Research map / Video / Action recognition
Multi-view Consistent Latent Action Learning for World Modeling and Control
Shenghua Wan, Xiaohai Hu, Xunlan Zhou, Lei Yuan, Le Gan, De-Chuan Zhan
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-view consistency loss (self-distillation) to latent action learning (ablation) | MuCoLA · Multi-view test sets, next-frame reconstruction | PSNR: 32.36 -> 37.97 | Improved |
| add cross-view consistency loss (ablation) | MuCoLA · Action regression on OpenX (test) | RMSE: 0.225 -> 0.214 | Improved |
| add explicit Gaussian prior (KL weight 0.5) on latent actions (ablation) | MuCoLA · Multi-view test sets, next-frame reconstruction | PSNR: 37.97 -> 35.42 | Hurt |
| add explicit Gaussian prior on latent actions (ablation) | MuCoLA · Action regression on OpenX (test) | RMSE: 0.214 -> 0.233 | Hurt |
| add Gaussian prior or drop consistency loss (ablation) | MuCoLA · VP2 visual planning and Meta-World MBRL | success rate | Hurt |
| use MuCoLA latent actions instead of AdaWorld | iVideoGPT world model · Multi-view test sets, next-frame reconstruction | LPIPS (%): 14.63 -> 9.42 | Improved |
| use MuCoLA instead of AdaWorld for behavior cloning | ViT-T policy · LIBERO (four suites) | success rate (%): Long 55.2 -> 65.4; Goal 76.1 -> 73.8 | Mixed |
| increase prototype vocabulary K to 8192 (and shrink below 2048) (ablation) | MuCoLA · reconstruction error | reconstruction error | Hurt |
| scale Spatial-Temporal Transformer parameter count (ablation) | MuCoLA · next-frame reconstruction | LPIPS | Improved |
| scale training data from 1% to 100% (ablation) | MuCoLA · next-frame reconstruction | LPIPS | Improved |
| increase labeled action samples for grounding (1K to 5K+) (ablation) | MuCoLA · LIBERO | success rate | 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.
- AdaWorld — AdaWorld: Learning Adaptable World Models with Latent Actions (ICML 2025) · better on 5 of 6 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
- Learning Latent Action World Models in the Wild (ICML 2026 · similarity 0.88)
- Learning Visual Feature-Based World Models via Residual Latent Action (NeurIPS 2026 · similarity 0.87)
- World Models as Group Actions (NeurIPS 2026 · similarity 0.87)
- Action Images: End-to-End Policy Learning via Multiview Video Generation (NeurIPS 2026 · similarity 0.86)
- PlaySlot: Learning Inverse Latent Dynamics for Controllable Object-Centric Video Prediction and Planning (ICML 2025 · similarity 0.86)
- Learning Latent Dynamic Robust Representations for World Models (ICML 2024 · similarity 0.86)
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.
