Towards Smart Transportation System

Abstract

Smart transportation system is a cross-field research topic that involves both the organizations that manage the large-scaled system and individual end-users who enjoy these services. Recent advancement of machine learning-based algorithms has either enabled or improved a wide range of applications due to its strength in making accurate predictions for complex problems with a minimal amount of domain knowledge and great ability of generalization. These nice properties imply potential to be explored for building smart transportation system. This paper studies how deep reinforcement learning (DRL) can be used to optimize the operating policy in modern bike sharing systems. As a case study, the authors demonstrate the potential power of the modern DRL by showing a policy-gradient-based reinforcement learning approach to the rebalancing problem in a bike sharing system, which can simultaneously improve both the user experience and reduce the operational expense.