Research map / Label-efficient and robust learning / Recognition and long-tail
MVP-LAM: Learning Action-Centric Latent Action via Cross-Viewpoint Reconstruction
Jung Min Lee, Dohyeok Lee, Seokhun Ju, Taehyun Cho, Jin Woo Koo, Li Zhao, Sangwoo Hong, Jungwoo Lee
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 cross-viewpoint reconstruction loss (ablation) | MVP-LAM · Bridge V2 latent action | MI (KSG): 0.27 -> 1.10 | Improved |
| add cross-viewpoint reconstruction loss (ablation) | MVP-LAM · Bridge V2 latent action | linear probe NMSE: 0.96 -> 0.73 | Improved |
| add human multi-view videos to training (ablation) | MVP-LAM · Bridge V2 latent action | MI (KSG): 0.50 -> 1.10 | Improved |
| add human multi-view videos to training (ablation) | MVP-LAM · Bridge V2 latent action | linear probe NMSE: 0.91 -> 0.73 | Improved |
| increase synchronization lag from 0 to 4 frames (ablation) | MVP-LAM · Bridge V2 latent action (sync lag) | MI (KSG): 1.02 -> 1.02 | No clear effect |
| increase synchronization lag from 0 to 4 frames (ablation) | MVP-LAM · Bridge V2 latent action (sync lag) | linear probe NMSE: 0.78 -> 0.80 | No clear effect |
| replace UniVLA latent actions with MVP-LAM | VLA pretraining · SIMPLER (WidowX) AVG | success rate (%): 39.6 -> 60.4 | Improved |
| replace UniVLA latent actions with MVP-LAM | VLA pretraining · LIBERO AVG (Bridge V2 pretrain) | success rate (%): 92.5 -> 94.1 | Improved |
| replace UniVLA latent actions with MVP-LAM | VLA pretraining · LIBERO-Object (Bridge V2 pretrain) | success rate (%): 95.4 -> 94.6 | Hurt |
| replace UniVLA latent actions with MVP-LAM | VLA pretraining · LIBERO-Long (Bridge V2 pretrain) | success rate (%): 87.5 -> 90.8 | Improved |
| replace UniVLA latent actions with MVP-LAM | latent action model · Bridge V2 inverse probe | NMSE: 0.875 -> 0.792 | Improved |
| replace UniVLA latent actions with MVP-LAM | latent action model · LIBERO-Goal inverse probe | NMSE: 0.587 -> 0.643 | Hurt |
Closest papers by meaning
- Latent Action Pretraining from Videos (ICLR 2025 · similarity 0.91)
- From Pixels to Tokens: A Systematic Study of Latent Action Supervision for Vision-Language-Action Models (ICML 2026 · similarity 0.91)
- RotVLA: Rotational Latent Action for Vision-Language-Action Model (NeurIPS 2026 · similarity 0.89)
- Latent Action Learning Requires Supervision in the Presence of Distractors (ICML 2025 · similarity 0.88)
- LAFP: Preserving Latent Action Structure in Latent Policy Learning via Flow Matching (NeurIPS 2026 · similarity 0.88)
- Videos are Sample-Efficient Supervisions: Behavior Cloning from Videos via Latent Representations (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.
