An unmanned aerial vehicle cluster autonomous cooperative control method and system

Abstract

The present application relates to a kind of unmanned aerial vehicle cluster autonomous cooperative control method, first identification static barrier and dynamic barrier and mark risk level;Second target is obscured, each unmanned aerial vehicle not being obscured with infrared camera tracking, obscured switching prediction mode, adopts distributed interactive multiple model algorithm to predict trajectory and fuses data;Based on artificial potential field method and height layer division strategy, avoid static barrier, different response strategies are taken for different risk dynamic barriers;Dynamic distribution is each unmanned aerial vehicle role, and task priority is adjusted according to positioning error and barrier distance, each unmanned aerial vehicle reports state in real time, and local re-planning is carried out when deviating from planned path exceeds threshold value;Finally, the role and path allocation strategy are optimized through reinforcement learning;This method improves task execution efficiency and cooperative ability, solves the problem of each unmanned aerial vehicle cluster in target positioning and dynamic avoidance in dense urban environment, and improves cooperative adaptive capacity.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a method and system for autonomous and collaborative control of UAV swarms. Background Technology

[0002] In today's complex urban environment, the demand for multi-drone collaborative operations is growing, such as in urban security monitoring and disaster relief. However, the urban environment is riddled with obstacles, including static buildings and trees, as well as dynamic vehicles and pedestrians, posing a significant challenge to the safe flight of each drone. Traditional single-drone obstacle avoidance methods are ill-suited to the complex and ever-changing urban environment and cannot fully leverage the advantages of multi-drone collaboration. While existing drone collaborative obstacle avoidance technologies consider obstacle information sharing to some extent, their comprehensive processing capabilities for both static and dynamic obstacles are limited, and they suffer from deficiencies in tracking and predicting when targets are occluded, making it difficult to meet the requirements of efficient and accurate collaborative obstacle avoidance in real-world missions.

[0003] With the continuous development of drone technology, collaborative obstacle avoidance among multiple drones has become a research hotspot. Currently, some collaborative obstacle avoidance methods are relatively fixed in role allocation and path planning, lacking dynamic adaptability. In actual flight, the positioning error of each drone and its distance from obstacles will constantly change. Fixed task priority allocation methods cannot respond to these changes in a timely manner, which may lead to untimely obstacle avoidance or low task execution efficiency. At the same time, existing role allocation and path allocation strategies are mostly based on preset rules, lacking self-learning and optimization capabilities, making it difficult to achieve optimal collaborative effects in different scenarios. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by proposing a more intelligent, flexible, and self-optimizing method for collaborative control of multiple unmanned aerial vehicles (UAVs).

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: S1: Each drone uses LiDAR point cloud stitching to generate a local 3D map of the city and marks static obstacles in the local 3D map; it uses millimeter-wave radar combined with Kalman filtering technology to identify dynamic obstacles and mark their risk level. S2, when the target is blocked, each unblocked drone continues to track the target's heat source information through infrared cameras, while each blocked drone switches to prediction mode, uses a distributed interactive multi-model algorithm to predict the target's trajectory, and integrates the prediction data of each drone. S3 uses an artificial potential field method combined with a height layer division strategy to avoid static obstacles; different response strategies are adopted for dynamic obstacles with different risks. S4 dynamically assigns roles to each drone and adjusts task priorities based on the positioning error of the target and the distance to obstacles; S5: Each drone reports its status information to the cluster control center in real time; if the actual path deviates from the planned path by more than a certain threshold, local replanning is triggered. S6 uses reinforcement learning to determine the role assignment rules and path assignment strategies for each drone.

[0006] Preferably, S2 specifically includes: Infrared cameras are continuously used to capture heat source information of the target, and image processing technology is used to extract the heat source area of ​​the target from the images captured by the cameras. When the target is detected to be occluded, each occluded UAV immediately switches from tracking mode to prediction mode. Multiple motion models are established to describe the different possible motion states of the target. The probability of each model is updated according to Bayes' theorem, and the state estimates of each model are fused according to the model probabilities to obtain the final target state estimate. Each occluded UAV sends its predicted target trajectory data to the cluster control center. The fusion center uses a weighted average method to fuse the prediction data of each UAV.

[0007] Preferably, step S4 adjusts the task priority based on the positioning error of the positioning target and the distance to obstacles, including: Each drone is assigned an initial role based on its performance parameters, and its performance parameters are continuously monitored. A comprehensive scoring function is set based on these parameters to evaluate the overall capabilities of each drone. The comprehensive capability scores of each drone are compared, and the roles of each drone are adjusted based on the comparison results. Each drone obtains the location information of the target through its own positioning system, and simultaneously uses other auxiliary means to reconfirm the target's location. Each drone uses sensors to monitor the distance to surrounding obstacles in real time and records the distance between each drone and the nearest obstacle. A task priority evaluation function is constructed by setting positioning error weights and obstacle distance weights. Different tasks are ranked according to the value of the task priority evaluation function P. The auxiliary means used are visual recognition or laser ranging.

[0008] Preferably, step S4, dynamically assigning roles to each drone, specifically includes: The system simultaneously receives signals from three systems, calculates the intensity ratio of direct and reflected signals in each frequency band in real time, and determines signal attenuation based on building material factors. Based on signal attenuation, when a reflected signal is detected to be stronger than the direct signal in a certain frequency band, it automatically switches to the optimal anti-multipath frequency band. It records positioning drift vectors from historical flights, stores typical error vectors categorized by motion state, and generates a reverse compensation vector based on the current velocity direction and acceleration matching database to compensate and correct the original positioning data. It marks static obstacles in the city's 3D map with 3D coordinates, clarifying their type and characteristics. It scans static obstacles along the working path of each UAV using LiDAR, compares the actual coordinates of the identified static obstacles with the pre-stored coordinates, and performs position calibration for each UAV based on the offset. It calculates the positioning reliability coefficient for each UAV in real time, dynamically adjusts the safety distance based on the positioning reliability coefficient, and determines whether identity switching is necessary. It summarizes the positioning error vectors of each UAV, generates a cluster spatial offset heatmap, and marks high-error-risk areas.

[0009] Preferably, the compensation and correction of the original positioning data includes: During each flight of each drone, the positioning drift vector is recorded using a positioning system. Based on the different motion states of each drone, typical error vectors are categorized and stored in the corresponding database. The current velocity direction and acceleration information of each drone are acquired in real time and matched against the motion states in the database. The error vector corresponding to the closest motion state is then found. Generate reverse compensation vector ; the reverse compensation vector Compared with the original location data The data is overlaid to obtain the compensated positioning data. .

[0010] Preferably, the dynamically adjusted safety distance is calculated using the following formula: , This is the baseline safety distance. It is an adjustment factor; The formula for the location credibility coefficient is: .

[0011] Preferably, the positioning drift vector includes: Each drone in the swarm collects its own electromagnetic signals and information on static and dynamic obstacles, extracts reflection interference features, and converts them into quantifiable phase offset vectors. The swarm shares information and collaboratively calibrates positioning data, using historical data after collaborative calibration to analyze the intrinsic relationship between the reflection interference features of different obstacles and the positioning drift error of each drone. Each drone monitors positioning drift in real time based on quantified relationships and provides early warnings. Based on the early warning signals, each swarm formulates corresponding preventative strategies. Based on the phase offset vectors of reflection interference detected by each drone and the overall collaborative strategy of the swarm, an attitude collaborative adjustment algorithm for the swarm's position is designed. Each drone adjusts its flight position and fuselage attitude according to the calculated position and attitude adjustment parameters, while simultaneously controlling the electromagnetic wave transmitting equipment to emit a reverse electromagnetic beam with a specific phase and amplitude. A dynamic cancellation effectiveness evaluation model is established to evaluate the effect of the reverse electromagnetic beam in canceling reflection interference in real time. Based on the evaluation results of the dynamic cancellation effectiveness evaluation model, the control parameters in the attitude collaborative adjustment algorithm for the swarm's position are optimized in real time.

[0012] Preferably, the step of extracting reflection interference features and converting them into quantifiable phase shift vectors specifically involves: Let the phase of the normal signal be... The phase of the interfered signal is Then the phase shift A coordinate system is established with each UAV as the origin. The direction of the phase offset is determined based on the positions of static and dynamic obstacles. The magnitude and direction of the phase offset are combined into a quantifiable phase offset vector. , where θ is the direction angle of the phase offset.

[0013] Preferably, the sharing of information and collaborative calibration of positioning data among each drone cluster includes: The collaborative calibration positioning data consists of each UAV's individual positioning information, sensor data, and reflection interference quantification results. Each UAV cluster shares its positioning information, sensor data, and reflection interference quantification results in real time via a wireless communication network. A weighted average method is used for collaborative calibration of the positioning data. Let the positioning data of each UAV of the i-th UAV be... The weight is Then the calibrated positioning data , where n is the number of drones in each drone swarm.

[0014] Preferably, an autonomous collaborative control system for unmanned aerial vehicle (UAV) swarms, the system comprising: The perception and recognition module is used by each drone to generate a local 3D map of the city by stitching together point clouds from LiDAR, and to mark static obstacles in the local 3D map; it uses millimeter-wave radar combined with Kalman filtering technology to identify dynamic obstacles and mark their risk levels.

[0015] The tracking and prediction module is used to track the heat source information of the target location through infrared cameras when the target location is blocked. The unblocked drones continue to track the target location heat source information through infrared cameras, while the blocked drones switch to prediction mode and use a distributed interactive multi-model algorithm to predict the target location trajectory and fuse the prediction data of each drone.

[0016] The avoidance response module is used to avoid static obstacles based on the artificial potential field method and combined with the height layer division strategy; different response strategies are adopted for dynamic obstacles with different risks.

[0017] The allocation module is used to dynamically assign roles to each drone and adjust task priorities based on the positioning error of the target and the distance to obstacles.

[0018] The path planning module is used for each drone to report its status information to the cluster control center in real time; if the actual path deviates from the planned path by more than a certain threshold, local replanning is triggered.

[0019] The strategy optimization module is used to learn the role assignment rules and path assignment strategies for each drone through reinforcement learning.

[0020] The beneficial effects of this invention are: By combining lidar point cloud and millimeter-wave radar technologies, accurate identification and risk level labeling of static and dynamic obstacles were achieved. Simultaneously, the DBSCAN clustering algorithm and artificial potential field method were used to optimize the path planning and dynamic response strategies of each UAV, ensuring flight safety. Furthermore, through dynamic role allocation and information fusion at the swarm control center, the mission execution efficiency and collaborative capabilities of each UAV swarm were improved, making it suitable for efficient monitoring and obstacle avoidance tasks in various complex scenarios. This effectively solved the key challenges of target localization and dynamic avoidance of complex 3D obstacles for each UAV swarm in dense urban environments, significantly improving the collaborative adaptability of each UAV in environments with positioning errors, dynamic target changes, and complex 3D obstacles. By differentiating responses to static and dynamic obstacles, trajectory conflicts between obstacle avoidance and pursuit were reduced, ensuring that each UAV swarm could execute missions efficiently and safely in complex urban environments.

[0021] By employing technologies such as multi-band positioning fusion, dynamic inertial navigation compensation, and landmark-assisted calibration, multipath interference and inertial navigation error accumulation are effectively suppressed, significantly improving the positioning accuracy and reliability of each UAV. At the same time, based on positioning reliability, the safety distance and identity switching are dynamically adjusted, and combined with the cluster spatial offset heat map, efficient collaborative control is achieved, ensuring the safe and stable operation and mission execution of each UAV cluster in complex urban environments.

[0022] By quantizing and modeling reflected interference and converting it into a phase offset vector, the interference can be accurately quantified. By using cluster collaborative calibration of positioning data and establishing quantization relationships, positioning drift can be prevented in advance. By designing an attitude collaborative adjustment algorithm for cluster position, reverse electromagnetic beams are emitted to cancel interference, and an evaluation model is established to optimize parameters, effectively improving the positioning accuracy and collaborative control capability of each UAV cluster in complex electromagnetic environments. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating an autonomous collaborative control method for unmanned aerial vehicle (UAV) swarms according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an autonomous collaborative control system for unmanned aerial vehicle (UAV) swarms according to the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0026] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0027] Example 1: Figure 1 This is a flowchart illustrating an autonomous collaborative control method for unmanned aerial vehicle (UAV) swarms according to Embodiment 1 of the present invention, comprising the following steps: S1: Each drone uses LiDAR point cloud stitching to generate a local 3D map of the city and marks static obstacles in the local 3D map; it uses millimeter-wave radar combined with Kalman filtering technology to identify dynamic obstacles and mark their risk level.

[0028] In this application, static obstacles and dynamic obstacles are collectively referred to as obstacles.

[0029] Specifically, 1a, each drone is equipped with a lidar that scans the surrounding environment to obtain point cloud data.

[0030] 1b. The iterative nearest-point algorithm is used to register point clouds in adjacent frames to minimize the distance error between point clouds.

[0031] 1c, which fuses the point clouds after registration of multiple frames to construct a local 3D map.

[0032] 1d. Cluster analysis is performed on the local 3D map. A density-based clustering algorithm is used to divide the point cloud into different clusters, each cluster representing a potential obstacle.

[0033] The density-based clustering algorithm (DBSCAN) was chosen, with two key parameters: neighborhood radius ϵ and minimum number of points. The neighborhood radius ϵ defines the range of points surrounding a point, while the minimum number of points specifies the minimum number of points a point's neighborhood must contain to be considered a core point. Dividing the point cloud into different clusters involves calculating the number of points in the neighborhood of each point in the local 3D map, i.e., the number of points contained within a sphere centered at p with radius ϵ. If the number of points in the neighborhood is greater than or equal to the minimum number of points, then point p is marked as a core point. Starting from any core point, all points density-reachable from it are found, and these points are grouped into a cluster. This process is repeated, selecting unvisited core points, until all core points have been visited.

[0034] 1e. Determine whether the cluster is a static obstacle based on its geometric features, and record its position and boundary information in the local 3D map.

[0035] Specifically, for each cluster, its geometric features (volume, aspect ratio, etc.) are calculated. The volume can be approximated by multiplying the number of voxels occupied by the cluster by the volume of a single voxel. Judgment conditions such as volume thresholds and aspect ratio threshold ranges are set (no fixed values, set according to the working scenario of each drone swarm). If the cluster's volume exceeds the volume threshold and its aspect ratio is within a reasonable range (e.g., for common static obstacles such as buildings and utility poles, the aspect ratio follows a certain pattern), then the cluster is determined to represent a static obstacle.

[0036] For clusters identified as static obstacles, calculate their centroid coordinates. As the location information of the obstacle, , It is a cluster The number of points in the cluster is determined, with p being the core point. Simultaneously, by finding the maximum and minimum values ​​of the cluster in the x, y, and z directions, the boundary information of the obstacle is determined and recorded in the local 3D map.

[0037] 1f, millimeter-wave radar emits electromagnetic waves and receives reflected signals. It calculates the distance and velocity of the target object by measuring the round-trip time and Doppler shift of the signal.

[0038] 1g, predict the trajectory of the target object through Kalman filtering, set a risk level threshold based on the speed of the target object, and mark the risk level of dynamic obstacles in the local 3D map based on the comparison between the speed of the target object and the risk level threshold.

[0039] Based on actual application scenarios and security requirements, different speed thresholds are set for different risk levels. For example, a high-risk speed threshold is set. Medium-risk speed threshold Low-risk speed threshold The calculated velocity v of the located target object is compared with the set risk level threshold. If... If so, the target object will be marked as a high-risk dynamic obstacle; if If it is, then it is marked as a medium-risk dynamic obstacle; if If a dynamic obstacle is identified as a low-risk obstacle, its risk level information will be recorded at the corresponding location on the local 3D map.

[0040] S2, when the target is blocked, each unblocked drone continues to track the target's heat source information through infrared cameras, while each blocked drone switches to prediction mode, uses a distributed interactive multi-model algorithm to predict the target's trajectory, and integrates the prediction data of each drone.

[0041] Specifically, 2a. Continuously use infrared cameras to capture and locate target heat source information, and extract the target heat source area from the images captured by the camera through image processing technology.

[0042] The threshold segmentation formula is as follows: If a pixel value I(x,y) is greater than a set threshold T (there is no fixed value; it depends on factors such as the performance of the infrared camera, ambient lighting conditions, and the thermal radiation characteristics of the target, etc. It can be determined experimentally. Given the existence of the target, the threshold T is adjusted to accurately extract the heat source area of ​​the target), then the pixel is identified as part of the target region. , where P(x,y) indicates whether pixel (x,y) belongs to the target region.

[0043] 2b. When the target is detected to be occluded (by the disappearance of the target in the image), each occluded UAV immediately switches from tracking mode to prediction mode.

[0044] 2c. Establish multiple motion models to describe the different possible motion states of the target. Update the probability of each model according to Bayes' theorem. Then, fuse the state estimates of each model according to the model probabilities to obtain the final target state estimate.

[0045] In 2D, each of the occluded drones sends its predicted target trajectory data (target position, speed, etc.) to the cluster control center.

[0046] 2e, the fusion center uses a weighted average method to fuse the prediction data for each drone.

[0047] Specifically, assuming there are M drones, the predicted target location for the j-th drone is: The weight is (The weights can be determined based on factors such as the prediction accuracy of each drone, and) The fused location target position .

[0048] S3 uses an artificial potential field method combined with a height layer division strategy to avoid static obstacles; for dynamic obstacles with different risks, different response strategies are adopted.

[0049] Specifically, 3a. Based on the flight environment characteristics and performance requirements of each UAV, the flight space is divided into multiple height layers in the vertical direction.

[0050] For example, based on the maximum flight altitude of each drone and minimum flight altitude Set the height layer interval Then the number of height layers Center height of each altitude level .

[0051] 3b. Based on the artificial potential field method, each UAV selects the altitude layer with the smallest sum of potential fields as the flight altitude layer, according to its current position and the position of the target, combined with the potential field conditions of each altitude layer.

[0052] In the artificial potential field method, the gravitational field is directed towards the target point (the location that each drone is expected to reach within the next 10 seconds), and its strength is inversely proportional to the distance; the repulsive field is directed away from the obstacle, and its strength is inversely proportional to the square of the distance from the obstacle.

[0053] 3c. Within the selected altitude layer, calculate the resultant force on each UAV within that layer, adjust the flight direction of each UAV according to the direction of the resultant force, and gradually fly towards the positioning target point.

[0054] 3D, based on the speed of dynamic obstacles Distance to each drone And assess the risk level based on the relative motion direction.

[0055] Among them, the risk assessment function is set. , That is the maximum speed of each drone. It is a safe distance threshold (which depends on the size of each drone, its flight speed, maneuverability, and the complexity of the surrounding environment). and It is the weighting coefficient ( Adjustments can be made based on the actual situation. Different risk levels are defined based on the R value, such as low risk, medium risk, and high risk.

[0056] 3e, Develop a response strategy for each drone based on the risk level.

[0057] Specifically, when the dynamic obstacle poses a low risk, each drone maintains its current flight status, appropriately adjusts its flight speed, and maintains a safe distance from the obstacle. This speed adjustment ensures that the relative speed between each drone and the obstacle is controlled. satisfy ,in This is the safe relative speed threshold (which is related to the braking performance of each drone and the characteristics of the obstacle. If each drone has good braking performance, the safe relative speed threshold can be set higher; if the obstacle is relatively fragile, the safe relative speed threshold should be set lower).

[0058] For medium-risk dynamic obstacles, each drone fine-tunes its flight direction while maintaining a safe distance from the obstacle. Based on the obstacle's direction and speed, an obstacle avoidance direction vector is calculated, causing each drone to gradually deviate from the obstacle's path. The obstacle avoidance direction can be determined by analyzing the obstacle's movement relative to each drone; for example, if the obstacle approaches from the right side of each drone, the obstacle avoidance direction vector will point to the left of each drone at a certain angle.

[0059] When a dynamic obstacle poses a high risk, each drone immediately takes emergency obstacle avoidance measures. The current flight mission is halted, and a new path is planned using the artificial potential field method. The high-risk obstacle is treated as a temporary static obstacle, and a strong repulsive field is constructed to quickly move each drone away from the obstacle. After moving away from the obstacle, a new path is planned to reach the target location.

[0060] S4 dynamically assigns roles to each drone and adjusts task priorities based on the positioning error of the target and the distance to obstacles.

[0061] Specifically, 4a. Assign an initial role to each drone based on its performance parameters and continuously monitor the various performance parameters of each drone.

[0062] These performance parameters include, but are not limited to, flight speed, endurance, payload capacity, and sensor accuracy.

[0063] 4b. Based on the performance parameters, a comprehensive scoring function is set to evaluate the overall capabilities of each UAV.

[0064] 4c. Compare the overall capability scores of each drone and adjust the role of each drone based on the comparison results.

[0065] For example, the comprehensive capability score of each unmanned aerial vehicle is compared. If the comprehensive capability score of an unmanned aerial vehicle originally serving as a task executor is greatly improved and is higher than that of the pilot unmanned aerial vehicle, and its flight speed and endurance also meet the pilot requirements, its role is adjusted to the pilot unmanned aerial vehicle, and the original pilot unmanned aerial vehicle is adjusted to a task executor or other suitable role.

[0066] In step 4d, each unmanned aerial vehicle acquires the position information of a positioning target through its own positioning system, and secondarily confirms the position of the positioning target through other auxiliary means.

[0067] Wherein, the auxiliary means may be visual recognition or laser ranging. The difference between the position of the positioning target obtained by the self-positioning system and the position obtained by the auxiliary means is calculated as a positioning error.

[0068] For example, if the position of the positioning target obtained by GPS positioning is , the position of the positioning target positioned by the auxiliary means is , then the positioning error is .

[0069] In step 4e, each unmanned aerial vehicle uses a sensor to monitor the distance from surrounding obstacles in real time, and records the distance between each unmanned aerial vehicle and the nearest obstacle.

[0070] In step 4f, a positioning error weight and an obstacle distance weight are set, and a task priority evaluation function is constructed.

[0071] Wherein, the priority evaluation function ( is the distance between each unmanned aerial vehicle and the nearest obstacle; is the positioning error weight, which is used to measure the proportion of the positioning error in the task priority evaluation; is the obstacle distance weight, which determines the influence degree of the obstacle distance on the task priority). When is very small, in order to avoid a zero denominator, a minimum distance threshold can be set (it is used to avoid the situation of zero denominator when calculating the task priority evaluation function, and is set according to the size and safety distance requirements of each unmanned aerial vehicle, and is set to be slightly smaller than the safety distance threshold. For example, when the safety distance threshold is 10 meters, can be set to 1 meter), when <d, let .

[0072] In step 4g, different tasks are sorted according to the value P of the task priority evaluation function.

[0073] For example, a higher priority value indicates a higher priority for the task. Each drone prioritizes executing the higher-priority task. If a drone has both a target location and obstacle avoidance task, the P-values ​​of the two tasks are calculated. If the obstacle avoidance task has a higher P-value, the drone will execute the obstacle avoidance task first, pausing or adjusting the execution of the target location task.

[0074] S5: Each drone reports its status information to the cluster control center in real time; if the actual path deviates from the planned path by more than a certain threshold, local replanning is triggered.

[0075] The status information includes position, speed, and obstacle avoidance status.

[0076] Specifically, 5a. Each UAV uses its own positioning system to obtain its own coordinates in three-dimensional space, and measures the velocity components of each UAV in the three coordinate axes through a velocity sensor, thereby calculating the resultant velocity of each UAV.

[0077] Among them, the velocity components of each drone in the three coordinate axes are: Combined speed .

[0078] 5b. Each drone uses obstacle avoidance sensors to detect surrounding obstacles. If an obstacle is detected and the distance is less than a safe distance threshold, the obstacle avoidance state is marked as triggered; otherwise, it is marked as normal flight.

[0079] 5c, Each drone will collect its position (x, y, z) and velocity. The obstacle avoidance status information is packaged into a data packet and sent to the cluster control center in real time.

[0080] The cluster control center stores the planned paths for each drone, which typically consist of a series of discrete waypoints. , composition.

[0081] For each UAV, the real-time reported position (x, y, z) is calculated by the cluster control center, and the distance d from that position to the nearest path point on the planned path is calculated.

[0082] 5e, Set a path deviation threshold. If the distance to the nearest path point is greater than the path deviation threshold, a local replanning will be triggered.

[0083] The path deviation threshold is set based on the positioning accuracy of each UAV, the accuracy requirements of path planning, and the importance of the mission. Local replanning starts from the current position of each UAV and, taking into account surrounding obstacle information and positioning target point information, replans a local path so that each UAV can return to the planned path or directly reach the positioning target point.

[0084] S6 optimizes the role assignment rules and path assignment strategies for each drone through reinforcement learning.

[0085] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: By combining lidar point cloud and millimeter-wave radar technologies, accurate identification and risk level labeling of static and dynamic obstacles were achieved. Simultaneously, the DBSCAN clustering algorithm and artificial potential field method were used to optimize the path planning and dynamic response strategies of each UAV, ensuring flight safety. Furthermore, through dynamic role allocation and information fusion at the swarm control center, the mission execution efficiency and collaborative capabilities of each UAV swarm were improved, making it suitable for efficient monitoring and obstacle avoidance tasks in various complex scenarios. This effectively solved the key challenges of target localization and dynamic avoidance of complex 3D obstacles for each UAV swarm in dense urban environments, significantly improving the collaborative adaptability of each UAV in environments with positioning errors, dynamic target changes, and complex 3D obstacles. By differentiating responses to static and dynamic obstacles, trajectory conflicts between obstacle avoidance and pursuit were reduced, ensuring that each UAV swarm could execute missions efficiently and safely in complex urban environments.

[0086] Example 2 In Example 1, autonomous collaborative control of each UAV swarm in a complex environment was achieved through various technical means. However, the multipath interference effect in densely populated urban areas with tall buildings posed a significant challenge to the positioning accuracy of each UAV. Traditional positioning methods in this environment suffer from uncontrollable errors and lack real-time discrimination mechanisms for signal reflection paths and dynamic compensation measures for inertial navigation errors. To more accurately suppress multipath interference, compensate for inertial navigation errors, and further improve the positioning accuracy and collaborative control capabilities of each UAV swarm, the positioning technology needs to be optimized and improved. Therefore, the technical solution of Example 2 is proposed.

[0087] In some embodiments, each drone role is dynamically assigned, and step S4 further includes: S41 synchronously receives signals from three systems, calculates the ratio of direct to reflected signal intensity in each frequency band in real time, and determines signal attenuation based on building material factors.

[0088] The three systems use GPS (L1 / L5 band), BeiDou (B1 / B2 band), and Galileo (E1 / E5a ​​band) signals.

[0089] Specifically, 41a synchronously receives signals from GPS (L1 / L5 band), BeiDou (B1 / B2 band), and Galileo (E1 / E5a ​​band) to acquire raw signal data from different frequency bands of each system.

[0090] 41b. For each frequency band signal, signal processing techniques are used to decompose the received mixed signal into direct signal and reflected signal, and their intensities are calculated separately.

[0091] 41c, For each frequency band signal, signal processing techniques are used to decompose the received mixed signal into direct signal and reflected signal, and the direct / reflected signal intensity ratio is calculated.

[0092] in, .

[0093] 41d. Considering the material of the buildings surrounding the flight area of ​​each UAV, a building material factor is introduced. By combining the ratio of direct signal intensity to reflected signal intensity and the building material factor, a signal attenuation evaluation index is constructed to further analyze the signal attenuation situation.

[0094] The formula for the signal attenuation evaluation index is as follows: , This is a building material factor, determined based on the intensity of reflection.

[0095] S42, based on signal attenuation, automatically switches to the optimal anti-multipath frequency band when a reflected signal in a certain frequency band is detected to be stronger than the direct signal.

[0096] Among them, the strategy of switching to the optimal frequency band for multipath resistance is to prioritize the BeiDou B2 band in high-rise areas and switch to the GPS L5 band in open areas.

[0097] S43 records the positioning drift vector during historical flights, stores typical error vectors according to motion state, and generates a reverse compensation vector based on the current velocity direction and acceleration matching database to compensate and correct the original positioning data.

[0098] Specifically, 43a, during each flight of each UAV, the positioning drift vector is recorded using the positioning system, and typical error vectors are classified and stored in the corresponding database according to the different motion states of each UAV.

[0099] The format for storing the positioning drift vector is as follows: , It represents the amount of positioning drift in the x-direction, reflecting the degree of deviation between the actual position of each UAV and the position determined by the positioning system in the horizontal direction (usually understood as the east-west direction, depending on the setting of the coordinate system); This represents the positioning drift in the y-direction, corresponding to the horizontal longitudinal direction (north-south direction), reflecting the positioning error of each UAV in this direction; It is the positioning drift in the z-direction, representing the positioning deviation in the vertical direction (up and down); the motion states include, but are not limited to, uniform straight line and sharp turn.

[0100] 43b: Real-time acquisition of the current speed, direction, and acceleration information of each drone, and matching it with the motion status in the database.

[0101] 43c, find the error vector corresponding to the closest motion state. Generate reverse compensation vector .

[0102] 43d, the reverse compensation vector Compared with the original location data The data is overlaid to obtain the compensated positioning data. .

[0103] S44: Mark the three-dimensional coordinates of static obstacles in the city's three-dimensional map to clarify their type and characteristics.

[0104] S45 uses LiDAR to scan static obstacles on the working path of each drone, compares the actual coordinates of the identified static obstacles with the pre-stored coordinates, and performs position calibration for each drone based on the offset.

[0105] S46: Each drone calculates its positioning reliability coefficient in real time, dynamically adjusts the safe distance based on the positioning reliability coefficient, and determines whether an identity switch is needed.

[0106] Specifically, in 46a, each drone calculates its positioning reliability coefficient in real time based on its own positioning error.

[0107] The formula for the location reliability coefficient is as follows: .

[0108] 46b: Dynamically adjust the safety distance based on the positioning reliability coefficient, and determine whether each drone needs to switch identities based on the magnitude of the positioning reliability coefficient.

[0109] The formula for dynamically adjusting the safety distance is: , This is the baseline safety distance. It is an adjustment factor.

[0110] S47, summarizes the positioning error vectors of each machine, generates a cluster spatial offset heat map, and marks high error risk areas.

[0111] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: By employing technologies such as multi-band positioning fusion, dynamic inertial navigation compensation, and landmark-assisted calibration, multipath interference and inertial navigation error accumulation are effectively suppressed, significantly improving the positioning accuracy and reliability of each UAV. At the same time, based on positioning reliability, the safety distance and identity switching are dynamically adjusted, and combined with the cluster spatial offset heat map, efficient collaborative control is achieved, ensuring the safe and stable operation and mission execution of each UAV cluster in complex urban environments.

[0112] Example 3 In Example 2, when each UAV swarm performs tasks in a complex electromagnetic environment, although each UAV can collect its own electromagnetic signals and obstacle information and extract reflection interference features, the positioning drift error analysis is based solely on individual data. However, the environments in which each UAV operates differ, and the obstacle reflection interference features it collects are significantly affected by local environmental factors, resulting in limited accuracy in the quantitative relationship between the reflection interference features obtained from individual analysis and the positioning drift error. Furthermore, when each UAV independently emits a reverse electromagnetic beam to cancel interference, there is a lack of overall consideration of the overall interference cancellation effect of the swarm, making it difficult to achieve globally optimal interference cancellation. Historically, the reflection interference situations faced by each UAV swarm in different locations and mission scenarios have been complex and diverse, and the single-machine processing method cannot fully tap the swarm's collaborative potential to cope with diverse interference situations. To further improve the positioning accuracy and interference cancellation capability of each UAV swarm in complex electromagnetic environments and obtain more accurate and reliable operational results, it is necessary to comprehensively consider the information of each UAV within the swarm and conduct further optimization and improvement.

[0113] In some embodiments, to locate the drift vector, step S43 further includes: S431: Each drone in each drone swarm collects its own electromagnetic signals and information on static and dynamic obstacles, extracts reflection interference features, and converts them into quantifiable phase offset vectors.

[0114] Let the phase of the normal signal be... The phase of the interfered signal is Then the phase shift A coordinate system is established with each UAV as the origin. The direction of the phase offset is determined based on the positions of static and dynamic obstacles. The magnitude and direction of the phase offset are combined into a quantifiable phase offset vector. , where θ is the direction angle of the phase offset.

[0115] S432 allows each drone cluster to share information and collaboratively calibrate positioning data. It then uses the historical data after collaborative calibration to analyze the intrinsic relationship between the reflection interference characteristics of different obstacles and the positioning drift error of each drone.

[0116] Among them, the collaborative calibration positioning data consists of each UAV's own positioning information, sensor data, and reflection interference quantification results.

[0117] Specifically, each UAV swarm shares its own positioning information (such as latitude, longitude, and altitude), sensor data (such as acceleration and angular velocity), and reflection interference quantification results (phase offset vector) in real time via a wireless communication network. A weighted average method is used for collaborative calibration of the positioning data. Let the positioning data of each UAV of the i-th swarm be... The weight is (Based on the positioning accuracy of each drone), the calibrated positioning data , where n is the number of drones in each drone swarm.

[0118] Using historical data after collaborative calibration, the relationship between reflection interference characteristics (phase offset vector) and positioning drift error of each UAV under different motion states and different types of static and dynamic obstacles (classified according to obstacle shape) is statistically analyzed. A quantitative relationship model between reflection interference characteristics and positioning drift error is established using methods such as linear regression. , where a, b, and c are coefficients obtained by fitting historical data.

[0119] S433: Each drone monitors and provides early warnings for positioning drift in real time based on quantified relationships, and each drone cluster formulates corresponding preventive strategies based on the early warning signals.

[0120] S434, based on the phase offset vector of the reflected interference detected by each UAV and the overall cooperative strategy of the cluster, an attitude cooperative adjustment algorithm for the cluster position is designed.

[0121] The attitude coordination adjustment algorithm for cluster position design employs an optimization-based algorithm. The position and attitude parameters of each UAV are used as optimization variables, with the objective function being to minimize the impact of reflection interference on the cluster and meet the requirements of the coordination strategy. The position and attitude adjustment parameters of each UAV are obtained through iterative optimization.

[0122] The S435 system adjusts the flight position and body attitude of each drone based on calculated position and attitude adjustment parameters, while simultaneously controlling the electromagnetic wave transmitting equipment to emit reverse electromagnetic beams with specific phases and amplitudes.

[0123] Specifically, each UAV determines the phase and amplitude of the reverse electromagnetic beam to be emitted based on calculated position and attitude adjustment parameters. Let the phase of the reflected interference electromagnetic wave be... , amplitude Then the phase of the reverse electromagnetic beam (Achieving phase reversal), amplitude (Adjustments may be made as needed based on actual conditions). Each UAV controls an electromagnetic wave transmitting device to emit a reverse electromagnetic beam according to the calculated phase and amplitude, so that it cancels out the reflected interference electromagnetic waves.

[0124] S436. Establish a dynamic cancellation effectiveness evaluation model to evaluate the effect of the reverse electromagnetic beam in canceling reflected interference in real time. Based on the evaluation results of the dynamic cancellation effectiveness evaluation model, optimize the control parameters in the attitude coordination adjustment algorithm of the cluster position in real time.

[0125] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: By quantizing and modeling reflected interference and converting it into a phase offset vector, the interference can be accurately quantified. By using cluster collaborative calibration of positioning data and establishing quantization relationships, positioning drift can be prevented in advance. By designing an attitude collaborative adjustment algorithm for cluster position, reverse electromagnetic beams are emitted to cancel interference, and an evaluation model is established to optimize parameters, effectively improving the positioning accuracy and collaborative control capability of each UAV cluster in complex electromagnetic environments.

[0126] Furthermore, this embodiment of the invention also provides an autonomous collaborative control system for unmanned aerial vehicle (UAV) swarms.

[0127] Figure 2 This is a schematic diagram of the structure of an autonomous collaborative control system for unmanned aerial vehicle (UAV) swarms according to an embodiment of the present invention.

[0128] like Figure 2 As shown, an autonomous collaborative control system for unmanned aerial vehicle (UAV) swarms includes: a perception and identification module, a tracking and prediction module, an avoidance and response module, an adjustment and allocation module, a path planning module, an adjustment module, and a strategy optimization module.

[0129] The perception and recognition module is used by each drone to generate a local 3D map of the city by stitching together point clouds from LiDAR, and to mark static obstacles in the local 3D map; it uses millimeter-wave radar combined with Kalman filtering technology to identify dynamic obstacles and mark their risk levels.

[0130] The tracking and prediction module is used to track the heat source information of the target location through infrared cameras when the target location is blocked. The unblocked drones continue to track the target location heat source information through infrared cameras, while the blocked drones switch to prediction mode and use a distributed interactive multi-model algorithm to predict the target location trajectory and fuse the prediction data of each drone.

[0131] The avoidance response module is used to avoid static obstacles based on the artificial potential field method and combined with the height layer division strategy; different response strategies are adopted for dynamic obstacles with different risks.

[0132] The allocation module is used to dynamically assign roles to each drone and adjust task priorities based on the positioning error of the target and the distance to obstacles.

[0133] The path planning module is used for each drone to report its status information to the cluster control center in real time; if the actual path deviates from the planned path by more than a certain threshold, local replanning is triggered.

[0134] The strategy optimization module is used to learn the role assignment rules and path assignment strategies for each drone through reinforcement learning.

[0135] It should be noted that other specific implementations of the UAV swarm autonomous collaborative control system of the present invention can refer to the above-described UAV swarm autonomous collaborative control method.

[0136] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0137] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0138] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0139] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0140] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0141] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0142] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for autonomous cooperative control of unmanned aerial vehicle (UAV) swarms, characterized in that, include: S1, each drone uses LiDAR point cloud stitching to generate a local 3D map of the city, and marks static obstacles in the local 3D map; Using millimeter-wave radar combined with Kalman filtering technology, dynamic obstacles are identified and their risk levels are marked; S2, when the target is blocked, each unblocked drone continues to track the target's heat source information through infrared cameras, while each blocked drone switches to prediction mode, uses a distributed interactive multi-model algorithm to predict the target's trajectory, and integrates the prediction data of each drone. S3, based on the artificial potential field method and combined with the height layer division strategy, avoids static obstacles; Different response strategies should be adopted for dynamic obstacles with different risks; S4 dynamically assigns roles to each UAV, adjusting task priorities based on target positioning errors and obstacle distances; it simultaneously receives signals from three systems: GPS L1 / L5 bands, BeiDou B1 / B2 bands, and Galileo E1 / E5a ​​bands, calculating the ratio of direct to reflected signal strength in each band in real time and determining signal attenuation based on building material factors; based on signal attenuation, when a reflected signal is detected to be stronger than the direct signal in a certain band, it automatically switches to the optimal anti-multipath band; it records positioning drift vectors during historical flights, classifies and stores typical error vectors according to motion states, and adjusts them based on current speed direction and acceleration. The system matches the database and generates a reverse compensation vector to compensate and correct the original positioning data; it marks the 3D coordinates of static obstacles in the city's 3D map to clarify their type and characteristics; it scans the static obstacles on the working path of each drone with LiDAR, compares the actual coordinates of the identified static obstacles with the pre-stored coordinates, and performs position calibration for each drone based on the offset; it calculates the positioning reliability coefficient of each drone in real time, dynamically adjusts the safety distance based on the positioning reliability coefficient, and determines whether role switching is necessary; it summarizes the positioning error vectors of each drone to generate a cluster spatial offset heatmap and marks high error risk areas. S5: Each drone reports its status information to the cluster control center in real time; if the actual path deviates from the planned path by more than a certain threshold, local replanning is triggered. S6 uses reinforcement learning to determine the role assignment rules and path assignment strategies for each drone.

2. The method for autonomous cooperative control of unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, S2 specifically includes: continuously capturing heat source information of the positioning target using an infrared camera, and extracting the heat source area of ​​the positioning target from the image captured by the camera using image processing technology; when the positioning target is detected to be occluded, each occluded UAV immediately switches from tracking mode to prediction mode; establishing multiple motion models to describe the possible different motion states of the positioning target, updating the probability of each model according to Bayes' theorem, and fusing the state estimates of each model according to the model probabilities to obtain the final positioning target state estimate; each occluded UAV sends its own predicted positioning target trajectory data to the cluster control center; the fusion center uses a weighted average method to fuse the prediction data of each UAV.

3. The method for autonomous cooperative control of unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, S4 adjusts task priorities based on the positioning error of the target and the distance to obstacles, including: assigning an initial role to each UAV based on its performance parameters, continuously monitoring the performance parameters of each UAV; setting a comprehensive scoring function based on the performance parameters to evaluate the comprehensive capabilities of each UAV; comparing the comprehensive capability scores of each UAV and adjusting the role of each UAV based on the comparison results; each UAV acquiring the location information of the target through its own positioning system, and simultaneously using other auxiliary means to reconfirm the location of the target; each UAV using sensors to monitor the distance to surrounding obstacles in real time, recording the distance between each UAV and the nearest obstacle; setting positioning error weights and obstacle distance weights to construct a task priority evaluation function; and sorting different tasks according to the value of the task priority evaluation function; wherein, the auxiliary means are visual recognition or laser ranging.

4. The method for autonomous cooperative control of unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, The compensation and correction of the original positioning data includes: recording the positioning drift vector using the positioning system during each flight of each UAV; classifying and storing typical error vectors in the corresponding database according to the different motion states of each UAV; acquiring the current velocity direction and acceleration information of each UAV in real time and matching them with the motion states in the database; and finding the error vector corresponding to the closest motion state. Generate reverse compensation vector ; the reverse compensation vector Compared with the original location data The data is overlaid to obtain the compensated positioning data. .

5. The method for autonomous cooperative control of unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, The formula for calculating the dynamically adjusted safety distance is as follows: , This is the baseline safety distance. It is an adjustment factor; The formula for the location credibility coefficient is: .

6. The method for autonomous cooperative control of unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, The positioning drift vector includes: each UAV in each UAV swarm collects its own electromagnetic signals and information on static and dynamic obstacles, extracts reflection interference features, and converts them into a quantifiable phase offset vector; each UAV swarm shares information and collaboratively calibrates positioning data, using historical data after collaborative calibration to analyze the intrinsic relationship between the reflection interference features of different obstacles and the positioning drift error of each UAV; each UAV monitors positioning drift in real time based on quantization relationships and issues early warnings, and each UAV swarm formulates corresponding early prevention strategies based on the early warning signals; based on the reflection interference phase offset vector detected by each UAV and the overall collaborative strategy of the swarm, an attitude collaborative adjustment algorithm for the swarm position is designed; each UAV adjusts its own flight position and fuselage attitude according to the calculated position and attitude adjustment parameters, while controlling the electromagnetic wave transmitting equipment to emit a reverse electromagnetic beam with a specific phase and amplitude; a dynamic cancellation effectiveness evaluation model is established to evaluate the effect of the reverse electromagnetic beam in canceling reflection interference in real time, and the control parameters in the attitude collaborative adjustment algorithm for the swarm position are optimized in real time based on the evaluation results of the dynamic cancellation effectiveness evaluation model.

7. The method for autonomous cooperative control of unmanned aerial vehicle (UAV) swarms according to claim 6, characterized in that, The extraction of reflection interference features and their conversion into a quantifiable phase offset vector specifically involves: assuming the phase of the normal signal is... The phase of the interfered signal is Then the phase shift ; A coordinate system is established with each UAV as the origin. The direction of the phase offset is determined based on the positions of static and dynamic obstacles. The magnitude and direction of the phase offset are combined into a quantifiable phase offset vector. , where θ is the direction angle of the phase offset.

8. The method for autonomous cooperative control of unmanned aerial vehicle (UAV) swarms according to claim 6, characterized in that, The shared information and collaborative calibration positioning data among each UAV cluster includes: the collaborative calibration positioning data being the positioning information, sensor data, and reflection interference quantification results of each UAV; each UAV cluster sharing its own positioning information, sensor data, and reflection interference quantification results in real time via a wireless communication network; and the collaborative calibration of the positioning data using a weighted average method, where the positioning data of the i-th UAV is... The weight is Then the calibrated positioning data , where n is the number of drones in each drone swarm.

9. An autonomous collaborative control system for unmanned aerial vehicle (UAV) swarms, applied to an autonomous collaborative control method for UAV swarms as described in any one of claims 1 to 8, characterized in that, The system includes: The perception and recognition module is used by each drone to generate a local 3D map of the city by stitching together point clouds from LiDAR, and to mark static obstacles in the local 3D map; it also uses millimeter-wave radar combined with Kalman filtering technology to identify dynamic obstacles and mark their risk levels. The tracking and prediction module is used to ensure that when the target is blocked, each unblocked drone continues to track the target's heat source information through an infrared camera, while each blocked drone switches to prediction mode, uses a distributed interactive multi-model algorithm to predict the target's trajectory, and integrates the prediction data of each drone. The avoidance response module is used to avoid static obstacles based on the artificial potential field method and combined with the height layer division strategy; different response strategies are adopted for dynamic obstacles with different risks. The allocation module is used to dynamically assign roles to each drone and adjust task priorities based on the positioning error of the target and the distance to obstacles. The path planning module is used for each drone to report its status information to the cluster control center in real time; if the actual path deviates from the planned path by more than a certain threshold, local replanning is triggered. The strategy optimization module is used to learn the role assignment rules and path assignment strategies for each drone through reinforcement learning.

Images