Imitation learning from human-generated spatial-temporal data

Abstract

Human dwellers make daily decisions by their own “strategies” governing their mobility dynamics (e.g., Uber drivers have preferred working regions and times, and urban commuters have preferred routes and transit modes). Understanding and characterizing the unique decision-making strategies of human agents has great potential in promoting their individual well-being. In this paper, we outline a novel spatial-temporal imitation learning (STIL) framework that defines, investigates, and addresses the emerging research challenges of analyzing and learning human decision-making strategies from human-generated spatial-temporal data. We present the state-of-the-art imitation learning algorithms, and the limitations of these algorithms in analyzing human-generated spatial-temporal data. Moreover, we present our preliminary studies, and outline the challenging open questions.