Human vs robot workers in fulfillment center pick processes | Amazon Consumer Science Summit
Gerard Geshon, a professor at the Wharton School at the University of Pennsylvania, discusses human and robot picking in fulfillment centers, presenting joint research with Omar Vespas of Columbia Business School to Amazon’s Consumer Science Summit.
In this talk
- 3:52
Comparing two picking systems
Geshon models a long warehouse storage area where picked items move to a drop-off station. In the manual system, people walk to inventory and select items; in the robot system, robots bring inventory pods to people who pick at the station. The analysis focuses on the picking process and the number of workers needed to meet an average response-time target.
- 8:23
How response time changes staffing
The initial comparison suggests that manual picking needs only modestly more workers as response times shorten. The robot system is more sensitive: at a one-hour target, it may need roughly four times as many robots as the manual system needs workers at a ten-hour target. The robot figure is a lower bound, while the manual figure comes from a feasible operating policy.
- 12:04
Batching drives capacity
The analysis treats workload as a combination of item selection and travel, with batch size as a key operating choice. For people, the proposed policy divides the warehouse into zones and sends a worker to pick after enough orders accumulate in a zone. In the robot model, the number of picks per pod depends on how long demand can accumulate before the pod is moved.
- 23:33
Why robots need more units at short targets
With the example parameters, robot batches grow slowly as the allowed response time increases, leaving robots to move pods that may contain only one or one and a half picks at the shortest targets. Faster inventory turnover can help pods accumulate demand, and tighter storage can reduce robot travel, but the comparison still shows a substantial robot staffing increase at short response times. Geshon says the crossover depends on operating conditions.
- 28:38
Different ways to build batches
People can gather items across warehouse space, while each robot is limited to moving one pod and must wait for demand to build on it over time. The example suggests an optimal range of about 20 to 50 zones for the human system, compared with roughly 500 pods in the robot system. This difference in batching explains why the robot system is more affected by tight response-time requirements.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.