Figure Tests Helix 2.5 in 30 Bay Area Homes
Figure says Helix 2.5 tidied toys, folded towels and made beds in 30 Bay Area homes it had never seen, succeeding 56% of the time zero-shot, up from 9% without its Index pretraining. Figure didn't name the homes or the robot, so this is an evaluation, not a customer pilot.
The test design is stricter than most home demos. The robot collected no data in the homes and wasn't fine-tuned there, the objects never appeared in its task training data, and partial completions earned no credit. That makes the 56% success rate a more honest number than a highlight reel, and it also means the robot failed almost half its tasks.
What the post leaves out matters as much. Figure doesn't say whose homes these were or how they were chosen, and it describes the robot without naming the model. Figure 03 is the company's home-focused robot, but the post doesn't link it to this test. Nothing here describes paying customers or deliveries. Read it as research progress toward household work, published by the company doing the research.