Analyzing the performance of distributed conflict resolution among autonomous vehicles

TL;DR

This study evaluates how cooperation, non-cooperation, centralization, and distribution affect autonomous vehicle traffic performance using a particle-based model that captures complex phenomena like fuel exhaustion and collisions.

Contribution

It introduces a particle-based modeling approach for autonomous vehicle traffic, incorporating energy dynamics and a distributed resource allocation protocol, to analyze performance impacts.

Findings

Cooperative resource allocation achieves optimal collective cost.

Distributed approaches can improve equity among vehicles.

Centralized methods optimize overall efficiency.

Abstract

This paper presents a study on how cooperation versus non-cooperation, and centralization versus distribution impact the performance of a traffic game of autonomous vehicles. A model using a particle-based, Lagrange representation, is developed, instead of a Eulerian, flow-based one, usual in routing problems of the game-theoretical approach. This choice allows representation of phenomena such as fuel exhaustion, vehicle collision, and wave propagation. The elements necessary to represent interactions in a multi-agent transportation system are defined, including a distributed, priority-based resource allocation protocol, where resources are nodes and links in a spatial network and individual routing strategies are performed. A fuel consumption dynamics is developed in order to account for energy cost and vehicles having limited range. The analysis shows that only the scenarios with cooperative resource allocation can achieve optimal values of either collective cost or equity coefficient, corresponding respectively to the centralized and to the distributed cases.