Data-Driven Robotics with Universal Robots AI Trainer Platform

[Erkan Teskancan]

## Data-Driven Robotics with Universal Robots AI Trainer Platform

Universal Robots took a significant step in AI-powered robotics with the AI Trainer platform, introduced at GTC 2026. This platform enables robots to learn tasks from real data rather than pre-programmed commands.

The platform was developed in collaboration with Scale AI, making it possible to directly collect high-quality datasets from industrial robots. This bridges the gap between AI research and real-world applications in the field.

### How the AI Trainer Platform Works

At its core is the leader-follower concept. A human operator guides the "leader" robot as it performs a task, while a synchronized "follower" robot replicates these movements in real-time. During this process, the system records multimodal data such as motion, force feedback, and visual data.

### AI Trainer Features

• Human-guided real-world data capture
• Recording of motion, force, and visual feedback data
• Support for training advanced AI models like Vision-Language-Action (VLA)
• Flexible automation capable of adapting to dynamic environments
• High-quality physical interaction data through direct torque control and force feedback
• Collection of training data on industrial-grade robots
• Continuous learning capability through data loop integration
• Support for simulation and synthetic data generation with NVIDIA

### Physical Interaction with High-Quality Data

One of the challenges in robotic AI development is the lack of quality training data. The AI Trainer, through force feedback and direct torque control, enables the learning of tasks requiring touch, pressure, and precise manipulation. This supports more reliable operation of robots in real-world applications.

### Continuously Learning Robotic Systems

The platform works in integration with Universal Robots’ AI Accelerator and Scale AI infrastructure, allowing data collected during operation to be reflected in the model and performance to be improved over time. This data-driven feedback loop enables the optimization of robotic systems and faster AI iterations.

### Simulation and Synthetic Data Generation

Universal Robots, in collaboration with NVIDIA, provides robot training in physics-based simulation environments using tools like Omniverse and Isaac Sim. These simulations allow for the safe testing of complex scenarios and the large-scale generation of synthetic data.

### Future General-Purpose Robots

At the event, a robotic foundation model developed by Generalist AI was also introduced. A demonstration where two robots independently performed a smartphone packaging task showcased their coordinated movement and precise manipulation capabilities.

This approach, supported by Universal Robots’ AI Trainer platform and simulation technologies, aims to accelerate the transition to AI-powered, experience-learning general-purpose robotic systems.