Startup Physical Intelligence raises $400 million in funding

The firm seeks to develop universally compatible software for robots, aiming to circumvent the need for task-specific programming

Physical Intelligence, a robotics AI startup, has raised $400 million in a funding round.

The investment was spearheaded by Amazon’s Jeff Bezos and venture capital firms Thrive Capital and Lux Capital, according to The New York Times.

OpenAI, Redpoint Ventures, and Bond also contributed to the round.

The funding positions Physical Intelligence at a post-money valuation of $2.4 billion, according to CNBC.

This valuation marks a substantial increase from the firm’s March seed round, which valued it at $400 million after a $70 million investment.

Physical Intelligence was co-founded in 2023 by Karol Hausman, previously a robotics scientist at Google; Sergey Levine, a professor at the University of California, Berkeley; and Lachy Groom, a former executive at Stripe.

The firm seeks to develop universally compatible software for robots, aiming to circumvent the need for task-specific programming.

This project requires extensive real-world operational data, which is currently scarce.

Consequently, Physical Intelligence is compiling its own datasets, supported by advancements in AI models that can process visual information.

The startup’s website states its goal is to bring “general-purpose AI into the physical world”.

Recently, Physical Intelligence published a paper demonstrating the capabilities of its software, π0 (pi-zero), which enabled robots to perform various household tasks.

Over the last eight months, the firm has been developing a “general-purpose” AI model for robots.

In its blog, the firm said: We believe this is a first step toward our long-term goal of developing artificial physical intelligence, so that users can simply ask robots to perform any task they want, just like they can ask large language models (LLMs) and chatbot assistants.

The firm explained that its model, like LLMs, is trained on diverse data and follows text instructions.

However, it also spans images, text, and actions, gaining physical intelligence through robotic training to output motor commands.

It can control various robots and be either prompted for tasks or fine-tuned for specific applications.