Remember that company, Figure, who decided to shell out a billion dollars over the next year, literally paying people to film themselves doing chores like folding laundry? Well, just one month into that ambitious plan, their investment is already paying off in a huge way. Their humanoid robot just saw its success rate for household tasks jump by an incredible 6X – from a modest 9% to a much more impressive 56%!
This isn’t some minor improvement; it’s a massive leap in what they call “zero-shot success,” meaning the robot tackled these chores in 30 rented Bay Area homes it had never been trained in. The secret sauce? Figure poured all that real-world “human experience” data, collected through their rapidly growing Index platform, into the robot’s brain. It’s the same robot, the same task data, the same architecture – they just changed that one variable, and boom, a 6X improvement. This kind of boost tells you exactly where the future of humanoid robotics is heading and why this data-driven approach is so crucial.
To back this incredible progress, Figure isn’t just dabbling; they’ve committed a staggering $3.5 billion in compute power for up to 100,000 Nvidia GPUs. Their Index platform is exploding, too, with roughly 55 minutes of new human experience uploaded every second by 115,000 weekly users! What’s even wilder is that Figure has discovered a scaling law for physical AI, allowing them to forecast performance gains from data to four decimal places before training even begins. The big question now isn’t if they’ll get better, but whether they can hit an astonishing 95% success rate in just a few more months.