Multi-Area Sampling-Based Spatiotemporal Trajectory Planning for Autonomous Driving in Dynamic On-Road Scenarios

This paper focuses on the spatiotemporal trajectory planning problem faced by autonomous driving with a dynamic on-road situation. To solve the swing problem which is caused by the motions of obstacles, a multi-area sampling method is proposed. The main idea is sampling endpoints in a series of defined areas at a fixed time interval, which will generate suitable trajectories with speed information to deal with complex maneuver tasks. Considering the driving safety and comfort, the cost function is designed deliberately for the generated trajectories in each area to evaluate the behaviors of the automobile. Then, the best trajectory in the whole course is found by the dynamic programming-based approach, which is presented to optimize the problem-solving process and at the same time reduce the computational burden which is brought about by the multi-area sampling method. Finally, the effectiveness of the proposed trajectory planning method is demonstrated in different overtaking scenarios of structured roads.