Research map / 3D vision / 3D generation
Learning to Act Robustly with View-Invariant Latent Actions
Youngjoon Jeong, Junha Chun, Taesup Kim
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 |
|---|---|---|---|
| pretrain encoder with VILA (view-invariant latent actions) vs ImageNet ResNet-18 baseline | Diffusion Policy · Lift unseen views (fine-tuned) | success rate (%): 77.00 -> 94.70 | Improved |
| pretrain encoder with VILA vs Vanilla | Diffusion Policy · Coffee unseen views (fine-tuned) | success rate (%): 0.35 -> 12.65 | Improved |
| use VILA vs Vanilla on real-world pick-and-place | Diffusion Policy · Real-world novel views | average success rate (%): 3.33 -> 63.33 | Improved |
| use VILA vs fine-tuned VLAs | real-world pick-and-place · Real-world novel views | average success rate (%): 36.67 -> 63.33 | Improved |
| add scene-level view-invariant encoders (CLASS) vs Vanilla in frozen setting on Lift | Diffusion Policy · Lift unseen views (frozen) | success rate (%): 47.30 -> 65.00 | Mixed |
| add image augmentation (crop, color jitter) to Vanilla | Vanilla encoder · Square unseen views (fine-tuned) | success rate (%): 8.70 -> 9.30 | No clear effect |
| add structural alignment loss on top of weighted InfoNCE (ablation) | VILA · Lift unseen views (fine-tuned) | unseen success rate (%): 91.70 -> 94.70 | Improved |
| add action-aware weighting to the contrastive loss (ablation) | VILA · Lift unseen views (fine-tuned) | unseen success rate (%): 84.30 -> 94.70 | Improved |
| add contrastive and structural losses on top of base latent-action loss (ablation) | VILA · Lift unseen views (fine-tuned) | unseen success rate (%): 79.50 -> 94.70 | Improved |
| replace distance-based structural loss with CKA alignment (ablation) | VILA · Lift unseen views (fine-tuned) | unseen success rate (%): 94.70 -> 92.00 | Hurt |
| add auxiliary action-regression head (ablation) | VILA · Lift unseen views (fine-tuned) | unseen success rate (%): 94.70 -> 86.70 | Hurt |
| use DTW instead of L2 for action-sequence distance (ablation) | VILA · Lift unseen views (fine-tuned) | unseen success rate (%): 94.70 -> 89.20 | Hurt |
Closest papers by meaning
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.89)
- RotVLA: Rotational Latent Action for Vision-Language-Action Model (NeurIPS 2026 · similarity 0.89)
- Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- SyncWorld: Visual Calibration Enables World Models as Zero-Shot Simulators (NeurIPS 2026 · similarity 0.88)
- OSVI-WM: One-Shot Visual Imitation for Unseen Tasks using World-Model-Guided Trajectory Generation (NeurIPS 2025 · similarity 0.88)
- Action Images: End-to-End Policy Learning via Multiview Video Generation (NeurIPS 2026 · similarity 0.88)
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.
