Research map / Reinforcement learning / Visual and representation RL
Diffusion Model is an Effective Planner and Data Synthesizer for Multi-Task Reinforcement Learning
Haoran He, Chenjia Bai, Kang Xu, Zhuoran Yang, Weinan Zhang, Dong Wang, Bin Zhao, Xuelong Li
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 GPT2 transformer backbone instead of U-Net (ablation) | MTDiff-p · MT50-rand near-optimal | success rate (%): 55.67 -> 59.53 | Improved |
| use GPT2 transformer backbone instead of U-Net (ablation) | MTDiff-p · MT50-rand sub-optimal | success rate (%): 47.42 -> 48.67 | Improved |
| replace one-hot task ID with trajectory prompt (ablation) | MTDiff-p · MT50-rand near-optimal | success rate (%): 61.32 -> 59.53 | No clear effect |
| replace one-hot task ID with trajectory prompt (ablation) | MTDiff-p · MT50-rand sub-optimal | success rate (%): 48.94 -> 48.67 | No clear effect |
| use diffusion planning with prompt instead of autoregressive PromptDT | Prompt-conditioned GPT2 policy · Meta-World MT50-rand near-optimal | success rate (%): 45.68 -> 59.53 | Improved |
| use diffusion planning instead of PromptDT | Prompt-conditioned GPT2 policy · Ant-dir | average score: 409.81 -> 602.17 | Improved |
| augment dataset with MTDiff-s synthetic data | TD3-BC · coffee-push | success rate (%): 28.60 -> 74.67 | Improved |
| augment dataset with MTDiff-s synthetic data | TD3-BC · hand-insert unseen | success rate (%): 30.60 -> 70.87 | Improved |
| augment dataset with MTDiff-s synthetic data | TD3-BC · sweep-into | success rate (%): 91.8 -> 92.87 | No clear effect |
| MTDiff-s synthetic data vs S4RL augmentation | TD3-BC · bin-picking unseen | success rate (%): 72.20 -> 55.73 | Hurt |
| train MTDiff-s on more tasks (10, 20, 30 vs single task) (ablation) | MTDiff-s · coffee-push | success rate | Improved |
Closest papers by meaning
- Task-Agnostic Pre-training and Task-Guided Fine-tuning for Versatile Diffusion Planner (ICML 2025 · similarity 0.91)
- Simple Hierarchical Planning with Diffusion (ICLR 2024 · similarity 0.91)
- Prior-Guided Diffusion Planning for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Structural Information-based Hierarchical Diffusion for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Score Regularized Policy Optimization through Diffusion Behavior (ICLR 2024 · similarity 0.88)
- Diffusion Models for Multi-Task Generative Modeling (ICLR 2024 · similarity 0.88)
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.
