Research on modeling end-to-end autonomous driving of UGV in temporary road environments
In recent years,deep learning based autonomous driving models have gained keen academic attention.Conventional autonomous driving models are mostly built in series using multi-level modules,which individually perform functions such as perception,planning and tracking.However,the problems of high coupling and poor risk resistance exist.To address these issues,this paper proposes an end-to-end model for autonomous driving of UGV on temporary roads.It employs three visual sensor images as inputs,GCViT as the backbone network for image feature extraction,transformer network and GRU network to output local planning paths,and PID algorithm to output corner information to achieve automatic tracking of UGV.Our experimental results show the single frame trajectory planning of the end-to-end model takes about 80ms,with an average trajectory deviation of 0.689 meters.Our model meets the real-time requirements and performs the tracking task of UGV on temporary roads.
国家科技期刊平台
NSTL
万方数据
