Robotic warehousing operations: a learn-then-optimize approach to large-scale neighborhood search

TL;DR

This paper supports robotic parts-to-picker operations in warehousing by optimizing order–workstation assignments, item–pod assignments, and the schedule of order fulfillment at workstations by optimizing order–workstation assignments, and the schedule of order fulfillment at workstations.

Abstract

The rapid deployment of robotics technologies requires dedicated optimization algorithms to manage large fleets of autonomous agents. This paper supports robotic parts-to-picker operations in warehousing by optimizing order-workstation assignments, item-pod assignments and the schedule of order fulfillment at workstations. The model maximizes throughput, while managing human workload at the workstations and congestion in the facility. We solve it via large-scale neighborhood search, with a novel learn-then-optimize approach to subproblem generation. The algorithm relies on an offline machine learning procedure to predict objective improvements based on subproblem features, and an online optimization model to generate a new subproblem at each iteration. In collaboration with Amazon Robotics, we show that our model and algorithm generate much stronger solutions for practical problems than state-of-the-art approaches. In particular, our solution enhances the utilization of robotic fleets by coordinating robotic tasks for human operators to pick multiple items at once, and by coordinating robotic routes to avoid congestion in the facility.

Figures (12)

• Figure 1: (a) Full-scale fulfillment center, with workstations in the foreground and inventory shelves stored behind. (b) Robotic agents bring shelves to the workstations in part-to-picker operations.
• Figure 2: Time-space network representation of fulfillment operations.
• Figure 3: Congestion contribution calculation for a traveling arc $a\in\mathcal{A}_{\text{tr}}$ that departs its origin at time $t=0$ and arrives at a workstation 60 seconds later. There are 20 shortest paths, shown in black. The congestion contribution for three intersections at time $t=30$ seconds is calculated by counting the fraction of those paths that pass through the intersection 30 seconds after leaving the origin.
• Figure 4: Learning-enhanced subproblem selection vs. learn-then-optimize subproblem generation.
• Figure 5: Trade-off between optimization error and approximation error, with $10^{50}$ candidate subproblems.