I. Introduction
In automated warehouse systems, a team of robots works together to fulfill a set of customer orders. Each order comprises one or more items found on the warehouse floor, which must be delivered to a picking station for consolidation and delivery. In automated sortation centres, meanwhile, a similar problem arises. Here, the robotic team is tasked with carrying mail tasks from one of several emitter stations, where new parcels arrive, to a bin of sorted tasks, all bound for the same processing facility where they will be dispatched for delivery. Illustrated in Fig. 1, such systems are at the heart of logistics operations for major online retailers such as Amazon and Alibaba. Practical success in both of these contexts depends on computing timely solutions to a challenging optimization problem known in the literature as Multi-agent Pickup and Delivery (MAPD) [1].