Historical Improvement Optimal Motion Planning with Model Predictive Trajectory Optimization for On-road Autonomous Vehicle

Abstract

This paper presents an efficient, robust, comfortable, and real-time motion planning framework for on-road autonomous vehicles. This proposed framework aims to enhance the performance of motion planning in complex environments such as driving in the urban area. It uses a path velocity decomposition method to separate the motion planning problem into path planning and velocity planning. The novelty lies in the use of Historical data in the $SL$ coordinate in the framework of a tree version of Rapidly-exploring Random Graph (RRT*) technique in path planner, called HSL-RRT*, which grows the path tree efficiently by the data from previous planning cycle. The velocity planner uses a Nonlinear Model Predictive Controller (NMPC) to generate optimal velocity along the path generated from the path planner, taking account of vehicle constraints and comfort. Analytic and simulation results are presented to validate the approach, with a special focus on the robustness and efficiency of the algorithm operating in complex scenarios.