Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Decentralized coordination of autonomous vehicles for Lane-Free driving under restricted communication

Geronymakis Pavlos

Full record

---

Summary

In recent years, there are significant advancements in the field of autonomousvehicles, due to progress in sensor technology, detection systems, computingpower, and artificial intelligence algorithms. All together generated a greateracceptance of autonomous vehicles in the public eye and have widen thehorizon of research in this field. One recently introduced concept currentlyexplored is lane-free traffic, which studies vehicles moving in roads withoutbeing subjected to traditional lane-based restrictions and are free to moveacross the width of the road. In this domain, existing research has employedthe Max-plus message-passing algorithm in order to enable coordination ofvehicles in the environment. For that research endeavor, it is assumed thatall vehicles moving along the lane-free highway are connected, can commu-nicate and coordinate with each other. Max-plus is an iterative algorithm forcoordination in multi-agent environments, where connected agents exchangelocal maximization messages until convergence or a timeout.However, when allowing for the realistic perspective that there exist inde-pendent agents (vehicles) that are unable or unwilling to communicate withothers, a certain level of uncertainty is created that renders the aforemen-tioned coordination approach ineffective. For the new independent agentsintroduced to the lane-free environment, we consider that their observationalinformation (position, speed) is known to the agents coordinating with Max-plus, but not their intended actions. To combat this, in the present diplomathesis we adjust the Max-plus algorithm accordingly, so that coordinatingagents can observe and take into consideration independent agents via emu-lated messages.These emulated messages are formed through the employment of vari-ous methods from the literature of Multiagent Decision Making under Un-certainty, namely the Maximax, Maximin, Hurwicz, Minimax Regret andLaplace criteria. Moreover, we formulate a simple opponent model that pre-dicts the actions of the observed agents. Finally, we provide a thorough eval-uation of our approach, including a detailed comparison of all criteria usedfor calculation of messages. Our experimental evaluation examines the per-formance across various situations, with ranging penetration rates for inde-pendent agents and under different levels of added noise in their behavior.

Available Files

Services