Machine Learning for Autonomous Vehicle Fleet Coordination

Abstract

In recent years, the relevance of autonomous vehicles has increased greatly. In combination with the increase in shared mobility services, autonomous vehicles could provide a shift towards more sustainable transport systems. However, typical transport-related fundamental challenges such as scalability, large and sudden demand peaks, regional seasonal variations, and high goals for traffic system efficiency have persisted. To efficiently realize autonomous demand-responsive mobility concepts, research is needed on innovative methods for the coordination of vehicle fleets, introducing new ideas.

The recent advances in computer technology have created applications for machine learning research in many digital platforms, including intelligent and autonomous transportation systems. The classroom implementation of a relatively small-scale demand-responsive autonomous vehicle fleet constitutes a complex optimization problem that can be tackled with a learned agent. Applications at this level have the ability to impact society broadly. In traffic psychology, it has been repeatedly hypothesized that traffic automation would be beneficial. Advancements in this area have great potential in improving road traffic efficiency, safety, and comfort. In addition, the above coordination solutions are generic, with no application-specific requirement for on-road evaluation. In this work, we specifically focus on ML-based techniques to solve the vehicle coordination problem. In a broader sense, we try to answer the generic challenges in autonomous vehicle fleet coordination at microscopic levels, involving the assumptions of autonomy and closed vehicle fleets in urban areas.