A drone swarm collaborative search method and system based on deep multi-agent reinforcement learning

Abstract

This invention discloses a collaborative search method and system for drone swarms based on deep multi-agent reinforcement learning. The method includes: defining a tracker drone, a scout drone, and an intelligent target drone; initializing an actor network, a critic network, and an experience replay buffer for each drone agent; each drone agent selects and executes an action based on its current state through its actor network, and when the environment returns to a new state and corresponding reward, stores the interaction experience in the experience replay buffer; randomly extracts a batch of experience samples from the experience replay buffer and updates the critic network; uses the critic network to calculate gradients, and updates the actor network of each agent based on these gradients; and continuously performs agent interaction, data collection, network training, and target network updates until a certain training cycle or performance standard is achieved. This invention can improve the efficiency and safety of drone swarms performing collaborative search tasks in complex environments.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for unmanned aerial vehicle (UAV) swarms, and in particular to a collaborative search method and system for UAV swarms based on deep multi-agent reinforcement learning. Background Technology

[0002] In collaborative search missions involving swarm drones, effectively coordinating the behavior of individual drones to achieve rapid response and accurate tracking of targets, while ensuring the ability to explore unknown environments, presents a significant challenge. Traditional centralized control methods, due to their inherent computational and communication bottlenecks, struggle to adapt to complex and ever-changing environments and the real-time control requirements of large-scale drone swarms. While model-based distributed control methods alleviate these problems to some extent, they are often limited by the accuracy and generalization ability of the models.

[0003] In recent years, breakthroughs in deep learning technology have provided new approaches to solving the aforementioned problems. Deep neural networks, with their powerful feature extraction and pattern recognition capabilities, have achieved remarkable results in visual tasks such as image classification, object detection, and semantic segmentation. However, in practical applications of UAV swarm control, they face a series of challenges, including insufficient data samples, dynamic environmental changes, and limited computing resources. Especially for specific tasks such as altitude-restricted scenarios, traditional deep learning methods struggle to achieve ideal performance due to the sparsity and diversity of samples.

[0004] Multi-agent reinforcement learning, as an emerging research direction, offers a new approach to solving the cooperative control problem of UAV swarms by introducing competition and cooperation mechanisms among multiple agents. Within the multi-agent reinforcement learning framework, each UAV can be considered an agent, learning the optimal strategy through interaction with the environment to achieve inter-agent cooperation and task completion.

[0005] For example, Chinese patent document CN113110582A discloses a control method for an intelligent system of unmanned aerial vehicle (UAV) swarms, and Chinese patent document CN115755949A discloses an autonomous aggregation method for multi-UAV swarms based on multi-agent deep reinforcement learning.

[0006] However, most existing multi-agent reinforcement learning algorithms focus on the synergy of policies while neglecting the interpretability and generalization ability of policies, which may lead to unpredictable behavior and risks in actual drone swarm control. Summary of the Invention

[0007] This invention provides a method and system for collaborative search of UAV swarms based on deep multi-agent reinforcement learning. By organically combining deep learning and reinforcement learning algorithms, it realizes intelligent control of UAV swarms, which is particularly suitable for multi-target tracking and exploration tasks. It can improve the efficiency and safety of UAV swarms performing collaborative search tasks in complex environments.

[0008] A collaborative search method for UAV swarms based on deep multi-agent reinforcement learning includes the following steps:

[0009] (1) Construct intelligent drone agents to form a cluster, including tracker drones for tracking and identifying specific targets, scout drones for exploring unknown areas in the environment, and intelligent target drones for simulating actual target tracking.

[0010] (2) Based on the deep multi-agent reinforcement learning algorithm, an actor network and a critic network are initialized for each UAV agent in the cluster, and an experience replay buffer is initialized to store the interaction experience of the UAV agents in the cluster.

[0011] (3) Each UAV agent in the cluster selects and executes an action through its actor network based on the current state. When the environment returns a new state and the corresponding reward, the interaction experience is stored in the experience replay buffer. The interaction experience includes the current state, action, reward, and new state.

[0012] (4) Randomly select a batch of experience samples from the experience replay buffer, use these samples and the commentator network to calculate the temporal difference error, and update the commentator network accordingly; where the temporal difference error is the difference between the actual return and the return predicted by the commentator network.

[0013] (5) Use the critic network to compute gradients, which represent the expected reward of the drone agent's actions under the current policy, and update the actor network of each drone agent based on these gradients;

[0014] (6) Continuously perform interaction, data collection, network training and target network updates of UAV agents. After reaching the set training cycle or performance standard, input the images captured by the UAV cluster, calculate the best action output of each UAV agent in the cluster, and realize the collaborative search of the target.

[0015] In step (1), the responsibilities and functions of the tracker drone, the scout drone, and the smart target drone are clearly defined as tracking and identifying specific targets, exploring unknown areas in the environment, and simulating actual target tracking.

[0016] In step (2), a deep neural network architecture is designed, including an input layer for receiving states and actions, multiple fully connected layers, and an output layer that outputs values ​​(commentator network) or actions (actor network); an experience replay buffer is implemented to store the experience sequences obtained by each UAV agent in the cluster through interaction with the environment; and a copy of the target network is created for stabilizing the training process.

[0017] In step (2), the deep multi-agent reinforcement learning algorithm adopts an improved multi-agent deep deterministic policy gradient algorithm. Each UAV agent in the cluster has its own actor network and critic network, but the critic network is centralized and can access the observations and actions of all agents, including a centralized critic mechanism and a decentralized actor mechanism.

[0018] The centralized commentator mechanism specifically involves processing the observations and actions of all agents; that is, updating the commentator network using the observations and actions of all drone agents in the cluster. This centralized commentator mechanism takes into account information from all agents, helping them learn how to collaborate within a team to achieve a common goal.

[0019] The distributed actor mechanism specifically allows each agent to learn independently; that is, each agent updates the actor network based on the output of its corresponding critic network. This mechanism enables each drone agent in the cluster to flexibly adapt to environmental changes, as each agent can quickly react to new observations.

[0020] Deep multi-agent reinforcement learning algorithms also include the following three mechanisms:

[0021] The reward policy mechanism based on Voronoi diagrams: The geometric properties of Voronoi diagrams are that a plane is divided into different regions based on a set of specific points, and each region contains only one specific point, and any location within that region is closer to that specific point than to any other specific point. For each reconnaissance drone agent in the environment, the goal is to minimize the area of ​​its Voronoi diagram region; therefore, the agent needs to be dispersed to cover a larger space.

[0022] Collision avoidance mechanism: Use sensors (such as radar and lidar) to detect the position of nearby drones and predict the flight trajectory of each drone agent in the cluster. When an agent gets too close, it receives a negative reward, thereby preventing collisions between drones.

[0023] Boundary constraint mechanism: Define the flight boundary of the agent and continuously monitor its position; when the agent's position exceeds this boundary, it receives a negative reward; thereby prompting the agent to stay within the task-related area and preventing it from getting lost in the environment or entering dangerous areas.

[0024] In steps (4) and (5), a centralized critic network is designed, which can access the observations and actions of all agents. Each drone agent in the cluster has its own actor network and makes decisions based on its own observations of the local environment. The strategies of different drone agents in the cluster are coordinated through the evaluation of the critic network to achieve overall optimization.

[0025] In step (6), the interaction of the UAV intelligent agent, data collection, network training, and target network updates are continuously carried out, as follows:

[0026] (6-1) Initialize an actor network and a critic network for each drone agent in the cluster; the actor network outputs the optimal action in a given state, while the critic network evaluates the value of the action under the current policy;

[0027] (6-2) Each drone agent in the cluster performs actions in the environment and collects data on its interactions with the environment, including state, actions, rewards and new states; this data is stored in an experience replay buffer for use during training;

[0028] (6-3) Update the critics' network;

[0029] (6-4) Update the actor network;

[0030] (6-5) Update the target network;

[0031] (6-6) Asynchronous update: In practical applications, agents can update their network parameters asynchronously, allowing them to learn independently at their own time steps;

[0032] (6-7) Exploration mechanism: The ε-greedy strategy is integrated into the agent network to ensure that the agent can explore enough environmental information in the early stages.

[0033] In step (6-3), the critic network is updated, and the specific process is as follows:

[0034] A batch of empirical samples is randomly drawn from the empirical replay buffer, and these samples are used to calculate the time difference error loss function of the critic network:

[0035] L critic =E (s，a，r，s') [r+γQ′ i (s′,μ′ i (s′))-Q i (s, a)] 2

[0036] Where E represents the expected value, s is the current state, a is the selected action, r is the immediate reward, s′ is the next state, γ is a discount factor used to measure the current value of the future reward, i represents the drone number, and Q′ is the next state. i (s′,μ′ i (s′)) is the value estimate of the next state, μ′ i (s′) are the parameters of the value function, Q i (s,a) is the value estimate of the current state;

[0037] Update the parameters of the critic network using gradient descent:

[0038]

[0039] Where ω is the parameter of the critic network, α ω It is the learning rate of the critic network. It is the gradient of the loss function L(ω) with respect to the parameter ω.

[0040] In step (6-4), the actor network is updated, and the specific process is as follows:

[0041] The critic network is used to compute the gradients of the actor network, which represent the expected reward of the agent's action under the current policy:

[0042]

[0043] Where i represents the UAV serial number, θ i The actor network parameters of drone i are J(θ). i Q is the expected return of drone i. i (s,a i ) is a state-action pair (s, a) i Value function estimation, It is the loss function μ i (s) with respect to parameter θ i The gradient;

[0044] Update the parameters of the actor network using gradient ascent:

[0045]

[0046] in, It is the learning rate of the actor network, θ i These are parameters of the actor network. The loss function J with respect to parameter θ i The gradient.

[0047] In step (6-5), the target network is updated, and the specific process is as follows:

[0048] To stabilize the training process, the parameters of the target network are gradually updated to approximate the parameters of the current network using a soft update method.

[0049] ω'←τωω+(1-τ)ω'θ′ i ←τθ i +(1-τ)θ′ i

[0050] Where ω is the parameter of the critic network, θ i τ is the parameter of the actor network, and τ is the soft update parameter, ranging from 0.0001 to 0.001.

[0051] A UAV swarm cooperative search system based on deep multi-agent reinforcement learning includes a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, they implement the aforementioned UAV swarm cooperative search method.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] 1. This invention significantly improves the tracking ability of UAV swarms of moving targets, especially in complex and dynamic environments, through a deep multi-agent reinforcement learning algorithm.

[0054] 2. This invention achieves effective collaboration among tracker drones, scout drones, and intelligent target drones, and optimizes resource allocation and role assignment in multi-agent systems.

[0055] 3. By introducing experience replay and target network, this invention reduces the dependence on real-time data and improves the stability and efficiency of the learning process.

[0056] 4. The drone swarm can adapt to environmental changes and flexibly adjust its strategies to cope with different tracking and exploration tasks, thus improving the robustness of the system.

[0057] 5. The method and system design of the present invention take scalability into account and can be easily integrated into existing UAV swarms or multi-agent systems. Attached Figure Description

[0058] Figure 1 This is a flowchart of a drone swarm cooperative search method based on deep multi-agent reinforcement learning, according to an embodiment of the present invention.

[0059] Figure 2 This is the training framework for the unmanned aerial vehicle (UAV) agent in this embodiment of the invention. Detailed Implementation

[0060] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be noted that the embodiments described below are intended to facilitate the understanding of the present invention and do not constitute any limitation thereof.

[0061] like Figure 1 As shown, a collaborative search method for UAV swarms based on deep multi-agent reinforcement learning includes the following steps:

[0062] S01 defines three roles in the system: tracker drone, reconnaissance drone, and intelligent target drone. Each drone is equipped with corresponding sensors and computing modules according to its functions.

[0063] In this embodiment of the invention, the tracker drone is designed to track and identify specific targets, equipped with a high-speed processor and advanced sensors, such as cameras and infrared sensors, to achieve accurate tracking; the scout drone is responsible for exploring unknown areas, collecting environmental information, and providing the tracker with possible target locations, and usually has a long flight time and a large detection range; the intelligent target drone simulates enemy targets, has the ability to evade tracking, and is equipped with obstacle avoidance technology and possible countermeasures to simulate evasion behavior in complex environments.

[0064] S02, design a deep neural network architecture for each agent, define the state space of each agent, implement an experience replay buffer to store the agent's interaction experience, and create a target network for each agent, which is a copy of its actor network, for stabilizing the training process.

[0065] In this embodiment of the invention, the deep neural network architecture designed for each agent specifically includes the following 5-layer structure:

[0066] A. Input Layer: This layer first receives data from the drone's sensors, including position, velocity, acceleration, gyroscope readings, and compass readings. This data is then fused to form a global and local representation of the environment.

[0067] B. Convolutional Layers: Multiple convolutional layers are used to extract spatial features from image or sensor data. Each convolutional layer is followed by a ReLU activation function to introduce non-linearity and enhance the network's expressive power.

[0068] C. Pooling layer: After the convolutional layer, a pooling layer is used for max pooling to reduce the dimensionality of the feature map while retaining the most important features.

[0069] D. Hidden Layers: Fully connected layers further combine and process the features extracted from previous layers. There can be multiple such layers, each of which increases the non-linear expressive power of the network through an activation function.

[0070] E. Output Layer: The last fully connected layer maps the feature vectors to the action space. For a continuous action space, the output of this layer can be a parameterized policy, such as the mean and variance of a normal distribution, used to determine the drone's next action.

[0071] In this embodiment of the invention, the state information of the intelligent agents corresponding to the tracker drone, the reconnaissance drone, and the intelligent target drone includes the following aspects:

[0072] A. Position and velocity: Where (x, y, z) represents the position coordinates of the UAV in three-dimensional space. This represents the velocity vector of the drone.

[0073] B. Environmental information: (V,E), where V is the set of obstacles in the environment and E is the environmental boundary information.

[0074] C. Locations of other drones: (x i ,y i ,z i |i∈teammates) represents the location of other drones of the same type in the same task.

[0075] D. Task-related parameters: (T,R), where T represents the task type and R represents the task-related rules and restrictions.

[0076] E. The state space of the tracker drone should also include the position and velocity of the target drone: This indicates the location and speed of the target drone being tracked by the tracker drone.

[0077] F. The state space of the reconnaissance drone should also contain information about unexplored areas: (U x U y U z ), indicating the location of an unexplored area or potential target.

[0078] In this embodiment of the invention, a deep neural network is used to approximate the policy function π and the value function Q. For each agent i, the actor network μ i and Value Network Q i Defined as:

[0079] μ i (s i = Neural Network Parameterization Strategy

[0080] Q i (s i ,a i ) = Neural Network Parameterized Value Function

[0081] In this embodiment of the invention, an experience replay mechanism is designed and implemented for storing the interactive experience e of the intelligent agent. t (s t ,a t ,r t ,s t+1 ).

[0082] In this embodiment of the invention, a target network μ is introduced. i ′ and Q i ′, to stabilize the training process:

[0083] μ i ′←τμ i +(1-τ)μ i ′

[0084] Q i ′←τQ i +(1-τ)Q i ′

[0085] Where, μ i and Q i It consists of the original actor network and the commentator network, and τ is the soft update parameter.

[0086] S03, each drone agent in the cluster selects and executes an action based on its current state through its actor network. When the environment returns a new state and corresponding reward, the interaction experience is stored in the experience replay buffer. The interaction experience includes the current state, action, reward, and new state.

[0087] S04, using the observations and actions of all agents in the experience replay buffer to update the critic network, and updating the actor network for each agent based on its corresponding critic network output, as shown in the following formula:

[0088] L critic =E( s，a，r，s′ )[r+yQ′ i (s',μ′ i (s'))-Q i (s, a)] 2

[0089] Where E represents the expected value, s is the current state, a is the selected action, r is the immediate reward, s′ is the next state, γ is a discount factor used to measure the current value of the future reward, i represents the drone number, and Q′ is the next state. i (s′,μ′ i (s′)) is the value estimate of the next state, μ′ i (s′) are the parameters of the value function, Q i (s,a) is the value estimate of the current state.

[0090]

[0091] Where, θ i The actor network parameters of drone i are J(θ). i Q is the expected return of drone i. i (s,a i ) is a state-action pair (s, a) i Value function estimation, It is the loss function μ i (s) with respect to parameter θ i The gradient.

[0092] S05 uses a specific loss function to train the model and optimize network parameters, as follows:

[0093] Update the parameters of the critic network using gradient descent:

[0094]

[0095] Where ω is the parameter of the critic network, α ω It is the learning rate of the critic network. It is the gradient of the loss function L(ω) with respect to the parameter ω.

[0096] Update the parameters of the actor network using gradient ascent:

[0097]

[0098] in, It is the learning rate of the actor network, θ i These are parameters of the actor network. The loss function J with respect to parameter θ i The gradient.

[0099] S06 continuously performs interaction, data collection, network training, and target network updates for the UAV intelligent agent, tests the effectiveness of the algorithm in a simulation environment, evaluates system performance, including tracking accuracy and exploration efficiency, and optimizes algorithm parameters based on feedback.

[0100] like Figure 2 As shown, its training framework is as follows: the actor network i receives observation i, performs action i, feeds back to the critic network i, and stores this information in the experience replay buffer. The critic network is then further updated by combining these actions with the observations.

[0101] This invention achieves effective control of drone swarms by introducing advanced deep learning technology and multi-agent reinforcement learning algorithms. The core of the system includes an improved multi-agent deep deterministic policy gradient algorithm, which achieves coordinated optimization of the behavior of each individual drone in the swarm through a centralized critic mechanism and a decentralized actor mechanism. Simultaneously, the system integrates a series of key technologies such as a Voronoi diagram-based reward policy, collision avoidance mechanism, and boundary constraint mechanism to enhance the environmental adaptability and task execution capability of the drone swarm.

[0102] The embodiments of this invention were tested in both simulated environments and real-world scenarios, demonstrating their effectiveness and robustness in collaborative target search tasks involving UAV swarms. By combining deep neural networks with multi-agent reinforcement learning techniques, this invention not only enhances the collaborative working capabilities of UAV swarms but also strengthens their adaptability to complex environments, possessing significant practical value and broad application prospects.

[0103] The embodiments described above provide a detailed explanation of the technical solutions and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for cooperative search of unmanned aerial vehicle (UAV) swarms based on deep multi-agent reinforcement learning, characterized in that, Includes the following steps: (1) Construct intelligent drone agents to form a cluster, including tracker drones for tracking and identifying specific targets, scout drones for exploring unknown areas in the environment, and intelligent target drones for simulating actual target tracking. (2) Based on the deep multi-agent reinforcement learning algorithm, an actor network and a critic network are initialized for each UAV agent in the cluster, and an experience replay buffer is initialized to store the interaction experience of the UAV agents in the cluster. (3) Each UAV agent in the cluster selects and executes an action through its actor network based on the current state. When the environment returns a new state and the corresponding reward, the interaction experience is stored in the experience replay buffer. The interaction experience includes the current state, action, reward, and new state. (4) Randomly select a batch of experience samples from the experience replay buffer, use these samples and the commentator network to calculate the temporal difference error, and update the commentator network accordingly; where the temporal difference error is the difference between the actual return and the return predicted by the commentator network. (5) Use the critic network to compute gradients, which represent the expected reward of the drone agent's actions under the current policy, and update the actor network of each drone agent based on these gradients; (6) Continuously perform interaction, data collection, network training and target network updates of UAV agents. After reaching the set training cycle or performance standard, input the images captured by the UAV cluster, calculate the best action output of each UAV agent in the cluster, and realize the collaborative search of the target.

2. The UAV swarm cooperative search method based on deep multi-agent reinforcement learning according to claim 1, characterized in that, In step (2), the deep multi-agent reinforcement learning algorithm adopts an improved multi-agent deep deterministic policy gradient algorithm. Each UAV agent in the cluster has its own actor network and critic network, but the critic network is centralized and can access the observations and actions of all agents, including a centralized critic mechanism and a decentralized actor mechanism.

3. The UAV swarm cooperative search method based on deep multi-agent reinforcement learning according to claim 2, characterized in that, The centralized commentator mechanism specifically involves processing the observations and actions of all agents, that is, using the observations and actions of all drone agents in the cluster to update the commentator network; The decentralized actor mechanism specifically allows each agent to learn independently, that is, for each agent, the actor network is updated based on the output of its corresponding critic network.

4. The UAV swarm cooperative search method based on deep multi-agent reinforcement learning according to claim 1, characterized in that, Deep multi-agent reinforcement learning algorithms also include the following three mechanisms: The reward strategy mechanism based on Voronoi diagrams: For each scout drone agent in the environment, the goal is to minimize the area of ​​its Voronoi diagram region. Therefore, the agents need to spread out to cover a larger space. Collision avoidance mechanism: Sensors are used to detect the positions of nearby drones and predict the flight trajectory of each drone agent in the cluster. When an agent gets too close, it receives a negative reward, thereby preventing collisions between drones. Boundary constraint mechanism: Define the flight boundary of the agent and continuously monitor its position; when the agent's position exceeds this boundary, it receives a negative reward; thereby prompting the agent to stay within the task-related area and preventing it from getting lost in the environment or entering dangerous areas.

5. The UAV swarm cooperative search method based on deep multi-agent reinforcement learning according to claim 1, characterized in that, The process of continuously engaging the drone's intelligent agent, collecting data, training the network, and updating the target network is as follows: (6-1) Initialize an actor network and a critic network for each drone agent in the cluster; the actor network outputs the optimal action in a given state, while the critic network evaluates the value of the action under the current policy; (6-2) Each drone agent in the cluster performs actions in the environment and collects data on its interactions with the environment, including state, actions, rewards and new states; this data is stored in an experience replay buffer for use during training; (6-3) Update the critics' network; (6-4) Update the actor network; (6-5) Update the target network; (6-6) Asynchronous update: In practical applications, agents can update their network parameters asynchronously, allowing them to learn independently at their own time steps; (6-7) Exploration mechanism: The ε-greedy strategy is integrated into the agent network to ensure that the agent can explore enough environmental information in the early stages.

6. The UAV swarm cooperative search method based on deep multi-agent reinforcement learning according to claim 5, characterized in that, In step (6-3), the critic network is updated, and the specific process is as follows: A batch of empirical samples is randomly drawn from the empirical replay buffer, and these samples are used to calculate the time difference error loss function of the critic network: L critic =E (s,a,r,s′) [r+γQ′ i (s′,μ′ i (s′))-Q i (s,a)] 2 Where E represents the expected value, s is the current state, a is the selected action, r is the immediate reward, s′ is the next state, γ is a discount factor used to measure the current value of the future reward, i represents the drone number, and Q′ is the next state. i (s′,μ′ i (s′)) is the value estimate of the next state, μ′ i (s′) are the parameters of the value function, Q i (s,a) is the value estimate of the current state; Update the parameters of the critic network using gradient descent: Where ω is the parameter of the critic network, α ω It is the learning rate of the critic network. It is the gradient of the loss function L(ω) with respect to the parameter ω.

7. The UAV swarm cooperative search method based on deep multi-agent reinforcement learning according to claim 5, characterized in that, In step (6-4), the actor network is updated, and the specific process is as follows: The critic network is used to compute the gradients of the actor network, which represent the expected reward of the agent's action under the current policy: Where i represents the UAV serial number, θ i The actor network parameters of drone i are J(θ). i Q is the expected return of drone i. i (s,a i ) is a state-action pair (s, a) i Value function estimation, It is the loss function μ i (s) with respect to parameter θ i The gradient; Update the parameters of the actor network using gradient ascent: in, It is the learning rate of the actor network, θ i These are parameters of the actor network. The loss function J with respect to parameter θ i The gradient.

8. The UAV swarm cooperative search method based on deep multi-agent reinforcement learning according to claim 5, characterized in that, In step (6-5), the target network is updated, and the specific process is as follows: To stabilize the training process, the parameters of the target network are gradually updated to approximate the parameters of the current network using a soft update method. ω′←τω+(1-τ)ω′θ′ i ←tth i +(1-τ)θ′ i Where ω is the parameter of the critic network, θ i τ is the parameter of the actor network, and τ is the soft update parameter, ranging from 0.0001 to 0.

001.

9. A collaborative search system for unmanned aerial vehicle (UAV) swarms based on deep multi-agent reinforcement learning, characterized in that, It includes a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the UAV swarm cooperative search method according to any one of claims 1-8.

Images