General Intuition raises $320M to build robotics foundation models trained on video game data

General Intuition, a robotics startup, raised $320 million at a $2.3 billion valuation last month, according to TechCrunch reporting. The company is developing foundation models for embodied AI using video game data rather than large volumes of real-world robotics data.

CEO Pim de Witte argues that robotics is poised to follow a similar trajectory to natural language processing, where general-purpose foundation models replace task-specific training. He contends that most current robotics work—focused on individual robots, environments, and tasks—will become redundant as general models improve.

The startup trained its foundation model on millions of hours of video game data, including controller inputs and timing, to encode human-like spatial-temporal reasoning. De Witte claims this approach reduces the need for hundreds of thousands or millions of hours of real-world robotics data, asserting that only a few minutes of real-world fine-tuning may suffice for specific applications.

General Intuition demonstrated its model’s capabilities by using it to play a video game for hours and to control a quadrupedal robot after fine-tuning on just eight minutes of real-world robotics data. The robot operated using only a front camera with no additional sensors in an office environment with dynamic objects and people moving nearby.

The company’s stated goal is not to build end-user robots but to provide a foundation model for other robotics companies to build upon. De Witte compared the strategy to enabling others to build self-driving car companies more easily rather than building one directly.