General Intuition Bets on Foundation Models for Robotics

The current robotics landscape is fragmented, with companies laboring to build specialized models for individual robots and environments. De Witte argues this approach is nearing obsolescence. By training on video game inputs—tracking precise controller actions against screen visuals—General Intuition has developed a model capable of transferring movement logic to physical hardware. The startup recently secured a $2.3 billion valuation after raising $320 million to prove this thesis.

The practical results have been striking. After fine-tuning on only eight minutes of real-world robotics data, the company’s model successfully navigated a quadrupedal robot through an office environment using only front-facing camera sensors. Rather than manufacturing its own hardware, General Intuition intends to provide the underlying intelligence for others to build upon. By offering a base model that understands physical space and time, the firm hopes to lower the barrier for developers building everything from self-driving cars to autonomous assistants.