Machine Learning, Waveless Picking, and E-commerce Warehouses


Speaker


Abstract

Amazon enjoyed its biggest day ever on Prime-Day 2016; a whopping 600 orders/second. Such a growth of online shopping next to piece picking, and same-day delivery require responsive and adaptive warehouses. Traditional warehousing is a poor option for handling the needs of e-commerce. E-commerce requires warehouses capable of picking, packing, and shipping many single-line orders. Add to this complex environment, the need of online retailers to support an ever-increasing selection of products. This has created a huge piles of data possibly with hidden relationships. In order to address these issues, the control software of warehouses are “re-programmed” constantly by humans in order capture the uncertainty and complexity in the environment. High investments are made on warehouse management systems with improved computational capability to capture the potential relationships in the piles of data. Alternatively, a warehouse which can adapt to constant changes in the environment would avoid the need to constant system redesign. In this research we study a machine learning approach to control the e-commerce warehouses: Reinforcement learning. Reinforcement learning results in a system which learns over time from changes to the environment, and automatically adapts itself to achieve better performance.