Multi-unmanned aerial vehicle cooperative encirclement method in obstacle environment

Abstract

The present application belongs to the technical field of multi-four-rotor cooperative hunting, and discloses a multi-unmanned aerial vehicle cooperative hunting method in an obstacle environment, which comprises the following steps: 1: three-dimensional decision space representation based on a Voronoi diagram: three-dimensional obstacle decision space is simplified and represented by constructing a safety priority reachable region; 2: constructing a distributed hunting decision based on area minimization: on the basis of the safety priority reachable region constructed by the three-dimensional decision space representation based on the Voronoi diagram, a cooperative hunting strategy based on area minimization is adopted to compress the movement region of an escaping unmanned aerial vehicle, and finally effective capture is realized. The three-dimensional decision space representation based on the Voronoi diagram can extract a convex bounded decision region in a typical non-convex region, i.e. a three-dimensional obstacle space, realize effective representation of the decision space, and lay a foundation for generation of a subsequent hunting strategy.

Description

Technical Field

[0001] This invention belongs to the field of multi-quadrotor cooperative encirclement technology, and particularly relates to a multi-UAV cooperative encirclement method in an obstacle environment. Background Technology

[0002] Multi-quadrotor cooperative pursuit tasks typically involve multiple quadrotors coordinating to track and surround one or more escaping targets, preventing their escape. These tasks fall under the category of multi-robot pursuit (MPE) games and are widely used in areas such as area search, surveillance, and target tracking. Existing solutions primarily rely on solving differential equations, such as differential game methods based on the Hamilton-Jacobi-Isaks (HJI) equations. While these methods can provide optimal control strategies in complex environments, their high dimensionality and computational complexity make them unsuitable for large-scale problems. Furthermore, although reinforcement learning methods can train cooperative pursuit strategies, their computational complexity is high, and they exhibit poor environmental adaptability. Summary of the Invention

[0003] The purpose of this invention is to provide a method for multi-UAV collaborative capture in obstacle environments to solve the above-mentioned technical problems.

[0004] To address the aforementioned technical problems, the specific technical solution of the multi-UAV cooperative encirclement and capture method under obstacle conditions according to the present invention is as follows:

[0005] A method for cooperative encirclement and capture of multiple drones in an obstacle environment, which obtains the spatial coordinate information of multiple drones, including defensive and offensive drones, based on relevant position capture devices and algorithms, includes the following steps:

[0006] Step 1: Representation of the 3D decision space based on the Veno diagram: A simplified representation of the 3D obstacle decision space is achieved by constructing a safe-priority reachable region;

[0007] Step 2: Constructing a distributed capture decision based on area minimization: Based on the safe priority reachable area constructed by the three-dimensional decision space representation module based on the Veno map, a cooperative capture strategy based on area minimization is adopted to compress the movement area of ​​the escaping drone and ultimately achieve effective capture.

[0008] Furthermore, step 1 includes the following steps:

[0009] Step 1.1: Construct the priority reachable region based on the Veno diagram: Based on the location of the capture drone cluster and the location of the escape drone, the priority reachable region of the capture drone in the three-dimensional scene is determined by the Veno partitioning method.

[0010] Step 1.2: Construction of safe priority reachable domain based on obstacle perception: Based on the priority reachable area, considering the safe distance between the capture drones and the minimum distance between the capture drones and obstacles, the safe priority reachable domain is constructed to ensure that these constraints are always satisfied throughout the capture process;

[0011] Step 1.3: Area Iteration and Update: Based on real-time scene changes, continuously adjust the safe-priority reachable area to lay the foundation for high-frequency real-time re-decision.

[0012] Furthermore, step 1.1 includes the following specific steps:

[0013] The escapee's priority reachable area is defined as the escapee's reachable area in a given environment. The collection of all locations that can be reached faster than any other pursuer;

[0014] Given a set of seed points P = {p1, p2, ..., p...} n The Vino diagram divides space into n convex polygonal regions, called Vino regions or Dirichlet regions, each Vino region V(p i ) by all points away from seed point p i The nearest points constitute a set of points that satisfy the following conditions:

[0015]

[0016] Where Q is the task scenario space, x is any point in the space, and p i It is a seed point, and points within the Vino region are connected to the seed point p. i The distance to it is smaller than the distance to other seed points;

[0017] In a two-dimensional plane, each Vino region in a Vino diagram is a convex polygon, while in three-dimensional space, the region is composed of polyhedra. The Vino edges in a Vino diagram are shared boundaries between adjacent seed points, defined as the set of all points equidistant from two adjacent seed points. In the three-scene representation, the shared Vino boundary is represented as the perpendicular bisector of the line connecting the two types of seed points. Mathematically, a point x on the shared Vino boundary... g The following conditions must be met:

[0018]

[0019] The priority reach areas for escapees and the priority reach areas for pursuers are defined as follows:

[0020]

[0021] in, and These represent the priority reachable areas for the pursuer and the escapee in an obstacle-free environment. In this environment, each agent can determine its reachable area in space by calculating the Voronoi diagram, thereby optimizing its path planning and control strategy.

[0022] By introducing the concept of a hyperplane, a parameterized expression of the preferred reachable region can be achieved:

[0023] In a three-dimensional environment, a hyperplane can be defined using a normal vector and the coordinate representation of a point on the hyperplane. Assume there is a known point P0(x0, y0, z0) on the hyperplane and a normal vector... The equation of the hyperplane is then expressed as:

[0024]

[0025] in, Let n be any point in space. The geometric object represented by this equation divides space into two half-spaces: one region that satisfies n1(x-x0)+n2(y-y0)+n3(z-z0)>0, and the other region that satisfies n1(x-x0)+n2(y-y0)+n3(z-z0)<0.

[0026] According to the hyperplane partitioning rule, each Voronoi partition of the pursuer can be represented by a set of hyperplanes. The combination of the internal regions of multiple hyperplanes can form a convex region, which represents the preferred reachable area of ​​either the pursuer or the escapee.

[0027] Therefore, the definition of the preferred reachable region PRR can be transformed from equation (3) into the following form:

[0028]

[0029] Here, X represents the vector from the origin to any point x in Q, and l1 and l2 refer to different intelligent agents, including the escapee and the hunter cluster. and The hyperplane parameters are defined by the positional differences between agents l1 and l2. This represents the normal vector of the corresponding hyperplane and the plane position calculated from the midpoint;

[0030] Calculate the hyperplane normal vector and corresponding plane parameters associated with each pair of agents. It is determined by the relative positional differences between the pursuers and the escapees, while the positional parameters of the plane... The calculation is performed using the midpoint between these two points, ensuring that the hyperplane separates the regions according to the definition of Vino partitions. The specific calculation formula is as follows:

[0031]

[0032] In the above formula, and Let l1 and l2 represent the position vectors of agents l1 and l2, respectively. These are the relative position vectors between them, and the planar position parameters. The midpoint between these two locations is used to determine the location, thus ensuring that each hyperplane accurately divides the space.

[0033] Furthermore, step 1.2 includes the following specific steps:

[0034] A convex polyhedron is constructed to represent the safe flight zone of the drone at its current position. This convex polyhedron consists of multiple separating hyperplanes. Each hyperplane isolates the drone from obstacles, forming a safe and unobstructed flight space. The mathematical representation of this space is as follows:

[0035]

[0036] in, Indicates the safe reach area of ​​the drone. Define the composition of v l The parameter set of the hyperplane;

[0037] The KD-Tree data structure is used to accelerate the obstacle point retrieval process and separate the hyperplane parameters. The calculation formula is as follows:

[0038]

[0039] Where, r s The safety radius is used to take into account the physical size of the UAV and ensures obstacle avoidance during flight by shifting the hyperplane a certain distance away from obstacles. By combining formula (5) and formula (7), the final safe priority reachable area is defined as follows:

[0040]

[0041] Furthermore, step 1.3 includes the following specific steps:

[0042] The safe, priority reachable area is recalculated based on real-time location information.

[0043] 1) Input information: Current location of the capture drone (x) i , i∈Γ, the current position of the target escaping drone x e Obstacle point cloud information Ω_obs in the environment;

[0044] 2) Initialize the remaining obstacle point cloud information Ω_remain = Ω_obs. Assign the obstacle point cloud information Ω_obs from the environment to the remaining obstacle point cloud information Ω_remain;

[0045] 3) Construct a separating hyperplane based on obstacle point clouds;

[0046] Enter a loop, and as long as the number of points in the remaining obstacle point cloud information is greater than a preset threshold, perform the following operations:

[0047] 3-1) Find the nearest obstacle point: Use the KD-tree algorithm to find the nearest obstacle point relative to the current position x of the drone being captured. i The nearest obstacle point o_near;

[0048] 3-2) Constructing the obstacle separation hyperplane: Calculating parameters These parameters are used to construct an obstacle separation hyperplane, and the relevant information of the hyperplane is stored in the safe-priority reachable region. middle;

[0049] 3-3) Update the remaining obstacle point cloud information: Remove points located outside the newly constructed hyperplane from the remaining obstacle point cloud information to prepare for the next loop;

[0050] 4) Dealing with other drones used for surveillance

[0051] By iterating through all the capture drones in a loop, where j ranges from 0 to N and j is not equal to i:

[0052] 4-1) Let l1 be the number of the current capture drone, and l2 be the number of the capture drone currently being traversed;

[0053] 4-2) Calculate parameters according to specific formulas <n ij p ij > and store these parameters in the security-priority reachable zone. middle;

[0054] 5) Dealing with escaping drones

[0055] Set l1 to the number of the current capture drone and l2 to the number of the escape target drone, and calculate the parameters according to a specific formula. <n ie p ie > and store these parameters in the security-priority reachable zone. middle;

[0056] 6) After the above steps, the safest priority reachable area is finally obtained. As the output of the algorithm.

[0057] Furthermore, step 2 includes the following specific steps:

[0058] Step 2.1: Construct a cooperative capture strategy based on area minimization: Based on the safe priority reachable domain constructed in Step 1.2, the safe priority reachable area of ​​the escaping drone is reduced to achieve effective capture of the escaping drone;

[0059] Step 2.2: Prove the effectiveness of the encirclement strategy: Prove the theoretical completeness of the cooperative encirclement strategy based on area minimization proposed in Step 2.1 through theoretical analysis;

[0060] Step 2.3: Distributed collaborative trapping decision framework: Lightweighting of the algorithm by distributing the computational load.

[0061] Furthermore, step 2.1 includes the following specific steps:

[0062] Area of ​​the safe priority access zone for escapees Defined as:

[0063]

[0064] in It is the safe priority reachable area for the escapee constructed in formula (9), according to the definition of formula (10), The derivative is:

[0065]

[0066] In equation (11), the drone is captured. The influence term can be decoupled as Through this decoupling, the predators employ a "minimum area" movement strategy, which is consistent with... The gradient descent direction is consistent, achieving a rapid reduction in the area of ​​the safe-accessible region for the escapee. The movement strategy of the pursuers is given by the following formula:

[0067]

[0068] The specific calculation method for the above strategy is derived using Leibniz's integral rule. It can be simplified to:

[0069]

[0070] Where, N s S represents the set of capture drones i that share a boundary with the escape drone e. i This represents the shared boundary between the two, and the area of ​​the shared boundary can be expressed as... The centroid of the shared boundary is defined as

[0071] By comparing equations (11) and (13), we can conclude that:

[0072]

[0073] Therefore, the area of ​​the safe priority reachable domain of the decoupled escape drone is affected by the capture drone as follows:

[0074]

[0075] Substituting equation (15) into equation (12), the final encirclement and capture strategy is simplified to:

[0076]

[0077] The "area minimization" collaborative encirclement strategy guides the pursuers to the centroid location where they share a boundary with the escapee. This strategy ultimately captures the escapee by gradually reducing the area of ​​the escapee's safe, priority-accessible zone.

[0078] Furthermore, step 2.2 includes the following specific steps:

[0079] Lemma 2: Consider a mission scenario where a single capture drone and a single escape drone move in region Q. According to the proposed capture strategy (16), for any allowed escape strategy, we have: Furthermore, if and only if the escapee employs the following movement strategies,

[0080]

[0081] Among them, C s It is tracking drone x i and escape drones x e The centroid of the shared boundary of the safety-priority reachable area proves that, for the case of a single tracker and a single escapee, Simplified to:

[0082]

[0083] Substituting the area-minimization-based containment strategy (16) obtained in step 2.1 into the above formula, we get:

[0084]

[0085] Since the shared boundary between the capture drone and the escape drone is represented by the centroid on the perpendicular bisector between them when they move at the same speed, Located on this shared boundary, therefore there are Furthermore, without loss of generality in step 2.1, it is assumed that v i =ve =1, that is Substituting the above conditions, we can see that, for a single tracker... And when At that time, the escapee's control strategy is as follows:

[0086]

[0087] Thus, Lemma 2 is proved. The movement strategy of the pursuers can guarantee that the area of ​​the safe reach of the escapee is strictly non-increasing, and the only escapee strategy that can keep the area of ​​the safe reach of the escapee unchanged is to move in the opposite direction of the centroid of the shared boundary between the two.

[0088] Define the square of the distance between the escapee and the pursuer as:

[0089] D = ||x i -x e || 2 =(x i -x e ) T (x i -x e ) (twenty one)

[0090] Lemma 3: The rate of change of the square of the distance when the pursuer employs a movement strategy (16) satisfy Furthermore, if and only if the escapee employs a control strategy as shown in equation (20),

[0091] Proof: According to the definition of D in equation (21), we have

[0092]

[0093] Substituting the movement strategy (16) of the pursuers into the above formula, we get:

[0094]

[0095] Similarly due to Located on the perpendicular bisector of the capture drone and the escape drone, therefore, The above formula can be further derived as follows:

[0096]

[0097] Without loss of generality, in step 2.1, the present invention assumes v i =v e =1, that is Therefore, there is and The motion strategy of the escape drone is shown in equation (20);

[0098] Thus, Lemma 3 is proved. The movement strategy of the pursuer can guarantee that the distance between it and the escapee is strictly non-increasing. For the escapee, the only escape strategy that can keep the distance constant is to move in the opposite direction of the centroid of the shared boundary between the two.

[0099] Based on Lemmas 2 and 3, we conclude that in the scenario of a single capture drone, when the capture drone adopts the movement strategy shown in Equation (16), the area of ​​the safe priority reachable region of the escapee and the distance of the escapee relative to the capture drone are both strictly non-increasing. Furthermore, the rate of change of both is zero only when the escapee moves in the opposite direction of the centroid of the shared boundary between the two. Otherwise, both are decreasing. Therefore, returning to the multi-drone cooperative capture scenario described in this invention, when the capture drone adopts the capture strategy shown in Equation (16), no matter what movement strategy the escapee adopts, it cannot guarantee that the condition that its distance relative to each adjacent capture drone is not increasing is satisfied simultaneously. Therefore, the distance between the escapee and the capture drone cluster will always be decreasing, and the capture drone cluster will eventually achieve final capture within a finite time.

[0100] Furthermore, step 2.3 includes the following specific steps:

[0101] 1) Initialization

[0102] For each capture drone, initialize the following variables:

[0103] A set of hyperplane parameters used to store securely accessible regions;

[0104] vertices_set: Used to store the set of vertices for each face of a convex polyhedron formed by a safe-priority reachable region;

[0105] shareplane: A flag indicating whether the current capture drone and the target share a boundary;

[0106] 2) Traverse each capture drone

[0107] In a loop, perform the following operations on each drone in the drone swarm: 2-1) Reset the shared boundary flag: Initialize the shareplane flag to True, indicating that it is assumed that the current drone shares a boundary with the target;

[0108] 2-2) Constructing a safe-priority reachable region: The SPRRConstruction algorithm based on obstacle perception is invoked, with the current location x of the capture drone as input. iThe location of the escaping drone (x) e In addition to the obstacle point cloud information Ω_obs in the environment, the set of hyperplane parameters for the safe-priority reachable area of ​​the current capture drone is obtained.

[0109] 2-3) Handling hyperplanes in areas with priority accessibility.

[0110] right Each hyperplane H in the process is processed as follows:

[0111] Calculate the intersection line: Calculate the intersection line between the current hyperplane H and the hyperplane H. The intersection lines of other hyperplanes are stored in lines;

[0112] Calculate and filter intersection points: Calculate the intersection points of each line and only select those located in the safest accessible area. The intersection points within the points are added to the points set;

[0113] Determine whether a valid plane is formed:

[0114] If the number of elements in the points set is less than 2, it means that these intersection points cannot form a valid plane. In this case, the hyperplane H is moved from... It is removed from the list because it does not contribute to the formation of the final convex polyhedron.

[0115] Meanwhile, if the hyperplane H is the current position x of the capture drone i Location x of the escaping drone e The shared boundary hyperplane is defined by setting the shareplane flag to False, indicating that the current capture drone and the target do not share a boundary.

[0116] If the number of elements in the points set is sufficient to form a valid plane, add the points set to the vertices_set.

[0117] 2-4) Determine the target location for the encirclement and capture.

[0118] The target point for capturing the drone is determined based on the value of the shareplane flag.

[0119] If shareplane is True, it means that the current capture drone and the target share a boundary. The centroid of the shared boundary is taken as the target point, i.e., T. i The centroid of the shared boundary;

[0120] If `shareplane` is False, it means that the current capture drone and the target do not share a boundary. The position x of the convex polyhedron corresponding to the escape drone's location will be determined. eThe nearest point is used as the target point for the encirclement, i.e., T. i For a convex polyhedron, x e The nearest point;

[0121] 3) After the above steps, the corresponding target point for each capture drone has been determined, and the final set of capture target points T is obtained. i As the output of the algorithm.

[0122] The multi-UAV cooperative encirclement method for obstacle environments of the present invention has the following advantages:

[0123] (1) The three-dimensional decision space representation method based on the Veno diagram of the present invention can extract the convex bounded decision region (referred to as the safe priority reachable region in the method of the present invention) in a typical non-convex region such as the three-dimensional obstacle space, thereby realizing the effective representation of the decision space and laying the foundation for the generation of subsequent capture strategies.

[0124] (2) The distributed encirclement decision-making method based on area minimization of the present invention can effectively protect escaped drones by reducing the area of ​​the safe priority reachable zone, and the present invention proves the theoretical completeness of this method. In addition, the present invention further realizes the lightweighting of the algorithm by effectively distributing the computational load by adopting a distributed computing method. Attached Figure Description

[0125] Figure 1 This is a schematic diagram of the preferred reachable area of ​​the present invention.

[0126] Figure 2 This is a schematic diagram of the safe priority reachable area of ​​the present invention.

[0127] Figure 3 This is a schematic diagram of the construction of a safe-priority reachable area in the simulation environment of this invention. Detailed Implementation

[0128] To better understand the purpose, structure, and function of this invention, the following detailed description of a multi-UAV cooperative encirclement method in an obstacle environment is provided in conjunction with the accompanying drawings.

[0129] This invention provides a multi-UAV cooperative encirclement method in an obstacle environment, which acquires the spatial coordinate information of multiple UAVs, including defensive and offensive UAVs, based on relevant position capture devices and algorithms.

[0130] Step 1: Representation of the 3D Decision Space Based on Veno Diagrams; This invention addresses bounded convex regions containing obstacles. Such a 3D obstacle space is a typical non-convex environment, making direct modeling and solving for control strategies very challenging. To address this challenge, this invention simplifies the representation of the 3D obstacle decision space by constructing a safety-priority reachable region. This method mainly includes the following steps:

[0131] Step 1.1: Construct the priority reachable region based on the Veno diagram: Based on the location of the capture drone cluster and the location of the escape drone, the priority reachable region of the capture drone in the three-dimensional scene is determined by the Veno partitioning method.

[0132] Although the escapee's movement strategies are inherently dynamic and unpredictable, their capture can still be achieved within a finite timeframe by effectively reducing their Priority Reachable Region (PRR). The escapee's Priority Reachable Region is defined as the area within which the escapee can reach an escapee in a given environment. In the middle, the collection of all locations that can be reached faster than any other hunter.

[0133] A Voronoi diagram, also known as a Thiessen polygon or Dirichlet diagram, is a geometric structure based on the partitioning of space using a set of points. Its core idea is to divide a planar or three-dimensional space into multiple regions, where the distance from every point within a region to the seed point (generated point) of that region is less than the distance to any other seed point. Specifically, given a set of seed points P = {p1, p2, ..., p...}, ... n The Voronoi diagram divides space into n convex polygonal regions, called Voronoi regions or Dirichlet regions. Each Voronoi region V(p i ) by all points away from seed point p i The nearest points constitute a set of points that satisfy the following conditions:

[0134]

[0135] Where Q is the task scenario space, x is any point in the space, and p i It is a seed point, and points within the Vino region are connected to the seed point p. i The distance to the seed point is smaller than the distance to other seed points.

[0136] In a two-dimensional plane, each Veno region of a Veno diagram is a convex polygon, while in three-dimensional space, the region is composed of polyhedra. Through this spatial division, the Veno diagram clearly shows the relative positional relationships between different seed points and their areas of influence. A Voronoi edge in a Veno diagram is a shared boundary between adjacent seed points, defined as the set of all points equidistant from two adjacent seed points. In a three-scene scenario, the shared Veno boundary is represented as the perpendicular bisector of the line connecting the two types of seed points. Mathematically, a point x on the shared Veno boundary... g The following conditions must be met:

[0137]

[0138] The construction of Veno diagram partitions is relatively simple and computationally efficient, making them very useful in many practical applications. In the multi-UAV cooperative manhunt problem described in this invention, the application of Veno diagrams helps to effectively partition the reachable area in the environment based on the positions of the manhunter and the escapee.

[0139] In an ideal, obstacle-free environment, for an agent system following a single integrator dynamics model and possessing equal maximum velocities, the escapee's preferred reach region can be represented by a Vino partition. Specifically, the escapee's preferred reach region and the pursuer's preferred reach region are defined as follows:

[0140]

[0141] in, and These represent the preferred reachable areas for the pursuer and the escapee in an obstacle-free environment. In this environment, each agent (whether pursuer or escapee) can determine its reachable area in space by calculating the Voronoi diagram, thereby optimizing its path planning and control strategy.

[0142] Based on the understanding that Vino partitions can be used to construct preferred reachable regions, this invention further introduces the concept of hyperplanes to achieve a parameterized expression of preferred reachable regions.

[0143] A hyperplane is an important geometric concept in high-dimensional space; it is a plane with one dimension lower than the space itself. Specifically, in d-dimensional space, a hyperplane is a (d-1)-dimensional subset. It can be used to divide the entire space into two half-spaces, making it a very effective geometric partitioning tool. The fundamental characteristic of a hyperplane is that it divides any point in space into two classes: one class lies on one side of the hyperplane, and the other class lies on the other side. In two-dimensional space, a hyperplane appears as a straight line; in three-dimensional space, it is a two-dimensional plane.

[0144] In a three-dimensional environment, a hyperplane can be defined using a normal vector and the coordinate representation of a point on the hyperplane. Assume there is a known point P0(x0, y0, z0) on the hyperplane and a normal vector... The equation of the hyperplane can then be expressed as:

[0145]

[0146] in, Let n be any point in space. The geometric object represented by this equation divides space into two half-spaces: one region satisfying n1(x-x0)+n2(y-y0)+n3(z-z0)>0, and the other region satisfying n1(x-x0)+n2(y-y0)+n3(z-z0)<0.

[0147] According to the hyperplane partitioning rule, the Voronoi partition of each predator can be represented by a set of hyperplanes. (See attached...) Figure 1 As shown, the combination of internal regions of multiple hyperplanes can form a convex region, which represents the preferred reachable area for either the hunter or the escapee. Through this geometric construction, we can efficiently describe and compute the reachable area of ​​each agent in the environment.

[0148] Therefore, the definition of PRR can be transformed from equation (3) into the following form:

[0149]

[0150] Here, X represents the vector from the origin to any point x in Q, and l1 and l2 refer to different intelligent agents, including the escapee and the hunter cluster. and The hyperplane parameters are defined by the positional differences between agents l1 and l2, for example: n ei and p ei The hyperplane parameters are defined by the positional difference between agents e and i. This represents the normal vector of the corresponding hyperplane and the plane position calculated from the midpoint.

[0151] To accurately describe the preferred reachable region for each agent, it is necessary to calculate the hyperplane normal vector and corresponding plane parameters associated with each pair of agents. (Hyperplane normal vector) It is determined by the relative positional differences between the pursuers and the escapees, while the positional parameters of the plane... The calculation is performed using the midpoint between these two points, ensuring that the hyperplane separates the regions according to the definition of Vino partitioning. The specific calculation formula is as follows:

[0152]

[0153] In the above formula, and Let l1 and l2 represent the position vectors of agents l1 and l2, respectively. These are the relative position vectors between them. Planar position parameters The midpoint between these two locations is used to determine the location, thus ensuring that each hyperplane accurately divides the space.

[0154] Step 1.2: Construction of safe priority reachable domain based on obstacle perception: Based on the priority reachable area, considering the safe distance between the capture drones and the minimum distance between the capture drones and obstacles, the safe priority reachable domain is constructed to ensure that these constraints are always satisfied throughout the capture process.

[0155] Unlike existing methods that primarily focus on assumptions about accessible scenarios, this invention extends the application to consider real-world environments that include obstacles. Based on the theoretical framework of priority safe accessibility zones, we further propose an obstacle-aware safety-priority accessibility zone model to address the interaction between drones and obstacles in obstacle-prone environments.

[0156] The separating hyperplane in formula (5) can effectively avoid collisions between multiple quadcopter drones. Therefore, we only need to further avoid collisions between the drone and obstacles during the capture phase to ensure flight safety. To achieve real-time obstacle detection and avoidance, we construct a convex polyhedron to represent the safe flight area of ​​the drone at its current position. This convex polyhedron is composed of multiple separating hyperplanes. Each hyperplane isolates the drone from obstacles, forming a safe and unobstructed flight space. The mathematical representation of this space is as follows:

[0157]

[0158] in, Indicates the safe reach area of ​​the drone. Define the composition The parameter set of the hyperplane. It is important to note that the number of hyperplanes is not fixed, but dynamically changes according to the geometry of the obstacle and the relative position of the UAV. Unlike some studies that assume that the obstacle is a strictly convex region with known vertices, we directly calculate the parameters of the separating hyperplane based on the point cloud map Ω_obs constructed by the quadcopter's visual perception system, without making strict geometric assumptions about the obstacle.

[0159] To ensure real-time decision-making, we employed a KD-Tree data structure to accelerate the obstacle point retrieval process. This method allows us to quickly obtain the nearest obstacle point (o_near) in the point cloud map to the current quadrotor position. Separating hyperplane parameters... The calculation formula is as follows:

[0160]

[0161] Where, r s This is the safety radius, used to take into account the physical dimensions of the drone, and ensures obstacle avoidance during flight by shifting the hyperplane a certain distance away from obstacles. For example... Figure 2 As shown, by combining formula (5) and formula (7), the final representation of the safe priority reachable region can be defined as follows:

[0162]

[0163] This representation method ensures that the quadcopter can effectively avoid obstacles in complex environments, and by dynamically adjusting the number and position of hyperplanes, it achieves adaptability to different obstacle shapes, thus providing a solid theoretical foundation and computational support for real-time obstacle avoidance and flight decision-making.

[0164] Step 1.3: Area Iteration and Update: Based on real-time scene changes (such as the relative position changes of the capture drone swarm and the escape drones), continuously adjust the safe priority reachable area to lay the foundation for high-frequency real-time re-decision.

[0165] As the relative positions of the capture drone swarm and the escape drones change in the environment, their safe priority reachable areas also change constantly. Therefore, it is necessary to recalculate the safe priority reachable areas based on the real-time changing position information. The specific calculation process is shown in Algorithm 1:

[0166] 1) Input information: Current location of the capture drone (x) i , i∈Γ, the current position of the target escaping drone x e Obstacle point cloud information in the environment Ω_obs

[0167] 2) Initialize the remaining obstacle point cloud information Ω_remain = Ω_obs. Assign the obstacle point cloud information Ω_obs in the environment to the remaining obstacle point cloud information Ω_remain as the basis for subsequent processing.

[0168] 3) Constructing a separating hyperplane based on obstacle point clouds

[0169] Enter a loop, and as long as the number of points in the remaining obstacle point cloud information is greater than a preset threshold, perform the following operations:

[0170] 3-1) Find the nearest obstacle point: Use the KD-tree algorithm to find the nearest obstacle point relative to the current position x of the drone being captured. i The nearest obstacle point is o_near.

[0171] 3-2) Constructing the obstacle separation hyperplane: Calculating parameters These parameters are used to construct an obstacle separation hyperplane, and the relevant information of the hyperplane is stored in the safe-priority reachable region. middle.

[0172] 3-3) Update the remaining obstacle point cloud information: Remove points located outside the newly constructed hyperplane from the remaining obstacle point cloud information to prepare for the next loop.

[0173] 4) Dealing with other drones used for surveillance

[0174] By iterating through all the capture drones (except the current i-th capture drone), i.e., j ranges from 0 to N and j is not equal to i:

[0175] 4-1) Set l1 to the number of the current capture drone and l2 to the number of the capture drone currently being traversed.

[0176] 4-2) Calculate parameters according to specific formulas <n ij P ij > and store these parameters in the security-priority reachable zone. middle.

[0177] 5) Dealing with escaped drones

[0178] Set l1 to the number of the currently captured drone, and l2 to the number of the escaped target drone. Calculate parameters according to a specific formula. <n ie p ie > and store these parameters in the security-priority reachable zone. middle.

[0179] 6) After the above steps, the safest priority reachable area is finally obtained. As the output of the algorithm.

[0180] Steps 1.2 and 1.3 are completed within milliseconds, successfully constructing a safe priority reachable area centered on each capture drone and escape drone, and then further distributed decision-making is carried out through step 2.

[0181] Step 2: Construct a distributed capture decision based on area minimization; based on the safe-priority reachable area constructed by the three-dimensional decision space representation module based on the Venn diagram, a cooperative capture strategy based on area minimization is adopted to compress the movement area of ​​the escaping UAV and ultimately achieve effective capture. This method mainly includes the following calculation process:

[0182] Step 2.1: Construct a collaborative capture strategy based on area minimization: Based on the safe priority reachable domain constructed in Step 1.2, the safe priority reachable area of ​​the escaping drone is reduced to achieve effective capture of the escaping drone.

[0183] In step 1, the safe-priority reachable area of ​​the escaping drone is defined as the set of spatial points that the escaping drone can reach before any of its pursuing drones. Therefore, the area of ​​the escapee's safe-priority reachable area is... Defined as:

[0184]

[0185] in It is the safe priority reachable area for the escapee constructed in formula (9), according to the definition of formula (10), The derivative is:

[0186]

[0187] From the above formula, we can see that The movement strategy of escape drones and the movement strategy of capturing drones It's a joint decision. The escape drone's movement strategy is unknown to the capture drone, but the capture drone can adjust its own movement strategy, gradually reducing the safe priority area accessible to the evader over time. Ultimately, capture is achieved. For further analysis, in equation (11), the capture of the drone... The influence term can be decoupled as Through this decoupling, the predator can employ a "minimum area" movement strategy, which is consistent with... The gradient descent direction is consistent, thus achieving a rapid reduction in the area of ​​the safe-accessible zone for the escapee. The movement strategy of the pursuers is given by the following equation:

[0188]

[0189] The above equation describes the encirclement and capture strategy for the drone. Next, this invention will derive the specific calculation method for this strategy. Using Leibniz's integral rule, It can be simplified to:

[0190]

[0191] Where, N s This represents the set of capture drones i that share a boundary with the escape drone e. i This represents the shared boundary between the two, and the area of ​​the shared boundary can be expressed as... The centroid of the shared boundary is defined as

[0192] By comparing equations (11) and (13), we can conclude that

[0193]

[0194] Therefore, the area of ​​the safe priority reachable domain of the decoupled escape drone is affected by the capture drone as follows:

[0195]

[0196] Substituting equation (15) into equation (12), the final encirclement and capture strategy is simplified to:

[0197]

[0198] From a physics perspective, the "area minimization" collaborative capture strategy proposed in this invention guides the capturer to the centroid location where it shares a boundary with the escapee. The core of this strategy lies in gradually reducing the area of ​​the escapee's safe, priority-accessible zone, ultimately achieving the capture of the escapee.

[0199] Step 2.2: Prove the effectiveness of the encirclement strategy: Prove the theoretical completeness of the cooperative encirclement strategy based on area minimization proposed in Step 2.1 through theoretical analysis;

[0200] Motion strategies in the capture of drones aim to reduce the area of ​​escapees. To demonstrate that the strategy guarantees capture, we demonstrate the safe reach area of ​​the escaping drone in a single capture drone scenario. The boundary conditions are always strictly non-increasing, and the distance between the capture drone and the escaping drone is also always strictly non-increasing. When extended to scenarios with multiple capture drone swarms, these dynamic boundary conditions will lead to capture within a finite time.

[0201] Lemma 2: Consider a mission scenario where a single capture drone and a single escape drone move in region Q. According to the proposed capture strategy (16), for any allowed escape strategy, we have... Furthermore, if and only if the escapee employs the following movement strategies,

[0202]

[0203] Among them, C s It is tracking drone x i and escape drones x e Security is the priority to reach the centroid of the shared boundary of the region.

[0204] Proof: For the case of a single tracker and a single escapee, Simplified to:

[0205]

[0206] Substituting the area-minimization-based containment strategy (16) obtained in step 2.1 into the above formula, we get:

[0207]

[0208] Since the shared boundary between the capture drone and the escape drone is represented by the centroid on the perpendicular bisector between them when they move at the same speed, Located on this shared boundary, therefore there are Furthermore, without loss of generality in step 2.1, the present invention assumes v i =v e =1, that is Substituting the above conditions, we can see that, for a single tracker... And when At that time, the escapee's control strategy is as follows:

[0209]

[0210] Lemma 2 is thus proved: the movement strategy of the pursuers can guarantee that the area of ​​the escapee's safe reach is strictly non-increasing, and the only escapee strategy that can keep the area of ​​the escapee's safe reach constant is to move in the opposite direction of the centroid of the shared boundary between the two.

[0211] Define the square of the distance between the escapee and the pursuer as:

[0212] D = ||x i -x e || 2 =(x i -x e ) T (x i -x e ) (twenty one)

[0213] Lemma 3: The rate of change of the square of the distance when the pursuer employs a movement strategy (16) satisfy Furthermore, if and only if the escapee employs a control strategy as shown in equation (20),

[0214] Proof: According to the definition of D in equation (21), we have

[0215]

[0216] Substituting the movement strategy (16) of the pursuers into the above formula, we get:

[0217]

[0218] Similarly due to Located on the perpendicular bisector of the capture drone and the escape drone, therefore, The above formula can be further derived as follows:

[0219]

[0220] Since the capture drone described in this invention does not have a speed advantage, without loss of generality in step 2.1, this invention assumes v i =v e =1, that is Therefore, there is and The motion strategy of the escape drone is shown in equation (20).

[0221] Lemma 3 is thus proved: the movement strategy of the pursuer ensures that the distance between it and the escapee is strictly non-increasing. For the escapee, the only escape strategy that can keep the distance constant is to move in the opposite direction of the centroid of the shared boundary between the two.

[0222] Based on Lemmas 2 and 3, this invention concludes that in a single capture drone scenario, when the capture drone adopts the movement strategy shown in Equation (16), both the safe priority reach area of ​​the escapee and the distance between the escapee and the capture drone are strictly non-increasing. Furthermore, their rates of change are zero only when the escapee moves in the opposite direction of the centroid of their shared boundary; otherwise, they are decreasing. Therefore, returning to the multi-drone cooperative capture scenario described in this invention, when the capture drone adopts the capture strategy shown in Equation (16), regardless of the escapee's movement strategy, the condition that its distance relative to each adjacent capture drone is not increasing cannot be guaranteed simultaneously. Therefore, the distance between the escapee and the capture drone swarm will always be decreasing, and ultimately, the capture drone swarm will achieve final capture within a finite time.

[0223] Step 2.3: Distributed collaborative trapping decision framework: Lightweighting of the algorithm by distributing the computational load.

[0224] In the multi-UAV cooperative encirclement decision-making method described in this invention, the encirclement UAVs rely on environmental information, the positions of other encirclement UAVs in the cluster, and the position of the escaped target UAV as input information to complete the encirclement decision-making process. This invention employs a distributed encirclement decision-making framework. After acquiring the input information, each encirclement UAV constructs only a safe-priority reachable region centered on itself. Based on this, the centroid of the shared boundary with the escaped UAV is calculated as the encirclement target point. Furthermore, considering that in the multi-UAV cooperative encirclement process, there may be situations where the safe-priority reachable regions of the current encirclement UAV and the escaped UAV do not share a boundary, the safe-priority reachable region of the current encirclement UAV degenerates into a safe-reachable area. A greedy method is used to calculate the closest point to the escaped target UAV in such a convex space as the target point, thereby maximizing the proximity to the escaped UAV.

[0225] The distributed collaborative encirclement decision-making algorithm framework of this invention is shown in Algorithm 2.

[0226] 1) Initialization

[0227] For each capture drone, initialize the following variables:

[0228] Used to store the set of hyperplane parameters for securely accessible regions.

[0229] vertices_set: Used to store the set of vertices for each face of a convex polyhedron formed by a safe-priority reachable region.

[0230] shareplane: A flag indicating whether the current capture drone and the target share a boundary.

[0231] 2) Traverse each capture drone

[0232] Perform the following operations on each drone in the drone swarm for the capture operation, using a loop:

[0233] 2-1) Reset the shared boundary flag: Initialize the shareplane flag to True, indicating that the current capture drone and the target share a boundary.

[0234] 2-2) Constructing a safe-priority reachable region: The SPRRConstruction algorithm based on obstacle perception is invoked, with the current location x of the capture drone as input. i The location of the escaping drone (x) e In addition to the obstacle point cloud information Ω_obs in the environment, the set of hyperplane parameters for the safe-priority reachable area of ​​the current capture drone is obtained.

[0235] 2-3) Handling hyperplanes in areas with priority accessibility.

[0236] right Each hyperplane H in the process is processed as follows:

[0237] Calculate the intersection line: Calculate the intersection line between the current hyperplane H and the hyperplane H. Find the intersection lines of other hyperplanes in the hyperplane and store these intersection lines in lines.

[0238] Calculate and filter intersection points: Calculate the intersection points of each line and only select those located in the safest accessible area. The intersection points within the points are added to the points set.

[0239] Determine whether a valid plane is formed:

[0240] If the number of elements in the points set is less than 2, it means that these intersection points cannot form a valid plane. In this case, the hyperplane H is moved from... It is removed because it does not contribute to the formation of the final convex polyhedron.

[0241] Meanwhile, if the hyperplane H is the current position x of the capture drone i Location x of the escaping drone e The shared boundary hyperplane is defined by setting the shareplane flag to False, indicating that the current capture drone and the target do not share a boundary.

[0242] If the number of elements in the points set is sufficient (i.e., enough to form a valid plane), add the points set to the vertices_set.

[0243] 2-4) Determine the target location for the encirclement and capture.

[0244] The target point for capturing the drone is determined based on the value of the shareplane flag.

[0245] If shareplane is True, it means that the current capture drone and the target share a boundary. The centroid of the shared boundary is taken as the target point, i.e., T. i The centroid of the shared boundary.

[0246] If `shareplane` is False, it means that the current capture drone and the target do not share a boundary. The position x of the convex polyhedron corresponding to the escape drone's location will be determined. e The nearest point is designated as the target point for the encirclement, i.e., T. i For a convex polyhedron, x e The nearest point.

[0247] 3) After the above steps, the corresponding target point for each capture drone has been determined, and the final set of capture target points T is obtained. i As the output of the algorithm.

[0248] Compared to centralized capture decision-making methods that construct all safe and prioritized reachable regions and calculate strategies on a single UAV before issuing them to other UAVs for execution, the distributed decision-making framework proposed in this invention distributes the computational load and reduces decision-making time. The safe and prioritized reachable regions constructed by the method of this invention in the simulation environment are shown in the attached figure. Figure 3 As shown, the area formed by blue squares represents the obstacle area Q. o The red spherical quadcopter represents the pursuer, the blue spherical quadcopter represents the escapee, and the colored convex areas represent the safe priority reach area (SPRR) for each quadcopter. and

[0249] Steps 2.1 to 2.3, immediately following the representation of the three-dimensional decision space based on the Venn diagram, are completed in polynomial time. The entire decision-making process can meet the requirement of a decision frequency of 20 Hz, satisfying the real-time requirements for deploying this method on an UAV.

[0250] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1.A method for cooperative hunting of multiple unmanned aerial vehicles (UAVs) in an obstacle environment, wherein spatial coordinate information of multiple UAVs including a defending UAV and an attacking UAV is obtained according to a relevant position capturing device and an algorithm, and the method is characterized in that, Comprising the following steps: Step 1: Three-dimensional decision space representation based on Voronoi diagram: simplified representation of three-dimensional obstacle decision space is realized by constructing safe priority reachable region; Step 1.1: Construction of priority reachable region based on Voronoi diagram: depending on the position of the trapping UAV cluster and the position of the escaping UAV, the priority reachable region of the trapping UAV in the three-dimensional space is determined by the Voronoi partition method; The priority reachable region of an escaper is defined as the set of all locations in a given environment that the escaper can reach faster than any other pursuer; ; Given a set of seed points P = {p1, p2, …, p n n}, a Voronoi diagram divides the space into n convex polygonal regions, called Voronoi regions, or Dirichlet regions, each Voronoi region V(p i ) is composed of all points closest to the seed point p i , satisfying the following conditions: wherein Q is a task scene space, x is an arbitrary point in the space, p i is a seed point, and points within the Veno region have a smaller distance to the seed point p i than to other seed points. In two-dimensional plane, each of the Venn region in the Venn diagram is a convex polygon, while in three-dimensional space, the region is composed of polyhedron. The Venn edge in the Venn diagram is the shared boundary between adjacent seed points, which is defined as the set of all points with equal distance to two adjacent seed points. In three-dimensional space, the shared Venn boundary is represented as the perpendicular bisector plane of the line connecting the two seed points. Mathematically, the point x on the shared Venn boundary satisfies the following condition: g satisfies the following condition: The priority reachable region of the escapee and the priority reachable region of the trapper are defined as: wherein, and respectively represent the priority reachable regions of the pursuer and the evader in an obstacle-free environment, in which each agent can determine its reachable region in space by calculating the Voronoi diagram, thereby optimizing its path planning and control strategy; The priority reachable region of the escapee and the priority reachable region of the trapper are defined as: A hyperplane in a three-dimensional environment can be defined by a normal vector and the coordinates of a point on the hyperplane, given a known point P0(x0, y0, z0) on the hyperplane and a normal vector The equation of the hyperplane is then given by: wherein is an arbitrary point in space, this equation represents a geometric object that divides the space into two half-spaces: one that satisfies n1(x-x0)+n2(y-y0)+n3(z-z0)>0 and the other that satisfies n1(x-x0)+n2(y-y0)+n3(z-z0)<0; The priority reachable region of the escapee and the priority reachable region of the trapper are defined as: According to the hyperplane separation rule, each trapper's Voronoi partition can be represented by a set of hyperplanes, and the combination of the inner regions of multiple hyperplanes can form a convex region, which represents the priority reachable region of the trapper or the escapee, Here, X denotes a vector from the origin to an arbitrary point x in Q, and l1 and l2 denote different intelligent agents including an escapee and a pursuer cluster, and is a hyperplane parameter defined by the positional difference between intelligent agents l1 and l2, representing a normal vector of the corresponding hyperplane and a plane position calculated through a midpoint; The hyperplane normal vector associated with each pair of agents and the corresponding plane parameter are calculated, the hyperplane normal vector is determined by the relative position difference between the pursuer or the evader, while the plane position parameter is calculated by the midpoint between the two, ensuring that the hyperplane separates the regions according to the definition of the Vornoi partition, the specific calculation formula is as follows: In the above formulae, and denote the position vectors of the agents li and l2, respectively, while is the relative position vector between them, and the planar position parameter is determined by the midpoint of these two positions, thus ensuring that each hyperplane accurately partitions the space; Thus, the definition of the priority reachable region PRR can be converted from equation (3) to the following form: Step 1.2: Construction of safe priority reachable region based on obstacle perception: based on the priority reachable region, the safety distance between the trapping UAVs and the minimum distance between the trapping UAVs and the obstacles are considered, and the safe priority reachable region is constructed to ensure that these constraints are always met during the entire capture process; wherein, represents a safe reachable region of the UAV, defines a parameter set of hyperplanes constituting a hyperplane. KD-Tree data structure is used to accelerate the searching process of obstacle points, and the formula of separating hyperplane parameter is as follows: The formula is as follows: where r s is the safety radius, used to consider the physical size of the UAV and to ensure the obstacle avoidance effect in flight by offsetting the hyperplane in the direction away from the obstacle by a certain distance, and by combining formula (5) with formula (7), the final safety priority reachable region is defined as follows: Step 1.3: Region iteration and update: According to the real-time scene changes, the safety priority reachable region is constantly adjusted, which lays the foundation for high-frequency real-time redetermination. A convex polyhedron is constructed to represent the safe flight region under the current position of the UAV, which is composed of multiple separate hyperplanes. Each hyperplane separates the UAV from the obstacles to form a safe and obstacle-free flight space, which is mathematically represented as follows: 1) input information: current position x of the trapping drone i , current position x of the target escape drone e , obstacle point cloud information Ω_obs in the environment; According to the real-time changing position information, the safe priority reachable region is recalculated, 2) Initialize the remaining obstacle point cloud information Ω_remain = Ω_obs; Assign the obstacle point cloud information Ω_obs in the environment to the remaining obstacle point cloud information Ω_remain; 3) Construct a separate hyperplane based on the obstacle point cloud; 3-1) Find the nearest obstacle point: use KD-tree algorithm to find the nearest obstacle point o_near to the current position x of the hunting drone i nearest obstacle point o_near; 3-2) Constructing the obstacle separating hyperplane: computation of parameters Using these parameters, the obstacle separating hyperplane is constructed and its relevant information is stored in the safety-priority reachable region ; Enter a loop, as long as the number of points in the remaining obstacle point cloud information is greater than a preset threshold, execute the following operations: 3-3) Update the remaining obstacle point cloud information: remove the points outside the just constructed hyperplane from the remaining obstacle point cloud information, and prepare for the next loop; 4) Process other trapping UAVs Traverse all trapping UAVs through a loop, i.e. j from 0 to N and j not equal to i: 4-2) Calculate parameters <n according to specific formula ij ,p ij > and store them in the secure-priority reachable area . 4-1) Set l1 as the number of the current trapping UAV, and l2 as the number of the currently traversed trapping UAV; Set l1 as the number of the current surrounding drone, set l2 as the number of the target escaping drone, calculate parameters <n ie ,p ie > according to specific formula, and store these parameters in the safety priority reachable area . 6) After the above steps, the final safety-priority reachable region is obtained As an output of the algorithm; 5) Process the target escaping UAV Step 2: Construction of distributed trapping decision based on area minimization: based on the safe priority reachable region constructed by the three-dimensional decision space representation module based on Voronoi diagram, the motion area of the escaping UAV is compressed through the area minimization based cooperative trapping strategy, and finally the effective capture is realized; Area of the safe-priority reachable region of the escapee is defined as: where is the safety-priority reachable region of the escapee constructed in formula (9), defined according to formula (10), The derivative of is: The influence term of the UAV pair in Equation (11) can be decoupled as By this decoupling, the pursuer adopts an "area minimization" motion strategy, which is consistent with the gradient descent direction of , achieving a rapid reduction of the area of the safe-priority reachable region of the evader. The motion strategy of the pursuer is given by​ The specific calculation method of the above strategy is derived by using Leibniz's rule of integration, which can be simplified as: where N s represents a set of pursuit drones i that share a boundary with an escape drone e, s i represents the shared boundary between the two, the area of the shared boundary can be represented as while the centroid of the shared boundary is defined as Step 2.1: Construction of area minimization based cooperative trapping strategy: based on the safe priority reachable region constructed in step 1.2, the effective trapping of the escaping UAV is realized by reducing the safe priority reachable region of the escaping UAV; By comparing equation (11) and equation (13), we get: Therefore, the area of the safety-priority reachable region of the escaping UAV decoupled from the surrounding UAVs is affected as follows: Substituting equation (15) into equation (12), the final pursuit motion strategy is simplified as: The "area minimization" cooperative hunting strategy guides the surrounding UAVs to the centroid position of the shared boundary with the escaping UAV; this strategy eventually achieves the capture of the escaping UAV by gradually reducing the area of the safety-priority reachable region of the escaping UAV; Step 2.2: Prove the effectiveness of the hunting strategy: theoretically analyze the theoretical completeness of the area minimization-based cooperative hunting strategy proposed in step 2.1; Lemma 2: Consider the task scenario of a single pursuer drone and a single evader drone moving in the region Q, according to the proposed pursuit strategy (16), for any admissible evader strategy, there exists Moreover, the evader is guaranteed to be captured if and only if the evader adopts the following motion strategy, where C s is the centroid of the shared boundary of the safe-priority reachable region of the pursuer drone x i and the escape drone x e It is shown that for the case of a single pursuer and a single escapee, simplifies to: Substituting the area-minimizing pursuit strategy (16) obtained from the analysis in Step 2.1 into the above equation, we obtain: Since the shared boundary between the pursuer drone and the evader drone represents the perpendicular bisector between the two when moving at the same speed, the center of mass lies on this shared boundary, therefore has And without loss of generality, assume v i = v e = 1, i.e. Substituting the above conditions, for the case of a single pursuer, and when the control strategy for the evader is: Thus, Lemma 2 is proved. The predator's movement strategy can ensure that the area of the safe reachable region of the evader is strictly non-decreasing, and the only strategy that can keep the area of the safe reachable region of the evader unchanged is to move in the opposite direction of the centroid of the shared boundary. Define the square of the distance between the escaping UAV and the pursuer as: D = || x i - x e || x 2 = (x i - x e ) T (x i - x e ) (21) Lemma 3: When the pursuer adopts the motion strategy (16), the rate of change of the squared distance satisfies Moreover, the escapee adopts the control strategy as shown in equation (20) if and only if Proof: According to the definition of D in equation (21), we have Substituting the predator's strategy (16) into the above equation gives: Also due to Located on the perpendicular bisector of the trapping drone and the escaping drone, therefore has The above formula is further derived as: Without loss of generality, in step 2.1, the present application assumes that v i = v e = 1, i.e. Thus, there is and The motion strategy of the escaping UAV when t = 1 is shown in equation (20); Thus, Lemma 3 is proved, and the movement strategy of the surrounding UAVs can ensure that the distance between the surrounding UAVs and the escaping UAV is strictly non-decreasing. The only escape strategy that can keep the distance unchanged for the escaping UAV is to move in the opposite direction of the centroid of the shared boundary between the two UAVs; According to Lemma 2 and Lemma 3, it is concluded that in the scenario of a single surrounding UAV, when the surrounding UAV adopts the movement strategy as shown in equation (16), the area of the safety-priority reachable region of the escaping UAV and the distance between the escaping UAV and the surrounding UAV are strictly non-decreasing, and the change rate of both is zero only when the escaping UAV moves in the opposite direction of the centroid of the shared boundary between the two UAVs, and in other cases, both are decreasing. Therefore, returning to the multi-UAV cooperative hunting scenario described in the present application, when the surrounding UAVs adopt the hunting strategy as described in equation (16), the escaping UAV cannot guarantee that the condition of non-increasing distance relative to each adjacent surrounding UAV is met simultaneously, regardless of the movement strategy adopted by the escaping UAV. Therefore, the distance between the escaping UAV and the surrounding UAV cluster will always be in a decreasing state, and eventually the surrounding UAV cluster will achieve final capture in a finite time; Step 2.3: Distributed cooperative hunting decision-making framework: achieve lightweight algorithm by distributing computing load; 1) Initialization For each surrounding UAV, initialize the following variables: a hyperplane parameter set for storing a security prioritized reachable region; vertices_set: a set of vertices of the convex polyhedron formed by the safety-priority reachable region; shareplane: a flag indicating whether the current surrounding UAV shares a boundary with the target; 2) Traverse each surrounding UAV Through a loop, perform the following operations for each UAV in the surrounding UAV cluster: 2-1) Reset the shared boundary flag: initially set the shareplane flag to True, indicating that the current surrounding UAV is assumed to share a boundary with the target; 2-2) Construction of safety-priority reachable region: call the safety-priority reachable region construction (SPRRConstruction) algorithm based on obstacle perception, input the position x of the current pursuit UAV i , the position x of the target escape UAV e , and the obstacle point cloud information Ω_obs in the environment, to obtain the hyperplane parameter set of the safety-priority reachable region of the current pursuit UAV 2-3) Process the hyperplane of the safety-priority reachable region Each hyperplane H in is processed as follows:​ Compute intersection lines: Compute the intersection lines of the current hyperplane H with all other hyperplanes in the center, and store these intersection lines in lines; Calculate intersections and filter: Calculate the intersection of each line and only add the intersection points to the points set that are inside the safety-priority reachable region . Determine whether an effective plane is formed: If the number of elements in the points set is less than 2, it means that these intersection points cannot form a valid plane, in which case the hyperplane H is rejected from since it does not contribute to the formation of the final convex polyhedron; at the same time, if the hyperplane H is the shared boundary hyperplane of the current predator drone position x i and the target escape drone position x e , the shareplane flag is set to False, indicating that the current predator drone does not share a boundary with the target. If the number of elements in the points set is sufficient, an effective plane can be formed, and the points set is added to the vertices_set; 2-4) Determine the hunting target point Determine the hunting target point of the current surrounding UAV according to the value of the shareplane flag: If shareplane is True, it means that the current hunting UAV and the target exist a shared boundary, and the centroid of the shared boundary is taken as the hunting target point, that is, T i is the centroid of the shared boundary. If shareplane is False, indicating that the current drone does not share a boundary with the target, the closest point in the convex polyhedron to the target drone position x e The most recent point as the target point, i.e., T i is the closest point in the convex polyhedron to x e ; 3) After the above steps, the corresponding trapping target point of each trapping UAV is determined, and finally the trapping target point set T is obtained i As the output of the algorithm.

Images