Algorithm for Robotic Picking in Amazon Fulfillment Centers Enables Humans and Robots to Work Together Effectively

Abstract

This paper describes how Amazon redesigned the robotic picking algorithm used in Amazon Robotics (AR) fulfillment centers (FCs) to enable humans and robots to work together effectively. In AR FCs, robotic drives fetch storage pods filled with inventory for associates to pick. The picking algorithm needs to decide which specific units of inventory on which pods should be picked to fulfill customer order shipments. We want to do so in a way that is most efficient and distance traveled by drives per unit picked is the key performance metric. This new algorithm reduced the distance traveled by drives per unit picked by 62% without negative operational impact and has since been implemented in all AR FCs. This improvement reduced the number of drives required in AR FCs by 31%, which amounted to half a billion dollars in savings. The redesigned algorithm enabled seamless collaboration between associates and robots, and its effectiveness in scaling up convinced Amazon to make AR FCs the standard for new FCs, allowing Amazon to reduce the storage footprint by about 29% compared with non-AR FCs. History: This paper was refereed.