Research map / Reinforcement learning / Visual and representation RL
Local Motion Matters: A Deconstruct-Recompose Paradigm for Reinforcement Learning Pre-training from Videos
Jinwen Wang, Youfang Lin, Xiaobo Hu, Shuo Wang, Kai Lv
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 pre-trained latent dynamics model (ablation) | DRP · DMCR Walker Run | episode return: 613 -> 681 | Improved |
| add pre-trained latent dynamics model (ablation) | DRP · DMCR Hopper Stand | episode return: 708 -> 796 | Improved |
| add pre-trained dynamics model (ablation) | IPV · DMCR Walker Run | episode return: 586 -> 595 | No clear effect |
| add pre-trained dynamics model (ablation) | IPV · DMCR Hopper Stand | episode return: 628 -> 634 | No clear effect |
| replace global flow encoder with local atomic-action motion representation (ablation) | DRP · DMCR Walker Run and Hopper Stand | episode return | Improved |
| add intra-frame attention to DAE (ablation) | DRP · DMCR and Meta-World | episode return | Improved |
| add inter-frame attention to DAE (ablation) | DRP · DMCR and Meta-World | episode return | Improved |
| add pre-training on SSV2 (ablation) | DRP · DMCR and Meta-World | episode return | Improved |
| use local-motion pre-training instead of IPV pre-training with same fine-tuning (ablation) | DRP · DMCR and Meta-World | episode return | Improved |
Closest papers by meaning
- StaMo: Unsupervised Learning of Generalizable Robot Motion from Compact State Representation (CVPR 2026 · similarity 0.89)
- DynaFLIP: Rethinking Robotics Perception via Tri-Modal-Dynamics Guided Representation (NeurIPS 2026 · similarity 0.88)
- Learning an Actionable Discrete Diffusion Policy via Large-Scale Actionless Video Pre-Training (NeurIPS 2024 · similarity 0.88)
- Videos are Sample-Efficient Supervisions: Behavior Cloning from Videos via Latent Representations (NeurIPS 2025 · similarity 0.87)
- LAFP: Preserving Latent Action Structure in Latent Policy Learning via Flow Matching (NeurIPS 2026 · similarity 0.87)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · 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.
