UR AI Trainer

UR AI Trainer is an imitation‑learning system integrated with Scale AI’s recording software, which enables automatic capture of synchronized motion, force, and visual data required to train Vision‑Language‑Action (VLA) models.

The system is designed to help AI teams to train AI models directly on industrial‑grade Universal Robots, bridging the gap between lab experimentation and real‑world factory deployment. It is built on production‑grade hardware and control architecture, ensuring data collected during training reflects real industrial conditions and supports faster, more reliable deployment of AI‑driven robotic applications. 

UR AI Trainer enables human operators to teach robots by demonstration using a leader–follower setup. While an operator guides a “leader” robot through a task, a synchronized “follower” robot mirrors the motion in real time, with high-quality multimodal data captured automatically during each demonstration. 

This allows for scalable data capture across live UR robot deployments and creates continuous feedback that accelerates the development and optimization of physical AI systems.

How Does it Work?:

• A human operator physically guides a UR “leader” robot through a task
• A UR “follower” robot mirrors the motion in real time
• On Scale AI software, the system captures synchronized:
  • Robot joint state and motion
  • Force and torque data
  • Multi‑camera visual data
• Data is structured into datasets ready for AI model training and fine‑tuning

This workflow creates a direct path from demonstration → dataset → AI model training, without custom data‑pipeline development.

• Force‑aware imitation learning using Direct Torque Control
• Leader–follower robot setup for human‑guided task demonstrations
• Synchronized capture of motion, force/torque, and multi‑camera vision
• Industrial‑grade UR robots used for both training and deployment
• Multi‑camera vision system aligned with robot state and force data
• On‑robot AI compute via the UR AI Accelerator
• Structured multimodal datasets ready for AI model training
• Integrated Scale AI recording and dataset management software
• Production‑ready hardware platform for multi‑robot configurations
• Modular system supporting flexible robot and hardware configurations
• Designed for training Vision‑Language‑Action (VLA) models
• Scalable data capture across real industrial environments

The Scale recording software enables:

• Flexible camera support: compatible with stereo and monocular camera setups
• Multimodal data capture: records joint states, force/torque signals, camera streams, and IMU data
• Built-in recording management: create, delete, replay, and quality-check recordings
• Task-level tracking: capture collector, task type, timing, and diversity metrics

• PolyScope X

• United States

Available by UR: 

• UR Collaborative robots: 4 robots, 2 leaders (UR3e) and 2 followers (UR7e)
• The UR AI Accelerator
• Recording software in collaboration with Scale AI

Follow our guidance for purchasing from partners: 

• Hardware setup in collaboration with Vention
• Robotiq 2F-85 grippers