Reproduce “Correct Robots Before They Make Mistakes: Proactive Human-in-the-Loop Intervention via Preference Learning” 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
Xinkai Jiang, Jing Jiang, Hongyi Zhou, Pankhuri Vanjani, Ahmad Baki, Jakub Suliga, Ge Li, Zhuoyue Li and 2 more
PHIL previews predicted motion in XR, lets operators intervene before anticipated errors, and pairs rejected predictions with executed corrections from the same state for preference training. Across four physical manipulation tasks (three for X-VLA), it reported higher average full-task success than HG-DAgger with five correction episodes: 44.0% versus 70.3% for Diffusion Policy and 57.8% versus 76.9% for X-VLA. The comparison changes both collection mode and learning objective.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.