Universal Robots and Scale AI have introduced a new system called the UR AI Trainer, designed to make it easier to train industrial robots using artificial intelligence. The platform was unveiled at NVIDIA GTC 2026 and aims to solve a major problem in robotics—how to move AI models from controlled lab environments into real factory settings.
The UR AI Trainer works using a “leader-follower” imitation learning setup. A human operator physically guides one robot (the leader) through a task, such as packaging a product, while a second robot (the follower) copies the motion in real time. During this process, the system captures detailed data including movement, force, and visual input, creating high-quality datasets used to train advanced AI models like vision-language-action systems.
A key innovation is that this data is collected directly on the same robots used in real production environments. Traditionally, AI models are trained on research machines that differ from factory robots, making deployment difficult. The UR AI Trainer bridges this “lab-to-factory gap”, allowing companies to train and deploy AI models on identical hardware, improving accuracy and reliability in real-world tasks.
The system also creates a continuous data feedback loop: robots learn from human demonstrations, improve through training, and generate better performance over time, which then feeds into further training. This approach marks a shift from rigid, pre-programmed automation to more flexible, AI-driven robots capable of handling complex tasks like assembly and manipulation. Overall, the UR AI Trainer represents a major step toward scalable physical AI, where robots can learn and adapt much like humans.
