AN UAV Dynamic Path Planning Method Based on Prior Information
LIN Jing
LI Chen
QIU Xingye
Abstract：Unmanned aerial vehicles(UAVs)face technical challenges of insufficient information utilisation and sparse rein-forcement learning rewards when performing path planning tasks in complex dynamic environments.Existing methods typically rely solely on locally perceived information for path decisions,making it difficult to effectively utilize global prior information acquired before the task.This results in inefficient algorithm training and limited path planning performance.To address the aforementioned issues,this paper proposes an UAV dynamic path planning method based on reinforcement learning with prior information,which achieves effective integration of heterogeneous information by designing a mechanism for collaborative fusion of multiple information sources.The A*algorithm is employed to preprocess known static environmental information,generating global navigation pathways as prior guidance data.Real-time local fields of view are converted into RGB three-channel image representations to accurately per-ceive dynamic obstacle states.A four-channel tensor input is constructed,fusing global navigation information with local perception data across channel dimensions to provide rich environmental state representations for reinforcement learning algorithms.A path planning decision model based on Deep Q-Networks is designed,employing a convolutional neural network architecture to extract multi-channel fusion features.An innovative hierarchical reward function based on global guidance is proposed,utilizing navigation path information to construct continuous learning signals.This effectively addresses the issue of reward sparsity,significantly en-hancing the algorithm's convergence efficiency.Compared to traditional reinforcement learning methods,it demonstrates notable im-provements in both training convergence speed and path quality.The experimental results validate the effectiveness of the prior infor-mation guidance mechanism in path planning for complex dynamic environments,providing new technical insights for the advance-ment of UAV autonomous navigation technology.
Keywords：Unmanned Aerial Vehicle(UAV)path planningreinforcement learningprior informationmulti-informa-tion fusion
Publication Date：2025-12-20
Online Publishing Date：2026-03-23(First online date of this platform, not the publication date of the document)
Pages：6( 67-72 )