Real-Time Graph-Based Path Planning for Autonomous Racecars

Abstract

This thesis presents a computationally efficient graph-based motion planner designed for high speed autonomous racing. The emergence of autonomous racing competitions, such as the Indy Autonomous Challenge (IAC), have sought to test the limits of autonomous vehicle tech- nology to accelerate development within the domain. A fast, safe, and reliable motion planning algorithm is developed for an autonomous vehicle operating under high-speed conditions such as the ones in the IAC. A variety of planning methods are investigated for this purpose, such as graph-based planning and sampling-based planning, among others. A graph-based method using the A* search algorithm is selected due to its computational efficiency, reliability, and predictability in structured environments. The proposed planner is augmented with techniques for integrating vehicular constraints with path smoothing and edge generation.

Two versions of the proposed path planner are presented. The version used for the 2021 and 2022 IAC competitions on oval tracks is developed and test results from simulation and running the planner in real time competition are presented. Additionally, improvements to the planner are implemented to enhance the dynamic feasibility of the planned path and allow for use on road courses. The improved planner is tested on an autonomous consumer sedan as well as in simulation. Both iterations of the proposed algorithm are shown to produce dynamically feasible maneuvers in the presence of a priori unknown obstacles while maintaining faster than real-time performance.