Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

VLA Models Are More Generalizable Than You Think: Revisiting Physical and Spatial Modeling

About

Vision-language-action (VLA) models achieve strong in-distribution performance but degrade sharply under novel camera viewpoints and visual perturbations. We show that this brittleness primarily arises from misalignment in Spatial Modeling, rather than Physical Modeling. To address this, we propose a one-shot adaptation framework that recalibrates visual representations through lightweight, learnable updates. Our first method, Feature Token Modulation (FTM), applies a global affine transformation to visual tokens and improves Libero viewpoint accuracy from 48.5% to 87.1% with only 4K parameters. Building on this, Feature Linear Adaptation (FLA) introduces low-rank updates to the ViT encoder, achieving 90.8% success with 4.7M parameters -- matching LoRA-scale finetuning at far lower cost. Together, these results reveal substantial untapped robustness in pretrained VLA models and demonstrate that targeted, minimal visual adaptation is sufficient to restore viewpoint generalization.

Weiqi Li, Quande Zhang, Ruifeng Zhai, Liang Lin, Guangrun Wang• 2025

Related benchmarks

TaskDatasetResultRank
Robot Policy LearningLIBERO--
73
Robot ManipulationLIBERO-V Across Novel Camera Viewpoints (unseen)
Spatial Success Rate96.5
14
Robot manipulation adaptationLIBERO novel viewpoints
Average Success Rate74.3
14
Robotic ManipulationLIBERO-V Visual
Performance (Camera)90.8
7
One-shot VLA AdaptationLIBERO
Success Rate (View)76.4
5
Robotic ManipulationReal-world UR10e Robot Novel Viewpoint (target domain)
Success Rate (Eggplant)58.3
5
Showing 6 of 6 rows

Other info

Follow for update