Research map / Reinforcement learning / Policy gradient and actor-critic
Dichotomous Diffusion Policy Optimization
Ruiming Liang, Yinan Zheng, Kexin ZHENG, Tianyi Tan, Jianxiong Li, Liyuan Mao, Zhihao Wang, Guang Chen, Hangjun Ye, Jingjing Liu, Jinqiao Wang, Xianyuan Zhan
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 |
|---|---|---|---|
| fine-tune with DIPOLE on navtrain | DP-VLA · NAVSIM | PDMS: 88.3 -> 89.7 | Improved |
| fine-tune with DIPOLE on navtest without ground truth | DP-VLA · NAVSIM | PDMS: 88.3 -> 94.8 | Improved |
| use DIPOLE instead of DPPO for fine-tuning | DP-VLA · NAVSIM navtest | PDMS: 89.0 -> 94.8 | Improved |
| add rejection sampling at inference (ablation) | DIPOLE · ExORL Walker stand | average return: 793 -> 953 | Improved |
| add rejection sampling at inference (ablation) | DIPOLE · ExORL Jaco reach-top-right | average return: 84 -> 117 | Improved |
| use dichotomous greedified weights instead of CFGRL (ablation) | DIPOLE w/o rs · ExORL Walker stand | average return: 782 -> 793 | Improved |
| use DIPOLE instead of FQL | DIPOLE · ExORL Jaco reach-top-right | average return: 224 -> 117 | Hurt |
| use DIPOLE instead of IFQL | DIPOLE · OGBench humanoidmaze-large-navigate | success rate: 11 -> 6 | Hurt |
| online fine-tuning with DIPOLE | DIPOLE · OGBench humanoidmaze-medium | success rate: 61 -> 97 | Improved |
| vary beta and greediness factor omega (ablation) | DIPOLE · ExORL | average return | No clear effect |
Closest papers by meaning
- Score Regularized Policy Optimization through Diffusion Behavior (ICLR 2024 · similarity 0.90)
- Diffusion-based Reinforcement Learning via Q-weighted Variational Policy Optimization (NeurIPS 2024 · similarity 0.90)
- Diffusion Policy Policy Optimization (ICLR 2025 · similarity 0.90)
- DIME: Diffusion-Based Maximum Entropy Reinforcement Learning (ICML 2025 · similarity 0.90)
- Diffusion Actor-Critic: Formulating Constrained Policy Iteration as Diffusion Noise Regression for Offline Reinforcement Learning (ICLR 2025 · similarity 0.89)
- Designing Reinforcement Learning for Diffusion Models: A Unified Path-Space View (NeurIPS 2026 · similarity 0.89)
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.
