Research map / Vision-language and multimodal / Robot action and navigation
VLA Models Are More Generalizable Than You Think: Revisiting Physical and Spatial Modeling
Weiqi Li, Quande Zhang, Ruifeng Zhai, Liang Lin, Guangrun Wang
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 FTM (global affine on visual tokens, 4K params) | π0.5 · LIBERO novel viewpoints | success rate (%): 48.5 -> 87.1 | Improved |
| add FLA (low-rank ViT adaptation, 4.7M params) instead of LoRA on whole VLA (467M) | π0.5 · LIBERO novel viewpoints | success rate (%): 90.3 -> 90.8 | No clear effect |
| FLA vs LoRA on LIBERO-V average | π0.5 · LIBERO-V | success rate (%): 94.6 -> 94.8 | No clear effect |
| use FTM instead of FLA (ablation) | π0.5 · LIBERO-V | success rate (%): 94.8 -> 90.5 | Hurt |
| use prompt learning instead of FTM (ablation) | π0.5 · LIBERO-V | success rate (%): 90.5 -> 75.1 | Hurt |
| increase FLA rank from 16 to 32 (ablation) | π0.5 FLA · LIBERO novel viewpoints | success rate (%): 90.8 -> 91.2 | No clear effect |
| apply FLA instead of LoRA on π0 (ablation) | π0 · LIBERO novel viewpoints | success rate (%): 83.6 -> 84.0 | No clear effect |
| use FLA on larger viewpoint shifts (ablation) | π0.5 · LIBERO large viewpoint change | success rate (%): 85.6 -> 87.9 | Improved |
| FLA on small viewpoint change vs LoRA (ablation) | π0.5 · LIBERO small viewpoint change | success rate (%): 94.8 -> 94.6 | No clear effect |
| FTM or FLA vs zero-shot on light, texture and noise perturbations | π0.5 · LIBERO-V light/texture/noise | success rate (%): 96.2 -> 96.8 (light) | No clear effect |
Closest papers by meaning
- SwiftVLA: Unlocking Spatiotemporal Dynamics for Lightweight VLA Models at Minimal Overhead (CVPR 2026 · similarity 0.90)
- EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models (NeurIPS 2025 · similarity 0.88)
- AdaptVision: Efficient Vision-Language Models via Adaptive Visual Acquisition (CVPR 2026 · similarity 0.88)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- CogVLA: Cognition-Aligned Vision-Language-Action Models via Instruction-Driven Routing & Sparsification (NeurIPS 2025 · similarity 0.88)
- Exploring the Better Multimodal Synergy Strategy for Vision-Language Models (AAAI 2025 · 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.
