作者: Constantin Hubmann , Jens Schulz , Gavin Xu , Daniel Althoff , Christoph Stiller
DOI: 10.1109/ITSC.2018.8569729
关键词: Computer science 、 Mathematical optimization 、 Belief state 、 Partially observable Markov decision process 、 Planner 、 State space 、 Monte Carlo method 、 Merge (version control) 、 Logistic regression
摘要: Autonomous driving in urban environments requires the capability of merging into narrow gaps. In cases high traffic density this becomes more complex since one must consider interaction with other vehicles. We formulate problem as a Partially Observable Markov Decision Process (POMDP) by including surrounding drivers state space to realize interactive behavior. The is solved online an anytime Monte Carlo sampling algorithm combination efficient A* rollout heuristic. This makes combined lateral and longitudinal optimization possible. resulting policy optimized regarding various future merge scenarios approaches most suitable gap while taking account uncertain behavior drivers. Therefore, we present novel motion model representing cooperation It based on logistic regression estimating probability for cooperative human driver given scene. demonstrate performance our simulated scenarios. approaching gaps, performing merges gaps are discussed.