Scale debuts UR AI Trainer at GTC
Scale AI demoed the UR AI Trainer with Universal Robots at GTC to capture force and visual data for model training — a practical move to accelerate robotics datasets and closed‑loop model iteration. That push into robotics data capture suggests growing demand for dedicated training clusters and inference fleets to handle sensory+vision workloads. (thenextweb.com) (advancedmanufacturing.org)
Universal Robots and Scale AI formally unveiled the UR AI Trainer at NVIDIA’s GTC on March 16, 2026. (finance.yahoo.com) The Trainer implements a leader‑follower imitation‑learning flow where a human guides a “leader” robot and a “follower” robot mirrors the motions while logging synchronized motion, force and visual sensor streams for model training. (ccstartup.com) Two UR7e cobots at UR’s GTC booth executed a staged smartphone‑packaging task driven by a robotic foundation model from Generalist AI, which UR lists as a preferred model partner for the demo. (robotics247.com) Universal Robots’ Anders Beck, VP of AI Robotics Products, described the UR AI Trainer as “the industry’s first direct lab‑to‑factory solution” and stressed the need for high‑fidelity, synchronized robot and vision datasets for production deployments. (advancedmanufacturing.org) Scale published a March 16, 2026 post announcing integration of its Physical AI Data Engine with UR hardware to provide on‑robot data capture, annotation pipelines and end‑to‑end datasets intended for embodied model training. (scale.com) NVIDIA used GTC to release an open Physical AI Data Factory blueprint that standardizes generation, augmentation and evaluation of physical training data—an ecosystem move that complements UR and Scale’s lab‑to‑factory data approach. (nvidianews.nvidia.com)
