Research map / Reinforcement learning / Offline RL
Model-Based Offline Planning
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 |
|---|---|---|---|
| Use MBOP instead of MOPO | D4RL medium and med-expert tasks · D4RL halfcheetah, hopper, walker2d | normalized return: halfcheetah medium: 40.2 -> 44.6; hopper medium: 26.5 -> 48.8; walker2d medium: 14.0 -> 41.0; halfcheetah med-expert: 57 | Mixed |
| Use MBOP instead of MOPO | D4RL random and mixed tasks · D4RL halfcheetah, hopper, walker2d | normalized return: random halfcheetah: 31.9 -> 6.3; hopper: 13.3 -> 10.8; walker2d: 13.0 -> 8.1; mixed halfcheetah: 54.0 -> 42.3; hopper: 9 | Mixed |
| Replace the learned behavior-cloning prior with a Gaussian prior (ablation) | MBOP · RLU Cartpole, 5000 points | mean return: 803.1 -> 605.8 | Hurt |
| Remove the estimated-return value function (ablation) | MBOP · RLU Cartpole, 5000 points | mean return: 803.1 -> 715.2 | Hurt |
| Remove both the behavior prior and value function (ablation) | MBOP · RLU Cartpole, 5000 points | mean return: 803.1 -> 726.6 | Hurt |
| Replace the learned behavior-cloning prior with a Gaussian prior (ablation) | MBOP · RLU Quadruped, 5000 points | mean return: 974.1 -> 561.6 | Hurt |
| Remove the estimated-return value function (ablation) | MBOP · RLU Quadruped, 5000 points | mean return: 974.1 -> 959.5 | No clear effect |
| Remove both the behavior prior and value function (ablation) | MBOP · RLU Quadruped, 5000 points | mean return: 974.1 -> 569.8 | Hurt |
| Replace the learned behavior-cloning prior with a Gaussian prior (ablation) | MBOP · RLU Walker, 41000 points | mean return: 752.0 -> 160.9 | Hurt |
| Remove the estimated-return value function (ablation) | MBOP · RLU Walker, 41000 points | mean return: 752.0 -> 700.6 | Hurt |
| Remove both the behavior prior and value function (ablation) | MBOP · RLU Walker, 41000 points | mean return: 752.0 -> 79.5 | Hurt |
| Use a learned behavior-cloning prior during planning (ablation) | MBOP · D4RL tasks | mean return | Mixed |
Closest papers by meaning
- MOPO: Model-based Offline Policy Optimization (NeurIPS 2020 · similarity 0.92)
- A Unified Framework for Alternating Offline Model Training and Policy Learning (NeurIPS 2022 · similarity 0.91)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.90)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Are Expressive Models Truly Necessary for Offline RL? (AAAI 2025 · similarity 0.89)
- Constrained Latent Action Policies for Model-Based Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
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.