Touch changes the warehouse manipulation problem
Amazon introduced Vulcan at its Delivering the Future event on May 7, 2025. The system is designed to pick products from and stow products into the fabric pods used in Amazon fulfillment centers. Those cubbies are densely packed, open only at the front and held by elastic bands, so a robot cannot reliably avoid contact with the pod and neighboring items.
Traditional industrial automation often treats unexpected contact as a fault: the machine stops or follows a rigid path despite the obstruction. Vulcan instead measures contact and uses it as part of the task. Amazon says the system completed a pilot and was moving into beta testing, with operating systems already present in Spokane, Washington, and Hamburg, Germany.
Two tools and six-axis force measurements
The stowing side uses an end-of-arm tool that Amazon compares to a ruler mounted on a hair straightener. It pushes items and elastic bands to make space while force and torque sensors measure interaction along six axes. Control software adjusts the motion so the tool stays below force limits intended to avoid damage.
The picking side combines a camera and suction cup. Stereo vision estimates free space and item geometry, while force feedback helps the arm manage inevitable contact. The perception system also estimates which items the robot can handle and sends unsuitable cases to a person instead of forcing an attempted pick.
Amazon reported that Vulcan could pick and stow about 75 percent of the item types in its fulfillment network at speeds comparable to front-line employees. That is a vendor-reported operational figure, not an independently audited benchmark. It also means the remaining item mix still needs other automation or human handling, and aggregate coverage does not describe error rates for every shape, material or packing condition.
Physical data rather than simulation alone
The company says it trained the manipulation system on thousands of physical examples that include touch and force measurements. Simulation is valuable for generating varied scenes, but deformable pod walls, elastic bands, crumpled packages and friction between arbitrary products are difficult to model precisely. Real interactions supply the contact data needed to decide when to push, back away or change the grasp.
Vulcan also records failures for later learning. That does not mean a deployed arm rewrites its behavior without controls during every attempt. In an industrial system, new policies need evaluation before release because an apparent improvement in one product configuration can create risk in another.
Deployment is targeted at an ergonomic bottleneck
Amazon initially focused Vulcan on the highest and lowest rows of storage pods. Those locations make employees climb step ladders or repeatedly bend, so automating them can reduce an identified ergonomic burden while people handle the more accessible rows and exceptions. This is a narrower and more measurable use case than claiming a general replacement for warehouse labor.
The system's significance lies in controlled contact at operational scale. Force sensing has a long history in robotics, but integrating it with perception, task planning, exception handling and a diverse inventory is the hard systems problem. Amazon planned additional deployment in Europe and the United States over the following years; the 2025 evidence established pilot operation and beta readiness, not network-wide availability.
Long-term assessment will need data on damaged items, interventions, throughput, maintenance and worker injury exposure. The launch material provides detailed mechanisms and intended benefits, but it does not publish a controlled before-and-after safety study. Vulcan is best understood as a specialized tactile warehouse system whose early deployment tests whether contact-rich manipulation can remain useful and predictable across millions of different products.