Universal Robots (UR) and Scale AI have launched UR AI Trainer, a system to train robots through human imitation. The platform allows operators to guide a “leader” robot through tasks while a “follower” robot records motion, force, and visual data.

The system aims to bridge the gap between laboratory research and factory deployment by capturing high-fidelity data on the same hardware used in production. This data is used to train Vision-Language-Action (VLA) models, allowing robots to move beyond rigid pre-programmed instructions to more flexible, AI-driven tasks.

“Our customers need a way to collect high-fidelity, synchronised robot and vision data to train AI models on the same robots they intend to deploy,” said Anders Beck, VP of AI Robotics Products at Universal Robots. The system utilises UR’s direct torque control and force feedback to record how a robot physically interacts with its environment.

The collaboration includes the following components:

• UR AI Accelerator platform for data capture and model deployment.
• Scale AI software stack for recording and replaying demonstration data.
• Integration with NVIDIA Isaac Sim and Omniverse for synthetic data generation.
• A planned release of a large-scale industrial dataset later this year.

Universal Robots have demonstrated the technology using a smartphone packaging task, which was previously difficult for robots to perform due to the requirement for delicate, contact-rich manipulation.