Agile Mixed-Integer-based Lane-Change MPC for Collision-Free and Efficient Autonomous Driving

Abstract

Obstacle avoidance and lane-change functionalities are crucial requirements for automated driving that are usually developed to excel in specific road scenarios and therefore, to some extent, lack agility and versatility. This work proposes a versatile vehicle motion controller for connected and autonomous vehicles (CAVs) that, independently of the encountered traffic scenario, realizes safe and efficient lane-change and overtaking maneuvers while implicitly guaranteeing precise obstacle avoidance at all times. This is achieved by introducing a two-layer model predictive control architecture that utilizes mixed-integer-based obstacle avoidance and lane selection formulations. The linear time-invariant optimal control problems are efficiently formulated in Frenet coordinates, incorporate the position predictions of surrounding traffic participants, and guarantee globally optimal solutions. The proposed architecture combines two structurally distinct policies, namely velocity and time-gap tracking, to achieve efficient and safe maneuvering. A method to reduce the computational complexity of the implemented mixed-integer quadratic programming (MIQP) problems is developed. The controller's agility and robustness with respect to multiple complex traffic scenarios are tested in detailed traffic simulations, as well as validated via high-fidelity co-simulations with the CARLA Simulator.