Reproduce “From Predictive Geometry to Action: Learning Complementary Predictive Representations for Embodied 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

Xiaoxu Xu, Hao Li, Hongtao Zhang, Yili Tang, Jiafei Cao, Junhao Cai, Haoxiang Ma, Weixin Li and 1 more

CPR helps a robot manipulate objects, such as removing a stopper from a flask, by separately learning scene geometry and action-relevant changes in geometry. During training, future observations supervise predictive representations, but are not required at inference. On WidowX in SimplerEnv, CPR raises average success across four Visual Matching tasks from 61.5% to 71.9%; results are limited to tested settings.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.