Data collection
UR AI Trainer (leader–follower data collection)
Universal Robots' kit for recording demonstrations on production arms: a person guides a leader UR arm and a follower UR arm mirrors it while logging motion, force and video.
How it works
A human moves a small 'leader' arm by hand. A second 'follower' arm copies the motion in real time, and the system records the follower's joint motion, force/torque and camera streams. Because the follower is the same industrial arm you would deploy, the data has the right dynamics and contact forces for training VLA or imitation policies. Scale AI handles dataset tooling. Announced at NVIDIA GTC in March 2026 with a public industrial dataset promised for later in 2026.
What it needs
Two UR e-Series arms (demo used UR3e leaders and UR7e followers), UR's AI Accelerator compute, and operators to perform demonstrations. Commercial availability and pricing were not stated at launch.
Strengths
- Data comes from the same industrial hardware you deploy on
- Force signals recorded, not just video
- Backed by the largest cobot vendor
Limits
- Announcement only; no deployment results yet
- Requires two arms per station
- Dataset release and pricing pending
Where it has been shown to work (1)
| Task | Robot | Maturity | Success | n | Grade | Year |
|---|---|---|---|---|---|---|
| — | UR7e | Lab | — | — | C | 2026 |