An Adaptive Trajectory Planning for Autonomous Driving Using Trapezoidal Decomposition

Trajectory planning is the core module of autonomous driving system and its main challenges lie in the decisions in non-convex space and the trade-off between safety, comfort and efficiency performances. This paper presents a noval framework that combines trapezoidal cell decomposition and convex optimization to generate optimal trajectories for autonomous vehicles in structured environment. The approach decoupes a 3D trajectory planning problem with timestamps information into 2D path planning and speed planning problems. Each 2D problem can be solved by a three-phase procedure. First, the trapezoidal cell decomposition is used to extract trapezoidal driving corridors. Second, the graph is constructed by adaptive sampling instead of inflexible uniform sampling, and the coarse path(or speed) and corresponding corridors are obtained by grpah search method. In the last phase, the coarse path(or speed) is optimized by piecewise Bézier polynomials and optimazation technique. The noval framework, called CB Planner, is especially useful for extracting corridors and generating coarse trajectory in complex environment. Experimental results show that compared with some state-of-art planners, this planner can generate higher quality trajectories and is more suitable for complex environments.