Efficient routing for multiple AGVs in container terminals using hybrid deep learning and metaheuristic algorithm

Abstract

This study aims to present the efficient routing of multiple agent systems by optimizing the automated guided vehicle (AGV) movement, waiting time, or container-lifting actions, providing optimal routing solutions in automated container terminals (ACTs). Through integrated scheduling, the AGV agent can determine the efficient route to the container, then pick it up and transport it to the final destination. Users can provide the starting position of the AGVs, container position, and drop-down location. The algorithm returns action lists for the AGV to perform. AGV route is implemented based on mapping the action in the action lists and the layout, preventing collisions and deadlocks among AGVs. We utilized the advantage actor-critic (A2C) reinforcement learning method combined with the ant colony optimization (ACO) of a swarm intelligence algorithm to solve the optimal routing problem in AGV-based ACTs. More specifically, this study presents the optimal action strategy that ACO-A2C finds for each AGV and a route scheme that each AGV can travel without colliding with other AGVs and obstacles. This novel method can potentially improve ACTs’ equipment utilization for efficient and competitive management.