Throughput optimization of multi-agent robotic automated warehouses

Abstract

In 2003 Kiva Systems (now Amazon Robotics) introduced a new type material handling automation to the world. The system is based on the principle that the physical infrastructure that contains inventory should be mobile. Kiva achieved this remarkable advancement by employing a fleet of robots to move shelving to human operators. Broadly, these types of systems are defined in the literature as multi-agent robotic systems. Amazon acquired Kiva Systems in 2012 to incorporate the technology into their operations. The goal of this thesis is to optimize the throughput of warehouses employing multi-agent robotic automation. It is assumed that extracting inventory from the automated system is the limiting factor in maximizing throughput (i.e. downstream process are unconstrained). Two strategies are advocated: 1) performing velocity segregation of inventory within the automation via a bifurcation between fast selling and slow selling inventory, 2) maximizing pick rates through policies that increase worker retention. It will be shown that velocity segregation increases machine efficiency by increasing the efficiency of delivering inventory to human operators. This assertion will be investigated by developing a theoretical understanding of how inventory velocity impacts machine efficiency and simulating different types of stow strategies impact on system efficiency. It is estimated that some stow strategies can increase machine efficiency by as much as 30%. It will also be shown that the number of man-hours worked by inexperienced pickers explains practically all of the variability of aggregate pick cycle times and hence pick rates, which motivates the argument for worker retention. Together, these two modifications are estimated to increase throughput by 10% over current baseline.