Sampling-Based Trajectory Repairing for Autonomous Vehicles

Ensuring the safety of autonomous vehicles is a challenging task, especially if the planned trajectories do not consider all traffic rules or they are physically infeasible. Since replanning the complete trajectory is often computationally expensive, efficient methods are necessary for resolving such situations. One solution is to deform or repair an initially-planned trajectory, which we call trajectory repairing. Our approach first detects the part of an invalid trajectory that can stay unchanged. Afterward, we use a hierarchical structure and our novel sampling-based algorithm informed closed-loop rapidly-exploring random trees (informed CL-RRTs) to efficiently repair the remaining part of the trajectory. We evaluate our approach with different traffic scenarios from the CommonRoad benchmark suite. The computational efficiency is demonstrated by comparing the computation times with those required when replanning the complete trajectory.