Reproduce “CFG-Policy: Unified Diffusion Policy Learning Framework via Classifier-free Guidance” 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
Shutong Ding, Zhaoyang Liu, Yimiao Zhou, Jingya Wang, Jiayuan Gu, Ye Shi
CFG-Policy blends two policies’ score or velocity predictions on the same noisy action during generation. In online reinforcement learning, a frozen RND exploration policy guides a trainable exploitation policy without changing objectives. On HalfCheetah-v4, QVPO with CFG-Policy reaches 12,990 ± 1,993 mean episodic return versus 11,166 ± 5,076, a 16.3% gain; returns do not directly measure diversity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.