Research map / Reinforcement learning / Offline RL
MOBODY: Model-Based Off-Dynamics Offline Reinforcement Learning
Yihong Guo, Yu Yang, Pan Xu, Anqi Liu
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 |
|---|---|---|---|
| learn target dynamics with separate action encoders and roll out the model, instead of reward regularization / filtering | off-dynamics offline RL · MuJoCo gravity/friction sum | normalized score: 986.13 -> 1427.26 | Improved |
| use dynamics trained on target data only instead of MOBODY dynamics (remove shared/separate-encoder model) (ablation) | MOBODY · Walker2d-Gravity 5.0 | normalized score: 46.05 -> 3.56 | Hurt |
| use CQL instead of target-Q-weighted BC policy learning (ablation) | MOBODY · Walker2d-Gravity 5.0 | normalized score: 46.05 -> 11.37 | Hurt |
| add cycle transition loss (VAE-style) (ablation) | MOBODY dynamics model · Walker2d-Gravity 0.1 | normalized score: 35.34 -> 65.85 | Improved |
| add target-Q weighting to behavior cloning (vs vanilla BC) (ablation) | MOBODY · Walker2d-Gravity 0.1 | normalized score: 19.53 -> 65.85 | Improved |
| target-Q weighting vs vanilla BC at gravity 2.0 (ablation) | MOBODY · Walker2d-Gravity 2.0 | normalized score: 47.13 -> 44.32 | No clear effect |
| learn dynamics with separate action encoders vs combined source+target data (ablation) | MOBODY dynamics · Walker2d-friction-0.5 | normalized score: 41.38 -> 76.96 | Improved |
| separate action encoders vs combined data for dynamics (ablation) | MOBODY dynamics · Walker2d-friction-0.5 | rollout MSE: 1.96 -> 1.25 | Improved |
| pretrain-finetune dynamics vs MOBODY dynamics (ablation) | MOBODY dynamics · Ant-friction-0.5 | normalized score: 51.09 -> 62.41 | Improved |
| MOBODY vs best baseline at larger dynamics shift | MOBODY · Walker2d-Gravity 5.0 | normalized score: 5.65 -> 46.05 | Improved |
| MOBODY vs SRPO at mild shift | MOBODY · Hopper-Gravity 0.5 | normalized score: 67.06 -> 33.57 | Hurt |
| MOBODY vs DARA on Hopper morph-foot Medium | MOBODY · Hopper morph-foot Medium | normalized score: 40.61 -> 13.05 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MOPO — MOPO: Model-based Offline Policy Optimization (NeurIPS 2020) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Beyond OOD State Actions: Supported Cross-Domain Offline Reinforcement Learning (AAAI 2024 · similarity 0.92)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.92)
- Localized Dynamics-Aware Domain Adaption for Off-Dynamics Offline Reinforcement Learning (NeurIPS 2026 · similarity 0.91)
- Reward-Consistent Dynamics Models are Strongly Generalizable for Offline Reinforcement Learning (ICLR 2024 · similarity 0.91)
- ODRL: A Benchmark for Off-Dynamics Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- Constrained Latent Action Policies for Model-Based Offline Reinforcement Learning (NeurIPS 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.
