Research map / Reinforcement learning / Offline RL
MOPO: Model-based Offline Policy Optimization
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 |
|---|---|---|---|
| add uncertainty reward penalty (learned variance) (ablation) | MBPO · halfcheetah-jump | average return: -980.8 -> 4016.6 | Helped |
| add uncertainty reward penalty (learned variance) (ablation) | MBPO · ant-angle | average return: 18.6 -> 2530.9 | Helped |
| add uncertainty reward penalty (ablation) | MBPO · halfcheetah-mixed | average return: 6409.1 -> 6405.8 | No clear effect |
| use learned-variance penalty instead of ensemble disagreement (ablation) | MOPO · halfcheetah-jump / ant-angle | average return: 3577.3 -> 4016.6; 2256.0 -> 2530.9 | Helped |
| add model ensemble (ablation) | MBPO · halfcheetah-mixed | average return: 2247.2 -> 5598.4 | Helped |
| replace max variance penalty with mean variance (ablation) | MOPO · halfcheetah-jump | average return: 4140.6 -> 4166.3 | No clear effect |
| use oracle true-error penalty instead of estimated (ablation) | MOPO · halfcheetah-mixed | average return: 6405.8 -> 6984.0 | Helped |
| use 0% vs 5% real data in SAC batches (ablation) | MBPO · D4RL 12 tasks | normalized score | No clear effect |
| model-based offline RL (MOPO) instead of model-free (BRAC-v) | offline RL · D4RL medium datasets | normalized score: 81.3 -> 17.8 | Hurt |
| model-based offline RL (MOPO) instead of model-free (BEAR) | offline RL · D4RL mixed datasets | normalized score: 10.8 -> 67.5 | Helped |
Closest papers by meaning
- OCEAN-MBRL: Offline Conservative Exploration for Model-Based Offline Reinforcement Learning (AAAI 2024 · similarity 0.94)
- A Minimalist Approach to Offline Reinforcement Learning (NeurIPS 2021 · similarity 0.94)
- COMBO: Conservative Offline Model-Based Policy Optimization (NeurIPS 2021 · similarity 0.93)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.93)
- Optimistic Model Rollouts for Pessimistic Offline Policy Optimization (AAAI 2024 · similarity 0.93)
- Model Selection for Off-policy Evaluation: New Algorithms and Experimental Protocol (NeurIPS 2025 · similarity 0.92)
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.