Reproduce “ΔWAM: Distilling Action Tangent Fields into World Action Models” 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
Ke Wu, Hanwen Huang, Bo Gu, Kaizhao Zhang, Xiangting Meng, Yupeng Zheng, Zijun Xu, Jieru Zhao and 1 more
ΔWAM trains a robot policy to predict future video while emphasizing changes that respond to nearby action changes. It preserves the real future in VAE latent space and uses an action-conditioned teacher’s local responses to weight prediction errors. On LIBERO-Plus, pooled success across perturbation instances is 87.8%, versus 85.7% for π0.5, but camera and layout shifts remain challenging.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.