The authors propose DriftSpread, a Drifting-based method for learning one-step generative policies in online MaxEnt RL.
One-step generative policy matches Boltzmann target for online MaxEnt RL
DriftSpread combines critic-guided drift and entropy-induced spread to learn expressive policies with single forward pass inference.
Top University
Chunyu Luo · Hongyu Zhao · Guangyi Ni · Yue Song · Lingkai Kong
University of Hong Kong · Beihang University · Harbin Institute of Technology (Shenzhen) · Independent Researcher · Tsinghua University
Research Digest··3 min read
Luo et al.
Why this paper
From University of Hong Kong and 4 others
In one line
DriftSpread learns expressive maximum-entropy RL policies that generate actions in one forward pass while outperforming evaluated diffusion and flow baselines on MuJoCo.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§