Amazon deployed 520,000 mobile robots (Kiva robots) in its warehouses between 2014 and 2022. These robots pick up shelves (pods) and transport them to stationary human pickers or to robotic picking stations. Why is this 'goods-to-robot' approach more efficient than 'robot-to-goods' (robots navigating to items, picking them, and transporting)?
AGoods-to-robot is always superior regardless of task
BGoods-to-robot minimizes robot travel distance: instead of many robots traveling long distances to fetch items, one robot delivers a shelf to a workstation, and a human or robot picks multiple items from it. Additionally, humans are faster at picking diverse items than robot grippers designed for few item types. This hybrid approach combines human and robot strengths
CGoods-to-robot doesn't work in real warehouses
DRobot-to-goods is always faster
The hybrid approach is the practical compromise in deployed warehouse systems. Pure centralization is too slow; pure decentralization gets stuck. The periodic replanning ensures the system doesn't degrade, while decentralized planning keeps latency low.