On-Road Motion Planning for Autonomous Vehicles

We present a motion planner for autonomous on-road driving, especially on highways. It adapts the idea of a on-road state lattice. A focused search is performed in the previously identified region in which the optimal trajectory is most likely to exist. The main contribution of this paper is a computationally efficient planner which handles dynamic environments generically. The Dynamic Programming algorithm is used to explore in spatiotemporal space and find a coarse trajectory solution first that encodes desirable maneuvers. Then a focused trajectory search is conducted using the "generate-and-test" approach, and the best trajectory is selected based on the smoothness of the trajectory. Analysis shows that our scheme provides a principled way to focus trajectory sampling, thus greatly reduces the search space. Simulation results show robust performance in several challenging scenarios.