Reproduce “EDITH: Teaching Robot to Follow Nonverbal Human Intent from Egocentric Vision and Gaze” 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

Dongjun Lee, Juheon Choi, Dong Kyu Shin, Sinjae Kang, Kimin Lee

EDITH lets a robot interpret brief spoken requests using the wearer’s first-person view and gaze to identify referenced objects. A high-level vision-language model pairs subtasks with keyframes from recent egocentric video, and a low-level policy executes them. In real-robot tests across three tasks, EDITH averaged 48.6% success and 75.9% task progress, outperforming tested baselines but not completing every trial.

Summary by alphaXiv.

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