Reinforcement Learning at Container Terminals

Seaport container terminals serve a crucial role in global supply chains. They must be capable of handling ever-larger ships in less time at competitive prices. As a response, terminals are seeking new approaches to optimize operational decisions to improve their container throughput and operational efficiency. The use of artificial intelligence (AI) methods keeps promising great potential for solving diverse and complex application cases in logistics planning. Especially reinforcement learning (RL) methods are increasingly being explored as machine learning modules no longer strictly follow a specific goal, but programmed agents act and self-optimize in a virtual training environment. A comprehensive review, classification, and discussion of relevant literature on RL at container terminals for operational decision problems is the subject of this paper. Thereby, the feasibility of RL approaches is shown, but also the hurdles and current limitations.