Multi-unmanned aerial vehicle cooperative path planning and obstacle avoidance method and device, and electronic equipment

Abstract

The invention discloses a multi-unmanned aerial vehicle cooperative path planning and obstacle avoidance method and device and electronic equipment. The method comprises the following steps: determining path length cost, threat cost and collision cost based on a multi-unmanned aerial vehicle path planning model comprising a flight constraint model, an environment constraint model and a dynamic threat source quantification model; determining an initial path by adopting a bidirectional greedy search algorithm, and constructing an elliptical sampling space; performing path optimization on the initial path in an elliptical sampling space by using an improved particle swarm algorithm to obtain a global optimization path of each unmanned aerial vehicle; processing the environment image of each unmanned aerial vehicle by using a depth extraction model to obtain a depth map of the unmanned aerial vehicle; converting the depth map into threat source distribution data; the angular velocity of the unmanned aerial vehicle is calculated based on the current orientation, the target orientation and the threat source distribution data of each unmanned aerial vehicle, the unmanned aerial vehicle is controlled to perform obstacle avoidance operation according to the angular velocity, and the efficient, reliable and flexible multi-unmanned aerial vehicle cooperative path planning and obstacle avoidance method is realized.

Description

Technical Field

[0001] The present invention belongs to the field of drone control technology, and specifically relates to a multi-drone collaborative path planning and obstacle avoidance method, device and electronic equipment. Background Art

[0002] In modern disaster relief work, traditional disaster relief models have exposed many limitations. Whether it is natural disasters such as earthquakes, floods, and mudslides, or man-made disasters such as chemical leaks and building collapses, they all pose huge challenges to rescue work.

[0003] After the earthquake, large swaths of urban buildings collapsed, roads were blocked amidst the rubble, and a large number of people were trapped. Traditional ground-based rescue forces struggled to quickly reach the core disaster area, and the risk of aftershocks posed a threat to the lives of rescue workers. In such a complex environment, accurately locating trapped people was extremely difficult, resulting in inefficient rescue efforts. When floods struck, large areas were inundated, and the currents were turbulent. Traditional rescue vessels struggled to navigate the complex waters safely and quickly, unable to reach affected people in a timely manner. The transportation of supplies also faced significant challenges. Debris flows, resulting in complex and unpredictable mountainous terrain, destroyed roads, disrupted communications, and made it difficult for rescue personnel and equipment to reach the affected population, making it difficult to accurately assess the situation.

[0004] Drone swarming technology offers new hope for disaster relief efforts. Drones, with their flexibility and maneuverability, can swiftly navigate complex disaster environments. Over earthquake debris, drones can easily fly over collapsed buildings, conduct a comprehensive survey of the rubble, quickly locate trapped people, and promptly transmit this information to rescuers. In flood-stricken areas, drones can swiftly reach affected people, unaffected by turbulent waters, and provide essential supplies and information. In mudslide-affected areas, drones can quickly capture topographical changes, helping rescuers plan safe routes.

[0005] However, the efficient execution of drones in disaster relief missions requires precise path planning. Actual disaster relief scenarios present complex and ever-changing environments, with numerous static and dynamic obstacles. Classic path planning algorithms are severely inadequate in handling these complexities. Graph-search-based algorithms, such as the A* algorithm and the Dijkstra algorithm, rely on relatively static map information. Once new obstacles appear at the disaster site, such as building debris from an earthquake or debris swept in by floodwaters, the original search map becomes invalid and needs to be rebuilt, which is unacceptable in time-sensitive disaster relief scenarios. While intelligent optimization algorithms, such as genetic algorithms and particle swarm optimization, possess certain global search capabilities, they often fail to plan safe and effective drone paths in a timely manner due to the complex iterative computations and slow convergence when dealing with unexpected obstacles, resulting in missed rescue opportunities.

[0006] Therefore, how to provide an efficient, reliable and flexible multi-UAV collaborative path planning and obstacle avoidance method for complex and dynamic environments has become an important issue. Summary of the Invention

[0007] In order to solve the above-mentioned problems existing in the prior art, the present invention provides a multi-UAV collaborative path planning and obstacle avoidance method, device and electronic equipment.

[0008] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0009] In a first aspect, the present invention provides a multi-UAV collaborative path planning and obstacle avoidance method, the multi-UAV collaborative path planning and obstacle avoidance method comprising:

[0010] Constructing a multi-UAV path planning model; the multi-UAV path planning model includes a flight constraint model, an environmental constraint model, and a dynamic threat source quantification model;

[0011] determining path length cost, threat cost, and collision cost based on the multi-UAV path planning model;

[0012] A bidirectional greedy search algorithm is used to determine an initial path, and an elliptical sampling space is constructed based on the initial path;

[0013] An improved particle swarm algorithm is used to optimize the initial path in the elliptical sampling space to obtain a global optimized path for each UAV; an initial population in the improved particle swarm algorithm is constructed based on the initial path, and a fitness function is constructed based on the path length cost, the threat cost, and the collision cost;

[0014] The environment image of each UAV is processed using a pre-trained depth extraction model to obtain a depth map containing the depth information of the UAV's surrounding environment; the environment image of each UAV is collected in real time while the UAV is flying along the UAV's globally optimized path;

[0015] Convert the depth map of each drone into threat source distribution data;

[0016] The angular velocity of each drone is calculated based on its current position, target position, and threat source distribution data, so as to control the drone to perform obstacle avoidance operations according to the angular velocity.

[0017] Optionally, the dynamic threat source quantification model includes:

[0018]

[0019] Among them, H(O h) represents the dynamic threat source quantification model of the h-th threat source; represents the mth state of the hth threat source; represents the probability of the mth state of the hth threat source appearing; M represents the total number of states of the hth threat source.

[0020] Optionally, determining a path length cost, a threat cost, and a collision cost based on the multi-UAV path planning model includes:

[0021] Determining a path length cost based on the multi-UAV path planning model and the fuel consumption cost and time consumption cost of each UAV;

[0022] Determining a threat cost based on the multi-UAV path planning model and the threat sources of each UAV's route through the threat area;

[0023] The collision cost is determined based on the multi-UAV path planning model and the safety distance between each UAV.

[0024] Optionally, the angular velocity of each UAV is calculated based on the current position, target position, and threat source distribution data of the UAV, including:

[0025] Calculate the obstacle avoidance behavior angular velocity and guidance behavior angular velocity of each drone based on its current position, target position, and threat source distribution data;

[0026] The angular velocity of each UAV is obtained by taking the weighted sum of the angular velocity of the obstacle avoidance behavior and the angular velocity of the guidance behavior.

[0027] Optionally, calculating the obstacle avoidance behavior angular velocity and the guidance behavior angular velocity of each UAV based on the current position, target position, and threat source distribution data of the UAV includes:

[0028] Using the obstacle avoidance behavior function, the obstacle avoidance behavior angular velocity of each UAV is calculated based on its current position, target position, and threat source distribution data.

[0029] The guidance behavior function is used to calculate the guidance behavior angular velocity of each UAV based on its current position and target position.

[0030] Optionally, the angular velocity of each UAV is obtained by performing a weighted summation of the angular velocity of the obstacle avoidance behavior and the angular velocity of the guidance behavior, including:

[0031]

[0032]

[0033] in, represents the angular velocity of the nth UAV; γ(E) represents the adjustment factor; E represents the regional entropy value; f c (φ n ) represents the angular velocity of the obstacle avoidance behavior of the n-th UAV or the angular velocity of the guidance behavior of the n-th UAV, c represents the obstacle avoidance behavior or guidance behavior of the n-th UAV; λ represents the entropy sensitivity coefficient.

[0034] In a second aspect, the present invention provides a multi-UAV collaborative path planning and obstacle avoidance device, the multi-UAV collaborative path planning and obstacle avoidance device comprising:

[0035] The first building module is used to build a multi-UAV path planning model; the multi-UAV path planning model includes a flight constraint model, an environmental constraint model and a dynamic threat source quantification model;

[0036] a determination module, configured to determine a path length cost, a threat cost, and a collision cost based on the multi-UAV path planning model;

[0037] A second construction module is used to determine an initial path using a bidirectional greedy search algorithm and construct an elliptical sampling space based on the initial path;

[0038] an optimization module, configured to optimize the initial path in the elliptical sampling space using an improved particle swarm algorithm to obtain a global optimized path for each UAV; an initial population in the improved particle swarm algorithm is constructed based on the initial path, and a fitness function is constructed based on the path length cost, the threat cost, and the collision cost;

[0039] A processing module is used to process the environment image of each UAV using a pre-trained depth extraction model to obtain a depth map containing the depth information of the environment around the UAV; the environment image of each UAV is collected in real time while the UAV is flying along the global optimization path of the UAV;

[0040] A conversion module, used to convert the depth map of each drone into threat source distribution data;

[0041] The calculation module is used to calculate the angular velocity of each drone based on its current position, target position and threat source distribution data, so as to control the drone to perform obstacle avoidance operations according to the angular velocity.

[0042] In a third aspect, the present invention provides an electronic device comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0043] Memory for storing computer programs;

[0044] The processor is used to implement the method steps described in any of the above-mentioned multi-UAV collaborative path planning and obstacle avoidance methods when executing the computer program stored in the memory.

[0045] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method steps described in any of the above-mentioned multi-UAV collaborative path planning and obstacle avoidance methods are implemented.

[0046] The present invention provides a multi-UAV collaborative path planning and obstacle avoidance method. By using a dynamic threat source quantification model within the multi-UAV path planning model, it can quantitatively analyze technical uncertainty in dynamic environments, avoiding decision redundancy or conflict in collaborative strategies in dynamic and complex scenarios. This improves the accuracy and dynamic adaptability of subsequent path optimization based on the multi-UAV path planning model. By employing a bidirectional greedy search algorithm to determine the initial path, the blindness of unilateral exploration in existing path planning methods is reduced, shortening the time it takes to discover the initial path. Furthermore, by constructing an elliptical sampling space based on the initial path and optimizing the initial path within the elliptical sampling space to obtain a globally optimized path, the search range for path optimization is reduced, the convergence of the improved particle swarm algorithm is accelerated, and computational complexity is effectively reduced, thereby achieving an efficient, reliable, and flexible multi-UAV collaborative path planning and obstacle avoidance method.

[0047] The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a multi-UAV collaborative path planning and obstacle avoidance method provided by an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of a scenario of constructing an initial path based on a bidirectional greedy search according to an embodiment of the present invention;

[0050] Figure 3 is a schematic diagram of an elliptical sampling space provided by an embodiment of the present invention;

[0051] Figure 4 Schematic diagram of the algorithm interaction process of the multi-UAV collaborative path planning and obstacle avoidance method provided by an embodiment of the present invention;

[0052] Figure 5 This is a schematic structural diagram of a multi-UAV collaborative path planning and obstacle avoidance device provided by an embodiment of the present invention;

[0053] Figure 6 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0055] The present invention is implemented in a scenario where multiple drones are conducting disaster relief operations in a specific area. Because the area is large, the search for an optimal path increases, so the drones are assigned to different areas. Multiple drones work collaboratively in this environment. Before flight, a path to the target area is pre-planned at the drone's starting position. This allows the drones to reach the target area along the optimal path while avoiding both static and dynamic threat sources. Considering this environment as a two-dimensional area, dynamic threats can affect the pre-planned path at the beginning of flight, forcing the drones to avoid dynamic threats.

[0056] In order to solve the technical problems that the existing path planning and obstacle avoidance methods are difficult to cope with complex dynamic environments, have complex iterative calculations, and have low planning efficiency, the embodiment of the present invention provides a multi-UAV collaborative path planning and obstacle avoidance method. Figure 1 , Figure 1 : This is a flow chart of a multi-UAV collaborative path planning and obstacle avoidance method provided by an embodiment of the present invention, which specifically includes the following steps:

[0057] Step S101: construct a multi-UAV path planning model; the multi-UAV path planning model includes a flight constraint model, an environmental constraint model, and a dynamic threat source quantification model.

[0058] In the embodiment of the present invention, a multi-drone refers to a drone cluster including multiple drones.

[0059] In an embodiment of the present invention, the flight constraint model includes a flight length constraint model and a flight angle constraint model.

[0060] The flight length constraint model specifically includes the minimum flight length constraint and the maximum flight length constraint, as follows:

[0061] The maneuverability of a drone determines the shortest straight-line distance it needs to fly to ensure a smooth transition before changing its flight attitude. Simply put, when a drone encounters a threat source, that is, an obstacle, and needs to change its flight path, the drone must fly in a straight line from the starting point of the threat source to the shortest distance it needs to change direction. This shortest distance is defined as the minimum flight length, also called the minimum step length. Usually, it is taken as the distance the drone flies in 1 second at cruising speed. Depending on the maneuverability of the drone, the minimum flight distance will also vary accordingly. Let the minimum flight length be l min , then each path in the plan must satisfy the following formula, where {l i|i=1,2,3,…,I} represents a certain track segment in the planned path, that is, the distance between two route nodes:

[0062] l i ≥l min ;

[0063] Among them, l i represents the i-th track segment in the planned path; I is the total number of track segments in the planned path.

[0064] If the UAV is to complete its flight mission to the maximum extent and return safely, its maximum total flight length is limited by the fuel it carries. The UAV has to carry a mission payload, and the fuel it carries for each mission is fixed, so the maximum total flight length of the UAV is fixed. The maximum total flight length is recorded as L max , l i represents the i-th track segment, L represents the total flight length, that is, the total track segment length, then L max Must meet:

[0065]

[0066] Where I represents the total number of track segments.

[0067] In the embodiment of the present invention, the flight angle constraint model specifically includes a maximum turning angle constraint and a pitch angle constraint, as follows:

[0068] The maximum angle a drone can turn horizontally is the maximum turning angle. When planning a route, the drone is constrained by its maneuverability when changing flight direction. Therefore, the turning angle must be kept within the maximum turning angle range. α represents the maximum horizontal turning angle of the drone. The horizontal angle α1 between each two track segments generated by the plan must satisfy the following formula:

[0069] α1≤α;

[0070] The maximum climb / dive angle is the maximum angle that the drone is allowed to climb or dive continuously in the vertical direction. When planning the drone route, the maximum climb or dive angle of the drone must be considered in the longitudinal direction, because if the angle of climb or dive is too large during an attitude transition, the drone will stall. Here we give the maximum climb angle γ of the drone. Usually, the dive angle is approximately equal to the climb angle. From the route node (x k-1 ,y k-1 ,z k-1 ) to route node (x k ,y k ,z k ) can be calculated according to the following formula:

[0071]

[0072] Among them, x k-1 Indicates the horizontal coordinate of the k-1th route node; y k-1 Indicates the vertical coordinate of the k-1th route node; z k-1 Indicates the vertical coordinate of the k-1th route node; x k Indicates the horizontal coordinate of the kth route node; y k Indicates the vertical coordinate of the kth route node; z k represents the vertical coordinate of the kth route node; K represents the total number of route nodes.

[0073] At this point, the construction of the flight constraint model is completed.

[0074] In one implementation, the environmental constraint model includes a landslide / collapsed object model, a terrain model, and a no-fly zone threat model.

[0075] The route of the UAV will be affected by the landslide / collapsed objects. Therefore, the model of the landslide / collapsed objects can be simplified to a circular area centered on the building. d1 represents the distance from the landslide / collapsed object to the UAV route node. Represents the radius of the landslide / collapsed object, threat probability P R The decrease with increasing distance can be expressed as:

[0076]

[0077] The route of a drone is affected by terrain, such as mountains. Therefore, the peaks are modeled. The modeling of the peak can be simplified as a circular area centered on the peak's vertex, with the radius representing the range of the peak's base. This model represents the projection of the peak on a horizontal plane, and the radius of the circle corresponds to the range of influence of the peak's base. This modeling method allows the drone's route to be planned more clearly, ensuring that the peak is avoided, thereby reducing the risk of collision. The distance from the peak center to the drone's route node is d2, and the peak radius is R P , then the mountain threat probability P P It can be expressed as:

[0078]

[0079] A no-fly zone is an area designated as a prohibited area for drones to fly over. Once a drone reaches the area, it will encounter an accident. Therefore, the threat of a no-fly zone is expressed as follows:

[0080]

[0081] At this point, the construction of the multi-UAV path planning environmental constraint model is completed.

[0082] In the implementation of this invention, both the flight constraint model and the environmental constraint model in the multi-UAV path planning model are static models. Because static models lack quantitative analysis of technical uncertainty in dynamic environments, collaborative strategies are prone to decision redundancy or conflict in complex scenarios (such as multiple threat source interactions). Therefore, in this embodiment of the invention, a dynamic threat source quantification model is introduced to quantify the position, velocity, and shape of each threat source.

[0083] In one implementation, the dynamic threat source quantification model includes:

[0084]

[0085] Among them, H(O h ) represents the dynamic threat source quantification model of the h-th threat source; represents the mth state of the hth threat source; represents the probability of the mth state of the hth threat source appearing; M represents the total number of states of the hth threat source.

[0086] In the embodiment of the present invention, the state of the threat source refers to the position, speed, etc. of the threat source; h ), the larger the value is, the higher the dynamic uncertainty of the threat source is.

[0087] Through the local communication network of the drone cluster, the entropy values of obstacles perceived by each drone are aggregated in real time to form a global entropy field E(x, y, t), which represents the complexity of dynamic threats in each area of the environment. x and y represent spatial coordinates, that is, the two-dimensional plane coordinates in the drone's flight environment, and t is a timestamp used to describe the dynamic changes in environmental information.

[0088] Step S102: determining the path length cost, threat cost, and collision cost based on the multi-UAV path planning model.

[0089] In an embodiment of the present invention, the objective function is determined based on the multi-UAV path planning model:

[0090] Assume that the path Q of the nth drone is n is an ordered sequence (S, P2, P3, ... P K-1 ,T), where S represents the known starting route node, and its coordinates are (x s ,y s ), T represents the known target route node, its coordinates are (x t ,y t ). The multi-UAV collaborative trajectory planning is modeled as a continuous optimization problem, and the overall objective function F is:

[0091] minF=J L +J T +JC ;

[0092] Among them, J L represents the path length cost of the UAV; J T represents the threat cost of drones; J C Represents the collision cost of the drone.

[0093] In an embodiment of the present invention, determining the path length cost, threat cost, and collision cost based on the multi-UAV path planning model includes:

[0094] Determining a path length cost based on the multi-UAV path planning model and the fuel consumption cost and time consumption cost of each UAV;

[0095] Determining a threat cost based on the multi-UAV path planning model and the threat sources of each UAV's route through the threat area;

[0096] The collision cost is determined based on the multi-UAV path planning model and the safety distance between each UAV.

[0097] Furthermore, the path length cost J L It is defined as the total flight distance of all drones, which describes the fuel consumption and time consumption of drones in the process of performing missions. K ) is the complete drone path, and the path length cost is calculated as follows:

[0098]

[0099] Among them, P1 is the starting route node S, P K is the target route node T; (x k ,y k ) is the kth route node P k coordinates; N′ is the total number of drones, n=1,2,…,N′.

[0100] Threat Cost J T It represents the cost required for the drone route to cross the threat zone. For the convenience of calculation, for a continuous route segment, for example, 5 discrete equidistant points are selected to calculate its threat cost. By accumulating the threat costs of all track segments, the total threat cost of the entire drone path can be obtained as follows:

[0101]

[0102] Where W represents the total number of different types of threat sources in the environment; d h represents the level of the hth threat source; K represents the total number of route nodes; P 0.1,h,k ,…P 0.9,h,kRespectively represent the hth threat source in P k and P k+1 The threat probability of the 1 / 10,…,9 / 10 positions in the track segment between them to the drone.

[0103] In the embodiment of the present invention, threat sources may include mountains, collapsed objects, and no-fly zones encountered during flight.

[0104] Different from single UAV trajectory planning, multi-UAV trajectory planning needs to consider the spatiotemporal coordination between multiple UAVs to avoid collisions between them. In other words, sufficient safety distance must be maintained between UAVs. Assuming the coordinates are Route nodes On the path of drone a, and the coordinates are Route nodes On the path of another drone b, drone a and drone b are any two drones among N drones, and the collision risk q ab and collision cost J C The calculation method can be as follows:

[0105]

[0106]

[0107] Where C represents the penalty factor; t a Indicates that drone a has reached the route node Time; t b Indicates that drone b has arrived at the route node Time; L ab represents the distance between drone a and drone b; L safe Indicates the safe distance between two drones; t safe represents the safe time to avoid collision; N represents the total number of drones.

[0108] Step S103: using a bidirectional greedy search algorithm to determine an initial path, and constructing an elliptical sampling space based on the initial path.

[0109] Since the ellipse sampling space of the Informed-RRT* algorithm (a sampling-based path planning algorithm) is mainly composed of the length c of the first initial path searched best Therefore, in order to improve the algorithm's operation speed, the length of the initial path is very important. In order to find a better path during the generation of the initial path, the introduction of bidirectional greedy search can solve the blindness of unilateral tree growth when generating the initial path. At the same time, it can effectively reduce the number of algorithm iterations, so that the tree storage capacity will not become too large.

[0110] See also Figure 2 , Figure 2 This is a schematic diagram of a scenario in which an initial path is constructed based on a bidirectional greedy search according to an embodiment of the present invention. Bidirectional greedy search involves simultaneously establishing two trees, Tree1 and Tree2, at the starting point and end point of the drone. The two trees grow with each other's starting point as the end point, using a greedy strategy during the growth process. The specific steps are as follows:

[0111] Provision P start and P goal Two points are each other's endpoints;

[0112] Generate a new route node P in Tree1 new Afterwards, from P new Continuously grow towards the node closest to itself in Tree2 to ensure the minimum path cost until it collides or reaches the target; the minimum path cost means that the drone travels the shortest distance on the path and is least threatened.

[0113] If the Euclidean distance between the newly generated route nodes of the two trees is less than the specified end value threshold l min , then connect the two points with the smallest straight-line distance and check for collision. If there is no collision, connect them to form the initial path.

[0114] Specifically, create two trees Tree1 (with the root node P start ) and Tree2 (with root node P goal ), stipulate P start and P goal The two points are each other's endpoints and the optimal path cost is initialized to c best =∞.

[0115] In each iteration, Tree1 and Tree2 are expanded alternately. When Tree1 generates a new route node P new Afterwards, from P new Continuously grow toward the nearest node of Tree2 (connect in a straight line) until it collides or reaches the target; vice versa for Tree2.

[0116] Random sampling P rand , and find the nearest neighbor in Tree1 or Tree2 rand The nearest route node P near ;

[0117] Along P near →P rand Direction expansion step size generates P new :

[0118]

[0119] Among them, η represents the step size parameter, which determines the length of expansion along this direction, that is, P new Relative to P near Along P near to P rand The distance moved in the direction.

[0120] When Tree1's P new Successfully connect to the node of Tree2, merge the paths and update c best =cost(σ0), and get the initial path.

[0121] To improve path search efficiency, this embodiment of the present invention optimizes the search range by excluding threat areas and defining the search range using an ellipse. In the initial stage, based on the location and range of fixed threats, areas that are clearly within or close to the threat area and do not meet the safety distance are directly eliminated from the search space. Simultaneously, an ellipse is constructed with the starting and target points as foci. Since the optimal path is mostly located near the straight-line distance between the starting and target points, this ellipse effectively narrows the search range and significantly reduces computational complexity.

[0122] See also Figure 3 , Figure 3 is a schematic diagram of an elliptical sampling space provided by an embodiment of the present invention, in which the focus is fixed at P start and P goal , the major axis is The short axis is

[0123] In step S104, the improved particle swarm algorithm is used to optimize the initial path in the elliptical sampling space to obtain the global optimized path of each UAV; the initial population in the improved particle swarm algorithm is constructed based on the initial path, and the fitness function is constructed based on the path length cost, threat cost and collision cost.

[0124] In an embodiment of the present invention, path search further employs an improved particle swarm algorithm, namely, a particle swarm algorithm that introduces inertia weights. First, initialization is completed by randomly generating a group of particles. The initial position of each particle must be ensured to be within the search space and not within the threat source. Then, during the iteration process of the particle swarm, the speed and position of the particles are adjusted according to the speed update formula with time-decreasing inertia weights to balance global search and local optimization. After each update, the fitness of the particles is evaluated, including path length, obstacle avoidance performance, etc., to dynamically update the individual optimal position and the global optimal position of the group. As the number of iterations increases, the particles gradually converge to a better path. Finally, by setting the maximum number of iterations or the termination condition of fitness stability, a global optimal path that meets the requirements, i.e., the global optimization path, is obtained.

[0125] In the embodiment of the present invention, the route node sequence {P1, P2, ..., P K} is encoded as a PSO particle, and finally each particle The position vector is:

[0126]

[0127] in, represents the particle corresponding to the rth path; the coordinates of the kth path node in the rth path are Indicates; K represents the total number of route nodes.

[0128] Therefore, the initial path σ0 is encoded as the particle position vector:

[0129]

[0130] in, Represents the particle corresponding to the initial path; the coordinates of the kth route node in the initial path are expressed as express.

[0131] The initial population is generated by adding Gaussian noise from σ0, where N(0,σ 2 ) means the mean is 0 and the variance is σ 2 The Gaussian distributed noise is used, and σ controls the disturbance intensity to avoid population homogeneity.

[0132]

[0133] in, represents the initial population.

[0134] In one implementation, the fitness function needs to integrate the total path length J L , Threat Cost J T , collision cost J C , and supports multi-UAV collaboration, the fitness function F(P) includes:

[0135] F(P)=J L +αJ T +βJ C ;

[0136] Among them, α and β are weight factors, and the values of α and β must be much greater than 1 to ensure safety priority.

[0137] In the improved particle swarm algorithm, the particle update formula includes:

[0138]

[0139] in, Represents particles Velocity in iteration t+1; ω is the inertia weight, using a dynamic decay strategy: ω = 0.9-0.4·t / T max , t is the current iteration number, T max is the preset maximum number of iterations. c1 and c2 are learning factors, r1 and r2 are random numbers. pbest represents the current optimal path, and gbest represents the global optimal path, i.e., the globally optimized path.

[0140] In the improved particle swarm optimization algorithm, the position update formula includes:

[0141]

[0142] in, Represents particles Position at iteration t+1; Represents particles Position in iteration t; The position of the particle should be in the elliptical area. If the path exceeds the preset flight area, that is, the elliptical sampling space, reflection is performed.

[0143] Finally, the particle swarm iteration reaches the preset maximum number of iterations, or the fitness value converges. For example, when the change in the global optimal solution for several consecutive generations is less than the threshold, the iteration is terminated and the global optimization path is obtained.

[0144] Step S105, using a pre-trained depth extraction model to process the environmental image of each UAV to obtain a depth map containing the depth information of the UAV's surrounding environment; the environmental image of each UAV is collected in real time during the UAV's flight along the UAV's global optimization path.

[0145] First, the training process of the depth extraction model is explained:

[0146] In an embodiment of the present invention, the KITTI dataset and a custom synthetic dataset are used. These data contain a large number of image samples and can be used to train deep convolutional neural networks. The KITTI dataset is a computer vision algorithm evaluation dataset for autonomous driving scenarios. In practical applications, the input of the depth extraction model is a real-time environmental image captured by a sensor (such as a camera), which reflects the environmental scene around the drone. During the training process, the real depth image y* is generally obtained by a high-precision depth sensor (such as a lidar) as the label data for training.

[0147] Construct an initial depth extraction model, specifically a deep CNN (Convolutional Neural Networks). A CNN architecture based on a global coarse-scale network is constructed, consisting of convolutional layers, pooling layers, ReLU activation layers, and fully connected layers. Convolutional layers extract image features, with different convolution kernels capturing different types of features, such as edges and textures. Pooling layers downsample feature maps to reduce data size while retaining important feature information. The ReLU activation function introduces nonlinearity into the network, enhancing its expressiveness. Fully connected layers integrate previously extracted features to output the final depth estimation result.

[0148] A batch normalization layer is added after each convolutional layer to normalize the output of the convolutional layer so that the data input to the next layer has a similar distribution, thereby suppressing overfitting and improving the training efficiency and generalization ability of the network.

[0149] The network is trained using the training dataset (KITTI dataset and custom synthetic dataset). The input image is input into the network, and after a series of layer processing, the depth estimation image y is obtained. Then the depth estimation image y and the real depth image y are calculated. * The loss function L(y,y * ).

[0150] Calculate the loss function L(y,y * ), measure the depth estimation image y and the real depth image y in the logarithmic space * The scale invariant error between . Perform normalization processing. is the logarithmic error of each pixel, which is done to make the formula scale invariant, where p is the number of pixels.

[0151] In this embodiment of the present invention, the loss function is calculated as follows:

[0152]

[0153] Based on the value of the loss function, an optimization algorithm (such as stochastic gradient descent) is used to update the network parameters. Training is iterated continuously, gradually reducing the loss function until the number of iterations reaches a preset number or the model converges, resulting in a trained depth extraction model. At this point, the pre-trained depth extraction model can be used to process the real-time environmental images collected by each drone as it flies along its global optimization path to generate a depth map containing depth information about the drone's surroundings.

[0154] Step S106: convert the depth map of each drone into threat source distribution data.

[0155] In this embodiment of the present invention, the generated depth map can be converted into 2.5D threat source distribution data in polar coordinates. The specific process is to obtain depth values in the vertical and horizontal directions from the center of the depth map. These depth values represent the depth information of the threat source observed from different angles around the drone.

[0156] The depth values of the n-th UAV in different directions are associated with the corresponding angles to form the threat source distribution data in polar coordinate form. The threat source distribution data specifically includes the n-th UAV at the g-th angle Above, corresponding to a depth value Used to indicate the distance of the threat source in the angle direction.

[0157] Step S107 : Calculate the angular velocity of each UAV based on the current position, target position, and threat source distribution data of the UAV, so as to control the UAV to perform obstacle avoidance operations according to the angular velocity.

[0158] In an embodiment of the present invention, a behavior control algorithm can be used to calculate the angular velocity of the drone. The algorithm takes as input information the threat source distribution data, the current position of each drone, the target position, the gain parameter, and the angle range. The threat source distribution data is the angle around the drone. And the corresponding depth value The target position is relative to the nth UAV relative to its current position φ n target direction.

[0159] In one implementation, calculating the angular velocity of each drone based on its current position, target position, and threat source distribution data includes:

[0160] Calculate the obstacle avoidance behavior angular velocity and guidance behavior angular velocity of each drone based on its current position, target position, and threat source distribution data;

[0161] The angular velocity of each UAV is obtained by taking the weighted sum of the angular velocity of the obstacle avoidance behavior and the angular velocity of the guidance behavior.

[0162] In this embodiment of the present invention, the angle around the drone is and the corresponding depth value This is input into the obstacle avoidance behavior function. The formula takes into account the angle, distance, and current orientation of the threat source. Through the attenuation effect of the exponential function, the farther the threat source is and the greater its deviation from the current orientation, the smaller its impact on the obstacle avoidance behavior.

[0163] In an embodiment of the present invention, the obstacle avoidance behavior angular velocity and the guidance behavior angular velocity of each drone are calculated based on the current position, target position, and threat source distribution data of the drone, including:

[0164] Using the obstacle avoidance behavior function, the obstacle avoidance behavior angular velocity of each UAV is calculated based on its current position, target position, and threat source distribution data.

[0165] In an embodiment of the present invention, the obstacle avoidance behavior function includes:

[0166]

[0167] Among them, f avoid (φ n ) represents the angular velocity of the nth UAV’s obstacle avoidance behavior; φ n represents the current position of the nth UAV; ψ goal Indicates the target position of the nth UAV relative to the current position; c obst represents the first gain coefficient; β represents the angle range.

[0168] The guidance behavior function is used to calculate the guidance behavior angular velocity of each UAV based on its current position and target position.

[0169] In the embodiment of the present invention, the guidance behavior function is used to calculate the guidance behavior angular velocity of the drone toward the target orientation. The formula is based on the current orientation of the drone φ n Target direction The sine value of the angle difference between the two, combined with the second gain coefficient λ goto To determine the intensity of the guided behavior. When it is positive, it means that the nth drone needs to rotate counterclockwise. When it is negative, the drone needs to rotate counterclockwise.

[0170] In an embodiment of the present invention, the guidance behavior function includes:

[0171]

[0172] Among them, f goto (φ n ) represents the angular velocity of the nth UAV’s guidance behavior; λ goto Represents the third gain coefficient.

[0173] In the embodiment of the present invention, the first gain coefficient, the second gain coefficient, the third gain coefficient and the angle range may all be set by those skilled in the art based on experience.

[0174] In an embodiment of the present invention, the angular velocity of each UAV is obtained by performing a weighted summation of the angular velocity of the obstacle avoidance behavior and the angular velocity of the guidance behavior, including:

[0175] The angular velocity of the nth UAV's guidance behavior and the angular velocity of the obstacle avoidance behavior are weighted and summed to obtain the control instruction of the nth UAV, that is, the angular velocity

[0176]

[0177]

[0178] Among them, γ(E) represents the adjustment factor; E represents the regional entropy value; f c (φ n ) represents the angular velocity f of the nth UAV’s obstacle avoidance behavior avoid (φ n ) or the angular velocity f of the nth UAV’s guidance behavior goto (φ n ); c represents the obstacle avoidance or guidance behavior of the nth UAV; λ represents the entropy sensitivity coefficient; when the regional entropy value E is high, γ(E) increases the weight of obstacle avoidance behavior to prioritize safety; in low entropy areas, the obstacle avoidance weight is reduced to improve guidance efficiency.

[0179] After calculating the angular velocity, the flight direction of the drone can be controlled according to the angular velocity, so that the drone can avoid the threat source and fly towards the target direction.

[0180] The drone adjusts its flight angle in the yaw direction based on the control command it receives, which is the angular velocity. At the same time, it maintains a certain forward speed and continuously repeats the above-mentioned process of depth map generation, conversion, and control command calculation during flight, adjusting its flight direction in real time to avoid threats and fly towards the target.

[0181] In an embodiment of the present invention, a dynamic threat source quantification model within a multi-UAV path planning model enables quantitative analysis of technical uncertainty in dynamic environments, avoiding decision redundancy or conflict in collaborative strategies in dynamic and complex scenarios, and improving the accuracy and dynamic adaptability of subsequent path optimization based on the multi-UAV path planning model. By employing a bidirectional greedy search algorithm to determine the initial path, the blindness of unilateral exploration in existing path planning methods is reduced, shortening the time it takes to discover the initial path. Furthermore, by constructing an elliptical sampling space based on the initial path, the initial path is optimized within the elliptical sampling space to obtain a globally optimized path. This reduces the search range for path optimization, accelerates the convergence of the improved particle swarm algorithm, and effectively reduces computational complexity, thereby achieving an efficient, reliable, and flexible multi-UAV collaborative path planning and obstacle avoidance method.

[0182] See also Figure 4 , Figure 4This diagram illustrates the algorithmic interaction process for the multi-UAV collaborative path planning and obstacle avoidance method provided by an embodiment of the present invention. After the multi-UAV path planning model is built, the modeling information is fully and accurately transmitted to the algorithm layer. This information includes the starting and ending coordinates of the UAV flight, as well as the strict safety constraints in disaster relief scenarios, such as no-fly zones and the distribution of dangerous threat sources. After receiving this critical information, the algorithm layer quickly conducts initial path planning based on the environmental characteristics presented by the model.

[0183] First, a bidirectional greedy search algorithm is used to search simultaneously from the starting point and the end point. By rapidly evaluating the costs between nodes, a preliminary path is constructed. While this path roughly connects the starting point and the end point, it may not be optimal. Next, based on this initial path, an elliptical search region is constructed using a pre-set formula. This elliptical region fully accounts for environmental uncertainty and the existence of potential feasible paths, providing a wider space for subsequent searches.

[0184] The improved particle swarm algorithm then searches within the elliptical search area. This algorithm simulates the swarming behavior of flocks of birds or fish, allowing multiple particles to continuously move and update their positions within the search area, with each particle representing a possible path. Through continuous iteration and optimization, the particle swarm algorithm finds the optimal path within the elliptical search area, ensuring that the drone reaches its target location with the shortest flight distance and the shortest time possible, while meeting safety constraints.

[0185] While the drone is flying along its globally optimized path, it may encounter unexpected threats, such as collapsed buildings after an earthquake or floating debris swept by floodwaters. When this happens, the drone's onboard sensors quickly capture information about the threat source, including its location, shape, and size, and perform a depth estimation to determine its impact on the drone's flight.

[0186] Once detailed information about the threat source is obtained, the behavioral arbitration control algorithm immediately activates. Taking into account factors such as the drone's current flight status, remaining energy, and mission priority, the algorithm quickly analyzes the optimal strategy for responding to the threat source and issues a corrective route command. This corrective route command replans the drone's flight path based on the threat source's real-time status, ensuring it safely and flexibly avoids the threat source and continues toward its target. The behavioral arbitration control algorithm also considers the coordination of the entire disaster relief mission, preventing a drone's route correction from impacting the missions of others.

[0187] Based on the same inventive concept, the present invention also provides a multi-UAV collaborative path planning and obstacle avoidance device, see Figure 5 , Figure 5: This is a schematic diagram of the structure of a multi-UAV collaborative path planning and obstacle avoidance device provided by an embodiment of the present invention. The multi-UAV collaborative path planning and obstacle avoidance device includes:

[0188] The first construction module 501 is used to construct a multi-UAV path planning model; the multi-UAV path planning model includes a flight constraint model, an environmental constraint model and a dynamic threat source quantification model;

[0189] a determination module 502 for determining a path length cost, a threat cost, and a collision cost based on the multi-UAV path planning model;

[0190] A second construction module 503 is configured to determine an initial path using a bidirectional greedy search algorithm, and construct an elliptical sampling space based on the initial path;

[0191] an optimization module 504 for optimizing the initial path in the elliptical sampling space using an improved particle swarm algorithm to obtain a global optimized path for each UAV; wherein an initial population in the improved particle swarm algorithm is constructed based on the initial path, and a fitness function is constructed based on the path length cost, the threat cost, and the collision cost;

[0192] Processing module 505 is used to process the environment image of each UAV using a pre-trained depth extraction model to obtain a depth map containing depth information of the environment surrounding the UAV; the environment image of each UAV is collected in real time while the UAV is flying along the global optimization path of the UAV;

[0193] A conversion module 506 is used to convert the depth map of each UAV into threat source distribution data;

[0194] The calculation module 507 is used to calculate the angular velocity of each drone based on the current position, target position and threat source distribution data of the drone, so as to control the drone to perform obstacle avoidance operations according to the angular velocity.

[0195] In an embodiment of the present invention, a dynamic threat source quantification model within a multi-UAV path planning model enables quantitative analysis of technical uncertainty in dynamic environments, avoiding decision redundancy or conflict in collaborative strategies in dynamic and complex scenarios, and improving the accuracy and dynamic adaptability of subsequent path optimization based on the multi-UAV path planning model. By employing a bidirectional greedy search algorithm to determine the initial path, the blindness of unilateral exploration in existing path planning methods is reduced, shortening the time it takes to discover the initial path. Furthermore, by constructing an elliptical sampling space based on the initial path, the initial path is optimized within the elliptical sampling space to obtain a globally optimized path. This reduces the search range for path optimization, accelerates the convergence of the improved particle swarm algorithm, and effectively reduces computational complexity, thereby achieving an efficient, reliable, and flexible multi-UAV collaborative path planning and obstacle avoidance method.

[0196] Optionally, the dynamic threat source quantification model includes:

[0197]

[0198] Among them, H(O h ) represents the dynamic threat source quantification model of the h-th threat source; represents the mth state of the hth threat source; represents the probability of the mth state of the hth threat source appearing; M represents the total number of states of the hth threat source.

[0199] Optionally, determine the module, specifically for:

[0200] The path length cost is determined based on the multi-UAV path planning model and the fuel consumption cost and time consumption cost of each UAV; the threat cost is determined based on the multi-UAV path planning model and the threat sources in the threat area crossed by each UAV route; and the collision cost is determined based on the multi-UAV path planning model and the safety distance between each UAV.

[0201] Optional computing module, specifically used for:

[0202] The obstacle avoidance angular velocity and guidance angular velocity of each UAV are calculated based on the current position, target position and threat source distribution data of each UAV; the angular velocity of the obstacle avoidance behavior and guidance behavior angular velocity of each UAV are weighted summed to obtain the angular velocity of the UAV.

[0203] Optionally, a calculation module calculates the obstacle avoidance behavior angular velocity and the guidance behavior angular velocity of each UAV based on the current position, target position, and threat source distribution data of the UAV, including:

[0204] Using the obstacle avoidance behavior function, the obstacle avoidance behavior angular velocity of each UAV is calculated based on its current position, target position, and threat source distribution data.

[0205] The guidance behavior function is used to calculate the guidance behavior angular velocity of each UAV based on its current position and target position.

[0206] Optionally, the calculation module performs a weighted sum of the obstacle avoidance behavior angular velocity and the guidance behavior angular velocity of each UAV to obtain the angular velocity of the UAV, including:

[0207]

[0208]

[0209] in, represents the angular velocity of the nth UAV; γ(E) represents the adjustment factor; E represents the regional entropy value; f c (φ n ) represents the angular velocity of the obstacle avoidance behavior of the n-th UAV or the angular velocity of the guidance behavior of the n-th UAV, c represents the obstacle avoidance behavior or guidance behavior of the n-th UAV; λ represents the entropy sensitivity coefficient.

[0210] The embodiment of the present invention further provides an electronic device, such as Figure 6 As shown, it includes a processor 601, a communication interface 602, a memory 603 and a communication bus 604, wherein the processor 601, the communication interface 602, and the memory 603 communicate with each other through the communication bus 604.

[0211] Memory 603, used for storing computer programs;

[0212] The processor 601 is configured to implement any of the above-mentioned multi-UAV collaborative path planning and obstacle avoidance method steps when executing the program stored in the memory 603.

[0213] The communication bus mentioned in the electronic device mentioned above may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or only one type of bus.

[0214] The communication interface is used for communication between the above electronic device and other devices.

[0215] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. Alternatively, the memory may be at least one storage device located away from the processor.

[0216] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0217] The present invention also provides a computer-readable storage medium having a computer program stored therein, which, when executed by a processor, implements the steps of any of the above-mentioned multi-UAV collaborative path planning and obstacle avoidance methods.

[0218] Optionally, the computer-readable storage medium may be a non-volatile memory (NVM), such as at least one disk memory.

[0219] Optionally, the computer-readable storage medium may also be at least one storage device located away from the processor.

[0220] In another embodiment of the present invention, a computer program product comprising instructions is provided, which, when executed on a computer, enables the computer to execute the method steps described in any of the above-mentioned multi-UAV collaborative path planning and obstacle avoidance methods.

[0221] It should be noted that the terms "first," "second," and the like are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention.

[0222] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification.

[0223] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art can understand and implement other variations of the disclosed embodiments by viewing the drawings and the disclosed content. In the description of the present invention, the word "comprising" does not exclude other components or steps, "one" or "a" does not exclude multiple situations, and "multiple" means two or more, unless otherwise clearly and specifically defined. In addition, certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0224] The method provided in the embodiments of the present invention can be applied to electronic devices. Specifically, the electronic devices can be desktop computers, portable computers, smart mobile terminals, servers, etc. This is not limited here; any electronic device that can implement the present invention falls within the scope of protection of the present invention.

[0225] As for the device / electronic device / storage medium embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0226] It should be noted that the device, electronic device and storage medium of the embodiments of the present invention are respectively the device, electronic device and storage medium for applying the above-mentioned multi-UAV collaborative path planning and obstacle avoidance method. All embodiments of the above-mentioned multi-UAV collaborative path planning and obstacle avoidance method are applicable to the device, electronic device and storage medium, and can achieve the same or similar beneficial effects.

[0227] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A multi-UAV collaborative path planning and obstacle avoidance method, characterized in that: The multi-UAV collaborative path planning and obstacle avoidance method includes: Constructing a multi-UAV path planning model; the multi-UAV path planning model includes a flight constraint model, an environmental constraint model, and a dynamic threat source quantification model; determining path length cost, threat cost, and collision cost based on the multi-UAV path planning model; A bidirectional greedy search algorithm is used to determine an initial path, and an elliptical sampling space is constructed based on the initial path; An improved particle swarm algorithm is used to optimize the initial path in the elliptical sampling space to obtain a global optimized path for each UAV; an initial population in the improved particle swarm algorithm is constructed based on the initial path, and a fitness function is constructed based on the path length cost, the threat cost, and the collision cost; The environment image of each UAV is processed using a pre-trained depth extraction model to obtain a depth map containing the depth information of the UAV's surrounding environment; the environment image of each UAV is collected in real time while the UAV is flying along the UAV's globally optimized path; Convert the depth map of each drone into threat source distribution data; The angular velocity of each drone is calculated based on its current position, target position, and threat source distribution data, so as to control the drone to perform obstacle avoidance operations according to the angular velocity.

2. The multi-UAV collaborative path planning and obstacle avoidance method according to claim 1, characterized in that: The dynamic threat source quantification model includes: Among them, H(O h ) represents the dynamic threat source quantification model of the h-th threat source; represents the mth state of the hth threat source; represents the probability of the mth state of the hth threat source appearing; M represents the total number of states of the hth threat source.

3. The multi-UAV collaborative path planning and obstacle avoidance method according to claim 1, characterized in that: Determining path length cost, threat cost, and collision cost based on the multi-UAV path planning model includes: Determining a path length cost based on the multi-UAV path planning model and the fuel consumption cost and time consumption cost of each UAV; Determining a threat cost based on the multi-UAV path planning model and the threat sources of each UAV's route through the threat area; The collision cost is determined based on the multi-UAV path planning model and the safety distance between each UAV.

4. The multi-UAV collaborative path planning and obstacle avoidance method according to claim 1, characterized in that: Calculate the angular velocity of each drone based on its current position, target position, and threat source distribution data, including: Calculate the obstacle avoidance behavior angular velocity and guidance behavior angular velocity of each drone based on its current position, target position, and threat source distribution data; The angular velocity of each UAV is obtained by taking the weighted sum of the angular velocity of the obstacle avoidance behavior and the angular velocity of the guidance behavior.

5. The multi-UAV collaborative path planning and obstacle avoidance method according to claim 4, characterized in that: The obstacle avoidance angular velocity and guidance angular velocity of each UAV are calculated based on the current position, target position, and threat source distribution data of the UAV, including: Using the obstacle avoidance behavior function, the obstacle avoidance behavior angular velocity of each UAV is calculated based on its current position, target position, and threat source distribution data. The guidance behavior function is used to calculate the guidance behavior angular velocity of each UAV based on its current position and target position.

6. The multi-UAV collaborative path planning and obstacle avoidance method according to claim 4, characterized in that: The angular velocity of each UAV is obtained by weighted summing the angular velocity of the obstacle avoidance behavior and the angular velocity of the guidance behavior, including: in, represents the angular velocity of the nth UAV; γ(E) represents the adjustment factor; E represents the regional entropy value; f c (φ n ) represents the angular velocity of the obstacle avoidance behavior of the n-th UAV or the angular velocity of the guidance behavior of the n-th UAV, c represents the obstacle avoidance behavior or guidance behavior of the n-th UAV; λ represents the entropy sensitivity coefficient.

7. A multi-UAV collaborative path planning and obstacle avoidance device, characterized in that: The multi-UAV collaborative path planning and obstacle avoidance device includes: The first building module is used to build a multi-UAV path planning model; the multi-UAV path planning model includes a flight constraint model, an environmental constraint model and a dynamic threat source quantification model; a determination module, configured to determine a path length cost, a threat cost, and a collision cost based on the multi-UAV path planning model; A second construction module is used to determine an initial path using a bidirectional greedy search algorithm and construct an elliptical sampling space based on the initial path; an optimization module, configured to optimize the initial path in the elliptical sampling space using an improved particle swarm algorithm to obtain a global optimized path for each UAV; an initial population in the improved particle swarm algorithm is constructed based on the initial path, and a fitness function is constructed based on the path length cost, the threat cost, and the collision cost; A processing module is used to process the environment image of each UAV using a pre-trained depth extraction model to obtain a depth map containing the depth information of the environment around the UAV; the environment image of each UAV is collected in real time while the UAV is flying along the global optimization path of the UAV; A conversion module, used to convert the depth map of each drone into threat source distribution data; The calculation module is used to calculate the angular velocity of each drone based on its current position, target position and threat source distribution data, so as to control the drone to perform obstacle avoidance operations according to the angular velocity.

8. The multi-UAV collaborative path planning and obstacle avoidance device according to claim 7, characterized in that: The dynamic threat source quantification model includes: Among them, H(O h ) represents the dynamic threat source quantification model of the h-th threat source; represents the mth state of the hth threat source; represents the probability of the mth state of the hth threat source appearing; M represents the total number of states of the hth threat source.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; The processor is configured to implement the multi-UAV collaborative path planning and obstacle avoidance method according to any one of claims 1 to 6 when executing a computer program stored in the memory.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the multi-UAV collaborative path planning and obstacle avoidance method according to any one of claims 1 to 6 is implemented.

Images