Reproduce “LaWAM: Latent World Action Models for Efficient Dynamics-Aware Robot Policies” 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
Jialei Chen, Kai Wang, Kang Chen, Shuaihang Chen, Feng Gao, Wenhao Tang, Zhiyuan Li, Weilin Liu and 4 more
LaWAM predicts a compact latent visual subgoal to guide robot action generation without generating future video. It retains a forward decoder that maps current visual features and a predicted transition code to endpoint features, then conditions action chunks on the subgoal. On LIBERO, it reports 98.6% success and 187 ms latency, versus LingBot-VA’s 98.5% and 4,482 ms, in stable-view manipulation settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.