A robot powered by Physical Intelligence’s AI folding laundryPhysical Intelligence
In San Francisco, inside a warehouse coated with gleaming steel panels, I am handed a fresh cup of coffee made entirely by a robot. This fact alone is unimpressive — robots have been making coffee for more than a decade — but the robotic brain that made this coffee is no one-trick pony. It has also learned how to do many other tasks, such as folding clothes, peeling vegetables and cleaning kitchens, in the time that most toddlers barely learn how to walk.
Physical Intelligence, a start-up founded in 2024, is betting that a robot brain that can learn how to do many different tasks will, in the not-too-distant future, enable robots to become enmeshed in our daily lives. Instead of focusing on a single machine, like the humanoid robots built by Tesla or Boston Dynamics or the factory robots used by Amazon, the company wants to build an adaptable control system that can perform many tasks with many different machines.
A general-purpose robotic intelligence isn’t a new idea: many roboticists would say it has been a long-term goal for decades. But, just as the early 2020s saw a flourishing of the large language models (LLMs) that power AI chatbots thanks to the right combination of computing power, data and algorithmic advances, Physical Intelligence is hoping to conjure a similar leap of progress in general robotics.
