Research map / Reinforcement learning / Offline RL
Diffusion Models as Optimizers for Efficient Planning in Offline RL
Renming Huang, Yunqiang Pei, Guoqing Wang, Yangming Zhang, Yang Yang, Peng Wang, Heng Tao Shen
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 |
|---|---|---|---|
| replace diffusion trajectory generation with autoregressive transformer plus few-step diffusion optimization | Trajectory Diffuser · D4RL Gym-locomotion | runtime per action (s): 4.61 -> 0.44 (vs DD) | Improved |
| replace diffusion trajectory generation with transformer plus diffusion optimization | Trajectory Diffuser · D4RL Gym-locomotion | average normalized return: 81.8 -> 84.2 (vs DD) | Improved |
| add diffusion-based trajectory optimization after transformer generation (ablation) | Trajectory Diffuser · D4RL Gym-locomotion | average normalized return: 79.1 -> 84.2 | Improved |
| use Transformer instead of GRU or LSTM for the feasible trajectory generator (ablation) | Trajectory Diffuser · D4RL Med-Expert HalfCheetah | normalized return: 80.8 / 83.1 -> 92.2 | Improved |
| increase diffusion optimization steps from 5 to 20 or more (ablation) | Trajectory Diffuser · D4RL Med-Expert | average normalized return: 105.0 -> 98.4 (20 steps) | Hurt |
| reduce diffusion optimization steps from 5 to 2 (ablation) | Trajectory Diffuser · D4RL Med-Expert | average normalized return: 105.0 -> 103.7 | No clear effect |
| apply Trajectory Diffuser on Adroit Pen versus Decision Diffuser | Trajectory Diffuser · Adroit Pen | average normalized return: 23.9 -> 96.7 | Improved |
| apply Trajectory Diffuser on Kitchen Mixed versus Decision Diffuser | Trajectory Diffuser · Franka Kitchen Mixed | normalized return: 65.0 -> 54.2 | Hurt |
| use Trajectory Diffuser versus DD on Medium HalfCheetah | Trajectory Diffuser · Medium HalfCheetah | normalized return: 49.1 -> 44.2 | Hurt |
Closest papers by meaning
- Prior-Guided Diffusion Planning for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.92)
- Simple Hierarchical Planning with Diffusion (ICLR 2024 · similarity 0.92)
- Score Regularized Policy Optimization through Diffusion Behavior (ICLR 2024 · similarity 0.91)
- Stitching Sub-trajectories with Conditional Diffusion Model for Goal-Conditioned Offline RL (AAAI 2024 · similarity 0.90)
- What Makes a Good Diffusion Planner for Decision Making? (ICLR 2025 · similarity 0.90)
- Diffusion-DICE: In-Sample Diffusion Guidance for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
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.
