HALO synthesizes vision-language-motion data for humanoid robots
Original title: Perceptive Humanoid Motion from Simulation
HALO achieves 91 percent success on composite tasks in unmapped real homes by scaling simulation data rather than human teleoperation.
HALO uses a pipeline that generates vision-language-motion data across a thousand procedurally created homes. It acquires skills like sitting on chairs, navigating objects and pushing boxes. The system orchestrates long-horizon tasks and transfers zero-shot to the physical world. Training never sees real homes yet reaches high success across 360 trials in six environments.
- 1,000 procedurally generated homes
- Zero-shot transfer to real homes
- 91 percent success across 360 trials
- Acquires repertoire of physical skills
- Scales with compute not human labor
From the original article
1 min read at arxiv.orgHumanoid robots have the potential to automate a wide range of physical tasks. Progress toward generalist humanoid autonomy, however, remains constrained by a data bottleneck: mapping pixels and free-form language to low-level joint commands requires paired supervision, and no corpus of such data exists at sufficiently large scale.
By Siyuan Hu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.