Sanctuary AI tests physical AI at factory speed
The Canadian company says it reached a 99.5%+ success rate on a wire-plugging task, with a 2.54-second cycle time, at a global Tier 1 automotive supplier.
Sanctuary AI announced on June 17 that it has validated an industrial wire-plugging task with a success rate above 99.5% and a 2.54-second cycle time, measured against the live production benchmarks of a global Tier 1 automotive supplier. The detail matters: this is not a slow lab demonstration, but a flexible-material manipulation task on a moving conveyor, at a pace comparable to an existing production line.
The announcement also signals a shift in strategy. Sanctuary AI, known for its work on general-purpose robots and robotic hands, says it wants to deploy its “Physical AI” on existing industrial robotic platforms rather than wait for humanoid hardware to reach mass commercialization. In this context, physical AI means models that can perceive, decide, and act in the real world, while respecting the contact, safety, and cycle-time constraints of a factory floor.
The task is more meaningful than it may sound. Inserting a flexible wire into a moving target is exactly the kind of operation that has often remained difficult for traditional automation: the part bends, the target position changes, contact has to be controlled, and every failed attempt costs time. Sanctuary AI says its system matched the customer’s line throughput, which moves the discussion away from humanoid showcase videos and toward a more practical benchmark. For manufacturers, the key question is not whether a robot looks human, but whether it can deliver performance, repeatability, and safety on a specific station.
This approach could shape how general robotics reaches production. By making its models work with commercial hardware that is already available, Sanctuary AI is looking for a shorter path into factories: start with precise cells, learn from real industrial constraints, then transfer those capabilities to future generations of robots, including humanoids. The claim still needs proof across more customers and tasks, but the signal is useful. Embodied AI becomes more credible when it is measured in seconds, success rates, and production lines that are already running.