Traffic Light Management for Automated Guided Vehicle Systems using Deep Reinforcement Learning

Abstract

Intersection management in the presence of automated guided vehicles (AGVs) in industrial settings is a major problem that the artificial intelligence community tries to tackle. In this paper, we perform a comparative analysis of three well-known reinforcement learning techniques that produce policies to control a signalized intersection to coordinate traffic flow among automated guided vehicles. We implemented and tested four scenarios with different numbers of lanes and traffic light phase configurations. The analysis allowed us to gain critical insight into ways to improve the coordination of single intersections operating AGVs. The results suggest that decreasing the number of phases and increasing the number of lanes can be beneficial. As for the algorithms used, the Deep Q-Networks (DQN) and Double DQN performed better in simpler scenarios, whereas Dueling DQN seems to be more appropriate for more complex intersection settings. (c) 2023 The Authors.