Reproduce “Decoupled Asynchronous System for Humanoids (DASH): Latency-Agnostic Scaling of Vision-Language-Action Models for Humanoid Manipulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bingzhang Wang, Nathan Zhao, Jialeng Ni, Jiajin Zheng, Yin Zhang, Kehua Chen, Jian Yao
DASH separates a slow vision-language model from a faster action head for humanoid pick-and-place. The model interprets instructions in the background, while the action head combines cached semantic features with fresh visual observations and updates movements. In RoboTwin, its 8B configuration runs at 20 Hz and succeeds on 65.9% of valid episodes, versus 62.7% synchronously; tasks were quasi-static.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.