A humanoid robot named HomeBody surfaced this week, with a pitch as bare-bones as its name: it explores, remembers, and acts on its own.
The project lives on a Stanford page at tml.stanford.edu, and beyond that one-line description, little else has surfaced. The submission hit Hacker News on September 26, 2026, and as of this writing has drawn 5 points and zero comments. No paper, no named researchers, and no technical specs have been published alongside the announcement, so what "acts on its own" actually means in practice - what sensors, what training approach, what hardware - is anyone's guess.
That gap matters because "long-horizon autonomy" is the hardest unsolved problem in home robotics. Humanoid projects from Tesla's Optimus to Figure and 1X have all leaned on flashy demo videos while keeping the actual decision-making architecture vague. A robot that can explore a house, build a persistent memory of it, and act without constant human prompting would be a genuine step past scripted demos - if the claim holds up.
For now, treat HomeBody as a headline, not a result. Five points and no comments on Hacker News just means almost nobody has looked at it yet - it says nothing about whether the underlying work is solid. Until there's a paper or a byline attached, the honest read is: promising three words, zero evidence.