An unmanned aerial vehicle cluster self-adaptive pathfinding method and system based on bionic intelligence

Abstract

The application provides an adaptive pathfinding method for unmanned aerial vehicle clusters, specifically including task initialization and environment perception, path planning based on bionic intelligence, multi-vehicle cooperation and dynamic adjustment. The method designs an information pheromone updating strategy with variation mechanism and dynamic obstacle avoidance capability, making the path search more adaptive and stable. At the communication level, a decentralized communication architecture is adopted between unmanned aerial vehicles, supplemented by a cloud synchronization mechanism, to realize real-time sharing of task information and efficient adjustment of path schemes. The pathfinding method is applicable to various practical scenarios, including post-disaster search and rescue, ecological environment monitoring, and large-area inspection, and has good scalability and robustness, effectively addressing the challenges brought by dynamic environments and task changes.

Description

Technical Field

[0001] This invention relates to adaptive pathfinding technology for unmanned aerial vehicle (UAV) swarms, and more particularly to an adaptive pathfinding method and system for UAV swarms based on biomimetic intelligence. Background Technology

[0002] While drone swarms are widely used in disaster relief and environmental monitoring, several technical bottlenecks remain in practical applications. Complex and ever-changing environments hinder efficient perception and rapid response in path planning and obstacle avoidance, particularly in areas with complex terrain and dense obstacles, often leading to path interruptions or safety hazards. Furthermore, limited inter-swarm communication capabilities and low information synchronization efficiency can cause coordination chaos due to delays or packet loss, impacting overall execution. Task allocation lacks flexibility and cannot be dynamically adjusted based on real-time status, frequently resulting in resource waste or uneven load distribution. In addition, existing systems generally lack real-time linkage mechanisms, causing a disconnect between path planning and task execution, making it difficult to respond promptly to sudden changes. Overall, current drone swarms still have significant room for improvement in environmental adaptability, communication coordination, dynamic task adjustment, and system stability, hindering their widespread adoption and efficient application in complex scenarios. Summary of the Invention

[0003] To address the aforementioned issues, this invention proposes a biomimetic intelligence-based adaptive pathfinding method for UAV swarms. It integrates a bee-foraging strategy for global exploration, an ant colony optimization algorithm for local adjustments, and reinforcement learning and the Bellman equation to optimize task allocation. The UAV swarm consists of reconnaissance aircraft, follower aircraft, and navigator aircraft, which respectively perform random search, pheromone guidance, and task scheduling. This enables the UAV swarm to efficiently and adaptively complete tasks in complex and dynamic environments, increasing the exploration rate of high-value areas.

[0004] An adaptive pathfinding method for UAV swarms based on biomimetic intelligence includes the following steps:

[0005] S1: Task initialization and environment awareness;

[0006] S2: Path planning based on biomimetic intelligence;

[0007] S3: Multi-machine collaboration and dynamic adjustment.

[0008] Furthermore, S1 specifically includes

[0009] S11: Divide the mission area into a gridded exploration space and set the starting point of the drone swarm based on historical data;

[0010] S12: Collect environmental data using sensors mounted on the drone, and calculate the pheromone concentration in each grid area. The formula for calculating the pheromone concentration is as follows:

[0011] ,

[0012] in, Environmental risk value, Due to terrain complexity, For historical data correlation, , , To adjust the weights.

[0013] Furthermore, S2 specifically includes:

[0014] S21 Global Path Exploration: Divide the drone swarm into three roles: reconnaissance aircraft, follower aircraft, and navigator aircraft, and perform the following operations:

[0015] The S211 reconnaissance aircraft employs a hybrid strategy combining random walks and distributed Monte Carlo search to generate candidate paths;

[0016] S212 follows the path selection based on path pheromone concentration and path heuristic factor. The path selection probability is determined by the following formula:

[0017] ,

[0018] in, For path pheromone concentration, As a path heuristic factor, , To adjust the weights;

[0019] The S213 navigator integrates data from multiple aircraft to generate a global pheromone map and optimizes the global path using a dynamic programming algorithm.

[0020] S22 Local Path Optimization: Ant colony optimization strategy is used to update path pheromones, with the following update rules:

[0021] ,

[0022] in, The pheromone evaporation coefficient, For the first The drone releases incremental pheromones and corrects its path using a dynamic obstacle avoidance algorithm; when the reconnaissance aircraft detects an abnormal pheromone concentration, it triggers local augmentation exploration and transmits high-value area data back to the navigator; the navigator dynamically allocates tasks through a reinforcement learning model and periodically updates the globally optimal path.

[0023] Furthermore, the candidate path generation strategy for the S211 reconnaissance aircraft specifically includes:

[0024] Using a probabilistic model Select the destination of the route, where Let j be the distance between the current drone and the destination. To explore intensity parameters; when an uncovered area or abnormal environmental parameters are detected, local augmentation exploration is triggered, and the path selection probability is adjusted to... .

[0025] Furthermore, the global path optimization of the navigator described in S213 includes:

[0026] Constructing a global pheromone graph ,in The data reliability weights for drone k are determined using the Bellman equation. Solve for the optimal path value function, where As a reward for action, This is the discount factor.

[0027] Furthermore, in the pheromone update rule described in S22, the pheromone increment... It is negatively correlated with the path efficiency of drone k, and the pheromone evaporation coefficient It adaptively adjusts based on the dynamic obstacle density of the task area.

[0028] Furthermore, the dynamic obstacle avoidance algorithm in S22 includes: introducing a real-time obstacle risk assessment model in the path optimization stage to correct the path selection probability; when using the A* algorithm to generate obstacle avoidance paths, the heuristic function combines the weighted sum of terrain complexity and environmental risk value.

[0029] Furthermore, the state-action value function update rule of the reinforcement learning model described in S22 is as follows:

[0030]

[0031] Where α is the preset weight, State-action value, For instant rewards, This is the discount factor.

[0032] An adaptive pathfinding system for unmanned aerial vehicle (UAV) swarms based on biomimetic intelligence, comprising:

[0033] The environmental perception module is used to divide the task area into grids and calculate the pheromone concentration using sensor data.

[0034] The path planning module integrates bee foraging strategies and ant colony optimization algorithms to control the path generation and optimization of the reconnaissance aircraft, follower aircraft, and navigator aircraft, respectively.

[0035] The dynamic adjustment module allocates tasks based on the reinforcement learning model and updates the global path according to real-time data;

[0036] The communication module enables pheromone sharing among drones and the issuance of commands by the navigator.

[0037] The beneficial effects of this invention are:

[0038] 1. Combining global and local optimization improves path planning efficiency; global exploration is performed using a bee foraging strategy, while local path adjustment is performed using an ant colony optimization algorithm, enabling efficient pathfinding for drone swarms in complex environments, avoiding the problems of traditional methods that only focus on local optima or have low global search efficiency.

[0039] It has strong dynamic adaptability and can cope with sudden environmental changes.

[0040] 2. By adopting a pheromone mechanism for adaptive path adjustment and combining it with dynamic path planning algorithms such as D Lite and A**, the drone swarm can quickly adjust its path in sudden situations such as fires and changes in obstacles, thereby improving environmental adaptability.

[0041] 3. Task collaboration optimization to improve the collaborative capabilities of UAV swarms: By using leader UAVs in conjunction with reinforcement learning (RL) and Bellman equations to dynamically adjust tasks, information sharing and efficient collaboration among UAV swarms can be achieved, effectively improving the overall task execution efficiency.

[0042] 4. Intelligent pheromone mechanism to increase the exploration ratio of high-value areas: By combining historical data, environmental perception and mission requirements to calculate pheromone concentration, the UAV is guided to focus on exploring high-value areas, thereby improving the quality of mission completion. Compared with traditional methods, it can cover more key areas in the same amount of time.

[0043] 5. Reduce computational complexity and improve real-time performance; adopt a hierarchical architecture (reconnaissance aircraft - follower aircraft - navigator aircraft) for task allocation, so that the computational burden is distributed within the UAV cluster, reducing global computational overhead and improving real-time performance and scalability.

[0044] 6. This invention combines biomimetic intelligent algorithms with reinforcement learning to significantly improve the efficiency of UAV swarms in terms of path planning, dynamic adaptability, task coordination, and computational optimization, and to enhance their practicality and stability in complex dynamic environments. Attached Figure Description

[0045] Figure 1 The process of this invention Figure 1

[0046] Figure 2 The process of this invention Figure 2 . Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings. The directional terms such as "front", "rear", "left", and "right" are all based on the figures shown and do not constitute a limitation on the scope of the present invention.

[0048] like Figures 1-2 As shown in the figure, this embodiment discloses an adaptive pathfinding method for UAV swarms based on biomimetic intelligence, which specifically includes the following steps: S1 Task initialization and environmental perception, S2 Path planning based on biomimetic intelligence, and S3 Multi-UAV collaboration and dynamic adjustment.

[0049] S1, Task Initialization and Environmental Awareness: First, a drone swarm is deployed, and a starting point is set based on historical data. The task area is divided into a grid to construct the exploration space. Next, the drones are equipped with relevant sensors according to the task requirements, and a local environmental assessment model is used to calculate the regional pheromone concentration. This concentration value is related to the region's environmental risk value, terrain complexity, and correlation with historical data.

[0050] S2. Bionic Intelligence-Based Path Planning: Drones are categorized into three types: reconnaissance drones, follower drones, and navigator drones. Reconnaissance drones are responsible for global random exploration; follower drones are guided by pheromones released by other drones; and navigator drones adjust path planning based on real-time and historical data. A bee-like foraging strategy is used for global path exploration, while an ant colony optimization strategy is employed to optimize local paths. During optimization, inefficient paths are fine-tuned, and a dynamic obstacle avoidance algorithm is incorporated to enhance the drone's ability to cope with unexpected obstacles.

[0051] S3, Multi-machine collaboration and dynamic adjustment:

[0052] S3.1 The reconnaissance aircraft uses a combination of random walk and distributed Monte Carlo exploration for path planning. If it detects an abnormal concentration of environmental pheromones, it will perform local enhanced exploration. While mainly exploring the global area, it will stay in local hotspot areas to collect information and then transmit it back to the navigator. It will also release more pheromones in high-value areas.

[0053] S3.2 The system uses ant colony optimization to select the optimal path. When the pheromone concentration decays to a certain threshold, it triggers a local search mode to find an alternative path and combines it with an obstacle avoidance algorithm to ensure the path is feasible.

[0054] S3.3 The navigator collects environmental data from all reconnaissance and follower aircraft, performs local information fusion, establishes an environmental pheromone map, solves for the optimal path using the Bellman equation, and uses a dynamic task adjustment strategy based on reinforcement learning to allocate tasks. The optimal path is recalculated at regular intervals to adjust the flight tasks of the swarm of UAVs. In emergency situations, it can directly command reconnaissance aircraft to conduct high-priority searches.

[0055] S3.4 The specific explanations of the relevant calculation processes involved in the above steps are as follows:

[0056] The formula for calculating the pheromone concentration in a region is designed as follows:

[0057]

[0058] in This represents the environmental risk value of the area. Represents terrain complexity. Represents the correlation of historical data. , , With preset weights, their respective weights affect the final calculation result of the regional pheromone concentration.

[0059] The formula for calculating the path selection probability (bee foraging model) is designed as follows:

[0060] .

[0061] in For path pheromone concentration, Representative path heuristic factor ( (where α, β, and γ represent the distance between two points), and preset weights are assigned to control pheromones and path quality to determine the probability of the UAV choosing different paths.

[0062] The formula for the pheromone update rule is designed as follows:

[0063] ,

[0064] The pheromone evaporation coefficient represents the degree to which pheromones evaporate over time. Represents the k-th drone on the path The incremental release of pheromones is used to continuously adjust the pheromone concentration along the path, thereby guiding the drone's path selection.

[0065] The formula used for path planning in the reconnaissance aircraft's exploration strategy is:

[0066]

[0067] For the current drone and the candidate path endpoint distance, To explore parameters (which tend to favor long-distance random jumps when larger), the formula for selecting the path endpoint is:

[0068] ;

[0069] in Let j be the current distance between the drone and the destination. To explore strength parameters

[0070] When an abnormal concentration of environmental pheromones is detected, the formula for local enhancement exploration is as follows:

[0071]

[0072] in Given the pheromone concentration of the current path, this strategy allows reconnaissance aircraft to focus their exploration on undetected areas.

[0073] The navigator uses the Bellman equation to solve for the optimal path during global path optimization:

[0074]

[0075] in Let be the optimal path value for state s. The reward for choosing action a and entering state s. This is a discount factor used to control the impact on future returns.

[0076] When assigning tasks, a dynamic task adjustment strategy based on reinforcement learning is adopted, with the following formula:

[0077]

[0078] Where α is the preset weight, State-action value, For instant rewards, This is the discount factor.

[0079] Constructing a global pheromone graph:

[0080]

[0081] in Data reliability weights for drone k

[0082] Use the above three steps to optimize the global path for the navigator.

[0083] Example 1

[0084] S1. Task Initialization and Environment Awareness

[0085] Before deploying the drone swarm, the system divides the mission area into a gridded exploration space. For example, a 10km×10km area is divided into 100×100 grids, with each grid having a side length of 100 meters. The initial deployment point is set based on the historical fire distribution, prioritizing locations around historically high-incidence areas.

[0086] Each drone collects environmental data using its onboard multispectral camera, lidar, and positioning module. The system calculates the following parameters for each grid:

[0087] Environmental risk value (r): calculated based on normalized smoke concentration and temperature indices;

[0088] Terrain complexity (c): Calculated from the slope and obstacle density extracted by lidar;

[0089] Historical data relevance (h): Set to 1 for records with high frequency, and 0 for the rest.

[0090] Based on the three indicators, the pheromone concentration P in the grid is calculated.

[0091] Where α=0.5, β=0.3, and γ=0.2 are preset weights.

[0092] S2. Path planning based on biomimetic intelligence

[0093] S2.1 Global Exploration Strategy

[0094] Drone swarms are divided into three roles:

[0095] Reconnaissance aircraft (approximately 30%): Candidate paths are generated using a combination of random walks and Monte Carlo search. The probability of selecting a path's endpoint is related to the distance to the current node and the exploration intensity. When localized high temperatures or dense smoke anomalies are detected, the strategy is adjusted to increase the intensity of the local search.

[0096] Follow-the-path (approximately 60%): Path selection is based on path pheromone concentration and path heuristic factors, using an improved ant colony algorithm to calculate the path selection.

[0097] Where τ is the pheromone concentration, η is the path heuristic factor, α=1.5, β=2.0.

[0098] Navigator (approximately 10%): Periodically collects data from the entire cluster to construct a global pheromone map. Iterative optimization of the path value function is performed using the Bellman equation, with a discount factor applied. =0.9, number of iterations ≥100.

[0099] S2.2 Local Optimization and Obstacle Avoidance Mechanism

[0100] All characters dynamically update pheromone concentrations based on the ant colony optimization strategy.

[0101] The volatility coefficient ρ is adjusted according to the local barrier density: when the barrier density is >30%, ρ is set to 0.8; otherwise, it is 0.5.

[0102] If the lidar detects an obstacle 50 meters ahead during path execution, the A* algorithm is triggered to replan the path, and the heuristic function takes into account the terrain and risk value weighted result.

[0103] If a path is blocked more than 3 times, the system will automatically reduce the pheromone concentration of that path by 50%.

[0104] S3. Multi-machine collaboration and dynamic adjustment

[0105] S3.1 Information Feedback and Global Update

[0106] The reconnaissance aircraft uploads high-value grid data (such as P > 0.8) to the navigator every 2 minutes, which is marked as a red area on the global map.

[0107] S3.2 Follow mode switching

[0108] If the pheromone concentration of the followed path is <0.3, the system switches to local search mode, adopts the D* Lite algorithm, expands the scanning range to 200 meters, and increases the path refresh rate to 1Hz.

[0109] S3.3 Dynamic Task Allocation

[0110] The navigator dynamically adjusts the proportions of the three roles based on a Q-learning model. The reward function is related to path efficiency and information coverage. The learning rate α = 0.2, the exploration rate ε = 0.1, and the policy is updated every 10 minutes.

[0111] Key parameter records:

[0112]

[0113] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features.

Claims

1. A biomimetic intelligence-based adaptive pathfinding method for unmanned aerial vehicle (UAV) swarms, characterized in that, Specifically, the following steps are included: S1: Task initialization and environment awareness; S2: Path planning based on biomimetic intelligence; S2 specifically includes: S21 Global Path Exploration: Divide the drone swarm into three roles: reconnaissance aircraft, follower aircraft, and navigator aircraft, and perform the following operations: The S211 reconnaissance aircraft employs a hybrid strategy combining random walks and distributed Monte Carlo search to generate candidate paths. The specific candidate path generation strategy of the S211 reconnaissance aircraft includes: Using a probabilistic model Select the destination of the route, where Let j be the current distance between the drone and the destination. To explore intensity parameters; when an uncovered area or abnormal environmental parameters are detected, local augmentation exploration is triggered, and the path selection probability is adjusted to... ; S212 follows the path selection based on path pheromone concentration and path heuristic factor. The path selection probability is determined by the following formula: ， in, For path pheromone concentration, As a path heuristic factor, , To adjust the weights; The S213 navigator integrates multi-aircraft data to generate a global pheromone map and optimizes the global path using a dynamic programming algorithm; the global path optimization of the S213 navigator includes: Constructing a global pheromone graph ,in The data reliability weights for drone k are determined using the Bellman equation. Solve for the optimal path value function, where As a reward for action, Discount factor; S22 Local Path Optimization: Ant colony optimization strategy is used to update path pheromones, with the following update rules: ， in, The pheromone evaporation coefficient, For the first The drone releases incremental pheromone levels, and the path is corrected using a dynamic obstacle avoidance algorithm; when the reconnaissance aircraft detects abnormal pheromone concentrations, it triggers local augmentation exploration and transmits high-value area data back to the navigator; the navigator dynamically allocates tasks through a reinforcement learning model and periodically updates the globally optimal path. S3: Multi-machine collaboration and dynamic adjustment.

2. The adaptive pathfinding method for UAV swarms based on biomimetic intelligence according to claim 1, characterized in that, S1 specifically includes S11: Divide the mission area into a gridded exploration space and set the starting point of the drone swarm based on historical data; S12: Collect environmental data using sensors mounted on the drone, and calculate the pheromone concentration in each grid area. The formula for calculating the pheromone concentration is as follows: ， in, Environmental risk value, Due to terrain complexity, For historical data correlation, , , To adjust the weights.

3. The adaptive pathfinding method for UAV swarms based on biomimetic intelligence according to claim 1, characterized in that, In the pheromone update rule described in S22, the pheromone increment... It is negatively correlated with the path efficiency of drone k, and the pheromone evaporation coefficient It adaptively adjusts based on the dynamic obstacle density of the task area.

4. The adaptive pathfinding method for UAV swarms based on biomimetic intelligence according to claim 3, characterized in that, The dynamic obstacle avoidance algorithm in S22 includes: introducing a real-time obstacle risk assessment model in the path optimization stage to correct the path selection probability; when using the A* algorithm to generate obstacle avoidance paths, the heuristic function combines the weighted sum of terrain complexity and environmental risk value.

5. The adaptive pathfinding method for UAV swarms based on biomimetic intelligence according to claim 3, characterized in that, The state-action value function update rule for the reinforcement learning model described in S22 is as follows: Where α is the adjustment weight, State-action value, For instant rewards, This is the discount factor.

6. A biomimetic intelligent UAV swarm adaptive pathfinding system according to any one of claims 1-5, characterized in that, include: The environmental perception module is used to divide the task area into grids and calculate the pheromone concentration using sensor data. The path planning module integrates bee foraging strategies and ant colony optimization algorithms to control the path generation and optimization of the reconnaissance aircraft, follower aircraft, and navigator aircraft, respectively. The dynamic adjustment module allocates tasks based on the reinforcement learning model and updates the global path according to real-time data; The communication module enables pheromone sharing among drones and the issuance of commands by the navigator.

Images