AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Generative Online Reinforcement Learning

Chubin Zhang, Zhenglin Wan, Feng Chen, Fuchao Yang, Lang Feng, Yaxin Zhou, Xingrui Yu, Yang You, Ivor Tsang, Bo An

ICML 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Regularized scalable deep value learning

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use diffusion decoder instead of Gaussian decoder in staged refinement (ablation)GORL · HopperStand stage 3converged return: 282 -> 869Improved
refine decoder on a fixed N(0,I) prior instead of evolving latents (ablation)GORL · FingerSpinreturn stabilityImproved
add stage-wise re-initialization of the latent policy (ablation)GORL · FingerSpinreturn stabilityImproved
add decoder refinement stages (Stage 0 to Stage 3) with diffusion decoder (ablation)GORL · HopperStandconverged return: 168, 604, 693, and 869Improved
Gaussian decoder across stages (ablation)GORL · HopperStandconverged return: 134, 208, 269, and 282No clear effect
GORL-DIFF vs strongest non-GORL baselineGORL-DIFF · HopperStandfinal return: 286.09 -> 874.63Improved
GORL-DIFF vs PPO on HumanoidStandGORL-DIFF · HumanoidStandfinal return: 74.97 -> 976.10Improved
GORL-FM vs PPO on HumanoidRunGORL-FM · HumanoidRunfinal return: 17.39 -> 337.61Improved
decoder refinement stages on HumanoidStand (S0 to S3)GORL-DIFF · HumanoidStandstage mean return: 89.7 -> 976.1Improved
direct flow-matching surrogate (FPO) instead of PPOGaussian PPO · FingerSpinfinal return: 539.03 -> 56.05Hurt
direct diffusion PPO (DPPO) instead of PPOGaussian PPO · HopperStandfinal return: 286.09 -> 2.14Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy