UR + Scale AI Trainer

Universal Robots and Scale AI issued a joint Business Wire press release on March 16, 2026 announcing the UR AI Trainer and positioning it as a “lab‑to‑factory” imitation‑learning system. (businesswire.com(businesswire.com)) (morningstar.com) The system captures synchronized multimodal streams — motion, force (torque) and vision — specifically to enable Vision‑Language‑Action model training from real production hardware rather than research rigs. (thenextweb.com(thenextweb.com)) (thenextweb.com) UR’s implementation uses a leader‑follower setup where an operator guides a “leader” robot while a synchronized “follower” records high‑fidelity telemetry, tapping the robot’s Direct Torque Control and force‑feedback capabilities for contact‑rich task data. (roboticsandautomationnews.com(roboticsandautomationnews.com)) (roboticsandautomationnews.com) Universal Robots says the Trainer runs on its AI Accelerator and PolyScope X stack — a platform that leverages NVIDIA Isaac libraries and Jetson AGX Orin in prior releases — tying the capture pipeline to the company’s production‑grade control software. (universal-robots.com(universal-robots.com)) (universal-robots.com) Scale AI is credited with the software stack and data tooling to label and operationalize the captured datasets, with Ben Levin, GM of Physical AI at Scale, saying UR’s global footprint creates an ideal foundation for scaled robotics data capture. (robotics247.com(robotics247.com)) (robotics247.com) At the GTC demo UR showed two UR7e cobots using a foundation model from partner Generalist AI to autonomously execute a complex smartphone packaging task, demonstrating contact‑rich, real‑world behavior learned from the Trainer’s captured datasets. (robotics24/7.com(robotics247.com)) (robotics247.com)