Research on Transformer Pyramid Networks Based Unmanned Aerial Vehicles Autonomous Navigation Algorithm
For addressing the problem of efficiently utilizing multi-sensor heterogeneous data in complex environ-ments for autonomous navigation,a new neural network architecture is proposed to achieve fast data fusion and trajectory planning.A lightweight pyramid feature network is designed for trajectory prediction,where an efficient Transformer net-work structure is introduced into the multi-level deep feature,and a pooling attention layer is constructed,which improves the feature representation ability while reducing the computational complexity and realizes the fast and robust trajectory pre-diction.Based on Unreal Engine and AirSim,the UAV simulation system is designed,and a large number of autonomous flight simulation experiments are conducted by controlling the UAV with the trained model.The experimental results show that the proposed algorithm can quickly achieve multi-sensor heterogeneous information fusion,generate feasible trajectories in real time and stably,realize autonomous obstacle avoidance in complex obstacle environments with a success rate of up to 81%,and effectively improve the autonomous navigation ability of UAVs.
NETL
NSTL
万方数据
