Reproduce “Emo-H2A: Driving a Facial Robot Directly from LLM Hidden States at Token Rate” 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
Yige Liu, Jingcheng Huang, Hao Sun, Min Li, Zhouping Yin
EmoManifold lets a robot move its face as it generates a reply, rather than waiting for a completed-turn emotion label. It reads each token’s language-model hidden state, maps coordinates to facial commands, and does not fine-tune the model. On a 29-degree-of-freedom robot, valence direction agreed with commanded shifts on 87.5% of detected events; this measures command tracking, not human emotion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.