Research map / Reinforcement learning / Visual and representation RL
Trajectory-wise Multiple Choice Learning for Dynamics Generalization in Reinforcement Learning
Younggyo Seo, Kimin Lee, Ignasi Clavera Gilaberte, Thanard Kurutach, Jinwoo Shin, Pieter Abbeel
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 |
|---|---|---|---|
| Use trajectory-wise oracle loss rather than single-transition MCL (ablation) | Multi-headed dynamics model · Unseen control environments | generalization performance | Improved |
| Add adaptive selection of the best prediction head instead of averaging head predictions (ablation) | T-MCL · Unseen control environments | average return | Improved |
| Add context learning to the multi-headed dynamics model (ablation) | T-MCL · Unseen HalfCheetah environments | generalization performance | Improved |
| Use a multi-headed architecture without trajectory-wise oracle loss, context learning, or adaptive planning (ablation) | PETS · Unseen HalfCheetah environments | generalization performance | No clear effect |
| Increase prediction-head count above 3 (ablation) | T-MCL · Unseen HalfCheetah environments | generalization performance | Hurt |
| Use more transitions in the trajectory-wise oracle loss (ablation) | T-MCL · Unseen control environments | generalization performance | Improved |
| Use more transitions for adaptive planning (ablation) | T-MCL · Unseen control environments | generalization performance | Improved |
| Use T-MCL instead of PETS | Model-based RL · HalfCheetah unseen environments | average return: 2223.9 -> 19280.1 | Improved |
| Use T-MCL instead of CaDM | Model-based RL · CrippledAnt | average return: 3.5x | Improved |
Closest papers by meaning
- Reward-Consistent Dynamics Models are Strongly Generalizable for Offline Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · similarity 0.87)
- Efficient Dynamics Modeling in Interactive Environments with Koopman Theory (ICLR 2024 · similarity 0.87)
- Any-step Dynamics Model Improves Future Predictions for Online and Offline Reinforcement Learning (ICLR 2025 · similarity 0.87)
- Model-Based Transfer Learning for Contextual Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.87)
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.
