Research map / Reinforcement learning / Offline RL
Efficient Diffusion Policies For Offline Reinforcement Learning
Bingyi Kang, Xiao Ma, Chao Du, Tianyu Pang, Shuicheng YAN
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 the Gaussian policy with EDP in TD3 | TD3 on Gym-locomotion D4RL · Gym-locomotion D4RL average | average normalized score (RAT): 75.3 -> 85.5 | Improved |
| Replace the Gaussian policy with EDP in IQL | IQL on Kitchen D4RL · Kitchen D4RL average | average normalized score (RAT): 53.3 -> 59.4 | Improved |
| Replace the Gaussian policy with EDP in IQL | IQL on Adroit D4RL · Adroit D4RL average | average normalized score (RAT): 54.4 -> 71.3 | Improved |
| Replace the Gaussian policy with EDP in IQL | IQL on AntMaze D4RL · AntMaze D4RL average | average normalized score (RAT): 63.0 -> 73.4 | Improved |
| Use action approximation instead of full DDPM-chain training (ablation) | EDP policy training · D4RL Gym-locomotion | training speed: 2.3x training | Improved |
| Use action approximation instead of full DDPM-chain training (ablation) | EDP policy training · D4RL Gym-locomotion | sampling speed: 3.3x sampling | Improved |
| Replace DDPM sampling with DPM-Solver (ablation) | EDP training · D4RL Gym-locomotion | training speed: 2.3x training speedup | Improved |
| Use action approximation at K=100 rather than full DDPM sampling (ablation) | EDP on walker2d-medium-v2 · walker2d-medium-v2 | normalized score (OMS): 86.9 -> 85.5 | Hurt |
| Use action approximation at K=100 rather than full DDPM sampling (ablation) | EDP on walker2d-medium-reply-v2 · walker2d-medium-reply-v2 | normalized score (OMS): 96.3 -> 93.3 | Hurt |
| Use action approximation at K=100 rather than full DDPM sampling (ablation) | EDP on walker2d-medium-expert-v2 · walker2d-medium-expert-v2 | normalized score (OMS): 111.5 -> 111.1 | No clear effect |
| Increase the number of EAS candidate actions (ablation) | EDP on Gym-locomotion · 8 of 9 locomotion tasks | normalized score | Improved |
| Use EAS for evaluation with a normal TD3+BC policy (ablation) | TD3+BC · Gym-locomotion D4RL average | normalized score | No clear effect |
Closest papers by meaning
- Diffusion Policies Creating a Trust Region for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.93)
- Entropy-regularized Diffusion Policy with Q-Ensembles for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- Behavior-Regularized Diffusion Policy Optimization for Offline Reinforcement Learning (ICML 2025 · similarity 0.91)
- DiffStitch: Boosting Offline Reinforcement Learning with Diffusion-based Trajectory Stitching (ICML 2024 · similarity 0.91)
- Prior-Guided Diffusion Planning for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Score Regularized Policy Optimization through Diffusion Behavior (ICLR 2024 · similarity 0.91)
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.
