LLM-based multi-unmanned aerial vehicle cooperative trajectory planning method and task management system

Abstract

The invention provides an LLM-based multi-unmanned aerial vehicle cooperative trajectory planning method and a task management system, and relates to the technical field of unmanned aerial vehicle control. The method comprises the following steps: S1, multi-source environment perception and dynamic map construction; s2, LLM intelligent task allocation and motion parameter optimization; s3, multi-unmanned aerial vehicle cooperative trajectory planning and optimization; s4, performing real-time security verification and track adjustment; and S5, track execution and dynamic re-planning. According to the invention, a dynamic map is constructed through multi-source environment perception, intelligent task allocation and motion parameter optimization are carried out by using LLM, a safe trajectory is generated by adopting cooperative trajectory planning based on unmanned aerial vehicle dynamics, and the reliability of the system is ensured through real-time safety verification and dynamic adjustment. According to the method, the motion law of the unmanned aerial vehicle is fused in trajectory planning, the optimal trajectory is directly calculated, the efficiency is high, and the dynamic environment adaptability is ensured through LLM real-time decision making.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a multi-UAV cooperative trajectory planning method and task management system based on LLM (Limited Least Mechanism). Background Technology

[0002] With the development of drone technology, multi-drone systems are widely used in smart construction site monitoring. However, existing drone management platforms have the following shortcomings: task allocation relies on human experience and lacks intelligence; trajectory planning does not consider drone dynamics constraints, resulting in low efficiency; the conflict avoidance mechanism in multi-drone collaboration is imperfect; and the real-time adjustment capability is poor, making it unable to adapt to dynamic environments. Summary of the Invention

[0003] This invention proposes a multi-UAV cooperative trajectory planning method and task management system based on LLM, which solves the problems of task allocation relying on manual intervention, trajectory planning not fully considering dynamic constraints, imperfect cooperative conflict avoidance mechanism, and poor real-time adjustment capability in existing multi-UAV systems.

[0004] This invention first proposes a cooperative trajectory planning method for multiple UAVs based on LLM, comprising the following steps: Step S1: Multi-source environmental perception and dynamic map construction S1.1: Environmental data is collected through multiple sensors, including at least an image sensor, a lidar and an inertial measurement unit. The collected environmental data is preprocessed, including noise reduction, distortion correction and feature extraction, to become preprocessed data, so as to extract effective sensing information. S1.2: Based on Kalman filtering, the preprocessed data is fused to form fused data, which is used to estimate the UAV's own state and the position and velocity of environmental targets; S1.3: Construct a dynamic occupancy grid map based on the fused data, and update the dynamic occupancy grid map in real time to reflect environmental changes, while tracking dynamic objects; Step S2: LLM Intelligent Task Allocation and Motion Parameter Optimization S2.1: Utilize LLM to analyze task requirements and understand scenarios. LLM receives dynamic environment maps, task descriptions, and UAV status as inputs and outputs task priorities and constraints. S2.2: Construct a multi-objective optimization problem based on the scenario understanding results. The multi-objective optimization problem considers multiple objectives, including task completion time, energy consumption, and safety. S2.3: A distributed task allocation algorithm is used to solve the multi-objective optimization problem, achieving efficient task allocation; S2.4: Obtain the motion parameters of each UAV through optimization, including velocity, acceleration and heading angle, and satisfy the UAV dynamics constraints; Step S3: Multi-UAV cooperative trajectory planning and optimization S3.1: Based on the UAV dynamics model, considering constraints including mass, inertia, and thrust, initialize the trajectory parameters; S3.2: Generates a smooth variable speed curve, enabling the drone to accelerate and decelerate smoothly; S3.3: Generates the three-dimensional trajectory prototype, which forms the basic segment of the UAV's flight path; S3.4: A distributed optimization method is used to collaboratively optimize the trajectories of multiple UAVs to avoid conflicts between UAVs; Step S4: Real-time security verification and trajectory adjustment S4.1: Calculate the collision probability between multiple UAVs based on a probabilistic model to assess trajectory safety; S4.2: Using Lyapunov stability theory, construct the Lyapunov function and verify the stability of the trajectory; S4.3: Construct a safe potential field to detect potential conflicts and provide obstacle avoidance information; S4.4: Based on the safety verification results, the trajectory is updated in real time using an adjustment strategy to ensure flight safety; Step S5: Trajectory Execution and Dynamic Replanning S5.1: The trajectory tracking control is achieved by using an adaptive PID control algorithm, and the controller dynamically adjusts the control gain according to the tracking error; S5.2: Real-time monitoring of UAV performance indicators, which include at least tracking error, energy consumption, and safety status; S5.3: Based on the LLM dynamic decision-making mechanism, replanning is triggered when environmental changes or performance anomalies are detected; S5.4: Online optimization and update of trajectories to adapt to dynamic environmental changes.

[0005] Furthermore, the specific implementation of step S1 is as follows: S1.1: Multi-sensor data acquisition and preprocessing Environmental data acquired by multiple sensors, including image sequences It LiDAR point cloud IMU data at , ωt and GPS location ; Image preprocessing, including denoising, distortion correction, and feature extraction. ,in It is a Gaussian filter kernel. This is the distortion correction amount; Point cloud preprocessing includes downsampling, ground segmentation, and obstacle clustering; ; S1.2: Data Fusion Based on Kalman Filtering State vector definition: ; Motion model (constant acceleration model): ; in: ; Observation model: ; in, ; The Kalman filter update steps include prediction and update: predict: ; renew: ; S1.3: Dynamic Environment Map Construction and Update Dynamically occupying grid map updates: ; Dynamic object tracking, using multi-object tracking algorithms: .

[0006] Furthermore, the specific implementation of step S2 is as follows: S2.1: Task Requirements Analysis and Scenario Understanding LLM Input: Environment Map M t Task Description D task Drone status ; LLM output: Task priority W =[ w 1 ,w 2 ,…,w K Constraints C ; S2.2: Modeling of Multi-Objective Optimization Problems Decision variables: Task assignment matrix Motion parameters ; Objective function: ; in: ; S2.3: Distributed Task Allocation Algorithm Task allocation based on auction algorithm: ; in It is a drone i For the task j The evaluation function; S2.4: Optimization Solution of Motion Parameters Using Sequential Quadratic Programming (SQP): ; The constraints include dynamic constraints and environmental constraints.

[0007] Furthermore, the specific implementation of step S3 is as follows: S3.1: Trajectory parameter initialization based on UAV dynamics UAV dynamics model: ; Trajectory parameters: ; S3.2: Generation of Variable Speed ​​Curve Based on trapezoidal velocity profile: ; The time points are obtained by solving the following equations: ; S3.3: Generation of Three-Dimensional Trajectory Primitives The extension of the Clothoid in 3D space: ; ,in It is a unit tangent vector that satisfies curvature constraints. , It is the maximum permissible curvature; S3.4: Cooperative Trajectory Optimization and Conflict Avoidance Distributed model predictive control formulation: ; Avoid conflicts between drones.

[0008] Furthermore, the specific implementation of step S4 is as follows: S4.1: Calculation of Collision Probability among Multiple Drones Collision probability based on Gaussian distribution: ; Using approximate calculations: ; S4.2: Lyapunov Stability Verification Define the Lyapunov function: ; Stability conditions: ; in x >0, verified by solving linear matrix inequalities. S4.3: Construction of Safe Potential Field and Collision Detection Definition of the safety potential field function: ; in: ; S4.4: Real-time Trajectory Adjustment Strategy Gradient descent-based adjustment: ; in P It is the learning rate.

[0009] Furthermore, the specific implementation of step S5 is as follows: S5.1: Adaptive PID Trajectory Tracking Control Control Law: ; Among them, adaptive gain adjustment: ; S5.2: Real-time monitoring of performance indicators Monitoring indicator: Tracking error e(t) Energy consumption E ( t Safety indicators S ( t ) Performance evaluation: ; S5.3: LLM Dynamic Decision Making and Replanning Triggering Replanning conditions: ; S5.4: Online trajectory optimization and update Predict path integrals using a model: ; in, It is the sampling trajectory. It is the cost of trajectory.

[0010] The present invention further proposes a multi-UAV cooperative trajectory planning task management system based on LLM, which adopts the above-described multi-UAV cooperative trajectory planning method based on LLM, including: The perception module is used to execute step S1 to realize multi-source environmental perception and dynamic map construction; The decision module is used to execute step S2 to realize LLM intelligent task allocation and motion parameter optimization; The planning module is used to execute step S3 to realize multi-UAV collaborative trajectory planning and optimization; The verification module is used to perform step S4 to achieve real-time security verification and trajectory adjustment; The execution module is used to execute step S5 to achieve trajectory execution and dynamic replanning.

[0011] The beneficial effects of this invention are as follows: The method constructs a dynamic environment map through multi-source sensor fusion; utilizes a large language model for high-level task understanding and scene analysis, and constructs a multi-objective optimization model to intelligently allocate tasks and optimize motion parameters; generates initial trajectories that satisfy constraints based on the UAV dynamics model, and employs a collaborative optimization algorithm to avoid conflicts between UAVs; ensures trajectory safety through formalized safety verification methods (such as collision probability calculation and Lyapunov stability analysis), and possesses real-time adjustment capabilities; finally, it executes the trajectory through adaptive control and achieves dynamic replanning based on performance monitoring and LLM decision-making. This invention integrates perception, decision-making, planning, verification, and execution modules, enabling efficient, safe, and autonomous collaborative operation of multiple UAV swarms in dynamic environments. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of the multi-UAV cooperative trajectory planning method based on LLM of the present invention; Figure 2 This is a flowchart of environmental perception and data fusion in this invention; Figure 3 This is a flowchart of LLM task allocation and decision-making in this invention; Figure 4 This is a flowchart of the multi-UAV cooperative trajectory planning in this invention; Figure 5This is a flowchart of the security verification and conflict avoidance process in this invention; Figure 6 This is a flowchart of trajectory execution and dynamic adjustment in this invention; Figure 7 This is a flowchart illustrating the transmission of parameters and data in this invention; Figure 8 This is a block diagram of the multi-UAV collaborative trajectory planning task management system based on LLM of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0015] Reference Figure 1-7 A cooperative trajectory planning method for multiple UAVs based on LLM includes the following steps: Step S1: Multi-source environmental perception and dynamic map construction S1.1: Environmental data is collected through multiple sensors, including at least an image sensor, a lidar and an inertial measurement unit. The collected environmental data is preprocessed, including noise reduction, distortion correction and feature extraction, to become preprocessed data, so as to extract effective sensing information. S1.2: Based on Kalman filtering, the preprocessed data is fused to form fused data, which is used to estimate the UAV's own state and the position and velocity of environmental targets; S1.3: Construct a dynamic occupancy grid map based on the fused data, and update the dynamic occupancy grid map in real time to reflect environmental changes, while tracking dynamic objects; Step S2: LLM Intelligent Task Allocation and Motion Parameter Optimization S2.1: Utilize LLM to analyze task requirements and understand scenarios. LLM receives dynamic environment maps, task descriptions, and UAV status as inputs and outputs task priorities and constraints. S2.2: Construct a multi-objective optimization problem based on the scenario understanding results. The multi-objective optimization problem considers multiple objectives, including task completion time, energy consumption, and safety. S2.3: A distributed task allocation algorithm is used to solve the multi-objective optimization problem, achieving efficient task allocation; S2.4: Obtain the motion parameters of each UAV through optimization, including velocity, acceleration and heading angle, and satisfy the UAV dynamics constraints; Step S3: Multi-UAV cooperative trajectory planning and optimization S3.1: Based on the UAV dynamics model, considering constraints including mass, inertia, and thrust, initialize the trajectory parameters; S3.2: Generates a smooth variable speed curve, enabling the drone to accelerate and decelerate smoothly; S3.3: Generates the three-dimensional trajectory prototype, which forms the basic segment of the UAV's flight path; S3.4: A distributed optimization method is used to collaboratively optimize the trajectories of multiple UAVs to avoid conflicts between UAVs; Step S4: Real-time security verification and trajectory adjustment S4.1: Calculate the collision probability between multiple UAVs based on a probabilistic model to assess trajectory safety; S4.2: Using Lyapunov stability theory, construct the Lyapunov function and verify the stability of the trajectory; S4.3: Construct a safe potential field to detect potential conflicts and provide obstacle avoidance information; S4.4: Based on the safety verification results, the trajectory is updated in real time using an adjustment strategy to ensure flight safety; Step S5: Trajectory Execution and Dynamic Replanning S5.1: The trajectory tracking control is achieved by using an adaptive PID control algorithm, and the controller dynamically adjusts the control gain according to the tracking error; S5.2: Real-time monitoring of UAV performance indicators, which include at least tracking error, energy consumption, and safety status; S5.3: Based on the LLM dynamic decision-making mechanism, replanning is triggered when environmental changes or performance anomalies are detected; S5.4: Online optimization and update of trajectories to adapt to dynamic environmental changes.

[0016] As a further implementation, step S1 is specifically implemented as follows: S1.1: Multi-sensor data acquisition and preprocessing Environmental data acquired by multiple sensors, including image sequences It LiDAR point cloud IMU data at , ωt and GPS location ; Image preprocessing, including denoising, distortion correction, and feature extraction. ,in It is a Gaussian filter kernel. This is the distortion correction amount; Point cloud preprocessing includes downsampling, ground segmentation, and obstacle clustering; ; S1.2: Data Fusion Based on Kalman Filtering State vector definition: ; Motion model (constant acceleration model): ; in: ; Observation model: ; in, ; The Kalman filter update steps include prediction and update: predict: ; renew: ; S1.3: Dynamic Environment Map Construction and Update Dynamically occupying grid map updates: ; Dynamic object tracking, using multi-object tracking algorithms: .

[0017] As a further implementation, step S2 is specifically implemented as follows: S2.1: Task Requirements Analysis and Scenario Understanding LLM Input: Environment Map M t Task Description D task Drone status ; LLM output: Task priority W =[ w 1 ,w 2 ,…,w K Constraints C ; S2.2: Modeling of Multi-Objective Optimization Problems Decision variables: Task assignment matrix Motion parameters ; Objective function: ; in: ; S2.3: Distributed Task Allocation Algorithm Task allocation based on auction algorithm: ; in It is a drone i For the task j The evaluation function; S2.4: Optimization Solution of Motion Parameters Using Sequential Quadratic Programming (SQP): ; The constraints include dynamic constraints and environmental constraints.

[0018] As a further implementation method, step S3 is specifically implemented as follows: S3.1: Trajectory parameter initialization based on UAV dynamics UAV dynamics model: ; Trajectory parameters: ; S3.2: Generation of Variable Speed ​​Curve Based on trapezoidal velocity profile: ; The time points are obtained by solving the following equations: ; S3.3: Generation of Three-Dimensional Trajectory Primitives The extension of the Clothoid in 3D space: ; ,in It is a unit tangent vector that satisfies curvature constraints. , It is the maximum permissible curvature.

[0019] S3.4: Cooperative Trajectory Optimization and Conflict Avoidance Distributed model predictive control formulation: ; Avoid conflicts between drones.

[0020] As a further implementation, step S4 is specifically implemented as follows: S4.1: Calculation of Collision Probability among Multiple Drones Collision probability based on Gaussian distribution: ; Using approximate calculations: ; S4.2: Lyapunov Stability Verification Define the Lyapunov function: ; Stability conditions: ; in x >0, verified by solving linear matrix inequalities. S4.3: Construction of Safe Potential Field and Collision Detection Definition of the safety potential field function: ; in: ; S4.4: Real-time Trajectory Adjustment Strategy Gradient descent-based adjustment: ; in P It is the learning rate.

[0021] As a further implementation method, step S5 is specifically implemented as follows: S5.1: Adaptive PID Trajectory Tracking Control Control Law: ; Among them, adaptive gain adjustment: ; S5.2: Real-time monitoring of performance indicators Monitoring indicator: Tracking error e(t) Energy consumption E ( t Safety indicators S ( t ) Performance evaluation: ; S5.3: LLM Dynamic Decision Making and Replanning Triggering Replanning conditions: ; S5.4: Online trajectory optimization and update Predict path integrals using a model: ; in, It is the sampling trajectory. It is the cost of trajectory.

[0022] The detailed parameter passing process is as follows: Inter-step parameter transfer relationships: 1. Parameter passing from step S1 to step S2: Input: Environment Map M t Drone status ; Output: Task assignment matrix A Motion parameters ; Transfer function: [ A,U ]= fLLM ( Mt,U,D task ).

[0023] 2. Parameter passing from step S2 to step S3: enter: A,U ; Output: Trajectory parameters ; Transfer function: .

[0024] 3. Parameter passing from step S3 to step S4: enter: ; Output: Verification results V Adjust trajectory ; Transfer function: .

[0025] 4. Parameter passing from step S4 to step S5: enter: ; Output: Control commands Performance indicators P ; transfer function .

[0026] 5. Feedback transmission from step S5 to S2: enter: ; Output: Replanning trigger signal; Transfer function: .

[0027] The detailed mathematical derivation is as follows: 1. Derivation of Kalman Filter in Environmental Perception Derivation of the state transition matrix: Considering UAV dynamics, a constant acceleration model is used: ; In this model, A is the system matrix. For a three-dimensional constant acceleration model: ; Theoretical derivation of process noise covariance: ; in, It is the noise intensity matrix for a continuous time process.

[0028] 2. Derivation of the optimization problem in task allocation Derivation of the dual problem of the task assignment problem: Original question: ; Duality problem: ; Subject to i+j·vij .

[0029] 3. Derivation of Differential Flatness in Trajectory Planning For unmanned aerial vehicle (UAV) systems, prove that they possess differential flatness: Flat output: All state and control inputs can be represented as a function of a flat output and its derivative: ; 4. Derivation of Collision Probability in Safety Verification Precise collision probability calculation: Assuming the drone is ellipsoidal in shape, the collision probability is: ; in B i It is a drone i The shape area.

[0030] 5. Derivation of Lyapunov Stability in Control Constructing Lyapunov functions: ; in, It is parameter error.

[0031] Differentiating, we get: ; By designing an adaptive law, .

[0032] Advantages of this method: 1. LLM Intelligent Decision Making: Adaptive task allocation and parameter optimization improve efficiency.

[0033] 2. Multi-drone collaboration: Distributed trajectory planning to avoid conflicts.

[0034] 3. Real-time security verification: Formal methods ensure security.

[0035] 4. Dynamic adjustment: LLM responds to environmental changes in real time.

[0036] 5. Energy consumption optimization: Considering energy consumption constraints, extend the drone's flight time.

[0037] In summary, the method of this invention constructs a dynamic environment map through multi-source sensor fusion; utilizes a large language model for high-level task understanding and scene analysis, and constructs a multi-objective optimization model to intelligently allocate tasks and optimize motion parameters; generates initial trajectories that meet constraints based on the UAV dynamics model, and employs a collaborative optimization algorithm to avoid conflicts between UAVs; ensures trajectory safety through formal safety verification methods (such as collision probability calculation and Lyapunov stability analysis), and possesses real-time adjustment capabilities; finally, executes the trajectory through adaptive control, and achieves dynamic replanning based on performance monitoring and LLM decision-making. This invention integrates perception, decision-making, planning, verification, and execution modules, enabling efficient, safe, and autonomous collaborative operation of multiple UAV swarms in dynamic environments. Example 2

[0038] This embodiment takes smart construction site monitoring as an example, assuming there are 3 drones and 2 monitoring tasks: 1. Step S1 is executed: Input: Sensor data; Processing: Kalman filter fusion to obtain an environment map. M t； Output: Drone status .

[0039] 2. Step S2 is executed: LLM analysis task requirements: Task 1 - Monitoring of Area A, Task 2 - Monitoring of Area B; Optimization solution: Obtain task allocation A； Motion parameters: , representing the velocity in the x-direction, the velocity in the y-direction, the acceleration in the x-direction, the acceleration in the y-direction, and the heading angle, respectively.

[0040] 3. Step S3 is executed: Trajectory planning: Generate the trajectory of UAV 1 from the starting point to area A; Trajectory parameters: .

[0041] 4. Step S4 is executed: Safety verification: Calculate the probability of collision. ; Stability verification: ; Output: Verification passed.

[0042] 5. Step S5 is executed: Trajectory tracking: PID control achieves trajectory tracking; Performance monitoring: Tracking error <0.1m; When a new obstacle is detected, replanning is triggered. Example 3

[0043] Reference Figure 8 A multi-UAV cooperative trajectory planning task management system based on LLM, employing the multi-UAV cooperative trajectory planning method based on LLM described in Embodiment 1 or 2, includes: The perception module is used to execute step S1 to realize multi-source environmental perception and dynamic map construction; The decision module is used to execute step S2 to realize LLM intelligent task allocation and motion parameter optimization; The planning module is used to execute step S3 to realize multi-UAV collaborative trajectory planning and optimization; The verification module is used to perform step S4 to achieve real-time security verification and trajectory adjustment; The execution module is used to execute step S5 to achieve trajectory execution and dynamic replanning.

[0044] The present invention has the following systematic advantages: 1. Modular design: Corresponding to different levels and functional modules of the system, which facilitates system integration and debugging.

[0045] 2. Clear data flow: Figure 7 The parameter passing relationships between steps are clearly shown, ensuring the consistency of the algorithm.

[0046] 3. Real-time performance guarantee: Optimized loop and parallel processing design to meet the real-time requirements of drone swarms.

[0047] 4. Excellent security: Multiple verification mechanisms and conflict avoidance strategies ensure system security and reliability.

[0048] 5. Highly adaptive: A complete feedback loop and dynamic adjustment mechanism adapt to complex environmental changes.

[0049] 6. Solid mathematical foundation: Each process is based on rigorous mathematical derivation and optimization theory.

[0050] This invention achieves high efficiency and security in smart construction site monitoring through LLM intelligent decision-making and multi-UAV collaborative trajectory planning, possessing significant theoretical value and practical application prospects. This method can effectively handle multi-UAV collaborative tasks in dynamic environments, providing a complete technical solution for UAV swarm applications.

Claims

1. A cooperative trajectory planning method for multiple unmanned aerial vehicles based on LLM, characterized in that: Includes the following steps: Step S1: Multi-source environmental perception and dynamic map construction S1.1: Environmental data is collected through multiple sensors, including at least an image sensor, a lidar and an inertial measurement unit. The collected environmental data is preprocessed, including noise reduction, distortion correction and feature extraction, to become preprocessed data, so as to extract effective sensing information. S1.2: Based on Kalman filtering, the preprocessed data is fused to form fused data, which is used to estimate the UAV's own state and the position and velocity of environmental targets; S1.3: Construct a dynamic environment map based on the fused data. The dynamic environment map is a dynamically occupied grid map, and the dynamic occupied grid map is updated in real time to reflect environmental changes and track dynamic objects. Step S2: LLM Intelligent Task Allocation and Motion Parameter Optimization S2.1: Utilize LLM to analyze task requirements and understand scenarios. LLM receives dynamic environment maps, task descriptions, and UAV status as inputs and outputs task priorities and constraints. S2.2: Construct a multi-objective optimization problem based on the scenario understanding results. The multi-objective optimization problem considers multiple objectives, including task completion time, energy consumption, and safety. S2.3: A distributed task allocation algorithm is used to solve the multi-objective optimization problem, achieving efficient task allocation; S2.4: Obtain the motion parameters of each UAV through optimization, including velocity, acceleration and heading angle, and satisfy the UAV dynamics constraints; Step S3: Multi-UAV cooperative trajectory planning and optimization S3.1: Based on the UAV dynamics model, considering constraints including mass, inertia, and thrust, initialize the trajectory parameters; S3.2: Generates a smooth variable speed curve, enabling the drone to accelerate and decelerate smoothly; S3.3: Generates the three-dimensional trajectory prototype, which forms the basic segment of the UAV's flight path; S3.4: A distributed optimization method is used to collaboratively optimize the trajectories of multiple UAVs to avoid conflicts between UAVs; Step S4: Real-time security verification and trajectory adjustment S4.1: Calculate the collision probability between multiple UAVs based on a probabilistic model to assess trajectory safety; S4.2: Using Lyapunov stability theory, construct the Lyapunov function and verify the stability of the trajectory; S4.3: Construct a safe potential field to detect potential conflicts and provide obstacle avoidance information; S4.4: Based on the safety verification results, the trajectory is updated in real time using an adjustment strategy to ensure flight safety; Step S5: Trajectory Execution and Dynamic Replanning S5.1: The trajectory tracking control is achieved by using an adaptive PID control algorithm, and the controller dynamically adjusts the control gain according to the tracking error; S5.2: Real-time monitoring of UAV performance indicators, which include at least tracking error, energy consumption, and safety status; S5.3: Based on the LLM dynamic decision-making mechanism, replanning is triggered when environmental changes or performance anomalies are detected; S5.4: Online optimization and update of trajectories to adapt to dynamic environmental changes.

2. The multi-UAV cooperative trajectory planning method based on LLM according to claim 1, characterized in that: The specific implementation of step S1 is as follows: S1.1: Multi-sensor data acquisition and preprocessing Environmental data acquired by multiple sensors, including image sequences It LiDAR point cloud IMU data at , ωt and GPS location ; Image preprocessing, including denoising, distortion correction, and feature extraction. ,in It is a Gaussian filter kernel. This is the distortion correction amount; Point cloud preprocessing includes downsampling, ground segmentation, and obstacle clustering; ； S1.2: Data Fusion Based on Kalman Filtering State vector definition: ； Motion model (constant acceleration model): ； in: ； Observation model: ; in, ; The Kalman filter update steps include prediction and update: predict: ； renew: ； S1.3: Dynamic Environment Map Construction and Update Dynamically occupying grid map updates: ； Dynamic object tracking, using multi-object tracking algorithms: 。 3. The multi-UAV cooperative trajectory planning method based on LLM according to claim 1, characterized in that: The specific implementation of step S2 is as follows: S2.1: Task Requirements Analysis and Scenario Understanding LLM Input: Environment Map M t， Task Description D task Drone status ; LLM output: Task priority W =[ w 1 ,w 2 ,…,w K Constraints C ; S2.2: Modeling of Multi-Objective Optimization Problems Decision variables: Task assignment matrix Motion parameters ; Objective function: ; in: ； S2.3: Distributed Task Allocation Algorithm Task allocation based on auction algorithm: ； in It is a drone i For the task j The evaluation function; S2.4: Optimization Solution of Motion Parameters Using Sequential Quadratic Programming (SQP): ； The constraints include dynamic constraints and environmental constraints.

4. The multi-UAV cooperative trajectory planning method based on LLM according to claim 1, characterized in that: The specific implementation of step S3 is as follows: S3.1: Trajectory parameter initialization based on UAV dynamics UAV dynamics model: ； Trajectory parameters: ; S3.2: Generation of Variable Speed ​​Curve Based on trapezoidal velocity profile: ； The time points are obtained by solving the following equations: ； S3.3: Generation of Three-Dimensional Trajectory Primitives The extension of the Clothoid in 3D space: ； ,in It is a unit tangent vector that satisfies curvature constraints. , It is the maximum permissible curvature; S3.4: Cooperative Trajectory Optimization and Conflict Avoidance Distributed model predictive control formulation: ； Avoid conflicts between drones.

5. The multi-UAV cooperative trajectory planning method based on LLM according to claim 1, characterized in that: The specific implementation of step S4 is as follows: S4.1: Calculation of Collision Probability among Multiple Drones Collision probability based on Gaussian distribution: ； Using approximate calculations: ； S4.2: Lyapunov Stability Verification Define the Lyapunov function: ； Stability conditions: ； in x >0, verified by solving linear matrix inequalities. S4.3: Construction of Safe Potential Field and Collision Detection Definition of the safety potential field function: ； in: ； S4.4: Real-time Trajectory Adjustment Strategy Gradient descent-based adjustment: ； in P It is the learning rate.

6. The multi-UAV cooperative trajectory planning method based on LLM according to claim 1, characterized in that: The specific implementation of step S5 is as follows: S5.1: Adaptive PID Trajectory Tracking Control Control Law: ； Among them, adaptive gain adjustment: ； S5.2: Real-time monitoring of performance indicators Monitoring indicator: Tracking error e(t) Energy consumption E ( t Safety indicators S ( t ) Performance evaluation: ; S5.3: LLM Dynamic Decision Making and Replanning Triggering Replanning conditions: ； S5.4: Online trajectory optimization and update Predict path integrals using a model: ； in, It is the sampling trajectory. It is the cost of trajectory.

7. A multi-UAV cooperative trajectory planning task management system based on LLM, characterized in that: The method described in any one of claims 1-6 includes: The perception module is used to execute step S1 to realize multi-source environmental perception and dynamic map construction; The decision module is used to execute step S2 to realize LLM intelligent task allocation and motion parameter optimization; The planning module is used to execute step S3 to realize multi-UAV collaborative trajectory planning and optimization; The verification module is used to perform step S4 to achieve real-time security verification and trajectory adjustment; The execution module is used to execute step S5 to achieve trajectory execution and dynamic replanning.

Images