A warehouse robot that knows when to stop squeezing.

The human problem

Picking and stowing items in high and low warehouse bins can involve ladders, reaching, and bending that are less ergonomic for workers.

The prior constraint

Many industrial robots could not reliably manipulate mixed items in crowded bins because they could not sense contact force well enough.

AI’s actual role

Physical-AI systems identify items and available space, while force-feedback sensors guide how the robot pushes, grips, and stops before damage.

The documented result

Amazon says Vulcan was working at fulfillment centres in Spokane, Washington, and Hamburg, Germany, on top storage rows. Amazon reports it can pick and stow about 75% of item types stored in its centres at speeds comparable to frontline employees, and can ask a human partner to take over an item it cannot move.

Why it may matter

The intended use is specific: shift awkward high- and low-row handling away from workers while retaining a human decision point for exceptions.

Limitations

The source does not display a specific publication day, so this artifact intentionally leaves the date null rather than infer one from secondary indexing.

Amazon's capability, ergonomics, and worker-development statements are company-reported; it provides no independent injury, error, pace, or job-quality results.

The company says broader U.S. and European deployment is planned, which is not evidence that it has occurred.

Handling 75% of item types is not the same as handling 75% of all tasks or proving a benefit for every worker.

Unresolved questions

What changes occur in injury rates, pace, and worker control at active sites?

What are the error, damage, and exception rates?

How are workers involved in decisions about deployment and retraining?