Physical Intelligence, an AI startup co-founded by former Google robotics scientist Karol Hausman, has successfully raised $400 million in funding to develop versatile AI 'brains' for robots.

The funding round, which values the company at approximately $2.4 billion post-money, was led by notable investors including Jeff Bezos, Thrive Capital, and Lux Capital.

The startup's flagship model, pi-zero, has demonstrated its ability to manage real-world tasks such as folding laundry and bussing tables, showcasing its capacity to recognize and manipulate various objects.

Despite these advancements, the company faces challenges due to a lack of large-scale multitask and multirobot data, although they are optimistic that growing datasets will enhance their robot models.

Physical Intelligence's model is trained on a diverse array of data, allowing it to process images, text, and actions, and learn to output motor commands through embodied experiences.

Recent advancements in AI models for visual data interpretation have bolstered the company's efforts to build a more capable robotic brain.

Hausman emphasized the ambitious goal of creating a generalist brain that can operate any robot, a task that requires extensive real-world operational data.

This trend is echoed by major tech companies, including Tesla, which is developing humanoid robots, and Meta, which is researching robotic touch perception.

The competitive landscape in the robotic AI sector is intensifying, with other startups like Vicarious, Universal Robots, and Seegrid also entering the market.

Overall, major tech firms such as Microsoft, Google, and Amazon are heavily investing in AI technologies, with projections indicating that funding for AI and cloud companies will reach $79.2 billion by the end of 2024.

Physical Intelligence acknowledges that while their models are still in early stages and face performance limitations, they are committed to refining their technology through collaboration within the robotics community.

The company's innovative model integrates diverse data types, enabling robots to perform complex tasks like sorting trash and making coffee, thus pushing the boundaries of what robotic systems can achieve.