Iterative LQR with Discrete Barrier States for Efficient Collision Avoidance of Autonomous Vehicle

Abstract

Ensuring safe motion planning of autonomous vehicles, especially on collision avoidance in an emergency, is challenging. Many optimal control methods have been proposed for safe and efficient nonlinear motion planning. It is well known that iterative linear quadratic regulator (iLQR) is particularly suitable for nonlinear optimization. Although many methods have been developed to solve the inequality constrained differential dynamic programming (DDP), this paper proposes a method for efficient vehicle motion planning by using discrete barrier states to iLQR. In the proposed method, multiple inequality constraints such as obstacle avoidance and road boundaries are reflected in a single barrier state in the collision avoidance scenario and applied to the iLQR/DDP framework. For comparison, both the unconstrained iLQR using the potential field method and the barrier state iLQR proposed in this study were simulated in the collision avoidance scenario. We tested on a real-time software-in-the-loop simulation using CarMaker and ROS. Compared with the unconstrained iLQR, our proposed method generated a safe and efficient optimized trajectory for autonomous vehicles in emergency situations such as fishhook tests.