Reinventing Order Picking at Amazon Robotics Fulfillment Centers
Algorithms and intelligence for large-scale robotic-assisted order fulfillment.
Since joining Amazon Fulfillment Technologies in 2024, my main line of work has been the new generation of large-scale online optimization models that decide, in real time, how picking work is assigned across machines and human associates in Amazon Robotics fulfillment centers. These are tri-partite assignment problems of enormous scale, solved continuously as orders arrive, and the resulting algorithms are deployed in North American and European fulfillment centers (see video), where they orchestrate very cool robots.
Two research threads have grown out of this work. The first uses machine learning to speed up large-scale optimization itself: learned models shrink the feasible region the solver must explore, improving convergence speed without degrading solution quality. The second, learning to configure optimization models, infers the best model parameters and configuration for online problems that must be solved in real time, improving resiliency and the main operational KPIs.
I have presented this work at the INFORMS Annual Meetings in Seattle (2024) and Atlanta (2025) under the title Reinventing Order Picking at Amazon Robotics Fulfillment Centers – Algorithms and Intelligence, and at Amazon’s Modeling and Optimization Seminar (2025). Two related peer-reviewed papers appeared at internal venues in 2025 (Amazon Consumer Science Summit and the Amazon Robotics Conference); their contents remain Amazon-confidential.