Physical Intelligence collects 400 million dollars for robotics with models to generalists AI | Google ai courses free | Google machine learning certification free | Generative ai in investment management | Turtles AI

Physical Intelligence has raised $400 million to develop foundation models that enable robots to perform complex physical tasks with greater autonomy and skill. Using an innovative approach, the company aims to create general-purpose robots that can learn faster and with less data.

Key points:

• Physical Intelligence has secured $400 million in funding.
• The company develops basic generalist models for robotics.
• Robots based on these models promise to learn faster and with fewer examples.
• Key investors include Jeff Bezos, Thrive Capital and Lux Capital.

Physical Intelligence announced today the completion of a funding round that raised $400 million, a key step in its ambitious project to develop basic AI models intended for robotics. These models, designed to be generalists, will enable robots to perform a wide range of physical tasks, including everyday tasks such as folding laundry or assembling boxes, with an unprecedented level of autonomy and adaptability. The California-based startup aims to reduce the need for massive data and specific examples, unlike traditional approaches, which require vast amounts of information to train systems. In a statement shared with The New York Times, Karol Hausman, co-founder and CEO of Physical Intelligence, emphasized that the company is not just creating a “brain” for each individual robot, but is working on a single “brain” that can control multiple robotic models, with a level of generalization that represents a significant leap from today’s technology. The initial prototype, called π 0, integrates large-scale multi-task and multi-robot data collection, using a new network architecture that aims to make robots more versatile, capable of performing a wide range of tasks with ease. The novelty of the proposed models lies precisely in their ability to apply learning to physical tasks that, while seeming simple to humans, are traditionally complex for robots, such as folding a shirt or rearranging a table, as evidenced by an analogy with Moravec’s paradox, which highlights how seemingly trivial tasks for humans turn out to be extremely difficult for AI. Early tests and demonstrations of the model, which include tasks such as dynamically filling containers or assembling boxes, were recently shown at RoboBusiness by Professor Sergey Levine of the University of California, one of the co-founders of Physical Intelligence. The model, while in its early stages, represents a promising step toward the future of highly adaptable and intelligent robotics. Physical Intelligence also confirmed that although the current models are still far from achieving perfect generalization capability, the goal is to continue with strengthening such capabilities through the use of data and strategic collaborations. The funding received, which brings the company’s valuation to $2.4 billion, was led by prominent names such as Jeff Bezos, Amazon’s executive chairman, and major venture capitalists such as Thrive Capital and Lux Capital. The company announced that it is continuing to expand its team to accelerate development.

Physical Intelligence’s success fits into a broader landscape in which other technology companies, such as Figure AI, Skild AI, and OpenAI, are focusing on developing foundation models for robotics, marking the beginning of a new era in which robots, thanks to AI, will become increasingly capable of performing complex tasks autonomously and adaptively.