Research map / Video / Video segmentation and tracking
Pre-training Contextualized World Models with In-the-wild Videos for Reinforcement Learning
Jialong Wu, Haoyu Ma, Chaoyi Deng, Mingsheng Long
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 |
|---|---|---|---|
| Pre-train ContextWM on Something-Something-v2 videos | ContextWM · Meta-world | success rate | Improved |
| Use ContextWM instead of a plain world model | IPV fine-tuning on Meta-world · Meta-world | sample efficiency | Improved |
| Pre-train ContextWM on Something-Something-v2 videos | ContextWM · DMControl Remastered | performance | Improved |
| Pre-train a plain world model on Something-Something-v2 videos | Plain WM · DMControl Remastered | performance | Mixed |
| Pre-train ContextWM on Something-Something-v2 videos | ContextWM · CARLA | episode return | Improved |
| Pre-train with Human3.6M instead of training from scratch | ContextWM · DMControl Remastered | performance | Hurt |
| Use the assembled three-dataset pre-training set instead of SSv2 | ContextWM · DMControl Remastered | performance | No clear effect |
| Pre-train on the more domain-similar RLBench videos instead of SSv2 | ContextWM · Meta-world | performance | Improved |
| Pre-train with assembled data from three in-the-wild datasets instead of SSv2 | ContextWM · Meta-world | performance | Improved |
| Pre-train on YouTube Driving or assembled videos instead of training from scratch | ContextWM · CARLA | performance | Improved |
| Pre-train on YouTube Driving or assembled videos instead of SSv2 | ContextWM · CARLA | performance | No clear effect |
| Add video pre-training (ablation) | ContextWM · Meta-world | aggregate performance | Improved |
Closest papers by meaning
- XLand-100B: A Large-Scale Multi-Task Dataset for In-Context Reinforcement Learning (ICLR 2025 · similarity 0.89)
- Zero-shot World Models via Search in Memory (NeurIPS 2025 · similarity 0.88)
- VideoWorld: Exploring Knowledge Learning from Unlabeled Videos (CVPR 2025 · similarity 0.88)
- Learning Visual Feature-Based World Models via Residual Latent Action (NeurIPS 2026 · similarity 0.88)
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- Solving New Tasks by Adapting Internet Video Knowledge (ICLR 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.
