TL;DR: A hot new startup called Physical Intelligence just dropped a foundational AI model that lets robots figure out brand new tasks completely on the fly.
Teaching a robot to fold laundry usually means writing thousands of lines of code specifically for folding laundry. But a startup named Physical Intelligence just released a new model—dubbed π0.7—that flips that entire process upside down.
They’ve essentially built a general-purpose brain. Instead of being hardcoded for one specific job, the system relies on deep reasoning to deduce how a task should be done, even if it’s never encountered the objects before. Hand it a strange tool, and it figures out the grip. Ask it to clean a mess, and it extrapolates the right physics to make it happen.
This is the holy grail for roboticists. We don’t want to program every single edge case. We want machines that can generalize. And with π0.7, we’re seeing the first real signs that robots are stepping out of the factory script and into unpredictable, real-world chaos.
Source: TechCrunch