Figure’s 30-home trial puts generalization to a harder test

Michael Cruz · Published 2026-10-19 · Event 2026-09-17

Figure reported testing Helix 2.5 in 30 previously unseen Bay Area homes on September 17, 2026. Its chosen tasks were room tidying, towel folding and bed making. This is a research update being added to the guide today, not a new announcement today.

The company reports a 56% complete-task success rate with Index pretraining, compared with 9% without that pretraining in its controlled comparison. These are Figure’s measurements, not independent results from this guide. “Unseen” refers to the evaluation homes and objects; task-specific training still occurred elsewhere.

For buyers, the important question is whether skills transfer beyond a prepared demonstration room. This evaluation tackles that question, but does not establish reliable completion of every household chore. The remaining gap is meaningful: success on just over half of the evaluated trials is different from dependable daily service.

We associate the update with Figure’s household platform for context. It does not change the guide’s untested rating, establish a retail offer or verify delivery.