Figure's Helix 2.5 achieves zero-shot whole-body generalization across 30 unseen homes

coreylynch · x · 2026-09-18

Figure unveiled Helix 2.5, its most advanced neural network, pretrained on Index, its global-scale human-behavior dataset. Highlights:

Zero-shot whole-body autonomy in 30 real Bay Area homes: tidying living rooms, folding towels, and making beds with no data collection, fine-tuning, or adaptation — billed as the first demonstration of zero-shot whole-body generalization at this scope on a humanoid.
One foundation model, three whole-body behaviors spanning locomotion, rigid and deformable manipulation, bimanual coordination, and active perception.
Index pretraining drives generalization: with everything else fixed, zero-shot success jumps from 8% to 56% (alternatively 9%→56%, 6x+).
Predictable scaling: doubling Index data let the team forecast the largest run's loss to four decimal places before training.

Figure has committed $3.5B of compute for the next Helix version.

More from Embodied