Research on temporal consistency and robustness in local planning of intelligent vehicles

Local planner can improve the capacity of planning system for intelligent vehicles by avoiding obstacles while tracking the reference path. Safety check is the one of the fundamental functions of a local planner. However, revealed by simulation and experiments, the widely used optimization-based local planning methods are unlikely to maintain temporal consistency without any accurate global positioning information. Temporal inconsistency will result in the deviation of the vehicle's actual trajectory from the original planned results, which will finally make the safety check invalid. This paper presents a forward prediction-based local planning algorithm which holds temporal consistency in the results without requiring any accurate global positioning information. Besides the temporal consistency issue, controlling error is another reason for safety check failure. Most current researches take error into consideration by enlarging the size of obstacles. Such methods are unable to prevent the vehicle from entering the dilated obstacle areas. In this paper, controlling error is introduced in the generation of the local paths. The traditional path with no width is replaced by a strip of path here. Based on the simulation results of the V-Rep virtual reality software, the forward prediction-based method features better safety check ability as compared with the optimization-based local planning methods. The proposed algorithm was applied to the intelligent vehicle by Beijing Institute of Technology which participated in The Future Challenge 2013. The vehicle succeeded to finish the event without human operation.