Artificial intelligence-based multi-unmanned aerial vehicle navigation method, device, equipment and medium

Abstract

The application mainly relates to the technical field of intelligent robots, and discloses a multi-unmanned aerial vehicle navigation method and device based on artificial intelligence, equipment and a medium, which comprises the following steps: acquiring current environment information of the position of each unmanned aerial vehicle, wherein the current environment information comprises position information of a target object, depth image information and current position information of each unmanned aerial vehicle; updating the moving route of each unmanned aerial vehicle based on the position information of the target object and the current position information of each unmanned aerial vehicle; and when the unmanned aerial vehicle moves to the target object according to the updated moving route, generating the corresponding obstacle avoidance strategy of the unmanned aerial vehicle based on the depth image information by using a preset obstacle avoidance strategy model, so that the unmanned aerial vehicle executes the obstacle avoidance strategy and avoids obstacles. The application can improve the navigation capability of the unmanned aerial vehicle and efficiently avoid obstacles during navigation.

Description

Technical Field

[0001] This invention relates to the field of intelligent robot technology, specifically to a multi-UAV navigation method, device, equipment, and medium based on artificial intelligence. Background Technology

[0002] With the rapid development of unmanned aerial vehicle (UAV) technology, UAVs are being applied in various fields, including agriculture, environmental monitoring, search and rescue, mining, and patrol inspection, due to their high flexibility, wide applicability, and economic controllability. In these applications, effective collaboration among multiple UAVs, especially when large-scale UAV swarms are performing missions, is crucial, as finding the optimal path to avoid obstacles and reach the target location is essential. However, traditional UAV obstacle avoidance methods rely on simultaneous localization and mapping (SMR) technology, which typically requires significant computational resources and is limited by prior map information, resulting in insufficient adaptability to unknown environments. To adapt to unknown environments and reduce reliance on computational resources, new UAV navigation and obstacle avoidance strategies need to be developed. Summary of the Invention

[0003] This invention provides a multi-UAV navigation method, device, equipment, and medium based on artificial intelligence, which can improve the navigation capabilities of UAVs and efficiently avoid obstacles during navigation.

[0004] This invention provides a multi-UAV navigation method based on artificial intelligence, the method comprising:

[0005] Obtain the current environmental information of each drone's location, including the target object's location information, depth image information, and the current location information of each drone;

[0006] Based on the location information of the target object and the current location information of each drone, the movement route of each drone is updated respectively;

[0007] When the drone moves toward the target object according to the updated movement route, an obstacle avoidance strategy corresponding to the drone is generated based on the depth image information using a preset obstacle avoidance strategy model, so that the drone executes the obstacle avoidance strategy and avoids obstacles.

[0008] Furthermore, the aforementioned AI-based multi-UAV navigation method also includes:

[0009] Based on the location information of the target object and the current location information of the drone, an artificial potential field is generated between the drone and the target object;

[0010] The artificial potential field is used to control the drone to move in the direction of the target object's location.

[0011] Furthermore, the obstacle avoidance strategy includes a second speed used by the UAV to avoid obstacles;

[0012] The aforementioned AI-based multi-UAV navigation method also includes:

[0013] As the drone moves, its first velocity is determined based on the gravitational effect of the artificial potential field.

[0014] The current speed of the drone is updated based on the first speed and the second speed.

[0015] Furthermore, the obstacle avoidance strategy model includes a first model network and a second model network;

[0016] When the drone moves toward the target object according to the updated movement route, based on the depth image information, a corresponding obstacle avoidance strategy is generated for the drone using a preset obstacle avoidance strategy model, so that the drone executes the obstacle avoidance strategy and avoids obstacles, including:

[0017] Based on the current environmental information, an observation vector is generated;

[0018] After the observation vector is input into the first model network, the first model network generates an action vector, and the obstacle avoidance strategy is generated based on the action vector.

[0019] After the action vector is input into the second model network, the score value corresponding to the action vector is determined by the second model network.

[0020] Based on the score, an optimized action vector is generated through the first model network, and the obstacle avoidance strategy is updated according to the optimized action vector.

[0021] Furthermore, generating the observation vector based on the current environmental information includes:

[0022] Depth images are acquired using the drone;

[0023] A spatial vector corresponding to the depth image is generated using a preset encoder;

[0024] Obtain the direction vector pointing to the target object and the current speed of the drone;

[0025] The observation vector is generated based on the spatial vector, the direction vector, and the current velocity.

[0026] Furthermore, the AI-based multi-UAV navigation method also includes training the obstacle avoidance strategy model;

[0027] Training the obstacle avoidance strategy model includes:

[0028] Generate multiple reward values;

[0029] A reward function is constructed based on each reward value and a preset weight coefficient corresponding to each reward value.

[0030] When the obstacle avoidance strategy model is trained according to a preset training dataset, feedback data of the obstacle avoidance strategy model training process is generated according to the reward function, and the obstacle avoidance strategy model is optimized according to the feedback data.

[0031] Furthermore, the generation of multiple reward values ​​includes:

[0032] Obtain the first distance value between the drone and the target object;

[0033] If a decrease in the first distance value is detected, a positive first reward value is generated; if an increase in the first distance value is detected, a negative first reward value is generated.

[0034] Obtain a second distance value between the drone and the obstacle. When the second distance is within a preset range, generate a negative second reward value.

[0035] The present invention also provides a multi-UAV navigation device based on artificial intelligence, the device comprising:

[0036] The acquisition module is used to acquire the current environmental information of the location of each drone, including the location information of the target object, depth image information, and the current location information of each drone.

[0037] The update module is used to update the movement route of each drone based on the location information of the target object and the current location information of each drone.

[0038] The generation module is used to generate an obstacle avoidance strategy corresponding to the UAV based on the depth image information and using a preset obstacle avoidance strategy model when the UAV moves toward the target object according to the updated movement route, so that the UAV executes the obstacle avoidance strategy and avoids obstacles.

[0039] The present invention also provides an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the artificial intelligence-based multi-UAV navigation method as described in any of the preceding claims.

[0040] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the AI-based multi-UAV navigation method as described in any of the preceding claims.

[0041] The present invention has at least the following beneficial effects:

[0042] In this technical solution, firstly, the UAV acquires current environmental information through sensors, including the location of the target object, depth image information, and the UAV's own current position. This real-time perception capability enables the UAV to navigate and avoid obstacles directly based on the current environment without pre-built maps. Then, based on the acquired target object location and UAV current position information, the movement route of each UAV is updated in real time. This dynamic path planning ensures that the UAV can adjust its flight trajectory according to the latest environmental information, thereby improving the flexibility and accuracy of navigation. Finally, when the UAV moves towards the target object according to the updated movement route, it generates corresponding obstacle avoidance strategies based on the depth image information using a pre-set obstacle avoidance strategy model. These strategies guide the UAV to effectively avoid obstacles during flight, avoiding the limitations of relying on prior map information and enabling the UAV to flexibly adapt to environmental changes. This technical solution enables UAVs to achieve efficient navigation and obstacle avoidance in complex environments, improving their adaptability and mission completion rate in unknown or dynamically changing environments. Attached Figure Description

[0043] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0044] Figure 1 This is a flowchart illustrating the steps of a multi-UAV navigation method based on artificial intelligence in this embodiment;

[0045] Figure 2 This is a flowchart of step S103 in a multi-UAV navigation method based on artificial intelligence in this embodiment;

[0046] Figure 3 This is a schematic diagram of the obstacle avoidance strategy generation model in an artificial intelligence-based multi-UAV navigation method according to this embodiment;

[0047] Figure 4 This is a flowchart of step S201 in an artificial intelligence-based multi-UAV navigation method of this embodiment;

[0048] Figure 5 This is a flowchart illustrating the depth image processing process in an artificial intelligence-based multi-UAV navigation method according to this embodiment.

[0049] Figure 6 This is a flowchart illustrating the steps involved in training an obstacle avoidance strategy model in an AI-based multi-UAV navigation method according to this embodiment.

[0050] Figure 7 This is a schematic diagram of the structure of a multi-UAV navigation device based on artificial intelligence according to this embodiment;

[0051] Figure 8 This is a schematic diagram of the structure of an electronic device. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] In this field, traditional multi-UAV obstacle avoidance methods primarily rely on real-time simultaneous localization and mapping (SLAM) technology. These methods typically use sensors such as LiDAR to perceive the surrounding environment and generate flight trajectories through path planning algorithms. To improve the performance of SLAM systems, prior map information is often introduced to assist UAV navigation in known environments. However, these traditional methods have some limitations. They usually require significant computational resources, which may be impractical in resource-constrained environments. Furthermore, these methods are limited by existing prior map information and have poor adaptability to unknown or dynamically changing environments. The paper "A UAV Navigation Approach Based on Deep Reinforcement Learning in Large Cluttered 3D Environments" proposes a UAV navigation method based on deep reinforcement learning, which achieves end-to-end obstacle avoidance navigation without an environment map. This method utilizes UAV sensor data, including ranging sensor data in seven horizontal and five vertical directions, as well as UAV attitude and target information, to train a reinforcement learning policy. Furthermore, a fast cyclic stochastic gradient algorithm was designed to solve reinforcement learning models online, thereby improving the obstacle avoidance capabilities of UAVs in complex environments. This method directly maps raw perception information into navigation signals, avoiding the memory and computational costs required for real-time map building, thus improving the navigation efficiency of UAVs in unknown environments.

[0054] However, the researchers in this application found that the aforementioned existing technical solutions mainly rely on data from seven horizontal and five vertical ranging sensors mounted on the UAV to identify obstacles. This method performs reasonably well when dealing with obstacles of specific shapes, but its recognition capability is insufficient when faced with diverse obstacle shapes, making it difficult to apply broadly to all types of obstacles. Since the UAV's obstacle perception primarily relies on data from the seven horizontal and five vertical ranging sensors, this configuration works well for recognizing obstacles with regular shapes such as cuboids, but its ability to recognize irregular shapes such as triangular pyramids or slender obstacles is insufficient, which limits the UAV's obstacle avoidance capabilities in complex environments.

[0055] When the navigation target changes, the navigation performance of the UAV, including its ability to find the shortest path and mission success rate, can significantly decline. This indicates that the flexibility and robustness of the existing technologies described above in adapting to target changes need improvement. In reinforcement learning training, the UAV's navigation target is pre-set and fixed. The UAV trains an optimal deep reinforcement learning strategy for this fixed target. However, once the target changes, the original strategy may fail to adapt to the new target, leading to decreased navigation performance and increased failure rate. This demonstrates that the UAV's navigation strategy lacks high flexibility and adaptability.

[0056] Within the existing technological framework described above, drones cannot independently acquire the position information of other drones, which limits their ability to achieve distributed control. Instead, they rely on a centralized control center to acquire and coordinate the positions of all drones in real time, which can lead to inefficiency and response delays in large-scale operations. Drones primarily acquire information about obstacles ahead through ranging sensors, but this method cannot acquire the position information of other drones. Therefore, a centralized controller is needed to acquire the positions of all drones in real time to achieve effective cooperative control. Furthermore, the existing technologies described above do not consider obstacle avoidance between drones, requiring additional control commands to be assigned to each drone to avoid mutual interference.

[0057] Training deep reinforcement learning policies for drones requires a considerable amount of time to achieve policy convergence. This time-consuming process can delay the deployment of drones in practical applications, especially in situations requiring rapid adaptation to new environments or tasks. The reward value of a reinforcement learning model is calculated based on the path connecting the start and end points and ranging sensor data, rather than the optimal path from the drone's current position to the end point. This results in the drone being unable to effectively find the optimal path based on its current position after obstacle avoidance, thus prolonging the learning process. This limits the drone's ability to quickly adapt and make decisions in dynamic environments.

[0058] To address the aforementioned technical problems, this application proposes the following embodiments, which can improve the navigation capabilities of drones and efficiently avoid obstacles during flight.

[0059] Please refer to Figure 1 , Figure 1 This is a flowchart of the steps of a multi-UAV navigation method based on artificial intelligence in this embodiment.

[0060] This embodiment provides a multi-UAV navigation method based on artificial intelligence, including:

[0061] S101. Obtain the current environmental information of each drone's location. The current environmental information includes the target object's location information, depth image information, and the current location information of each drone.

[0062] S102. Based on the location information of the target object and the current location information of each UAV, update the movement route of each UAV.

[0063] S103. When the UAV moves toward the target object according to the updated movement route, an obstacle avoidance strategy corresponding to the UAV is generated based on the depth image information and using the preset obstacle avoidance strategy model, so that the UAV can execute the obstacle avoidance strategy and avoid obstacles.

[0064] In step S101 of some embodiments, the drone may be equipped with various sensors, such as GPS, LiDAR, visual sensors, and ultrasonic sensors. These sensors provide information such as position, obstacle detection, and altitude. By integrating the data acquired by different sensors through a data fusion algorithm, the current environmental information of the drone's location can be obtained. The current environmental information includes detailed information about obstacles, such as their location and size.

[0065] In step S102 of some embodiments, a dynamic window method (DWA) is used for local path planning. Given the current state and the dynamic window, all possible movements are evaluated, and an optimal movement strategy is selected to allow the UAV to avoid obstacles and move towards the target. Alternatively, improved A* or RRT algorithms can be used for path planning. These algorithms can dynamically adjust the flight path based on real-time updated environmental information and the UAV's state, ensuring the UAV reaches the target point safely and efficiently.

[0066] In step S102 of some embodiments, in a multi-UAV cooperative mission, each UAV can utilize a distributed path planning algorithm. For example, the Hybrid A* algorithm can be used for distributed path planning. Each UAV independently plans its path based on its own target point and environmental information, and continuously detects obstacles and collision risks during flight. If a risk exists, a safe flight speed is calculated based on a three-dimensional speed obstacle model, and the path is adjusted accordingly.

[0067] In some embodiments, an AI-based multi-drone navigation method further includes:

[0068] Based on the location information of the target object and the current location information of the drone, an artificial potential field is generated between the drone and the target object; through the gravitational effect of the artificial potential field, the drone is controlled to move in the direction of the target object's location.

[0069] Furthermore, the drone's navigation speed is calculated by combining its current position, target position, and the attractive force of the artificial potential field to determine its velocity (velocity1) toward the target. The specific calculation method is as follows:

[0070]

[0071]

[0072] in, Indicates attractiveness, Indicates the attractiveness coefficient. Indicates the coordinates of the target point. This represents the drone's coordinates, and V represents its velocity. This represents the distance between two coordinate points.

[0073] Understandably, through the gravitational effect of an artificial potential field, drones can move quickly and accurately toward the location of their target, reducing the time spent searching for targets in complex environments and improving overall navigation efficiency. Furthermore, the basic principle of the artificial potential field method is intuitive; it guides drone movement by simulating the interaction of gravity and repulsion, resulting in a simple algorithm structure that is easy to implement in drone systems. In addition, the artificial potential field method does not require pre-generating a complete path; instead, it calculates the drone's direction of motion in real time based on the current potential field state. This results in relatively low computational complexity, meeting the real-time control requirements of drones and handling dynamic obstacle environments well. When the position or state of obstacles changes, the drone can quickly adjust its flight path based on the new potential field information, exhibiting strong dynamic adaptability. Moreover, the generated path is relatively smooth, avoiding frequent sharp turns or abrupt changes, which helps the drone maintain a stable flight attitude, reducing energy consumption and flight time.

[0074] In some embodiments, an AI-based multi-drone navigation method further includes:

[0075] When the drone moves, a first velocity is determined based on the gravitational effect of the artificial potential field; the drone's current velocity is updated based on the first and second velocities; the obstacle avoidance strategy includes the second velocity used by the drone to avoid obstacles.

[0076] It is understood that this embodiment combines the velocity calculated by the artificial potential field method with the velocity and parameters generated by the reinforcement learning model to enable the UAV to complete obstacle avoidance and navigation tasks.

[0077] One specific implementation of this embodiment is to use the speed of artificial potential field calculation as the first speed, the speed of reinforcement learning model generation as the second speed, and update the current speed of the drone based on the first and second speeds:

[0078]

[0079] in, It is the speed of the drone at the next moment. These are the parameters generated by the reinforcement learning model. The velocity is calculated from the artificial potential field. It refers to the speed at which reinforcement learning models are generated.

[0080] It is understood that this embodiment effectively combines control navigation speed and obstacle avoidance speed through a parameter fusion strategy to improve the obstacle avoidance and navigation capabilities of the UAV in unknown environments. It also possesses high generalization ability, enabling it to adapt to various unknown environments and providing the UAV with a flexible and effective obstacle avoidance and navigation solution. By effectively combining control navigation speed and obstacle avoidance speed, the UAV can autonomously learn and optimize its obstacle avoidance strategy in complex environments while maintaining navigation efficiency towards targets.

[0081] In some embodiments, the obstacle avoidance strategy model includes a first model network and a second model network.

[0082] Please refer to Figure 2 , Figure 2 This is a flowchart of step S103 in a multi-UAV navigation method based on artificial intelligence in this embodiment.

[0083] In some embodiments, step S103 includes:

[0084] S201. Generate observation vectors based on current environmental information.

[0085] S202. After inputting the observation vector into the first model network, the first model network generates action vectors, and an obstacle avoidance strategy is generated based on the action vectors.

[0086] S203. After inputting the action vector into the second model network, the score value corresponding to the action vector is determined through the second model network.

[0087] S204. Based on the score, generate an optimized action vector through the first model network, and update the obstacle avoidance strategy according to the optimized action vector.

[0088] In some embodiments, the obstacle avoidance strategy model is a policy model based on the DQN algorithm, comprising an evaluation network and a target network. The first model network is the policy network, and the second model network is the value function network. The first model network generates action vectors. It takes observation vectors as input and outputs the Q-value for each possible action. During training, the parameters of the evaluation network are continuously updated using gradient descent to better approximate the true Q-value. The second model network determines the score value corresponding to the action vector. The parameters of the target network are not updated in real time but are periodically copied from the evaluation network. This is done to stabilize the training process and reduce network training instability caused by frequent changes in the target Q-value.

[0089] Specifically, observation vectors are generated based on current environmental information, including the current position and speed of the UAV, the position of the target object, and the position of obstacles.

[0090] The observation vector is input into the evaluation network (first model network). The evaluation network outputs the Q-value for each possible action and selects the action with the highest Q-value as the current action vector.

[0091] The action vector is input into the target network (second model network). The target network is used to calculate the score of the action vector, which is the expected reward value in the current state.

[0092] The parameters of the evaluation network are updated based on the score, a loss function (such as mean squared error) is used to measure the difference between the predicted Q value and the target Q value, and optimization is performed using gradient descent.

[0093] When the score meets the preset conditions, the optimized motion vector is output.

[0094] The obstacle avoidance strategy is determined based on the optimized motion vectors. These motion vectors contain the optimal obstacle avoidance actions for the UAV in the current environment, such as adjusting flight direction and speed. The UAV executes this obstacle avoidance strategy to avoid obstacles and move towards the target object.

[0095] Please refer to Figure 3 , Figure 3 This is a schematic diagram of the obstacle avoidance strategy generation model in an artificial intelligence-based multi-UAV navigation method of this embodiment.

[0096] In some embodiments, the obstacle avoidance strategy model is a policy model based on the Soft Actor-Critic algorithm, comprising a policy network (Actor Network) and a value function network (Critic Networks). The first model network is the policy network, and the second model network is the value function network. In one specific embodiment, both the first and second model networks adopt a three-layer structure, with each layer containing a 1024-dimensional hidden layer, forming a multilayer perceptron (MLP). The output of the policy network is the parameters of a Gaussian action distribution; for example, it outputs a 4-dimensional vector corresponding to the three control velocities of the UAV. A speed control parameter These parameters, including the mean and variance, define the probability distribution of the UAV's action choices when performing obstacle avoidance and navigation tasks. The value function network outputs an evaluation of that action to train the policy.

[0097] It is understood that this embodiment generates a corresponding obstacle avoidance strategy for each UAV through an obstacle avoidance strategy model. It does not rely on the centralized controller to allocate control commands separately, nor does it rely on the global coordinates of other UAVs. It is treated like other obstacles in the environment and does not require additional obstacle avoidance algorithms.

[0098] Please refer to Figure 4 , Figure 4 This is a flowchart of step S201 in an artificial intelligence-based multi-UAV navigation method of this embodiment.

[0099] In some embodiments, step S201 includes:

[0100] S301. Acquire depth images using drones.

[0101] S302. Generate the spatial vector corresponding to the depth image using a preset encoder.

[0102] S303, Obtain the direction vector pointing to the target object and the current speed of the drone.

[0103] S304. Generate observation vectors based on spatial vectors, direction vectors, and current velocity.

[0104] In this embodiment, the drone is equipped with a depth camera capable of capturing depth images in front, providing distance information to surrounding objects. Using a regularized autoencoder (RAE), the drone can process depth images captured at three consecutive time points. The RAE encodes these images into a 50-dimensional spatial vector. Then, the 50-dimensional spatial vector , Direction vector pointing to the target object and current speed Integrate to form a comprehensive observation vector .

[0105] In some embodiments, this observation vector can be used as input to a Soft Actor-Critic (SAC) algorithm to train the drone's obstacle avoidance navigation strategy.

[0106] It is understandable that when generating spatial vectors corresponding to depth images using a preset encoder, the image goes through a process of encoding it into spatial vectors and then reconstructing these representations back into the original image using a decoder. This process effectively extracts and highlights key visual information in the environment relevant to the drone's mission, thereby enhancing the drone's ability to understand its surroundings.

[0107] Please refer to Figure 5 , Figure 5 This is a flowchart illustrating the depth image processing process in an artificial intelligence-based multi-UAV navigation method according to this embodiment.

[0108] like Figure 5 As shown, in an encoder-decoder architecture, after a depth image is input, the encoder transforms the original image into a more easily processed feature representation, while the decoder may then reconstruct these features into an image form. During this process, spatial vectors are extracted.

[0109] It is understood that this embodiment utilizes a regularized autoencoder to encode the depth image in front of the UAV, extracting a low-dimensional representation vector from the image. This process effectively transforms complex environmental information into a concise latent representation vector, enabling the UAV to automatically identify and filter visual features closely related to the obstacle avoidance task. Since the environmental information is acquired through images, the UAV can learn obstacle avoidance capabilities for obstacles of any shape. By combining an autoencoder and depth images, this invention significantly improves the generalization ability of UAV obstacle avoidance.

[0110] Please refer to Figure 6 , Figure 6 This is a flowchart illustrating the steps involved in training an obstacle avoidance strategy model in an AI-based multi-UAV navigation method according to this embodiment.

[0111] In some embodiments, an AI-based multi-UAV navigation method further includes a step of training an obstacle avoidance strategy model, with specific implementations including:

[0112] S401, Generate multiple reward values.

[0113] S402. Construct a reward function based on each reward value and the preset weight coefficient corresponding to each reward value.

[0114] S403. When training the obstacle avoidance strategy model based on the preset training dataset, feedback data of the obstacle avoidance strategy model training process is generated according to the reward function, and the obstacle avoidance strategy model is optimized based on the feedback data.

[0115] In some embodiments, step S401 includes:

[0116] Obtain the first distance value between the drone and the target object; if a decrease in the first distance value is detected, generate a positive first reward value; if an increase in the first distance value is detected, generate a negative first reward value.

[0117] In some embodiments, step S401 includes:

[0118] Obtain a second distance value between the drone and the obstacle. When the second distance value is within a preset range, generate a negative second reward value.

[0119] In some embodiments, step S401 includes:

[0120] When the drone is detected to be in a stopped state, a negative third reward value is generated.

[0121] In one specific implementation, the obstacle avoidance strategy model is a strategy model based on the Soft Actor-Critic algorithm. During the training of the obstacle avoidance strategy model, the drone's reward calculation involves three key aspects: the distance to the target, the distance to the obstacle, and its own motion state. When the drone approaches the target point, it receives a positive reward; conversely, it receives a negative reward. When the drone is within a certain range of the obstacle, it receives a negative reward. When the drone stops moving, it receives a negative reward.

[0122]

[0123]

[0124]

[0125]

[0126]

[0127] in, It is the reward value for moving towards the target. It is the distance the drone moves relative to the target within one time step. It is the change in distance moved relative to the target. It is the attraction coefficient. It is the repulsive force exerted by the i-th obstacle point on the drone. It is the repulsive force coefficient. It is the minimum distance at which the repulsive force formula takes effect. It is a safe distance. It is the distance between two coordinate points. It is the direction between two coordinate points. It is the reward value for getting closer to an obstacle. The drone receives reward points when it stops moving. and is the reward value coefficient, and r is the reward value obtained by the drone after performing the action.

[0128] Understandably, this embodiment utilizes a reward function to train the obstacle avoidance strategy model. This function incorporates the principle of artificial potential fields, effectively reducing the learning time of deep reinforcement learning by combining linear target movement reward values ​​with nonlinear obstacle avoidance reward values. This design not only improves learning efficiency but also enhances the adaptability and flexibility of the UAV in complex environments.

[0129] In any of the above method embodiments, firstly, the UAV acquires current environmental information through sensors, including the location of the target object, depth image information, and the UAV's own current location. This real-time perception capability enables the UAV to navigate and avoid obstacles directly based on the current environment without pre-built maps. Then, based on the acquired target object location and UAV current location information, the movement route of each UAV is updated in real time. This dynamic path planning ensures that the UAV can adjust its flight trajectory according to the latest environmental information, thereby improving the flexibility and accuracy of navigation. Finally, when the UAV moves towards the target object according to the updated movement route, a corresponding obstacle avoidance strategy is generated based on the depth image information using a preset obstacle avoidance strategy model. These strategies guide the UAV to effectively avoid obstacles during flight, avoiding the limitations of relying on prior map information, and enabling the UAV to flexibly adapt to environmental changes. This technical solution enables UAVs to achieve efficient navigation and obstacle avoidance in complex environments, improving their adaptability and mission completion rate in unknown or dynamically changing environments.

[0130] Please refer to Figure 7 , Figure 7 This is a schematic diagram of the structure of a multi-UAV navigation device based on artificial intelligence in this embodiment.

[0131] This embodiment also provides a multi-UAV navigation device based on artificial intelligence, including:

[0132] The acquisition module 501 is used to acquire the current environmental information of the location of each drone. The current environmental information includes the location information of the target object, the depth image information, and the current location information of each drone.

[0133] The update module 502 is used to update the movement route of each drone based on the location information of the target object and the current location information of each drone.

[0134] The generation module 503 is used to generate an obstacle avoidance strategy corresponding to the UAV based on depth image information and a preset obstacle avoidance strategy model when the UAV moves toward the target object according to the updated movement route, so that the UAV can execute the obstacle avoidance strategy and avoid obstacles.

[0135] It will be understood by those skilled in the art that all or some of the steps and apparatuses in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. As is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0136] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0137] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement any of the above-mentioned AI-based multi-UAV navigation methods.

[0138] refer to Figure 8 , Figure 8 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0139] The processor 601 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0140] The memory 602 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 602 can store operating devices and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 602 and is called and executed by the processor 601 using the artificial intelligence-based multi-UAV navigation method of the embodiments of this application.

[0141] The input / output interface 603 is used to implement information input and output;

[0142] The communication interface 604 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0143] Bus 605 transmits information between various components of the device (e.g., processor 601, memory 602, input / output interface 603, and communication interface 604);

[0144] The processor 601, memory 602, input / output interface 603, and communication interface 604 are connected to each other within the device via bus 605.

[0145] It is understood that the content of the above method embodiments is applicable to the embodiments of this electronic device. The specific functions implemented by the embodiments of this electronic device are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0146] This application also provides a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to implement the AI-based multi-UAV navigation method as described in any of the above specific embodiments.

[0147] This application also discloses a computer program product, including a computer program or computer instructions, which are stored in a computer-readable storage medium. The processor of a computer device reads the computer program or computer instructions from the computer-readable storage medium and executes the computer program or computer instructions, causing the computer device to perform the AI-based multi-UAV navigation method as described in any of the preceding embodiments.

[0148] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0149] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices. It should be understood that in this application, “at least one” means one or more, and “more than one” means two or more.

[0150] In the several embodiments provided in this application, it should be understood that the disclosed apparatus, devices, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0151] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0152] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0153] If the integrated unit is implemented as 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 this application, 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. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0154] Although the description of this application has been quite detailed and particularly focused on several of the described embodiments, it is not intended to limit itself to any of these details or embodiments or any particular embodiment. Rather, it should be considered as effectively covering the intended scope of this application by referring to the appended claims and taking into account the prior art, which provides for a broad possible interpretation of these claims. Furthermore, the foregoing description of this application with respect to embodiments foreseeable by the inventors is intended to provide a useful description, and non-substantial modifications to this application that have not yet been foreseen may still represent equivalent modifications.

Claims

1. A multi-UAV navigation method based on artificial intelligence, characterized in that, The method includes: Obtain the current environmental information of each drone's location, including the target object's location information, depth image information, and the current location information of each drone; Based on the location information of the target object and the current location information of each drone, the movement route of each drone is updated respectively; Based on the location information of the target object and the current location information of the drone, an artificial potential field is generated between the drone and the target object; The artificial potential field is used to control the drone to move in the direction of the target object's location; When the UAV moves toward the target object according to the updated movement route, an obstacle avoidance strategy corresponding to the UAV is generated based on the depth image information using a preset obstacle avoidance strategy model, so that the UAV executes the obstacle avoidance strategy and avoids obstacles; the obstacle avoidance strategy includes a second speed adopted by the UAV to avoid obstacles; The method further includes: As the drone moves, its first velocity is determined based on the gravitational effect of the artificial potential field. The current speed of the drone is updated based on the first speed and the second speed, wherein the speed calculated by the artificial potential field is used as the first speed and the speed generated by the reinforcement learning model is used as the second speed. The first velocity is calculated based on the current position, the target position, and the attractive force of the artificial potential field.

2. The multi-UAV navigation method based on artificial intelligence according to claim 1, characterized in that, The obstacle avoidance strategy model includes a first model network and a second model network; When the drone moves toward the target object according to the updated movement route, based on the depth image information, a corresponding obstacle avoidance strategy is generated for the drone using a preset obstacle avoidance strategy model, so that the drone executes the obstacle avoidance strategy and avoids obstacles, including: Based on the current environmental information, an observation vector is generated; After the observation vector is input into the first model network, the first model network generates an action vector, and the obstacle avoidance strategy is generated based on the action vector. After the action vector is input into the second model network, the score value corresponding to the action vector is determined by the second model network. Based on the score, an optimized action vector is generated through the first model network, and the obstacle avoidance strategy is updated according to the optimized action vector.

3. The multi-UAV navigation method based on artificial intelligence according to claim 2, characterized in that, The step of generating the observation vector based on the current environmental information includes: Depth images are acquired using the drone; A spatial vector corresponding to the depth image is generated using a preset encoder; Obtain the direction vector pointing to the target object and the current speed of the drone; The observation vector is generated based on the spatial vector, the direction vector, and the current velocity.

4. The multi-UAV navigation method based on artificial intelligence according to claim 1, characterized in that, The method also includes training the obstacle avoidance strategy model; Training the obstacle avoidance strategy model includes: Generate multiple reward values; A reward function is constructed based on each reward value and a preset weight coefficient corresponding to each reward value. When the obstacle avoidance strategy model is trained according to a preset training dataset, feedback data of the obstacle avoidance strategy model training process is generated according to the reward function, and the obstacle avoidance strategy model is optimized according to the feedback data.

5. The multi-UAV navigation method based on artificial intelligence according to claim 4, characterized in that, The generation of multiple reward values ​​includes: Obtain the first distance value between the drone and the target object; If a decrease in the first distance value is detected, a positive first reward value is generated; if an increase in the first distance value is detected, a negative first reward value is generated. Obtain a second distance value between the drone and the obstacle. When the second distance is within a preset range, generate a negative second reward value.

6. A multi-UAV navigation device based on artificial intelligence, characterized in that, The device includes: The acquisition module is used to acquire the current environmental information of the location of each drone, including the location information of the target object, depth image information, and the current location information of each drone. The update module is used to update the movement route of each UAV based on the location information of the target object and the current location information of each UAV; specifically, based on the location information of the target object and the current location information of the UAV, an artificial potential field is generated between the UAV and the target object; through the gravitational effect of the artificial potential field, the UAV is controlled to move in the direction of the target object's location; The generation module is used to generate an obstacle avoidance strategy corresponding to the UAV based on the depth image information and using a preset obstacle avoidance strategy model when the UAV moves toward the target object according to the updated movement route, so that the UAV executes the obstacle avoidance strategy and avoids obstacles; the obstacle avoidance strategy includes a second speed adopted by the UAV to avoid obstacles; As the drone moves, its first velocity is determined based on the gravitational effect of the artificial potential field. The current speed of the drone is updated based on the first speed and the second speed, wherein the speed calculated by the artificial potential field is used as the first speed and the speed generated by the reinforcement learning model is used as the second speed. The first velocity is calculated based on the current position, the target position, and the attractive force of the artificial potential field.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the artificial intelligence-based multi-UAV navigation method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the AI-based multi-UAV navigation method as described in any one of claims 1 to 5.

Images