Reproduce “VTAM: Efficient Visuo-Tactile World-Action Models for Contact-Rich Manipulation without Tactile Pretraining” 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
Haoran Yuan, Weigang Yi, Zhenyu Zhang, Wendi Chen, Yuchen Mo, Jiashi Yin, Xinzhuo Li, Yue Luo and 6 more
VTAM adapts a video-pretrained model to predict future camera and tactile observations, then uses features to generate actions. It uses about 100 demonstrations per task, without tactile or action pretraining. On UniVTAC’s eight-task benchmark, it averages 83.4%, versus 48.0% for baselines without tactile pretraining and 84.5% for N₀-TWAM, pretrained on 30,000+ hours of tactile data; this is not zero-shot transfer to new tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.