Reproduce “Sensorimotor Alignment: A Platonic Proxy for Embodied Policy Success” 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
Yunfei Xie, Howard H. Qian, Shiyi Lan, Kaiyu Hang, Chen Wei
The authors introduce S², an offline measure of whether a model’s sensory features reflect robot-motion geometry. It compares nearest neighbors across demonstrations in sensory and trajectory spaces, using cosine distance and dynamic time warping, then computes a centered, normalized CKNNA score. Across 12 fine-tuned VLAs, S² correlated with SimplerEnv success (r = +0.717), but does not establish causation or replace rollouts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.