Game Theory Models for the Verification of the Collective Behaviour of Autonomous Cars

TL;DR

This paper discusses multi-agent game theory models for verifying collective autonomous car behavior, highlighting the limitations of non-cooperative strategies and proposing intention-aware adaptation as a promising approach.

Contribution

It introduces a formal framework using online routing game models to analyze and potentially guarantee optimal collective behavior of autonomous cars.

Findings

Non-cooperative autonomous adaptation cannot ensure optimal traffic flow.

Intention-aware adaptation with decision constraints may prevent unwanted behaviors.

Online routing game models are promising for formal verification.

Abstract

The collective of autonomous cars is expected to generate almost optimal traffic. In this position paper we discuss the multi-agent models and the verification results of the collective behaviour of autonomous cars. We argue that non-cooperative autonomous adaptation cannot guarantee optimal behaviour. The conjecture is that intention aware adaptation with a constraint on simultaneous decision making has the potential to avoid unwanted behaviour. The online routing game model is expected to be the basis to formally prove this conjecture.