Key facts
- General Intuition has raised $320 million at a $2.3 billion valuation.
- The company is developing foundation models for robotics, aiming to generalize movement and interaction across environments.
- CEO Pim de Witte believes this approach will reduce the need for extensive, specialized real-world data collection.
- The model was trained on millions of hours of video game data, including controller inputs.
- General Intuition's model can power a quadrupedal robot after fine-tuning on just eight minutes of real-world robotics data.
General Intuition, a startup focused on artificial intelligence for robotics, has secured $320 million in funding at a $2.3 billion valuation. CEO Pim de Witte believes the field of embodied AI is on the cusp of a transformative shift, akin to the 'ChatGPT moment' experienced by natural language processing (NLP).
De Witte argues that instead of building specialized robot models from scratch using vast amounts of task-specific data, the industry should focus on creating general-purpose foundation models. These models, he posits, can learn transferable intuition about movement and interaction across diverse environments with significantly less data.
"A lot of companies right now are doing lots of specialized work focused on individual embodiments, individual environments, and individual robots," de Witte told TechCrunch. He anticipates much of this specialized work will become redundant with the advent of general models like the one General Intuition is developing.
The company's approach involves training a foundation model on millions of hours of video game data, including controller inputs, to develop a human-like intuition for spatial-temporal reasoning. This data, according to de Witte and lead investor Vinod Khosla, is key to achieving this goal.
General Intuition has demonstrated its model's capabilities, showing it can play video games for extended periods and, after fine-tuning on just eight minutes of real-world robotics data, power a quadrupedal robot. De Witte expressed surprise at the robot's ability to operate in a dynamic office environment with only a front camera and no other sensors.
The company's ultimate aim is not to manufacture robots but to provide the foundational AI models that other robotics companies can utilize to build their own machines more easily. De Witte likened this to making it significantly simpler for others to establish companies like self-driving car manufacturers.