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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use diffusion decoder instead of Gaussian decoder in staged refinement (ablation) | GORL · HopperStand stage 3 | converged return: 282 -> 869 | Improved |
| refine decoder on a fixed N(0,I) prior instead of evolving latents (ablation) | GORL · FingerSpin | return stability | Improved |
| add stage-wise re-initialization of the latent policy (ablation) | GORL · FingerSpin | return stability | Improved |
| add decoder refinement stages (Stage 0 to Stage 3) with diffusion decoder (ablation) | GORL · HopperStand | converged return: 168, 604, 693, and 869 | Improved |
| Gaussian decoder across stages (ablation) | GORL · HopperStand | converged return: 134, 208, 269, and 282 | No clear effect |
| GORL-DIFF vs strongest non-GORL baseline | GORL-DIFF · HopperStand | final return: 286.09 -> 874.63 | Improved |
| GORL-DIFF vs PPO on HumanoidStand | GORL-DIFF · HumanoidStand | final return: 74.97 -> 976.10 | Improved |
| GORL-FM vs PPO on HumanoidRun | GORL-FM · HumanoidRun | final return: 17.39 -> 337.61 | Improved |
| decoder refinement stages on HumanoidStand (S0 to S3) | GORL-DIFF · HumanoidStand | stage mean return: 89.7 -> 976.1 | Improved |
| direct flow-matching surrogate (FPO) instead of PPO | Gaussian PPO · FingerSpin | final return: 539.03 -> 56.05 | Hurt |
| direct diffusion PPO (DPPO) instead of PPO | Gaussian PPO · HopperStand | final return: 286.09 -> 2.14 | Hurt |
Closest papers by meaning
- Flow-Based Policy for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.92)
- Generative Actor-Critic with Soft Bridge Policies (NeurIPS 2026 · similarity 0.89)
- Aligning Flow Map Policies with Optimal $Q$-Guidance (NeurIPS 2026 · similarity 0.88)
- Generative Control as Optimization: Time Unconditional Flow Matching for Adaptive and Robust Robotic Control (NeurIPS 2026 · similarity 0.88)
- GenPO: Generative Diffusion Models Meet On-Policy Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Consistency Models as a Rich and Efficient Policy Class for Reinforcement Learning (ICLR 2024 · similarity 0.87)
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.
