MPC-Based Routing and Tracking Architecture for Safe Autonomous Driving in Urban Traffic

Abstract

This paper presents a configurable routing and tracking architecture that uses multi-objective Model Predictive Control (MPC) as its driving algorithm to guarantee safe autonomous driving of different vehicle types. The architecture consists of three main components and primarily relies on labeled map data to generate optimal path and velocity trajectories in accordance with the vehicle type and the desired control objectives. We begin with introducing the overall system architecture and its different inputs, outputs, and components. We also briefly explain the open-source services utilized in this work for trajectory generation, namely OpenStreetMap and GraphHopper. We then focus on formulating the multi-objective MPC problem and its vehicle-specific constraints, which is solved offline to generate the reference path and velocity trajectories. Afterwards, we discuss some adaptions to the system model and the controller operating strategy to incorporate real-time tracking of these trajectories while guaranteeing collision avoidance. Finally, we successfully demonstrate the system’s feasibility by numerically evaluating its performance in a typical urban driving scenario for different vehicles.