CN-121994232-A - LLM-based multi-unmanned aerial vehicle collaborative track planning method and task management system

CN121994232ACN 121994232 ACN121994232 ACN 121994232ACN-121994232-A

Abstract

The invention provides a collaborative trajectory planning method and a task management system of multiple unmanned aerial vehicles based on LLM, and relates to the technical field of unmanned aerial vehicle control. The method comprises the steps of S1, multisource environment perception and dynamic map construction, S2, LLM intelligent task allocation and motion parameter optimization, S3, multi-unmanned-plane collaborative track planning and optimization, S4, real-time safety verification and track adjustment, S5, track execution and dynamic re-planning. According to the invention, a dynamic map is constructed through multisource environment awareness, intelligent task allocation and motion parameter optimization are performed by using LLM, a safety track is generated by adopting collaborative track 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 rule of the unmanned aerial vehicle is fused in track planning, the optimal track is directly calculated, the efficiency is high, and the dynamic environment adaptability is ensured through LLM real-time decision.

Inventors

GUAN ZHENCHANG
ZHANG HE
Gui Yunquan
LI DAJUN

Assignees

福州大学
中建三局第三建设工程有限责任公司

Dates

Publication Date: 20260508
Application Date: 20260116

Claims (7)

1. A collaborative trajectory planning method of a multi-unmanned aerial vehicle based on LLM is characterized by comprising the following steps: Step S1, multisource environment perception and dynamic map construction S1.1, acquiring environmental data through a multi-sensor, wherein the multi-sensor at least comprises an image sensor, a laser radar and an inertial measurement unit, and preprocessing the acquired environmental data, including denoising, distortion correction and feature extraction, into preprocessed data so as to extract effective perception information; S1.2, fusing the preprocessed data based on Kalman filtering to form fused data, and estimating the state of the unmanned aerial vehicle and the position and speed of an environmental target; S1.3, constructing a dynamic environment map according to the fusion data, wherein the dynamic environment map is a dynamic occupation grid map, and updating the dynamic occupation grid map in real time to reflect environment changes and simultaneously tracking dynamic objects; step S2, LLM intelligent task allocation and motion parameter optimization S2.1, analyzing task demands and understanding scenes by using LLM, wherein the LLM receives a dynamic environment map, task descriptions and unmanned aerial vehicle states as input and outputs task priorities and constraint conditions; S2.2, constructing a multi-objective optimization problem based on a scene understanding result, wherein the multi-objective optimization problem simultaneously considers a plurality of objectives in task completion time, energy consumption and safety; s2.3, solving a multi-objective optimization problem by adopting a distributed task allocation algorithm to realize efficient allocation of tasks; S2.4, obtaining motion parameters of each unmanned aerial vehicle, including speed, acceleration and course angle, through optimization solution, and meeting the unmanned aerial vehicle dynamic constraint; Step S3, planning and optimizing collaborative trajectories of multiple unmanned aerial vehicles S3.1, initializing track parameters based on an unmanned aerial vehicle dynamic model by considering mass, inertia and thrust constraints; S3.2, generating a smooth variable speed curve so that the unmanned aerial vehicle can accelerate and decelerate smoothly; S3.3, generating a three-dimensional track primitive to form a basic section of the unmanned aerial vehicle flight path; s3.4, adopting a distributed optimization method to carry out collaborative optimization on the tracks of the multiple unmanned aerial vehicles, so as to avoid conflict among the unmanned aerial vehicles; Step S4, real-time security verification and track adjustment S4.1, calculating collision probability among multiple unmanned aerial vehicles based on a probability model, and evaluating track safety; s4.2, constructing a Lyapunov function and verifying the stability of the track by utilizing a Lyapunov stability theory; s4.3, constructing a safety potential field to detect potential conflict and providing obstacle avoidance information; S4.4, updating the track by adopting a real-time adjustment strategy according to the safety verification result, so as to ensure the flight safety; Step S5, track execution and dynamic re-planning S5.1, adopting a self-adaptive PID control algorithm to realize track tracking control, and dynamically adjusting control gain by a controller according to tracking errors; S5.2, monitoring performance indexes of the unmanned aerial vehicle in real time, wherein the performance indexes of the unmanned aerial vehicle at least comprise tracking errors, energy consumption and safety states; S5.3, based on an LLM dynamic decision mechanism, triggering re-planning when detecting environmental change or abnormal performance; and S5.4, online optimizing the update track to adapt to dynamic environment changes.
2. The collaborative trajectory planning method for multiple unmanned aerial vehicles based on LLM according to claim 1, wherein the implementation of step S1 is as follows: s1.1 Multi-sensor data acquisition and preprocessing Environmental data acquired by multiple sensors, including image sequence It, liDAR point cloud IMU data at, ωt and GPS position ; Image preprocessing including denoising, distortion correction, and feature extraction Wherein In the form of a gaussian filter kernel, Is a distortion correction amount; The point cloud preprocessing comprises downsampling, ground segmentation and obstacle clustering; ; S1.2. Kalman filtering-based data fusion State vector definition: ; motion model (constant acceleration model): ; Wherein: ; observation model: ; Wherein, the ; The Kalman filtering updating step includes prediction and updating: And (3) predicting: ; Updating: ; S1.3 dynamic Environment map construction and update Dynamically occupying grid map updates: ; Dynamic object tracking, using a multi-target tracking algorithm: 。
3. The collaborative trajectory planning method for multiple unmanned aerial vehicles based on LLM according to claim 1, wherein the specific implementation of step S2 is as follows: S2.1 task demand analysis and scene understanding LLM inputs environmental map M t, task description D task , unmanned plane status ; LLM outputs task priority W= [ W 1 ,w 2 ,…,w K ], constraint condition C; s2.2 Multi-objective optimization problem modeling Decision variable task allocation matrix Motion parameters ; Objective function: ; Wherein: ; s2.3 distributed task allocation algorithm Task allocation based on auction algorithm: ; Wherein the method comprises the steps of The evaluation function of the unmanned plane i to the task j; S2.4 motion parameter optimization solving Sequence Quadratic Programming (SQP) was used: ; Wherein the constraints include dynamic constraints and environmental constraints.
4. The collaborative trajectory planning method for multiple unmanned aerial vehicles based on LLM according to claim 1, wherein the specific implementation of step S3 is as follows: S3.1 track parameter initialization based on unmanned aerial vehicle dynamics Unmanned aerial vehicle dynamic model: ; Track parameters: ; S3.2 variable speed Curve Generation Based on trapezoidal velocity profile: ; wherein the time node is obtained by solving the following equation: ; s3.3 three-dimensional track primitive generation Expansion of the gyrate primitive (Clothoid) in 3D space: ; Wherein Is a unit tangent vector, meets curvature constraint , Is the maximum allowable curvature; s3.4 collaborative trajectory optimization and Conflict avoidance Distributed model predictive control formulation: ; And collision among unmanned aerial vehicles is avoided.
5. The collaborative trajectory planning method for multiple unmanned aerial vehicles based on LLM according to claim 1, wherein the implementation of step S4 is as follows: s4.1, calculating collision probability of multiple unmanned aerial vehicles Probability of collision based on gaussian distribution: ; using approximation calculations: ; S4.2 Lyapunov stability verification Definition of the lyapunov function: ; stability conditions: ; Where x >0, verified by solving the linear matrix inequality S4.3 safety potential field construction and Conflict detection Safety potential field function definition: ; Wherein: ; S4.4 track real-time adjustment strategy Adjustment based on gradient descent: ; where P is the learning rate.
6. The collaborative trajectory planning method for multiple unmanned aerial vehicles based on LLM according to claim 1, wherein the implementation of step S5 is as follows: S5.1 self-adaptive PID track tracking control Control law: ; Wherein the gain is adaptively adjusted: ; S5.2 real-time monitoring of Performance index Monitoring index, tracking error E (t), energy consumption E (t), safety index S (t) Performance evaluation: ; S5.3 LLM dynamic decision and re-programming trigger Reprofiling conditions: ; S5.4 on-line track optimization update Predicting path integration using a model: ; Wherein, the Is the sampling trajectory of the sample, Is the track cost.
7. A collaborative trajectory planning task management system for a LLM-based multi-unmanned aerial vehicle, comprising: the sensing module is used for executing the step S1 to realize multi-source environment sensing and dynamic map construction; The decision module is used for executing the step S2 to realize LLM intelligent task allocation and motion parameter optimization; The planning module is used for executing the step S3 to realize the planning and optimization of the collaborative trajectories of the multiple unmanned planes; the verification module is used for executing the step S4 to realize real-time safety verification and track adjustment; and the execution module is used for executing the step S5 to realize track execution and dynamic re-planning.

Description

LLM-based multi-unmanned aerial vehicle collaborative track planning method and task management system Technical Field The invention relates to the technical field of unmanned aerial vehicle control, in particular to a collaborative track planning method and a task management system of multiple unmanned aerial vehicles based on LLM. Background With the development of unmanned aerial vehicle technology, a multi-unmanned aerial vehicle system is widely applied to intelligent building site monitoring. However, the existing unmanned aerial vehicle management platform has the defects that task allocation depends on manual experience, intelligence is lacked, unmanned aerial vehicle dynamic constraint is not considered in track planning, efficiency is low, a collision avoidance mechanism in multi-unmanned aerial vehicle cooperation is imperfect, and real-time adjustment capability is poor, so that the unmanned aerial vehicle management platform cannot adapt to a dynamic environment. Disclosure of Invention The invention provides a collaborative track planning method and a task management system of a multi-unmanned aerial vehicle based on LLM, which solve the problems that in the prior art, task allocation of the multi-unmanned aerial vehicle system depends on manpower, dynamic constraint is not fully considered in track planning, a collaborative collision avoidance mechanism is imperfect and real-time adjustment capability is poor. The invention firstly provides a collaborative trajectory planning method of a plurality of unmanned aerial vehicles based on LLM, which comprises the following steps: Step S1, multisource environment perception and dynamic map construction S1.1, acquiring environmental data through a multi-sensor, wherein the multi-sensor at least comprises an image sensor, a laser radar and an inertial measurement unit, and preprocessing the acquired environmental data, including denoising, distortion correction and feature extraction, into preprocessed data so as to extract effective perception information; S1.2, fusing the preprocessed data based on Kalman filtering to form fused data, and estimating the state of the unmanned aerial vehicle and the position and speed of an environmental target; s1.3, constructing a dynamic occupation grid map according to the fusion data, updating the dynamic occupation grid map in real time to reflect environmental changes, and simultaneously tracking dynamic objects; step S2, LLM intelligent task allocation and motion parameter optimization S2.1, analyzing task demands and understanding scenes by using LLM, wherein the LLM receives a dynamic environment map, task descriptions and unmanned aerial vehicle states as input and outputs task priorities and constraint conditions; S2.2, constructing a multi-objective optimization problem based on a scene understanding result, wherein the multi-objective optimization problem simultaneously considers a plurality of objectives in task completion time, energy consumption and safety; s2.3, solving a multi-objective optimization problem by adopting a distributed task allocation algorithm to realize efficient allocation of tasks; S2.4, obtaining motion parameters of each unmanned aerial vehicle, including speed, acceleration and course angle, through optimization solution, and meeting the unmanned aerial vehicle dynamic constraint; Step S3, planning and optimizing collaborative trajectories of multiple unmanned aerial vehicles S3.1, initializing track parameters based on an unmanned aerial vehicle dynamic model by considering mass, inertia and thrust constraints; S3.2, generating a smooth variable speed curve so that the unmanned aerial vehicle can accelerate and decelerate smoothly; S3.3, generating a three-dimensional track primitive to form a basic section of the unmanned aerial vehicle flight path; s3.4, adopting a distributed optimization method to carry out collaborative optimization on the tracks of the multiple unmanned aerial vehicles, so as to avoid conflict among the unmanned aerial vehicles; Step S4, real-time security verification and track adjustment S4.1, calculating collision probability among multiple unmanned aerial vehicles based on a probability model, and evaluating track safety; s4.2, constructing a Lyapunov function and verifying the stability of the track by utilizing a Lyapunov stability theory; s4.3, constructing a safety potential field to detect potential conflict and providing obstacle avoidance information; S4.4, updating the track by adopting a real-time adjustment strategy according to the safety verification result, so as to ensure the flight safety; Step S5, track execution and dynamic re-planning S5.1, adopting a self-adaptive PID control algorithm to realize track tracking control, and dynamically adjusting control gain by a controller according to tracking errors; S5.2, monitoring performance indexes of the unmanned aerial vehicle in real time, wherein the performance indexes of the unmanned