Collaboration Between the City and Machine Learning Community is Crucial to Efficient Autonomous Vehicles Routing

TL;DR

Effective autonomous vehicle routing requires collaboration between city authorities and the machine learning community to ensure traffic stability, fairness, and environmental sustainability, given the limitations of current MARL algorithms and models of human behavior.

Contribution

This paper emphasizes the importance of city-ML collaboration for developing safe, fair, and efficient AV routing algorithms amidst current technical and modeling challenges.

Findings

MARL algorithms often fail to converge or need long training periods

Simulated training is limited due to lack of accurate human behavior models

Real-world training risks traffic destabilization and increased emissions

Abstract

Autonomous vehicles (AVs), possibly using Multi-Agent Reinforcement Learning (MARL) for simultaneous route optimization, may destabilize traffic networks, with human drivers potentially experiencing longer travel times. We study this interaction by simulating human drivers and AVs. Our experiments with standard MARL algorithms reveal that, both in simplified and complex networks, policies often fail to converge to an optimal solution or require long training periods. This problem is amplified by the fact that we cannot rely entirely on simulated training, as there are no accurate models of human routing behavior. At the same time, real-world training in cities risks destabilizing urban traffic systems, increasing externalities, such as $CO_2$ emissions, and introducing non-stationarity as human drivers will adapt unpredictably to AV behaviors. In this position paper, we argue that city authorities must collaborate with the ML community to monitor and critically evaluate the routing algorithms proposed by car companies toward fair and system-efficient routing algorithms and regulatory standards.