Fully Distributed Model Predictive Control of Connected Automated Vehicles in Intersections: Theory and Vehicle Experiments

TL;DR

This paper presents a fully distributed model predictive control framework for coordinating connected automated vehicles at intersections, validated through vehicle experiments demonstrating safety and efficiency.

Contribution

It introduces a novel distributed control architecture using convexification techniques for real-time vehicle coordination at intersections, suitable for in-vehicle implementation.

Findings

Experimental validation shows effective vehicle coordination.

The control system ensures safety at intersections.

Convexification enables fast solution of nonconvex control problems.

Abstract

We propose a fully distributed control system architecture, amenable to in-vehicle implementation, that aims to safely coordinate connected and automated vehicles (CAVs) at road intersections. For control purposes, we build upon a fully distributed model predictive control approach, in which the agents solve a nonconvex optimal control problem (OCP) locally and synchronously, and exchange their optimized trajectories via vehicle-to-vehicle (V2V) communication. To accommodate a fast solution of the nonconvex OCPs, we apply the penalty convex-concave procedure which solves a convexified version of the original OCP. For experimental evaluation, we complement the predictive controller with a localization layer, being in charge of self-localization, and an estimator, which determines joint collision points with other agents. Experimental tests reveal the efficacy of the proposed control system architecture.