Towards safe, smooth, and stable path planning for on-road autonomous driving under uncertainty

Abstract

For on-road autonomous driving, path planner needs to generate a target path which is not only safe and smooth, but also stable under uncertainty propagated from sensing such as localization error and perception error. In this paper, we propose a path planning method which generates a target path in two steps: reference path planning and local path planning. The reference path planner keeps stability of lateral positing in a lane and also narrows down solution space efficiently. The local path planner ensures safety and smoothness of the target path by generating thousands of Bezier curves that follows the reference path with various shape and evaluating them comprehensively in highly parallelized processing. The proposed path planner has been integrated into our autonomous driving system and tested extensively on surface roads and freeways in real-world, showing capability to handle complex scenarios and stability under uncertain environment sensing.