Research map / Vision-language and multimodal / Human and embodied modeling
VER: Vision Expert Transformer for Robot Learning via Foundation Distillation and Dynamic Routing
Yixiao Wang, Mingxiao Huo, Zhixuan Liang, Yushi Du, Lingfeng Sun, Haotian Lin, Jinghuan Shang, Chensheng Peng, Mohit Bansal, Mingyu Ding, Masayoshi Tomizuka
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 |
|---|---|---|---|
| use VER-B instead of Theia-B | policy head on frozen encoder · 11 manipulation tasks | success rate (%): 67.1 -> 74.7 | Improved |
| use VER-T instead of Theia-T | flow-matching head · cross->bin | success rate: 0.65 -> 0.95 | Improved |
| use VER-T instead of Theia-T | diffusion head · Real-world pour | success rate: 0.45 -> 0.90 | Improved |
| use VER instead of fine-tuned GR00T N1.5 | flow-matching head · cube->cup | success rate: 0.73 -> 0.75 | Improved |
| add curriculum Top-K annealing to patchwise expert routing (ablation) | VER · Relocate | success rate (%): 47.6 -> 56.4 | Improved |
| add curriculum Top-K annealing to patchwise expert routing (ablation) | VER · Pen | success rate (%): 78.0 -> 80.8 | Improved |
| use patchwise expert routing instead of a single frozen teacher router (ablation) | VER · Relocate | success rate (%): 38.4 -> 47.6 | Improved |
| use patchwise instead of layerwise teacher routing (ablation) | VER · Relocate | success rate (%): 36.4 -> 47.6 | Improved |
| increase Top-K from 1 to 3 (ablation) | VER-Tiny · Relocate+Pen average | success rate (%): 60.0 -> 68.0 | Improved |
| add one train-from-scratch expert to distilled experts (ablation) | VER · Relocate+Pen average | success rate (%): 72.0 -> 78.7 | Improved |
| increase CTA schedule S from 0 to 60 (ablation) | VER · pen-v0 | success rate (%): 77.3 -> 81.3 | Improved |
| scale experts from (K,L)=(1,3) to (2,6) (ablation) | VER-Tiny · pen-v0 | success rate (%): 44.0 -> 69.6 | Improved |
Closest papers by meaning
- Swiss Army Knife: Synergizing Biases in Knowledge from Vision Foundation Models for Multi-Task Learning (ICLR 2025 · similarity 0.87)
- Train the Agent, Not the Expert: Learning to Harness Heterogeneous Experts for Multi-Turn Visual Reasoning (NeurIPS 2026 · similarity 0.87)
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.87)
- ROVER: Recursive Reasoning Over Videos with Vision-Language Models for Embodied Tasks (NeurIPS 2025 · similarity 0.86)
- Transforming Vision Transformer: Towards Efficient Multi-Task Asynchronous Learner (NeurIPS 2024 · similarity 0.86)
- MENTOR: Mixture-of-Experts Network with Task-Oriented Perturbation for Visual Reinforcement Learning (ICML 2025 · 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.
