Optimizing the spatio-temporal resource search problem with reinforcement learning (GIS Cup)

Abstract

Collecting spatio-temporal resources is an important goal in many real-world use cases such as finding customers for taxicabs. In this paper, we tackle the resource search problem posed by the GIS Cup 2019 where the objective is to minimize the average search time of taxicabs looking for customers. The main challenge is that the taxicabs may not communicate with each other and the only observation they have is the current time and position. Inspired by radial transit route structures in urban environments, our approach relies on round trips that are used as action space for a downstream reinforcement learning procedure. Our source code is publicly available at https://github.com/Fe18/TripBanditAgent.