VLS: A Reinforcement Learning-Based Value Lookahead Strategy for Multi-product Order Fulfillment
Abstract
The fast-paced growth of online ordering comes with additional challenges that are not as prevalent in the traditional brick-and-mortar retail sector. For one, e-commerce retailers have the challenge and flexibility of selecting which warehouses and stores should fulfill online orders. This is known as the omnichannel order fulfillment problem. In this work, we combine a lookahead search tree strategy with a reinforcement learning-based cost-to-go estimator to produce an effective cost-saving order fulfillment strategy, named the value lookahead strategy (VLS). Furthermore, we design and implement a simulator with the capabilities to simulate a wide variety of order fulfillment scenarios which can allow for developing, training, and evaluating order fulfillment strategies, even in the presence of limited data. We show that using these in conjunction can produce an order fulfillment strategy with lower total fulfillment costs than all other order fulfillment strategies we compared against.
Publication Title
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Recommended Citation
Wickman, R., Li, J., & Zhang, X. (2023). VLS: A Reinforcement Learning-Based Value Lookahead Strategy for Multi-product Order Fulfillment. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 14177 LNAI, 518-534. https://doi.org/10.1007/978-3-031-46664-9_35
