TL;DR
The hardest part of building a robot butler is teaching it the thousand small chores of a real home, and the cheapest teacher on Earth is ordinary video of people doing things. NVIDIA says its new robot brain, GR00T 2, soaked up about 400,000 hours of human video and now runs three very different robot bodies, from a simple gripper arm to a full humanoid. Given five short demonstrations of a new task, under three minutes of robot footage each, it averaged 95% success on the tasks NVIDIA tested, after about half an hour of extra training on a single graphics card (company figures, on tasks NVIDIA chose). That is the shape of a household robot people could teach by showing rather than programming. Earlier today we covered a Japanese robot brain that learned to fold towels from human hands (our story). To see our odds on when a home robot handles everyday chores on its own, and the timeline of breakthroughs that leads there, subscribe.
Below the line, for paid subscribers: The full story: where it stands, how we got here and what's dated next; our labeled speculation on what it adds up to; and every source.