Research map / Reinforcement learning / Offline RL
Model-Based Offline Meta-Reinforcement Learning with Regularization
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Adapt the behavior-versus-meta-policy regularization weight across tasks and training iterations instead of fixing it (ablation) | MerPO · robotic locomotion meta-environments | average return | Helped |
| Add behavior-policy regularization to within-task policy optimization (ablation) | MerPO · robotic locomotion meta-environments | average return | Helped |
| Use a learned dynamics model for within-task optimization rather than model-free CQL (ablation) | MerPO · robotic locomotion meta-environments | average return | Helped |
| Use both meta-policy and behavior-policy regularizers with α=0.4 rather than behavior-policy regularization only (ablation) | MerPO · unseen offline tasks with different data qualities | average return | Helped |
| Use the meta-policy to guide RAC and initialize the task policy from it (ablation) | MerPO · unseen offline tasks | average return | Helped |
| Remove meta-policy network initialization while retaining meta-policy regularization (ablation) | MerPO · unseen offline tasks with good offline datasets | average return | No clear effect |
| Use a more conservative model utilization by increasing the real-data ratio (ablation) | MerPO · robotic locomotion meta-environments | average return | Helped |
| Use meta-policy-only regularization instead of RAC's interpolated behavior and meta-policy regularizers (ablation) | MerPO · robotic locomotion meta-environments | average return | Hurt |
| Use MerPO with meta-policy regularization on an expert-data task rather than COMBO | MerPO · new offline task with expert data | average return | Helped |
| Use FOCAL rather than COMBO across datasets of different quality | offline meta-RL · Half-Cheetah-Fwd-Back and Ant-Fwd-Back | average return | Mixed |
| Regularize policy improvement with a poor-quality meta-policy only rather than using RAC's behavior-policy safeguard (ablation) | RAC · D4RL continuous-control tasks | average return | Hurt |
Closest papers by meaning
- Offline Meta-Reinforcement Learning with Advantage Weighting (ICML 2021 · similarity 0.94)
- Iteratively Refined Behavior Regularization for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.92)
- Entropy Regularized Task Representation Learning for Offline Meta-Reinforcement Learning (AAAI 2025 · similarity 0.92)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Towards an Information Theoretic Framework of Context-Based Offline Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- Scrutinize What We Ignore: Reining In Task Representation Shift Of Context-Based Offline Meta Reinforcement Learning (ICLR 2025 · 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.