Research map / Reinforcement learning / Visual and representation RL
Structured Expert Routing with Multi-View Task Priors for Offline Meta-Reinforcement Learning
Yisen Zhao, Peixi Peng, Xinyu Hu, Cong Li, Zhan Su, Zhuojian Li
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 |
|---|---|---|---|
| add dynamics prior (ablation) | Prompt-DT · Cheetah-Vel | test return: -135.89 -> -112.51 | Improved |
| add dynamics prior (ablation) | Prompt-DT · Hopper-Param | test return: 309.90 -> 340.87 | Improved |
| add semantic prior on top of dynamics (ablation) | Prompt-DT + Dyn · Hopper-Param | test return: 340.87 -> 352.19 | Improved |
| add dynamics prior to MoE routing (ablation) | MoE (M3DT) · Cheetah-Vel | test return: -107.19 -> -106.49 | No clear effect |
| add semantic prior on top of dynamics (ablation) | MoE + Dyn · Hopper-Param | test return: 348.05 -> 364.59 | Improved |
| add trajectory prior on top of dynamics and semantic (ablation) | MoE + Dyn + Sem · Cheetah-Vel | test return: -102.88 -> -92.64 | Improved |
| combine Content and Global routing scores instead of Content only (ablation) | TGR · Cheetah-Vel | test return: -110.48 -> -92.64 | Improved |
| combine Content and Global routing scores instead of Global only (ablation) | TGR · Hopper-Param | test return: 362.45 -> 370.52 | Improved |
| add semantic input to routing (vs w/o Sem) (ablation) | TGR · Cheetah-Vel | test return: -105.75 -> -92.64 | Improved |
| add dynamics input to routing (vs w/o Dyn) (ablation) | TGR · Hopper-Param | test return: 360.18 -> 370.52 | Improved |
| add dynamics and trajectory inputs (vs w/o Dyn&Traj) (ablation) | TGR · Cheetah-Vel | test return: -120.18 -> -92.64 | Improved |
| use TGR instead of scaling Meta-DT to ~34M params | Meta-DT (Large) · Hopper-Param | test return: 350.47 -> 370.52 | Improved |
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.
- Meta-DT — Meta-DT: Offline Meta-RL as Conditional Sequence Modeling with World Model Disentanglement (NeurIPS 2024) · better on 3 of 3 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
- Pessimistic Latent Task-aware Optimization for Robust Offline Meta-Reinforcement Learning (NeurIPS 2026 · similarity 0.89)
- MetaCARD: Meta-Reinforcement Learning with Task Uncertainty Feedback via Decoupled Context-Aware Reward and Dynamics Components (AAAI 2024 · similarity 0.89)
- Entropy Regularized Task Representation Learning for Offline Meta-Reinforcement Learning (AAAI 2025 · similarity 0.88)
- Towards an Information Theoretic Framework of Context-Based Offline Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.88)
- Meta-DT: Offline Meta-RL as Conditional Sequence Modeling with World Model Disentanglement (NeurIPS 2024 · similarity 0.88)
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.
