General Intuition raises $320M to use video game data to train robots

June 27, 2026 - 07:50

A startup called General Intuition has secured $320 million in fresh funding to pursue an unusual approach to robotics training. Instead of relying solely on real-world demonstrations or simulated environments, the company is pulling clips from video games that already contain embedded action labels. These labels, which track what a character does at each moment, can be repurposed to teach AI models how to move and interact with objects.

The idea is that modern video games offer vast amounts of structured data. Every jump, grab, or combat move in a game is already recorded alongside the game's physics and timing. General Intuition believes this data can help robots learn basic motor skills faster than traditional methods, which often require expensive, hand-labeled footage or slow reinforcement learning in custom simulators.

The $320 million round suggests strong investor confidence in this data-centric strategy. While the company has not disclosed which games it uses, the approach could apply to anything from open-world adventures to sports simulations. Critics note that game physics do not always match the real world, but the company argues that the sheer volume of data helps models generalize better.

General Intuition plans to use the funds to expand its engineering team and build a large-scale data pipeline. The goal is to create a foundation model for robotics that understands common actions like pushing, pulling, and grasping, learned directly from gameplay footage.