Vehicle trajectory prediction of multi-vehicle interaction scene based on Transformer
Vehicle trajectory prediction was a key technology in autonomous driving,which was of great significance to improved the safety of the path planning function of autonomous vehicles.The historical vehicle tracks of multiple vehicles in the lane-changing scene were modeled first,scene samples were extracted based on the NGSIM data set,and an attention-mechanism-based Transformer encoder-decoder model was constructed.By capturing potential relationships between tracks,the corresponding predictive tracks were generated by a recursive method.Experiments showed that Transformer model could better capture the spatio-temporal interaction characteristics between adjacent trajectories,and present better prediction results in complex trajectory prediction tasks.
NETL
NSTL
万方数据
维普
