Oversized Package Placement Optimization in Warehouses

Abstract

Amazon Logistics’ delivery stations process two types of packages: oversized packages (OV) and non-OV packages (non-OV). Currently, the placement of OV to OV racks is coupled with non-OV placement to non-OV racks because Amazon’s algorithm does not explicitly assign OV to a rack but places it in the rack where the non-OV package of the closest delivery destination is. The result is frequent overflow from OV rack to floor for some racks while very few packages in others, leading to severely inefficient use of space and uneven labor distribution.

This project identifies the root cause of the problem and develops 4 solutions to eliminate overflow and evenly distribute the packages within a delivery station. Different configurations and current package placement assignment methodologies are considered and current industry work on optimally assigning and picking products of different types is reviewed. Based on this research, we modeled package placement using integer linear programming, dynamic averaging, and pooling.

This thesis provides general frameworks for optimizing placing and picking different types of products in a warehouse setting through the case of two product types scenario. It can be further expanded to multiple product type scenarios in a general supply chain and logistics system where efficient and fair use of resources has been a constant challenge.