AI training

Inka Krischke,

Universal Robots and Scale AI rely on imitation learning

Universal Robots (UR) presented the 'UR AI Trainer' at GTC 2026 in Silicon Valley. The solution was developed together with Scale AI and, according to the company, marks a turning point: robots are evolving from rigidly programmed systems to flexibly learning, AI-controlled applications.

© Universal Robots

The solution is based on high-quality training data from special training environments in which robots imitate human movements.

"Our customers - from large enterprises to AI research labs - are no longer just asking for AI capabilities," explains Anders Beck, VP of AI Robotics Products at Universal Robots. "They need a way to capture highly accurate, synchronized robotics and image data to train AI models directly on the robots that will actually be used.Our AI Trainer is the industry's first solution that takes AI model training directly from the lab to the factory."

Inconsistent hardware and insufficient data quality often make it difficult to train AI robots. Training data is often collected with research robots that are not suitable for industrial environments. These are based solely on visual information, which is particularly problematic for sensitive or contact-intensive tasks. Beck says: "This is exactly where the AI Trainer comes in: Force feedback and direct torque control give developers precise control over how robots interact with their environment."

The AI Trainer allows human operators to guide robots in a "leader-follower" setup. One robot is guided manually while a second mirrors the movement in real time. At the same time, movement, force and image data are recorded synchronously to generate structured training data for modern AI models (VLA).

Advertisement

On the UR AI Accelerator platform, the AI Trainer combines robot hardware with Scale AI software. This creates a continuous data cycle that enables AI models to be trained, improved and scaled directly in production.

  • Xing Icon
  • LinkedIn Icon
Advertisement
Advertisement

You might also be interested in

Advertisement
Advertisement
Advertisement
Advertisement
Advertisement
Advertisement
Advertisement
Subscribe to our newsletter
Advertisement
Back to home