Unmanned aerial vehicle group path planning method and device in communication limited environment

Abstract

The invention discloses an unmanned aerial vehicle group path planning method and device in a communication limited environment, and relates to the technical field of unmanned aerial vehicles or other related fields, and the method comprises the steps: deploying an unmanned aerial vehicle group in the communication limited environment, collecting an environment image of the communication limited environment, carrying out the feature extraction of the environment image through employing a ResNet network model, and obtaining a path of the unmanned aerial vehicle group; the DQN network model is trained in a communication limited simulation environment to obtain an environment feature vector, the environment feature vector is input into the trained DQN network model to obtain a target path of the unmanned aerial vehicle group, the DQN network model is trained in the communication limited simulation environment, and the training process comprises the steps of initializing the simulation environment and the DQN network model, alternately training a main network and a target network of the DQN network model based on an experience playback mechanism, and obtaining a target path of the unmanned aerial vehicle group. And obtaining a target path, performing performance evaluation on the target path, and iteratively updating the DQN network model based on an evaluation result. According to the invention, the technical problem of low adaptability and accuracy of path planning under the condition of limited communication in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of drone technology or other related fields, and in particular to a method and device for drone group path planning in a communication-restricted environment. Background Art

[0002] Drone swarm path planning is an important research direction in the field of drone technology. Especially in communication-restricted environments, how to achieve effective path planning for drone swarms is of great significance to improving the autonomy and safety of drone swarms. With the continuous development of drone applications, the demand for their application in civil and environmental monitoring fields is growing. However, in urban environments, complex terrains and other scenarios, drone swarms face the challenge of limited communication, which brings great difficulties to the collaborative work and safe flight of drones.

[0003] In the related technology, the traditional UAV swarm path planning strategy mainly relies on the central control unit for unified scheduling and path allocation, but this strategy is not applicable in communication-restricted environments. The UAV swarm path planning strategy is usually divided into global path planning strategy and local path planning strategy. The global path planning strategy focuses on planning a conflict-free path in advance based on known environmental information before the UAV performs a mission, but this strategy has poor adaptability and is difficult to cope with dynamic environments; the local path planning strategy is suitable for navigation in unknown environments and can cope with uncertainties in dynamic environments, but this strategy is limited by communication conditions. When the environment is a communication-restricted environment, the applicability and accuracy of the local path planning strategy will be reduced.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a method and device for path planning of a drone swarm in a communication-restricted environment, so as to at least solve the technical problem of low adaptability and accuracy of path planning under communication-restricted conditions in the related art.

[0006] According to one aspect of an embodiment of the present invention, a method for path planning of a swarm of unmanned aerial vehicles in a communication-constrained environment is provided, comprising: deploying a swarm of unmanned aerial vehicles in a communication-constrained environment, and collecting an environmental image of the communication-constrained environment; using a ResNet network model to perform feature extraction on the environmental image to obtain an environmental feature vector; inputting the environmental feature vector into a trained DQN network model to obtain a target path of the swarm of unmanned aerial vehicles; wherein the DQN network model is trained in a communication-constrained simulation environment, and the training process comprises: initializing the simulation environment and the DQN network model; based on an experience replay mechanism, alternately training a main network and a target network of the DQN network model to obtain a target path, wherein the update frequency of the target network is lower than the update frequency of the main network; performing performance evaluation on the target path, and iteratively updating the DQN network model based on the performance evaluation result.

[0007] Optionally, the step of extracting features from the environment image using a ResNet network model to obtain an environment feature vector includes: storing the environment image in a drone memory, and expanding the environment image data set using a data enhancement strategy; defining a structure of the ResNet network model, the structure including a network depth; inputting the expanded environment image data set into the ResNet network model to obtain an environment feature vector, wherein the ResNet network model includes: a convolutional layer, an activation function, a pooling layer, and a fully connected layer.

[0008] Optionally, the step of initializing the simulation environment and the DQN network model includes: starting the simulation software, establishing an urban environment model, establishing a flight area in the urban environment model, and deploying a swarm of drones in the flight area, wherein the urban environment model includes at least buildings and roads; defining the structure of the DQN network model, the structure including an input layer, multiple hidden layers, and an output layer; defining hyperparameters of the DQN network model, the hyperparameters including a learning rate, a discount factor, and an exploration rate; defining a loss function of the DQN network model; defining an experience replay buffer, the experience replay buffer being used to store drone state tuples, the drone state tuples including the current state of the drone, the current action, the current reward, and the new state.

[0009] Optionally, based on the experience replay mechanism, the steps of alternately training the main network and the target network of the DQN network model include: obtaining the environmental feature vector of the simulation environment, and storing the preprocessed environmental feature vector in the memory of the drone; inputting the current state of the drone into the main network in the DQN network model, executing the current action, and obtaining the new state and current reward of the drone; inputting the new state into the target network in the DQN network model, calculating the Q values ​​of all new actions, and determining the target Q value, wherein the Q value is the value of the new action or the current action; updating the loss function based on the target Q value, the current reward, and the Q value in the current state; updating the DQN network parameters using the gradient descent strategy; alternately training the main network and the target network, and when the convergence conditions are met, determining the target predicted Q value, and obtaining the target path corresponding to the target predicted Q value.

[0010] Optionally, the step of performing a performance evaluation on the target path and adjusting the DQN network model based on the performance evaluation results includes: defining performance evaluation indicators, the performance evaluation indicators at least including: a path length indicator, a flight time indicator, and a safety indicator; based on the performance evaluation results, adjusting the hyperparameters of the DQN network model, the update frequency of the target network, and the capacity of the experience replay buffer.

[0011] Optionally, the step of deploying a swarm of drones in a communication-restricted environment and collecting environmental images of the communication-restricted environment includes: deploying a swarm of drones in the communication-restricted environment, collecting environmental images through sensors and camera modules on the drones, and preprocessing and normalizing the environmental images, wherein the preprocessing includes removing noise and abnormal pixels from the environmental images, and the abnormal pixels include: pixels with sudden brightness changes, pixels with distorted colors, and pixels that are inconsistent with the expected physical environment.

[0012] Optionally, after obtaining the target path of the drone swarm, the method includes: continuously monitoring the DQN network model, collecting flight data of the drone swarm, abnormal conditions, and environmental data of the communication restricted environment, wherein the flight data includes: obstacle avoidance success rate, the difference between the flight time and the expected time, and the abnormal conditions include: the drone fails to effectively avoid obstacles and the drone flight path is long.

[0013] According to another aspect of an embodiment of the present invention, a drone swarm path planning device in a communication-restricted environment is also provided, including: a collection unit, used to deploy a drone swarm in a communication-restricted environment and collect environmental images of the communication-restricted environment; an extraction unit, used to use a ResNet network model to extract features of the environmental image to obtain an environmental feature vector; an acquisition unit, used to input the environmental feature vector into a trained DQN network model to obtain a target path of the drone swarm; a training unit, the DQN network model is trained in a communication-restricted simulation environment, and the training process includes: initializing the simulation environment and the DQN network model; based on an experience replay mechanism, alternately training a main network and a target network of the DQN network model to obtain a target path, wherein the update frequency of the target network is lower than the update frequency of the main network; performing performance evaluation on the target path, and iteratively updating the DQN network model based on the performance evaluation result.

[0014] Optionally, the extraction unit includes: an expansion module, used to store the environmental image in the drone memory, and use a data enhancement strategy to expand the environmental image data set; a first definition module, used to define the structure of the ResNet network model, the structure including the network depth; an input module, used to input the expanded environmental image data set into the ResNet network model to obtain an environmental feature vector, wherein the ResNet network model includes: a convolutional layer, an activation function, a pooling layer, and a fully connected layer.

[0015] Optionally, the training unit includes: a startup module, used to start the simulation software, establish an urban environment model, establish a flight area in the urban environment model, and deploy a drone swarm in the flight area, wherein the urban environment model includes at least buildings and roads; a second definition module, used to define the structure of the DQN network model, the structure includes an input layer, multiple hidden layers and an output layer; a third definition module, used to define the hyperparameters of the DQN network model, the hyperparameters include a learning rate, a discount factor and an exploration rate; a fourth definition module, used to define the loss function of the DQN network model; a fifth definition module, used to define an experience replay buffer, the experience replay buffer is used to store drone state tuples, the drone state tuples include: the current state of the drone, the current action, the current reward and the new state.

[0016] Optionally, the training unit also includes: a storage module, which is used to obtain the environmental feature vector of the simulation environment and store the preprocessed environmental feature vector in the drone memory; an execution module, which is used to input the current state of the drone into the main network in the DQN network model, execute the current action, and obtain the new state and current reward of the drone; a calculation module, which is used to input the new state into the target network in the DQN network model, calculate the Q values ​​of all new actions, and determine the target Q value, where the Q value is the value of the new action or the current action; a first update module, which updates the loss function based on the target Q value, the current reward and the Q value in the current state; a second update module, which is used to update the DQN network parameters using a gradient descent strategy; a determination module, which is used to alternately train the main network and the target network, determine the target predicted Q value when the convergence conditions are met, and obtain the target path corresponding to the target predicted Q value.

[0017] Optionally, the training unit also includes: a sixth definition module, used to define performance evaluation indicators, the performance evaluation indicators include at least: a path length indicator, a flight time indicator, and a safety indicator; an adjustment module, based on the performance evaluation results, adjusts the hyperparameters of the DQN network model, the update frequency of the target network, and the capacity of the experience replay buffer.

[0018] Optionally, the acquisition unit includes: a processing module, used to deploy a drone swarm in a communication-restricted environment, collect environmental images through sensors and camera modules on the drones, and preprocess and normalize the environmental images, wherein the preprocessing includes removing noise and abnormal pixels from the environmental images, and the abnormal pixels include: pixels with sudden brightness changes, pixels with distorted colors, and pixels that are inconsistent with the expected physical environment.

[0019] Optionally, the acquisition unit includes: a monitoring module, which is used to continuously monitor the DQN network model, collect flight data of the drone group, abnormal conditions and environmental data of the communication restricted environment, wherein the flight data includes: obstacle avoidance success rate, the difference between the flight time and the expected time, and the abnormal conditions include: the drone fails to effectively avoid obstacles and the drone flight path is long.

[0020] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is also provided, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned methods for drone group path planning in a communication restricted environment.

[0021] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the above-mentioned methods for drone swarm path planning in a communication restricted environment.

[0022] According to another aspect of an embodiment of the present invention, a computer program product is also provided, including a computer program, which, when executed by a processor, implements the steps of any one of the above-mentioned methods for path planning of a drone swarm in a communication-restricted environment.

[0023] In the present disclosure, a swarm of drones is deployed in a communication-restricted environment, environmental images of the communication-restricted environment are collected, and then features of the environmental images are extracted by using a ResNet network model to obtain an environmental feature vector. Finally, the environmental feature vector is input into a trained DQN network model to obtain a target path of the swarm of drones. The DQN network model is trained in a communication-restricted simulation environment, and the training process includes: initializing the simulation environment and the DQN network model; based on an experience replay mechanism, alternately training the main network and the target network of the DQN network model to obtain a target path; performing performance evaluation on the target path, and iteratively updating the DQN network model based on the performance evaluation results. The present invention adopts the ResNet network model to process environmental images, enhances the perception and utilization of environmental features, and further improves the accuracy and response speed of the drone swarm path planning. By adopting the DQN network model and experience replay mechanism, each drone is given the ability to make autonomous decisions, ensuring that each drone can autonomously navigate to the target location and avoid collisions with other drones. At the same time, the adaptability and accuracy of path planning are improved, so that the drone swarm can operate stably in a complex and changeable environment with limited communication, thereby solving the technical problem of low adaptability and accuracy of path planning under limited communication conditions in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0025] Figure 1 is a flow chart of an optional method for UAV swarm path planning in a communication-constrained environment according to an embodiment of the present invention;

[0026] Figure 2 is a schematic diagram of an optional drone swarm path planning system in a communication-constrained environment according to an embodiment of the present invention;

[0027] Figure 3 is a schematic diagram of an optional UAV swarm path planning device in a communication-restricted environment according to an embodiment of the present invention;

[0028] Figure 4 It is a hardware structure block diagram of an electronic device (or mobile device) for a method for drone swarm path planning in a communication-restricted environment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] It should be noted that the drone swarm path planning method and device in a communication-restricted environment in the present disclosure can be used in the drone technology field when path planning is performed on a drone swarm in a communication-restricted environment, and can also be used in any field other than the drone technology field when path planning is performed on a drone swarm in a communication-restricted environment. The present disclosure does not limit the application field of the drone swarm path planning method and device in a communication-restricted environment.

[0032] The following embodiments of the present invention can be applied to systems / applications / devices for drone swarm path planning in various communication-restricted environments. The present invention deploys a drone swarm in a communication-restricted environment, collects environmental images of the communication-restricted environment, and then extracts features of the environmental images by using a ResNet network model to obtain an environmental feature vector. Finally, the target path of the drone swarm is obtained by inputting the environmental feature vector into a trained DQN network model. The environmental images are processed by using a ResNet network model to enhance the perception and utilization of environmental features, thereby improving the accuracy and response speed of the drone swarm path planning. By using a DQN network model and an experience playback mechanism, each drone is given the ability to make autonomous decisions, ensuring that each drone can autonomously navigate to the target location and avoid collisions with other drones. At the same time, the adaptability and accuracy of path planning are improved, so that the drone swarm can operate stably in a communication-restricted and complex and changeable environment.

[0033] The present invention is described in detail below in conjunction with various embodiments.

[0034] Embodiment 1

[0035] According to an embodiment of the present invention, an embodiment of a method for path planning of a drone swarm in a communication-restricted environment is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0036] Figure 1 is a flow chart of an optional method for UAV group path planning in a communication-constrained environment according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0037] Step S101, deploying a drone swarm in a communication-restricted environment to collect environmental images of the communication-restricted environment.

[0038] A communication restricted environment refers to an environment where the freedom and efficiency of communication are restricted under certain conditions. This environment may be caused by a variety of factors, including but not limited to policy restrictions and geographical location restrictions. For example, during the college entrance examination, an institution will turn on wireless signal jammers at various test sites to affect the normal communication of customers in the test site area and surrounding areas, such as the inability to make and receive calls, the inability to access the Internet, and the failure to send and receive messages; for geographical location restrictions, in remote areas or underground parking lots where signal coverage is incomplete, IoT cards are likely to lose connection due to weak signals. This is because the propagation of signals is restricted by ground infrastructure, such as insufficient construction density of base stations, resulting in weak signal coverage or no signal coverage in some areas.

[0039] The drone swarm path planning strategy of the embodiment of the present invention can be applied to various communication restricted environments. The embodiment of the present invention first deploys a drone swarm in a communication restricted environment, collects environmental images, then extracts features from the environmental images, and inputs the extracted feature vectors into the trained DNS network model to obtain the target path of the drone swarm.

[0040] Optionally, the step of deploying a swarm of drones in a communication-restricted environment and collecting environmental images of the communication-restricted environment includes: deploying a swarm of drones in the communication-restricted environment, collecting environmental images through sensors and camera modules on the drones, and preprocessing and normalizing the environmental images, wherein the preprocessing includes removing noise and abnormal pixels from the environmental images, and the abnormal pixels include: pixels with sudden brightness changes, pixels with distorted colors, and pixels that are inconsistent with the expected physical environment.

[0041] In an embodiment of the present invention, the environmental image is collected by the sensor and camera module on the drone. The environmental image can be a gray image or a color image. The environmental image collected by the embodiment of the present invention is an RGB image. After the environmental image is collected, it is preprocessed and normalized. The noise and abnormal pixels of the environmental image are removed during the preprocessing. The abnormal pixels include pixels with sudden brightness changes, pixels with color distortion, and pixels inconsistent with the expected physical environment. The normalization process normalizes all pixel values ​​of the environmental image to the range of [0,1]. The environmental image after preprocessing and normalization will be input into the ResNet network model to extract the environmental feature vector.

[0042] Step S102: Use the ResNet network model to extract features from the environment image to obtain an environment feature vector.

[0043] The ResNet network is a deep neural network that solves the gradient vanishing and gradient exploding problems in the deep network training process by introducing the concept of residual connection, effectively deepening the depth of the network. Traditional convolutional neural networks or fully connected networks will more or less have problems of information loss and loss during information transmission, and will also cause gradient vanishing or gradient explosion, making it impossible to train very deep networks. The ResNet network solves this problem to a certain extent by proposing the idea of ​​residual learning. In an embodiment of the present invention, the collected environmental image is input into the ResNet network model, the features of the environmental image are extracted, and the environmental feature vector is obtained. The environmental feature vector will be input into the DQN network model to obtain the target path of the drone swarm.

[0044] Optionally, the step of extracting features from the environment image using a ResNet network model to obtain an environment feature vector includes: storing the environment image in a drone memory, and expanding the environment image data set using a data enhancement strategy; defining a structure of the ResNet network model, the structure including a network depth; inputting the expanded environment image data set into the ResNet network model to obtain an environment feature vector, wherein the ResNet network model includes: a convolutional layer, an activation function, a pooling layer, and a fully connected layer.

[0045] In the embodiment of the present invention, before inputting the environment image into the ResNet network model, it is necessary to perform data enhancement processing on the environment image, and expand the environment image data set and enhance the diversity of the data through rotation, scaling, and cropping. Then, the structure of the ResNet network model is defined, including determining the depth of the ResNet network model. In the embodiment of the present invention, the depth of the ResNet network model is set to 100 layers.

[0046] The ResNet network model also includes convolutional layers, activation functions, pooling layers, and fully connected layers. By inputting the environment image into the ResNet network model, the environment features are gradually extracted through the forward propagation of the convolutional layers, activation functions, pooling layers, and fully connected layers. By utilizing the residual connection in the ResNet network model, the stability and performance of the deep neural network are improved, and the problems of gradient vanishing and gradient exploding in the deep neural network are avoided. The formula of the residual connection is expressed as:

[0047] Output=ReLU(x+Conv2D(ReLU(Conv2D(x)));

[0048] Among them, Output is the output information, x is the input information, Con2D represents the convolution operation, and ReLU is the activation function.

[0049] The last convolutional layer of the ResNet network model is connected to the global average pooling layer. By applying global average pooling, a three-dimensional feature map is obtained. After being processed by the last fully connected layer, the three-dimensional feature map is converted into a two-dimensional feature vector. This two-dimensional feature vector is used as the environment feature vector and input into the DQN network model to obtain the target path of the drone swarm.

[0050] Step S103, input the environmental feature vector into the trained DQN network model to obtain the target path of the drone swarm.

[0051] The DQN network is a neural network that combines deep learning and reinforcement learning. Its main idea is to use neural networks to approximate the Q-value function and maximize long-term cumulative rewards through interaction with the environment. The definition of the Q-value function is the basis of the DQN network, which represents the value of taking an action in a given state. The basic principle of the DQN network is to estimate the value of taking action a in state s by learning a value function Q(s,a), and then select the action that maximizes the Q value to achieve the optimal strategy.

[0052] In an embodiment of the present invention, by inputting the environmental feature vector into the DQN network model that has been trained in a simulation environment, the target path of the drone swarm can be obtained. In an actual communication-restricted environment, the drone swarm flies according to the target path, which can realize the autonomous navigation function of the drones and avoid collisions between drones and other drones or obstacles.

[0053] Optionally, after obtaining the target path of the drone swarm, the method includes: continuously monitoring the DQN network model, collecting flight data of the drone swarm, abnormal conditions, and environmental data of the communication restricted environment, wherein the flight data includes: obstacle avoidance success rate, the difference between the flight time and the expected time, and the abnormal conditions include: the drone fails to effectively avoid obstacles and the drone flight path is long.

[0054] In an embodiment of the present invention, after obtaining the target path of the drone group, the DQN network model is continuously monitored, and the DQN network model of the embodiment of the present invention can be improved by collecting the flight data, abnormal conditions and environmental data of the drone group in the communication-restricted environment. Among them, the flight data includes the obstacle avoidance success rate and the difference between the flight time and the expected time. These data can be used to evaluate the effectiveness of the current path planning strategy; abnormal conditions include the failure of the drone to effectively avoid obstacles and the lengthy flight path of the drone. These abnormal conditions can be used to indicate the optimization direction of the DQN network model; the environmental data of the communication-restricted environment can reflect the problems of increased obstacle avoidance failures and longer flight time when the drone is operating in a new or changing environment. By recording these data, the optimization direction of the path planning strategy can be indicated to adapt to the new environment.

[0055] The embodiments of the present invention ensure that the path planning strategy of the drone swarm can be continuously optimized by continuously monitoring and iteratively improving the DQN network model in practical applications, thereby continuously improving the adaptability and accuracy of path planning.

[0056] Step S104, wherein the DQN network model is trained in a communication-restricted simulation environment, and the training process includes: initializing the simulation environment and the DQN network model; based on the experience replay mechanism, alternately training the main network and the target network of the DQN network model to obtain a target path, wherein the update frequency of the target network is lower than the update frequency of the main network; performing performance evaluation on the target path, and iteratively updating the DQN network model based on the performance evaluation result.

[0057] In an embodiment of the present invention, the DQN network model is trained in a simulation environment with limited communication. During the training process, the simulation environment and the DQN network model are first initialized, and then the main network and the target network of the DQN network model are alternately trained based on the experience replay mechanism to obtain the target path. The DQN network model uses two neural networks with the same structure but different parameters, one is the main network for predicting Q values, and the other is the target network for calculating the target Q value. The weights of the target network are regularly copied from the main network, but the update frequency is much lower than that of the main network. This approach helps to reduce the correlation between the target Q value and the current Q value, thereby improving the stability of training.

[0058] After alternately training the main network and the target network and reaching the convergence condition, the target path of the unmanned cluster is obtained, and then the performance of the target path is evaluated. The DQN network model is iteratively updated based on the performance evaluation results to optimize the DQN network model. The optimized DQN network model can be used in actual communication-constrained environments to realize autonomous navigation of the drone swarm.

[0059] Optionally, the step of initializing the simulation environment and the DQN network model includes: starting the simulation software, establishing an urban environment model, establishing a flight area in the urban environment model, and deploying a swarm of drones in the flight area, wherein the urban environment model includes at least buildings and roads; defining the structure of the DQN network model, the structure including an input layer, multiple hidden layers, and an output layer; defining hyperparameters of the DQN network model, the hyperparameters including a learning rate, a discount factor, and an exploration rate; defining a loss function of the DQN network model; defining an experience replay buffer, the experience replay buffer being used to store drone state tuples, the drone state tuples including the current state of the drone, the current action, the current reward, and the new state.

[0060] In an embodiment of the present invention, when initializing the simulation environment, the simulation software is first started to establish an urban environment model, a flight area is established in the urban environment model, and a swarm of drones is deployed in the flight area. The urban environment model is used to simulate the real environment, and the flight area simulates the flight area deployed for the swarm of drones in reality.

[0061] When initializing the DQN network model, the structure of the DQN network model is first defined, including an input layer, multiple hidden layers and an output layer, wherein the input layer is used to receive the environmental feature vector, the hidden layer is used to process the vector, the hidden layer contains an activation function, and the embodiment of the present invention sets the activation function to the ReLU function, and the output layer is used to predict the action value and output the target Q value.

[0062] Redefine the hyperparameters of the DQN network model, including the learning rate α, the discount factor γ, and the exploration rate ε. The three hyperparameters are used to update the Q value so that the Q value gradually increases during the training process.

[0063] Then define the loss function L of the DQN network model to measure the difference between the current Q value and the target Q value. The expression of the loss function is:

[0064]

[0065] Among them, L(θ) is the loss function, θ represents the DQN network, and s t is the current state, a t is the current action, r t is in state s t Execute a t The current reward obtained by the action, γ is the discount factor, s t+1 is the new state (i.e., the next state), θ target Represents the target network of the DQN network model, maxQ(s t+1 θ target ) is in the next state s t+1 The maximum Q value of all possible actions under the target network θ target It is calculated that Q(s t , a t ; θ) is the current network prediction in state s t Next, perform action a t Q value, E represents the mathematical expectation value.

[0066] Initializing the DQN network model also includes defining an experience replay buffer, which is used to store drone state tuples. The drone state tuple includes: the current state of the drone, the current action, the current reward, and the new state. The experience replay buffer is specifically expressed as:

[0067] D={(s 1 , a 1 , r 1 ,s 2 ), ..., M};

[0068] Among them, s 1 is the current state of the drone, a 1is the current action of the drone, r 1 is the current reward of the drone, s 2 is the new state of the drone, and M is the capacity of the experience replay buffer. After initializing the simulation environment and the DQN network model, you can start training the DQN network model.

[0069] It should be noted that s t and t+1 are the current state and the new state respectively. State s represents the current position information of the UAV in a communication-restricted environment and the surrounding environment perception data, including but not limited to the coordinates of the UAV, flight speed, flight altitude, and distance to obstacles. t is the current action. Action a represents the operations that the drone can perform in the current state, including but not limited to adjusting the flight direction, changing the flight speed, and changing the flight altitude. t is in state s t Execute a t The current reward obtained by the action. The reward r represents the reward obtained by the drone after executing the current action. The reward value can be positive or negative. For example, when the reward indicator is whether to approach the target point, the reward value for the drone to approach the target after executing the current action can be positive, and the reward value for the drone to move away from the target after executing the current action can be negative; when the reward indicator is whether to avoid obstacles, the reward value for the drone to avoid obstacles after executing the current action can be positive, and the reward value for the drone to collide with obstacles after executing the current action can be negative.

[0070] Optionally, based on the experience replay mechanism, the steps of alternately training the main network and the target network of the DQN network model include: obtaining the environmental feature vector of the simulation environment, and storing the preprocessed environmental feature vector in the memory of the drone; inputting the current state of the drone into the main network in the DQN network model, executing the current action, and obtaining the new state and current reward of the drone; inputting the new state into the target network in the DQN network model, calculating the Q values ​​of all new actions, and determining the target Q value, wherein the Q value is the value of the new action or the current action; updating the loss function based on the target Q value, the current reward, and the Q value in the current state; updating the DQN network parameters using the gradient descent strategy; alternately training the main network and the target network, and when the convergence conditions are met, determining the target predicted Q value, and obtaining the target path corresponding to the target predicted Q value.

[0071] The experience replay mechanism is an important component module in the DQN network model. This mechanism sets up an experience replay buffer, stores past experience data (including the current state of the drone, current action, current reward, and new state) in a buffer, and randomly extracts experience data for training during the training process, thereby improving data utilization and reducing the correlation between data.

[0072] In an embodiment of the present invention, when alternately training the main network and the target network of the DQN network model, the environmental feature vector of the simulation environment is first obtained, the preprocessed environmental feature vector is stored in the drone memory, and the environmental feature vector is input into the DQN network model in the drone memory to realize the training of the DQN network model.

[0073] During the training process, first the current state of the drone s t Input the main network, the main network performs the current action a t , returns a new state s t+1 and the current reward r t , and then the new state s t+1 Input the target network and calculate in the new state s t+1 The Q values ​​of all possible actions (i.e., all new actions) are calculated, and then the largest Q value among all Q values ​​is selected as the target Q value, which is maxQ(s t+1 θ target ), which represents the value in the new state s t+1 The maximum value that can be obtained under the condition of the training is obtained, and then the target Q value, the current reward, and the Q value in the current state are input into the loss function L(θ), the loss function is updated, and the DQN network parameters are updated by adopting the gradient descent strategy to realize the iterative update of the DQN network. Finally, when the convergence condition is met, the last target Q value is determined as the target predicted Q value, and the target path corresponding to the target predicted Q value is obtained as the optimal planning path. After training, the structure and hyperparameters of the DQN network model are determined, and the DQN network model is used in applications in actual scenarios.

[0074] Optionally, the step of performing a performance evaluation on the target path and adjusting the DQN network model based on the performance evaluation results includes: defining performance evaluation indicators, the performance evaluation indicators at least including: a path length indicator, a flight time indicator, and a safety indicator; based on the performance evaluation results, adjusting the hyperparameters of the DQN network model, the update frequency of the target network, and the capacity of the experience replay buffer.

[0075] In an embodiment of the present invention, the target path can also be performance evaluated during the training process of the DQN network model, and the DQN network model can be adjusted based on the performance evaluation results to optimize the DQN network model. The performance evaluation indicators include path length indicators, flight time indicators, and safety indicators. Based on the performance evaluation results, the hyperparameters of the DQN network model, the update frequency of the target network, and the capacity of the experience replay buffer can be adjusted.

[0076] The embodiments of the present invention can balance the updating frequency of the DQN network model by adjusting the learning rate. If the learning rate is too large, the DQN network model may not meet the convergence condition. If the learning rate is too small, the convergence speed of the DQN network model may be very slow. By adjusting the discount factor, the reward in the current state and the reward in the new state can be balanced, so that the DQN network model pays more attention to the short-term or long-term value. By adjusting the exploration rate, the optimization direction of the DQN network model can be controlled. By adjusting the updating frequency of the target network, the stability of the DQN network model can be enhanced. By adjusting the capacity of the experience replay buffer, the efficiency and stability of the DQN network model can be improved.

[0077] Figure 2 is a schematic diagram of an optional drone group path planning system in a communication-constrained environment according to an embodiment of the present invention, Figure 2 In the paper, the UAV swarm path planning system in a communication-constrained environment includes three modules: data acquisition and preprocessing module, environment perception and state assessment module, and path planning and decision-making module.

[0078] The data acquisition and preprocessing module is used to collect detailed urban 3D model data and the initial position information of the drone in the communication-restricted environment, and receive the environmental image data collected by the drone during flight in real time. This module realizes the digital modeling of the communication-restricted environment and performs necessary data preprocessing operations. At the same time, this module is also responsible for converting the environmental image data into environmental feature vectors, providing high-quality input data for subsequent environmental perception, state assessment and path planning.

[0079] The environment perception and state assessment module is used to collect environmental images in real time through the sensors and cameras carried by the drone, namely RGB environmental images. The ResNet deep learning model is used to extract features from the collected environmental images to obtain environmental feature vectors. This module evaluates the current state of the drone and environmental image data, including obstacle recognition, target positioning, and position perception of other drones, to achieve autonomous obstacle avoidance and navigation of the drone in a communication-restricted environment.

[0080] The path planning and decision-making module is used to perform path planning and decision-making based on the current state of the drone and the environmental feature vector using the DQN network model. Each drone has an independent path planning and decision-making module, which uses an independent control strategy and does not require a central control unit for centralized control. By combining reinforcement learning networks and deep learning networks, dynamic path planning is achieved to ensure that the drone can autonomously navigate to the target location while avoiding collisions with other drones. At the same time, by applying the experience replay mechanism and target network, the module can continuously optimize the path planning strategy to achieve the best path planning effect.

[0081] The above-mentioned UAV swarm path planning method in a communication-constrained environment provides the steps of deploying a UAV swarm in a communication-constrained environment, collecting environmental images of the communication-constrained environment, and then extracting features of the environmental images by using a ResNet network model to obtain an environmental feature vector. Finally, the target path of the UAV swarm is obtained by inputting the environmental feature vector into a trained DQN network model. The DQN network model is trained in a communication-constrained simulation environment. During the training process, the simulation environment and the DQN network model are first initialized, and then based on the experience replay mechanism, the main network and the target network of the DQN network model are alternately trained to obtain the target path. Finally, the performance of the target path is evaluated, and the DQN network model is iteratively updated based on the performance evaluation results to obtain a DQN network model with better performance.

[0082] The embodiment of the present invention uses the ResNet network model to process environmental images, thereby enhancing the perception and utilization of environmental features, and further improving the accuracy and response speed of the path planning of the drone swarm. By adopting the DQN network model and the experience replay mechanism, each drone is given the ability to make autonomous decisions, ensuring that each drone can autonomously navigate to the target location and avoid collisions with other drones. At the same time, dynamic path planning is realized, and the adaptability and accuracy of path planning are improved, so that the drone swarm can operate stably in a communication-restricted and complex and changeable environment, thereby solving the technical problem of low adaptability and accuracy of path planning under communication-restricted conditions in related technologies.

[0083] The following is a detailed description in conjunction with another embodiment.

[0084] Embodiment 2

[0085] A drone swarm path planning device in a communication-restricted environment provided in this embodiment includes multiple implementation units, each implementation unit corresponding to each implementation step in the above-mentioned embodiment one.

[0086] Figure 3 is a schematic diagram of an optional UAV group path planning device in a communication-constrained environment according to an embodiment of the present invention, such as Figure 3 As shown, the drone group path planning device in the communication restricted environment may include: a collection unit 30, an extraction unit 31, an acquisition unit 32, and a training unit 33.

[0087] The collection unit 30 is used to deploy a group of drones in a communication-restricted environment and collect environmental images of the communication-restricted environment.

[0088] The extraction unit 31 is used to extract features from the environment image using a ResNet network model to obtain an environment feature vector.

[0089] The acquisition unit 32 is used to input the environmental feature vector into the trained DQN network model to obtain the target path of the drone swarm.

[0090] The training unit 33 trains the DQN network model in a communication-restricted simulation environment, and the training process includes: initializing the simulation environment and the DQN network model; based on the experience replay mechanism, alternately training the main network and the target network of the DQN network model to obtain the target path, wherein the update frequency of the target network is lower than the update frequency of the main network; performing performance evaluation on the target path, and iteratively updating the DQN network model based on the performance evaluation result.

[0091] The above-mentioned drone swarm path planning device in a communication-restricted environment can deploy a drone swarm in a communication-restricted environment through a collection unit 30, collect environmental images of the communication-restricted environment, and then use a ResNet network model to extract features of the environmental images through an extraction unit 31 to obtain an environmental feature vector. Finally, the environmental feature vector is input into a trained DQN network model through an acquisition unit 32 to obtain a target path of the drone swarm, and the DQN network model is trained through a training unit 33. The DQN network model is trained in a communication-restricted simulation environment, and the training process includes: initializing the simulation environment and the DQN network model; based on an experience replay mechanism, alternately training the main network and the target network of the DQN network model to obtain a target path, performing a performance evaluation on the target path, and iteratively updating the DQN network model based on the performance evaluation result. This embodiment uses the ResNet network model to process environmental images, thereby enhancing the perception and utilization of environmental features, and further improving the accuracy and response speed of the drone swarm path planning. By adopting the DQN network model and experience replay mechanism, each drone is given the ability to make autonomous decisions, ensuring that each drone can autonomously navigate to the target location and avoid collisions with other drones. At the same time, the adaptability and accuracy of path planning are improved, so that the drone swarm can operate stably in a complex and changeable environment with limited communication, thereby solving the technical problem of low adaptability and accuracy of path planning under limited communication conditions in related technologies.

[0092] Optionally, the extraction unit 31 includes: an expansion module, used to store the environmental image in the drone memory, and use a data enhancement strategy to expand the environmental image data set; a first definition module, used to define the structure of the ResNet network model, the structure including the network depth; an input module, used to input the expanded environmental image data set into the ResNet network model to obtain an environmental feature vector, wherein the ResNet network model includes: a convolutional layer, an activation function, a pooling layer, and a fully connected layer.

[0093] Optionally, the training unit 33 includes: a startup module, used to start the simulation software, establish an urban environment model, establish a flight area in the urban environment model, and deploy a drone swarm in the flight area, wherein the urban environment model includes at least buildings and roads; a second definition module, used to define the structure of the DQN network model, the structure includes an input layer, multiple hidden layers and an output layer; a third definition module, used to define the hyperparameters of the DQN network model, the hyperparameters include a learning rate, a discount factor and an exploration rate; a fourth definition module, used to define the loss function of the DQN network model; a fifth definition module, used to define an experience replay buffer, the experience replay buffer is used to store drone state tuples, the drone state tuples include: the current state of the drone, the current action, the current reward and the new state.

[0094] Optionally, the training unit 33 also includes: a storage module, which is used to obtain the environmental feature vector of the simulation environment and store the preprocessed environmental feature vector in the drone memory; an execution module, which is used to input the current state of the drone into the main network in the DQN network model, execute the current action, and obtain the new state and current reward of the drone; a calculation module, which is used to input the new state into the target network in the DQN network model, calculate the Q values ​​of all new actions, and determine the target Q value, where the Q value is the value of the new action or the current action; a first update module, which updates the loss function based on the target Q value, the current reward and the Q value in the current state; a second update module, which is used to update the DQN network parameters using a gradient descent strategy; a determination module, which is used to alternately train the main network and the target network, and when the convergence conditions are met, determine the target predicted Q value and obtain the target path corresponding to the target predicted Q value.

[0095] Optionally, the training unit 33 also includes: a sixth definition module, used to define performance evaluation indicators, the performance evaluation indicators include at least: a path length indicator, a flight time indicator, and a safety indicator; an adjustment module, based on the performance evaluation results, adjusts the hyperparameters of the DQN network model, the update frequency of the target network, and the capacity of the experience replay buffer.

[0096] Optionally, the acquisition unit 30 includes: a processing module, used to deploy a drone swarm in a communication-restricted environment, collect environmental images through sensors and camera modules on the drones, and preprocess and normalize the environmental images, wherein the preprocessing includes removing noise and abnormal pixels from the environmental images, and the abnormal pixels include: pixels with sudden brightness changes, pixels with distorted colors, and pixels that are inconsistent with the expected physical environment.

[0097] Optionally, the acquisition unit 32 includes: a monitoring module, which is used to continuously monitor the DQN network model, collect flight data of the drone group, abnormal conditions and environmental data of the communication restricted environment, wherein the flight data includes: obstacle avoidance success rate, the difference between the flight time and the expected time, and the abnormal conditions include: the drone fails to effectively avoid obstacles and the drone flight path is long.

[0098] The above-mentioned drone swarm path planning device in the communication restricted environment may also include a processor and a memory. The above-mentioned collection unit 30, extraction unit 31, acquisition unit 32, training unit 33, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize the corresponding functions.

[0099] The above processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the path planning of the drone group can be realized in a communication-restricted environment by adjusting the kernel parameters.

[0100] The above-mentioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one storage chip.

[0101] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the method for drone group path planning in a communication restricted environment of any one of the above-mentioned embodiments one.

[0102] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors implement the drone swarm path planning method in a communication restricted environment of any one of the above-mentioned embodiments one.

[0103] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the drone swarm path planning method in a communication-restricted environment described in each embodiment of the present application.

[0104] The present application also provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the drone group path planning method in a communication restricted environment described in each embodiment of the present application are implemented.

[0105] Figure 4 1 is a hardware structure block diagram of an electronic device (or mobile device) for a method for planning a path for a drone swarm in a communication-restricted environment according to an embodiment of the present invention. Figure 4 As shown, the electronic device may include one or more ( Figure 4 402a, 402b, ..., 402n are used to illustrate) processor 402 (processor 402 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), memory 404 for storing data. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a keyboard, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 4 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 4 More or fewer components as shown, or with Figure 4 Different configurations are shown.

[0106] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0107] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0109] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0110] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0111] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-On lyMemory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0112] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for UAV swarm path planning in a communication-constrained environment, characterized in that: include: Deploy a swarm of drones in a communication-restricted environment to collect environmental images of the communication-restricted environment; Using a ResNet network model to extract features from the environment image to obtain an environment feature vector; Inputting the environmental feature vector into the trained DQN network model to obtain the target path of the drone swarm; The DQN network model is trained in a communication-restricted simulation environment, and the training process includes: initializing the simulation environment and the DQN network model; based on an experience replay mechanism, alternately training a main network and a target network of the DQN network model to obtain a target path, wherein the update frequency of the target network is lower than that of the main network; performing performance evaluation on the target path, and iteratively updating the DQN network model based on the performance evaluation result.

2. The path planning method according to claim 1, characterized in that: The step of extracting features from the environment image using a ResNet network model to obtain an environment feature vector includes: The environment image is stored in the memory of the drone, and a data enhancement strategy is used to expand the environment image data set; Defining the structure of the ResNet network model, wherein the structure includes network depth; The expanded environment image data set is input into the ResNet network model to obtain the environment feature vector, wherein the ResNet network model includes: a convolutional layer, an activation function, a pooling layer and a fully connected layer.

3. The path planning method according to claim 1, characterized in that: The steps of initializing the simulation environment and the DQN network model include: Starting simulation software, establishing an urban environment model, establishing a flight area in the urban environment model, and deploying the drone swarm in the flight area, wherein the urban environment model at least includes: buildings and roads; Defining the structure of the DQN network model, the structure comprising: an input layer, a plurality of hidden layers and an output layer; Defining hyperparameters of the DQN network model, wherein the hyperparameters include: learning rate, discount factor, and exploration rate; Defining the loss function of the DQN network model; An experience replay buffer is defined, where the experience replay buffer is used to store drone state tuples, where the drone state tuples include: the current state of the drone, the current action, the current reward, and the new state.

4. The path planning method according to claim 3, characterized in that: The steps of alternately training the main network and the target network of the DQN network model based on the experience replay mechanism include: Obtaining an environmental feature vector of the simulation environment, and storing the preprocessed environmental feature vector in a memory of the drone; Input the current state of the drone into the main network in the DQN network model, execute the current action, and obtain the new state and current reward of the drone; Input the new state into the target network in the DQN network model, calculate the Q value of all new actions, and determine the target Q value, wherein the Q value is the value of the new action or the current action; Based on the target Q value, the current reward, and the Q value in the current state, updating the loss function; Adopting a gradient descent strategy to update the DQN network parameters; The main network and the target network are trained alternately, and when convergence conditions are met, a target predicted Q value is determined, and the target path corresponding to the target predicted Q value is obtained.

5. The path planning method according to claim 1, characterized in that: The step of performing a performance evaluation on the target path and adjusting the DQN network model based on the performance evaluation result includes: Defining performance evaluation indicators, the performance evaluation indicators at least including: path length indicator, flight time indicator, and safety indicator; Based on the performance evaluation result, the hyperparameters of the DQN network model, the update frequency of the target network, and the capacity of the experience replay buffer are adjusted.

6. The path planning method according to claim 1, characterized in that: The steps of deploying a drone swarm in a communication-restricted environment and collecting an environmental image of the communication-restricted environment include: The drone swarm is deployed in the communication restricted environment, the environment image is collected through sensors and camera modules on the drones, and the environment image is preprocessed and normalized, wherein the preprocessing includes removing noise and abnormal pixels of the environment image, and the abnormal pixels include: pixels with sudden brightness changes, pixels with distorted colors, and pixels inconsistent with the expected physical environment.

7. The path planning method according to claim 1, characterized in that: After obtaining the target path of the drone group, it includes: Continuously monitor the DQN network model, collect flight data, abnormal conditions, and environmental data of the communication-restricted environment of the drone group, wherein the flight data includes: obstacle avoidance success rate, the difference between the flight time and the expected time, and the abnormal conditions include: the drone fails to effectively avoid obstacles and the drone flight path is long.

8. A drone group path planning device in a communication-restricted environment, characterized in that: include: A collection unit, used to deploy a drone swarm in a communication-restricted environment and collect environmental images of the communication-restricted environment; An extraction unit, used to extract features from the environment image using a ResNet network model to obtain an environment feature vector; An acquisition unit, used for inputting the environmental feature vector into the trained DQN network model to obtain the target path of the drone swarm; A training unit, wherein the DQN network model is trained in a communication-constrained simulation environment, and the training process includes: initializing the simulation environment and the DQN network model; based on an experience replay mechanism, alternately training a main network and a target network of the DQN network model to obtain a target path, wherein an update frequency of the target network is lower than an update frequency of the main network; and performing a performance evaluation on the target path, and iteratively updating the DQN network model based on a performance evaluation result.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the drone group path planning method in a communication restricted environment as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the drone swarm path planning method in a communication restricted environment as described in any one of claims 1 to 7.

Images