Figure AI just unveiled Helix 2.5 to get humanoids working in everyday homes. Humanoids have largely relied on training for specific environments. Helix 2.5’s biggest advance is zero-shot generalization: using what it already knows to navigate unfamiliar rooms and handle new objects, without retraining. Figure rented 30 Bay Area homes the robot had never seen, where Figure 03 autonomously tidied living rooms, made beds and folded towels. Figure calls this a first for humanoids at this scale. Behind it is Index, a global program turning human experience into robot training data, with more than 90,000 contributors each week. In a controlled test, Index pretraining alone pushed zero-shot task success from 9% to 56%. More human data also brought predictable gains in robot-action prediction, suggesting humanoids could scale like LLMs. Robot learning still has hard problems to solve, but signs of general physical intelligence are emerging. Figure has committed $3.5B in compute to scale up Helix training. Humanoids working in everyday homes no longer seem so far off. Chinese media report that Tesla’s Optimus team has also begun supplier audits in China, with production orders already placed.