Research map / Reinforcement learning / Offline RL
When to Trust Your Simulator: Dynamics-Aware Hybrid Offline-and-Online Reinforcement Learning
Haoyi Niu, Shubham Sharma, Yiwen Qiu, Ming Li, Guyue Zhou, Jianming HU, 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 |
|---|---|---|---|
| replace adaptive value regularization weight with uniform weight (add adaptive omega) (ablation) | H2O · HalfCheetah Medium Replay - Gravity | average return: 6675±179 -> 6813±289 | Improved |
| remove dynamics-ratio importance weight in Bellman error (add it) (ablation) | H2O · HalfCheetah Medium Replay - Gravity | average return: 4721±196 -> 6813±289 | Improved |
| add dynamics ratio to uniform-weight variant (ablation) | H2O · HalfCheetah Medium Replay - Gravity | average return: 5223±198 -> 6675±179 | Improved |
| remove value regularization entirely (add it) (ablation) | H2O · HalfCheetah Medium Replay - Gravity | average return: 6501±147 -> 6813±289 | Improved |
| add dynamics ratio to variant without value regularization (ablation) | H2O · HalfCheetah Medium Replay - Gravity | average return: 5290±356 -> 6501±147 | Improved |
| add value regularization when dynamics ratio is absent (ablation) | H2O · HalfCheetah Medium Replay - Gravity | average return: 5290±356 -> 4721±196 | Hurt |
| naively add offline real data to policy evaluation and improvement of DARC (DARC+) | DARC · HalfCheetah Medium Expert - Gravity | average return: 4759±353 -> 72±109 | Hurt |
| use H2O instead of CQL trained only on real offline data | CQL · HalfCheetah Medium - Gravity | average return: 6066±73 -> 7085±416 | Improved |
| use H2O instead of SAC trained in imperfect simulator | SAC(sim) · HalfCheetah Medium - Friction | average return: 2684±2646 -> 6848±445 | Improved |
| DARC versus H2O on Medium Expert Friction | H2O · HalfCheetah Medium Expert - Friction | average return: 9038±1480 vs 6745±562 | Hurt |
Closest papers by meaning
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Beyond OOD State Actions: Supported Cross-Domain Offline Reinforcement Learning (AAAI 2024 · similarity 0.90)
- OCEAN-MBRL: Offline Conservative Exploration for Model-Based Offline Reinforcement Learning (AAAI 2024 · similarity 0.90)
- Uni-O4: Unifying Online and Offline Deep Reinforcement Learning with Multi-Step On-Policy Optimization (ICLR 2024 · similarity 0.90)
- From Static Policies to Adaptive Priors in Offline Reinforcement Learning (NeurIPS 2026 · similarity 0.90)
- Hybrid Reinforcement Learning from Offline Observation Alone (ICML 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.
