Optimizing the Performance of Robotic Mobile Fulfillment Systems

A robotic mobile fulfillment system is a novel type of automated part-to-picker material handling system. In this type of system, robots transport mobile shelves, called pods, containing items between the storage area and the workstations. It is well suited to e-commerce, due to its modularity and it’s ability to adapt to changing orders patterns. Robots can nearly instantaneously switch between inbound and outbound tasks, pods can be continually repositioned to allow for automatic sorting of the inventory, pods can contain many different types of items, and unloaded robots can drive underneath pods, allowing them to use completely different routes than loaded robots.

This thesis studies the performance of robotic mobile fulfillment systems by solving decision problems related to warehouse design, inventory and resource allocation, and real-time operations. For warehouse design, a new queueing network is developed that incorporates realistic robot movement, storage zones, and multi-line orders. For inventory allocation, we develop a new type of queueing network, the cross-class matching multi-class semi-open queueing network, which can be applied to other systems as well. Resource (re)allocation is modeled by combining queueing networks with Markov decision processes while including time-varying demand. This model compares benchmark policies from practice with the optimal policy. Lastly, we study decision rules for real-time operations by using detailed, large scale simulations.