A probabilistic optimization approach for motion planning of autonomous vehicles

摘要

This paper presents a novel probabilistic approach for improving the motion planning performance of autonomous driving. The proposed approach is based on the sampling-based planning algorithm, which generates an optimal trajectory from a set of trajectory candidates. In order to treat the uncertainty in the perception data and the vehicle system, the particle filter framework is applied to the motion planning algorithm with four main steps: the time update of the trajectory candidates, the perception measurement update, the trajectory selection and the motion goal resampling. Since the proposed planning algorithm recursively generates an optimal trajectory, the time update of the trajectory candidate updates the motion goals of the trajectory candidates in the previous step using the vehicle model, and it also generates a new set of candidates. In order to evaluate the optimality of each candidate with regard to the safety and the reliability, a perception measurement update is performed. In this step, the importance weight of each candidate is computed using perception data and its adaptive likelihood function. Based on the candidates with updated importance weights, an optimal trajectory is determined in the trajectory selection. Then, the motion goal resampling modifies the set of motion goals based on the importance weights for efficient management of the motion goals in the iterative planning algorithm. The developed algorithm is validated using various types of test. The results show that the proposed method not only provides an integrated probabilistic interface between the perception and the planning but also results in an excellent performance in terms of the computation efficiency.