to improve the safety and efficiency of path planning.

[0089] Specifically, the multi-UAV dynamic threat environment path planning method disclosed in this application includes the following steps:

[0090] Modeling the environment, including the impact of threat zones, dynamic obstacles, and buildings on the drone’s path;

[0091] Construct a three-dimensional path planning problem for multiple UAVs based on constraints and objective functions;

[0092] For the three-dimensional path planning problem of multiple UAVs, an improved RRT* algorithm is used to simultaneously generate flight paths for multiple UAVs. The improved RRT* algorithm includes: dividing the entire environment into regions according to the x and y axes and assigning obstacles to corresponding sub-regions; first performing collision detection when generating new path segments to ensure that UAVs avoid each other during flight; adaptively adjusting the step size based on the region where the parent node of the sampling point is located and the distribution of obstacles; generating multiple candidate nodes using a node offset strategy, and selecting the optimal node using a candidate node evaluation strategy; setting a gravitational potential field for the end point to interact with the sampling point to guide the generation of candidate nodes; using a candidate node evaluation strategy to evaluate the quality of candidate nodes; and adopting a pruning strategy to remove redundant nodes in the path.

[0093] The Dubins method is used to optimize the flight paths of multiple drones. The Dubins method uses geometric analysis to provide the shortest feasible path for constrained kinematic systems. It is a fundamental planning tool in fields such as robotics and drones and is currently used in this field, so this application will not elaborate further.

[0094] In one embodiment, the present application requires modeling the environment before performing drone path planning. Through environmental modeling, the actual environment is converted into a computer-processable mathematical representation, providing an operating space for the path planning algorithm.

[0095] Buildings are constructed as cubes, and a safety distance is set for them. Drones cannot pass through them. Since a safety radius is set for drones, a safety distance is also set for buildings. Dynamic obstacles are set as spheres. Dynamic obstacles move regularly throughout the environment, moving at a constant speed along a path. Their specific properties are shown in Tables 1 and 2. Drones must always maintain a safe distance from them. Threat points are modeled as cone-like objects. Their specific properties are shown in Table 3. They have threat direction, threat range angle, and threat distance. Although drones can pass through their coverage area, they are at risk of being destroyed by interference. The closer the distance, the stronger the interference, while the farther the distance, the weaker the interference.

[0096] Table 1 Threat types

[0097] Threat Sources Are drones allowed? building Impassable Dynamic obstacles Impassable Threat Points Passable

[0098] Table 2 Dynamic obstacle attributes

[0099] symbol significance <![CDATA[Path d ]]> Waypoints <![CDATA[v d ]]> Flight speed <![CDATA[r d ]]> Threat Radius

[0100] Table 3 Firepower point attributes

[0101] symbol significance <![CDATA[P t ]]> Threat point location <![CDATA[dir t ]]> Threat Direction Vector <![CDATA[θ max ]]> Threat range angle <![CDATA[r t ]]> Threat distance

[0102] definition is the working space, W o ,W t ,W f ∈W represents the impassable area, the threatening but passable area, and the safe passable area. The impassable area includes static buildings and dynamic obstacles W O =W b ∪W d (t). This study defines m dynamic obstacles, and the position of the kth dynamic obstacle is The present invention studies n drones A={1,2,...,n}, for drone a i Its path can be expressed as

[0103] P i (t) = {p i,0 (t),p i,1 (t),p i,2 (t),…p i,k (t),…,p i,n (t),p i,n+1 (t)}t∈[0,T i ]

[0104] where p i,0 (t) represents drone ai Starting point, p i,n+1 (t) indicates the end point. i,k (t) = {x i,k (t),y i,k (t),z i,k (t)} represents drone a i Intermediate path points at time t. In the present invention, these path points are connected by straight lines. Current path representations have certain shortcomings. For example, a path constructed by directly connecting path points may result in an uneven flight trajectory, which may limit the performance of the drone. This application uses the Dubins algorithm to construct a three-dimensional path between two obstacle-free points. This method can both meet the drone's own performance constraints and minimize changes to the drone's flight trajectory.

[0105] In one embodiment, the present application generates a path for a drone in three-dimensional space that satisfies curvature and angle constraints. Given a starting point and an end point, as well as the drone's minimum turn R, the optimal flight path between the two points can be determined by simply providing the heading and pitch angles of the starting and end points.

[0106] The azimuth angle of the UAV represents its steering ability in the horizontal plane, that is, the lateral deviation angle of the UAV relative to the forward direction, which is represented by the symbol γ. In the assumption of the Dubins curve, the UAV is restricted to flying only in the direction of its forward movement. Therefore, the range of the azimuth angle is Correspondingly, the pitch angle of the drone describes its attitude adjustment in the vertical plane, that is, the tilt angle upward or downward relative to the horizontal plane, which is represented by ψ. Similarly, the range of the pitch angle is also

[0107] There is a continuous path point P on a path k , P k+1 , the UAV is at the path point P k The flight direction is calculated as follows:

[0108]

[0109] In the above formula, V k Indicates that the drone is in P k In particular, the UAV is at the end point P n+1 The flight direction remains the same as the flight direction of the previous waypoint.

[0110] Assume V k =(v xk ,v yk ,v zk ), according to the three-dimensional dynamic model of the UAV, the vector relationship between the direction angle and the pitch angle is as follows:

[0111] v xk =||V k ||cosγcosψ

[0112] v yk =||V k ||cosγsinψ

[0113] v zk =||V k ||sinγ

[0114] Therefore, the calculation formula for the direction angle and pitch angle can be obtained:

[0115]

[0116] In one embodiment, the multi-UAV path planning model includes:

[0117] Multi-UAV 3D path planning involves finding flight paths for multiple drones in three-dimensional space, taking into account multiple constraints and mission requirements. In this application, multi-UAV collaborative path planning is modeled as a minimization optimization problem. The core goal is to minimize flight path cost and drone flight risk while meeting the complex requirements and constraints associated with the mission. The objective function design primarily considers factors such as path length and path risk. Constraints include collision avoidance, flight area restrictions, drone physical performance, and safe distances between drones.

[0118]

[0119]

[0120] d in the objective function ij Indicates drone a i The length of the j-th path; similarly, w ij Indicates drone a i The closer it is to 1, the greater the threat level. The specific calculation is as follows:

[0121]

[0122] In the above formula, w k (p) represents the threat level of the kth threat source to p, d kp represents the distance between the kth threat source and point p, and los = 1 means that there is no building blocking the threat source and point p.

[0123] Constraints:

[0124] (1) The first constraint is to ensure that the path does not exceed the maximum turning angle γ of the UAV during traversal max , which is the key to ensure the safe and stable flight of the drone.

[0125] C1=γ―γ max ≤0

[0126] (2) The second constraint is that the climb angle cannot exceed the predetermined maximum angle ψ max , which is expressed as follows:

[0127] C2=ψ―ψ max ≤0

[0128] (3) The third constraint is that the UAV cannot collide with static buildings during flight. For static buildings, the present invention sets a safety distance around them, and the UAV only needs to avoid collision with them in space, which can be expressed as

[0129]

[0130] B represents the static building set W b At any point, r i,safe Indicates drone a i safe distance.

[0131] (4) The fourth constraint is that the UAV cannot collide with dynamic obstacles, which is expressed as follows:

[0132]

[0133] r k,safe Indicates the threat radius of the kth dynamic obstacle

[0134] (5) The fifth constraint is that collisions between drones are prohibited, which is expressed as follows:

[0135]

[0136] In one embodiment, the present application proposes a Multi-Unmanned Aerial Vehicle Time Rapidly-exploring Random Tree star (MUTE-RRT*) algorithm based on the RRT* algorithm. The overall framework of the algorithm is described below, and the pseudo code of the algorithm is shown in Algorithm 1.

[0137] First, the vertex set, edge set, path set, and flag are initialized. This invention also extends the time dimension to the vertex set. The time at which a drone reaches a vertex is obtained (line 1). Next, a loop is executed, using the node generation strategy node_generation to obtain new sampling points (line 9). Next, the growth tree is updated according to the original RRT* algorithm (lines 10-12). The goal_connection function is then used to determine whether the drone has reached the target node and can connect to it. If so, the flight path of the drone is obtained (lines 13-15). A pruning strategy is then used to reduce redundant nodes (line 22). Finally, the trajectory is optimized using the Dubins trajectory function (line 23).

[0138]

[0139]

[0140] Compared with the traditional RRT* algorithm, the MUTE-RRT* algorithm has the following improvements:

[0141] First, to improve path detection efficiency, the present invention divides the entire environment into regions along the x and y axes and assigns obstacles to corresponding sub-regions. During path detection, only obstacles within the region passed by the path segment need to be checked, thus reducing computational complexity.

[0142] Second, the algorithm simultaneously plans paths for multiple drones, generating multiple expansion trees simultaneously. To avoid path conflicts, the algorithm introduces a time dimension based on RRT* and proposes a spatiotemporal safety detection mechanism. When generating new path segments, this mechanism first performs collision detection to ensure that drones can effectively avoid each other during flight, ensuring flight safety.

[0143] Third, the algorithm proposes an improved node generation strategy. First, the step size is adaptively adjusted based on the location of the sampling point's parent node and the distribution of obstacles. Then, a node offset strategy is used to generate multiple candidate nodes, and a candidate node evaluation strategy is used to select the optimal node to improve path quality.

[0144] Fourth, drawing on the artificial potential field method, a gravitational potential field is set for the end point, which interacts with the sampling points to guide the generation of candidate nodes and improve the goal orientation of path planning.

[0145] Fifth, referring to the fitness function in the A* algorithm, a candidate node evaluation strategy is proposed to evaluate the quality of candidate nodes. This strategy helps improve path quality and accelerate algorithm convergence.

[0146] Sixth, a pruning strategy is used to remove redundant nodes in the path, reducing unnecessary path bends while ensuring safety, thereby optimizing path quality.

[0147] Seventh, for the final path, the Dubins curve is used to optimize the trajectory so that it meets the UAV's self-generated performance constraints.

[0148] In one embodiment, during the generation of a new node, it is necessary to detect whether the newly generated path segment collides with an obstacle. However, since the path segment is shorter than the entire environment space, checking its collision with all obstacles one by one is computationally intensive and inefficient. To this end, the present application proposes a region division mechanism that improves the efficiency of collision detection by zoning the environment. Specifically, the entire environment is first divided into multiple sub-areas according to the x-axis and y-axis, and classified according to the coverage of the obstacles. Static buildings are divided according to the space they occupy, threat points are divided according to their affected areas, and dynamic obstacles are classified according to their range of motion. It should be noted that a single obstacle may be assigned to multiple adjacent sub-areas to ensure the integrity of the detection. Subsequently, during the path generation process, the sub-area in which the path segment is located is quickly determined by the spatial position and time information of the path segment, thereby accurately screening obstacles that may affect the path segment. This method can effectively reduce the computational complexity of collision detection and greatly improve the efficiency of path planning.

[0149] In one embodiment, during multi-UAV path planning, this application proposes a spatiotemporal safety detection mechanism to avoid collisions between UAVs in time and space. During the construction of the expansion tree, each time a new path segment is added, it is not only necessary to check whether it will collide with obstacles in the environment, but also to perform spatiotemporal collision detection on other expansion tree edges within the same time period to ensure safety between UAVs.

[0150] Specifically, as shown in Algorithm 2, the time range of the current path segment is first determined. Path edges within this time range are then filtered from all expansion trees. Based on the region partitioning mechanism, path edges that fall within the same region as the current path segment are further filtered to reduce computational complexity. Next, the shortest distance between these path edges and the current path segment is calculated. If this distance is less than the safe distance, further spatiotemporal testing is performed. This involves recalculating the shortest distance between the two paths at the same time point. If this distance is still less than the safe distance, the connection is deemed unsafe. If all tests pass, the path segment can be safely added to the expansion tree.

[0151]

[0152] In one embodiment, to improve the random sampling quality and planning efficiency of the RRT* algorithm, this application proposes an improved node generation strategy during the node generation process. This strategy generates multiple candidate nodes at once and combines this with a candidate node evaluation mechanism to screen them, thereby improving node quality. Furthermore, an artificial potential field method is introduced to accelerate algorithm convergence.

[0153] Specifically, as shown in Algorithm 3, multiple sampling points are first generated, and then multiple candidate nodes are generated using the artificial potential field method. Subsequently, the step size is adaptively adjusted based on the obstacle density in the area of ​​the nearest parent node to optimize the distribution of candidate nodes. Next, a candidate node evaluation strategy is used to rank all candidate nodes, and a spatiotemporal safety detection mechanism is used to select the optimal candidate node to ensure the safety and feasibility of the path.

[0154]

[0155]

[0156] In one embodiment, the adaptive step size generation strategy includes:

[0157] In the original RRT* algorithm, the step size is always fixed. However, in real-world environments, obstacles are often unevenly distributed. In densely built-up areas, the step size should be shortened to improve path feasibility and enable the expanded tree to more flexibly avoid obstacles. In sparsely built-up areas, the step size can be appropriately increased to accelerate path search and improve planning efficiency.

[0158] To this end, this application proposes an adaptive step size generation strategy. The core idea is to dynamically adjust the step size according to the obstacle distribution in the environment, thereby optimizing the path extension process. Specifically, the strategy includes the following steps:

[0159] First, determine the child region where the nearest parent node of the current expansion tree is located.

[0160] The generation direction of the new node is calculated based on the direction of the nearest parent node pointing to the sampling point and the direction of the nearest parent node pointing to the target point.

[0161] Based on the above direction, three adjacent sub-regions of the sub-region along the direction are further determined to form a detection range including four sub-regions.

[0162] The building densities of these four sub-areas are comprehensively considered, and the step size is dynamically adjusted so that the path shrinks in areas with dense obstacles and expands in open areas, thereby improving the adaptability and efficiency of path search.

[0163] In one embodiment, the present application constructs an attractive potential field for the endpoint. When a new node is generated, the new node is obtained by using the potential field and the direction of the sampling point. The artificial potential field function set by the present application is as follows, where ζ is the gravitational potential field coefficient.

[0164]

[0165] This application combines conical and quadratic potential fields. At long distances, the conical potential field provides a fixed, large attraction, enabling the agent to quickly move toward the target. At close range, the quadratic potential field provides an attraction that decreases with distance, helping the object approach the target more precisely. Combining these two potential fields creates a potential field that provides sufficient attraction when the object is far from the target, while providing more precise control when the agent approaches the target, thus avoiding overshooting.

[0166] The gradient corresponding to the potential field is shown in Eq.

[0167]

[0168] in, is the boundary distance between the two potential fields, when When , the gradients of the quadratic potential field and the conical potential field are equal. Set to three times the step size and ζ to 1 / 3.

[0169] Next, we calculate the gravitational force on the nearest parent node based on the random sampling point and the end point position. The calculation formula is as follows:

[0170]

[0171] In one embodiment, a candidate node evaluation strategy is designed to evaluate and rank multiple candidate tree nodes. This strategy uses an optimized heuristic function to evaluate the value of each candidate node. After the evaluation is completed, all candidate tree nodes are ranked from low to high based on their fitness values.

[0172] In the A* algorithm, its heuristic function is defined as follows:

[0173] f(x)=h(x)+g(x)

[0174] Where f(x) represents the value information of the current node x, h(x) represents the estimated cost from the current node to the end point, and g(x) represents the actual cost from the starting point to the current node.

[0175] This application uses the following formula to quantify the fitness value of x_new

[0176] f(x new )=g(x parent )+βh(x new )+d i s(x parent ,x new )

[0177] f(x new ) consists of three parts. The first part is from the starting point to x parent The path length consumed is given by g(x parent ) indicates that the second part is from the candidate node x new The distance to the end point is calculated using the Euclidean distance, using h(x new ) indicates that the third part is from x parent to x new The distance is also calculated using the Euclidean distance, using d i s(x paren t,x new ) is represented by β. β is the balancer of this function, which coordinates the dependence on the historical path and the response to the target proximity.

[0178] Although the improved algorithm performs well in terms of quality and efficiency, the generated path still contains redundant nodes, which increases the path length and affects the flight efficiency of the drone. To further optimize the path, in one embodiment, this application introduces a pruning strategy to reduce unnecessary nodes while ensuring safety, thereby improving the smoothness and quality of the path.

[0179] Starting with the first generated path, with the starting point as the starting node, the system checks whether it can safely connect to subsequent nodes. If the check passes, it directly crosses the intermediate nodes until it cannot continue to cross. Then, it repeats the same operation from the node that cannot be connected until the end of the path. It is important to note that safety checks must meet spatiotemporal safety requirements. That is, the path must not collide with other drones or obstacles in time and space to ensure flight safety.

[0180] This application has the following beneficial effects:

[0181] First, the multi-UAV path planning environment is expanded to complex urban environments containing threat areas and dynamic obstacles, making it closer to real-life application scenarios and providing important support for the development of multi-UAV path planning. In response to the challenges of multi-UAV path planning, the present invention introduces the time dimension based on the Rapidly-exploring Random Tree star (RRT*) algorithm and proposes a spatiotemporal safety detection mechanism to ensure the safe flight of UAVs. In addition, an environmental area segmentation strategy is proposed to improve the efficiency of safety detection. At the same time, the present invention improves the RRT* algorithm and proposes an adaptive step size strategy, a candidate node generation and evaluation strategy, and a path cutting strategy to optimize the path search process. These improvements effectively reduce planning time and improve path quality, thereby enhancing the path planning performance of the multi-UAV system.

[0182] As used herein, the word "preferred" is intended to serve as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as advantageous over other aspects or designs. Rather, the use of the word "preferred" is intended to present concepts in a concrete manner. As used in this application, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified or clear from the context, "X employs A or B" is intended to mean any of the naturally inclusive permutations. That is, if X employs A; X employs B; or X employs both A and B, then "X employs A or B" is satisfied in any of the foregoing examples.

[0183] Moreover, although the present disclosure has been shown and described with respect to one or implementation, those skilled in the art will think of equivalent variations and modifications based on reading and understanding of this specification and the accompanying drawings. The present disclosure includes all such modifications and variations and is limited only by the scope of the appended claims. In particular, with respect to the various functions performed by the above-mentioned components (such as elements, etc.), the terms used to describe such components are intended to correspond to any component (unless otherwise indicated) that performs the specified function of the component (such as it is functionally equivalent), even if structurally different from the disclosed structure that performs the function in the exemplary implementation of the present disclosure shown herein. In addition, although the specific features of the present disclosure have been disclosed with respect to only one of several implementations, such features can be combined with one or other features of other implementations that can be desired and advantageous for a given or specific application. Moreover, insofar as the terms "including", "having", "containing" or their variations are used in specific embodiments or claims, such terms are intended to be included in a manner similar to the term "comprising".

[0184] The functional units in the embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or multiple or more units may be integrated into a single module. The aforementioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. The aforementioned storage medium may be a read-only memory, a magnetic disk, or an optical disk, etc. The aforementioned devices or systems may execute the storage method in the corresponding method embodiment.

[0185] In summary, the above embodiment is one implementation method of the present invention, but the implementation method of the present invention is not limited to the described embodiment. Any other changes, modifications, substitutions, combinations, and simplifications that deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A multi-UAV path planning method in a dynamic threat environment based on MUTE-RRT*, characterized by: The following steps are involved: Modeling the environment, including the impact of threat zones, dynamic obstacles, and buildings on the drone’s path; Construct a three-dimensional path planning problem for multiple UAVs based on constraints and objective functions; For the three-dimensional path planning problem of multiple UAVs, an improved RRT* algorithm is used to simultaneously generate flight paths for multiple UAVs. The improved RRT* algorithm includes: dividing the entire environment into regions according to the x and y axes and assigning obstacles to corresponding sub-regions; first performing collision detection when generating new path segments to ensure that UAVs avoid each other during flight; adaptively adjusting the step size based on the region where the parent node of the sampling point is located and the distribution of obstacles; generating multiple candidate nodes using a node offset strategy, and selecting the optimal node using a candidate node evaluation strategy; setting a gravitational potential field for the end point to interact with the sampling point to guide the generation of candidate nodes; using a candidate node evaluation strategy to evaluate the quality of candidate nodes; and adopting a pruning strategy to remove redundant nodes in the path. The Dubins method is used to optimize the flight paths of multiple UAVs.

2. The multi-UAV dynamic threat environment path planning method based on MUTE-RRT* according to claim 1 is characterized in that: Environmental modeling includes: The building is constructed as a cube, and a safety distance is set for the building so that the drone cannot pass through; a safety radius is set for the drone; the dynamic obstacle is set as a sphere so that the drone always keeps a safe distance from it; the threat point is modeled as a cone-like object, which has the threat direction, threat range angle and threat distance; for is the working space, W o ,W t ,W f ∈W represents the impassable area, the threatening but passable area, and the safe passable area; the impassable area includes static buildings and dynamic obstacles W O =W b ∪W d (t); W b represents a static building, W d Represents a dynamic obstacle; there are m dynamic obstacles in the environment, and the position of the kth dynamic obstacle at time t is There are n drones A={1,2,...,n} in the environment. For drone a i Its path is represented as P i (t)={p i,0 (t),p i,1 (t),p i,2 (t),…p i,k (t),…,p i,n (t),p i,n+1 (t)}t∈[0,T i ] where p i,0 (t) represents drone a i Starting point, p i,n+1 (t) represents the end point, n+1 represents the number of trajectory points, p i,k (t) represents drone a i At time t, the intermediate path point, T i Indicates drone a i flight time.

3. The multi-UAV dynamic threat environment path planning method based on MUTE-RRT* according to claim 2 is characterized in that: The Dubins algorithm is used to construct a three-dimensional path between two obstacle-free points; The optimal flight path between the start and end points is determined based on the heading and pitch angles of the start and end points, the minimum turning R of the drone, and the three-dimensional path planning of multiple drones, including: The drone's heading angle γ represents its steering ability in the horizontal plane, that is, the lateral deviation angle of the drone relative to the forward direction; the drone's pitch angle ψ describes its attitude adjustment in the vertical plane, that is, the upward or downward tilt angle relative to the horizontal plane; For continuous path points P on a path k , P k+1 , the UAV is at the path point P k The flight direction is calculated as follows: In the above formula, V k Indicates that the drone is in P k The flight direction on the drone is at the end point P n+1 The flight direction is consistent with the flight direction of the previous waypoint; V k =(v xk ,v yk ,v zk ), according to the three-dimensional dynamic model of the UAV, the vector relationship between the direction angle and the pitch angle is as follows: in xk =||V k ||cosγcosψ v yk =||V k ||cosγsinψ in zk =||V k ||sinγ The calculation formula for the direction angle and pitch angle is obtained:

4. The multi-UAV dynamic threat environment path planning method based on MUTE-RRT* according to claim 2 is characterized in that: The multi-UAV collaborative path planning is modeled as a minimization optimization problem, with the goal of minimizing the flight path cost and the flight risk of the UAVs while meeting the complex requirements and constraints related to the mission; The objective function is as follows: Among them, N u Indicates the number of drones, N i Indicates drone a i The number of path segments, C m represents the constraint condition, x max Indicates the maximum coordinate of the drone in the x-axis direction, y max Indicates the maximum coordinate of the drone in the y-axis direction, z max Indicates the maximum coordinate of the drone in the z-axis direction, x i (t) represents drone a i The x-axis coordinate, y-axis coordinate of the j-th path i (t) represents drone a i The y-axis coordinate of the j-th path, z i (t) represents drone a i The z-axis coordinate of the j-th path, d ij Indicates drone a i The length of the j-th path; w ij Indicates drone a i The degree to which the j-th path is threatened, the closer it is to 1, the greater the degree of threat it is threatened. The specific calculation is as follows: In the above formula, w k (p) represents the threat level of the kth threat source to p, d kp represents the distance between the kth threat source and point p, los = 1 means there is no building blocking the distance between the threat source and point p, θ kp represents the threat angle of the kth threat source to p, θ max Indicates the threat angle range, r f represents the safe traversable radius, r k represents the threat distance of the kth threat source, N t represents the number of navigable paths with threats, p i,j Indicates drone a i At the path point of the jth path; The constraints are expressed as follows: The first constraint is to ensure that the path does not exceed the maximum turning angle γ of the drone during traversal max , which is the key to ensure the safe and stable flight of the drone: C1=γ―γ max ≤0 The second constraint is that the climb angle cannot exceed a predetermined maximum angle ψ max , which is expressed as follows: C2=ψ―ψ max ≤0 The third constraint is that the drone cannot collide with static buildings during flight. For static buildings, a safe distance is set around them, and the drone avoids collision with them in space. It can be expressed as: B represents the static building set W b At any point, r i,safe Indicates drone a i The safety distance, P i (t) indicates; The fourth constraint is that the drone cannot collide with dynamic obstacles, which is expressed as follows: r k,safe represents the threat radius of the kth dynamic obstacle, is the position of the kth dynamic obstacle at time t; The fifth constraint is that collisions between drones are prohibited, which can be expressed as follows: j represents the jth drone.

5. The multi-UAV dynamic threat environment path planning method based on MUTE-RRT* according to claim 1 is characterized in that: When generating a new path segment, collision detection is first performed to ensure that drones avoid each other during flight. During the construction of the expansion tree, each time a new path segment is added, it is not only checked whether it will collide with obstacles in the environment, but also spatiotemporal collision detection is performed on other expansion tree edges in the same time period to ensure safety between drones. Specifically, it includes: First, the time range of the current path segment is determined, and then the path edges within this time period are filtered out from all expansion trees. Based on the divided areas, the path edges in the same area as the current path segment are further filtered to reduce the amount of calculation. Next, the shortest distance between these path edges and the current path segment is calculated. If this distance is less than the safe distance, further spatiotemporal detection is required, that is, the shortest distance between the two is recalculated at the same time point. If it is still less than the safe distance, it is determined that the connection cannot be made safely. If all tests pass, the path segment is added to the expansion tree.

6. The multi-UAV dynamic threat environment path planning method based on MUTE-RRT* according to claim 1 is characterized in that: The method includes adaptively adjusting the step size based on the area where the parent node of the sampling point is located and the obstacle distribution, generating multiple candidate nodes using the node offset strategy, and selecting the optimal node through the candidate node evaluation strategy, including: First, multiple sampling points are generated, and multiple candidate nodes are generated by combining the artificial potential field method; Adaptively adjust the step size based on the obstacle density in the area where the nearest parent node is located to optimize the distribution of candidate nodes; All candidate nodes are ranked using the candidate node evaluation strategy, and the optimal candidate nodes are screened out through the spatiotemporal security detection mechanism to ensure the safety and feasibility of the path.

7. The multi-UAV dynamic threat environment path planning method based on MUTE-RRT* according to claim 6 is characterized in that: The adaptive step size dynamically adjusts the step size according to the obstacle distribution in the environment, thereby optimizing the path extension process; specifically, the following steps are included: Determine the sub-region where the nearest parent node of the current expansion tree is located; Calculate the generation direction of the new node based on the direction of the nearest parent node pointing to the sampling point and the direction of the nearest parent node pointing to the target point; Based on the above direction, three adjacent sub-regions of the sub-region along the direction are further determined to form a detection range including four sub-regions; The building densities of these four sub-areas are comprehensively considered, and the step size is dynamically adjusted so that the path shrinks in areas with dense obstacles and expands in open areas, thereby improving the adaptability and efficiency of path search.

8. The multi-UAV dynamic threat environment path planning method based on MUTE-RRT* according to claim 1 is characterized in that: The process of setting a gravitational potential field for the endpoint, interacting with the sampling point, and guiding the generation of candidate nodes specifically includes: The artificial potential field function is set as follows, where ζ is the gravitational potential field coefficient: x rand is the sampling point location, x goal is the end position, is the boundary distance between the two potential fields; The above formula combines conical and quadratic potential fields. At long distances, the conical potential field provides a fixed, large attraction, allowing the agent to move quickly toward the target. At close range, the quadratic potential field provides an attraction that decreases as the distance decreases, helping the object approach the target more accurately. Combining these two potential fields creates a potential field that provides sufficient attraction when the object is far away from the target, while providing more precise control when the agent approaches the target. The gradient corresponding to the potential field is shown as: Among them, x nearest is the position of the point closest to the end point, is the boundary distance between the two potential fields, when When , the gradients of the quadratic potential field and the conical potential field are equal; According to the random sampling point and the end point position, the gravitational force of the nearest parent node is calculated. The calculation formula is as follows: F t (x nearest ) is the gravitational force exerted on the nearest parent node, x rand is the sampling point location, F rand is the gravitational force on the sampling point.

9. The multi-UAV dynamic threat environment path planning method based on MUTE-RRT* according to claim 1 is characterized in that: Use the candidate node evaluation strategy to evaluate the quality of candidate nodes, including: The definition of the heuristic function is as follows: f(x)=h(x)+g(x) Where f(x) represents the value information of the current node x, h(x) represents the estimated cost from the current node to the end point, and g(x) represents the actual cost from the starting point to the current node; The fitness value of x_new is quantified by the following formula: f(x new )=g(x parent )+βh(x new )+dis(x parent ,x new ) g(x parent ) is the starting point to the parent node x parent The path length consumed; h(x new ) is from candidate node x new Euclidean distance to the end point; dis(x parent ,x new ) is from x parent to x new The Euclidean distance of ; β is the balancer of the function, coordinating the dependence on the historical path and the response to the target proximity; The pruning strategy specifically includes: Starting from the first generated path, with the starting point as the starting node, check whether it can be safely connected to the subsequent nodes in turn; if the test passes, directly cross the intermediate nodes until it can no longer cross; then, re-execute the same operation from the node that cannot be connected until the end of the path; during the test, ensure that the path does not collide with other drones or obstacles at the time and space levels to ensure flight safety.

10. A multi-UAV dynamic threat environment path planning device based on MUTE-RRT* according to any one of claims 1 to 9, characterized in that: include: Environmental Modeling Module: used to build models of the impact of threat areas, dynamic obstacles, and buildings on the drone's path; Building module: Constructing a multi-UAV 3D path planning problem based on constraints and objective functions; Path generation module: For the multi-UAV three-dimensional path planning problem, the improved RRT* algorithm is used to simultaneously generate flight paths for multiple UAVs. The improved RRT* algorithm includes: dividing the entire environment into regions according to the x and y axes and assigning obstacles to corresponding sub-regions; when generating new path segments, collision detection is first performed to ensure that UAVs avoid each other during flight; the step size is adaptively adjusted based on the region where the parent node of the sampling point is located and the distribution of obstacles; a node offset strategy is used to generate multiple candidate nodes, and the optimal node is selected through a candidate node evaluation strategy; a gravitational potential field is set for the end point to interact with the sampling point to guide the generation of candidate nodes; a candidate node evaluation strategy is used to evaluate the quality of candidate nodes; and a pruning strategy is adopted to remove redundant nodes in the path; Optimization module: Use the Dubins method to optimize the flight paths of multiple UAVs.

Images