[2J4-GS-2] Machine learning: Deep reinforcement learning

[2J4-GS-2-04] Analysis of Service area Adaptation for a Ride-share by using Multi-agent deep reinforcement learning

〇Naoki Yoshida¹, Itsuki Noda², Toshiharu Sugawara¹ (1. Waseda University, 2. National Institute of Advanced Industrial Science and Technology)

Keywords:Multi-agent Reinforcement Leraning, Ride-share service, Deep Q-network

An excessive number of (taxi) drivers have participated in the ride-share, but they may result in many unassigned empty cars in a city due to the concentration, leading to traffic jams and the waste of energy. Therefore, an effective strategy to appropriately decide the service areas where agents have to wait for passengers is a crucial issue for easing such problems and for achieving quality service. For this purpose, we use the method called the SAAMS to allocate service areas to agents by using deep reinforcement learning based on demand prediction data. In this paper, we compare the performances and the characteristics of specified service areas generated by a joint action learner (JAL) and independent leaners (ILs). On experimental results suggested that ILs are more likely to decide their individually specific areas through learning and could adapt to dynamic changing of passengers’ demands than the JAL.