UR AI Trainer launched

Universal Robots and Scale AI revealed the UR AI Trainer at NVIDIA GTC in San Jose on March 16, 2026 as part of the conference’s demo lineup running March 16–19, 2026. (businesswire.com) Two UR7e cobots running a robotic “embodied foundation” model from Generalist AI performed an on‑stage smartphone‑packaging workflow in the first public demo of Generalist’s embodied foundation models. (robotics247.com) Universal Robots says the UR AI Trainer is delivered on its AI Accelerator/PolyScope X stack, which already integrates NVIDIA Isaac libraries and Jetson modules for on‑device robotics inference and control. (universal-robots.com) The product uses a leader‑follower imitation‑learning setup where an operator guides a “leader” arm while a synchronized “follower” mirrors and replays demonstrations, with recordings written to Scale’s tooling for immediate replay and training iterations. (roboticsandautomationnews.com) Scale AI confirmed the Trainer plugs into its Data Engine for Physical AI to annotate and operationalize collected workcell recordings, and the partners say they will publish a large‑scale industrial dataset sourced from UR systems later this year. (scale.com) (rockingrobots.com) Universal Robots highlighted that its control features are already deployed across roughly 100,000 industrial cobots worldwide, a footprint the company says enables direct “lab‑to‑factory” model iteration at scale. (roboticsandautomationnews.com) Anders Beck of Universal Robots framed the Trainer as the industry’s “first direct lab‑to‑factory solution,” while Ben Levin at Scale AI described the collaboration as enabling continuous model refinement from operational deployments. (finance.yahoo.com) Generalist AI’s CEO Pete Florence characterized the booth demo as evidence that embodied foundation models can deliver the dexterity and reliability required for contact‑rich manufacturing tasks. (greenstocknews.com)