Reproduce “LAPT-Scale: Scaling Embodied Policies with Latent Action Pre-Training on Human Videos” 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
Zhiyuan Xiang, Peidong Liu, Mingyang Li, Jiawei Li, Wenhow Li, Wenbo Ding, Xiang Deng
LAPT-Scale turns frame-to-frame changes in human videos into continuous latent actions for policy pre-training, then aligns the policy with robot actions and task demonstrations. Its visual model combines reconstruction with perceptual, semantic, and optical-flow signals. In real-robot tests, average success across three tasks rose from 63.3% with 1K video hours to 85.0% with 5K, on one platform.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.