VGAS: Value-Guided Action-Chunk Selection for Few-Shot Vision-Language-Action Adaptation

Vision–Language–Action (VLA) models bridge multimodal reasoning with physical control, but adapting them to new tasks with scarce demonstrations remains unreliable. While fine-tuned VLA policies often produce semantically plausible trajectories, failures often arise from unresolved geometric ambiguities, where near-miss actions lead to divergent execution outcomes under limited supervision. We study few-shot VLA adaptation from a \emph{generation–selection} perspective and propose a novel framework $\mathbf{\text{VGAS}}$ ($\mathbf{\text{V}}$alue-$\mathbf{\text{G}}$uided $\mathbf{\text{A}}$ction-chunk $\mathbf{\text{S}}$election). It performs inference-time best-of-$N$ selection to identify action chunks that are both semantically faithful and geometrically precise. Specifically, $\mathbf{\text{VGAS}}$ employs a finetuned VLA as a high-recall proposal generator and introduces the $\text{Q-Chunk-Former}$, a geometrically grounded Transformer critic to resolve fine-grained geometric ambiguities. In addition, we propose $\mathit{\text{Explicit}}\mathit{\text{Geometric}}\mathit{\text{Regularization}}$ ($\mathtt{\text{EGR}}$), which shapes a discriminative value landscape to preserve action ranking resolution among near-miss candidates while mitigating value instability under scarce supervision. Experiments and theoretical analysis demonstrate that $\mathbf{\text{VGAS}}$ consistently improves success rates and robustness under limited demonstrations and distribution shifts. Our code is available at https://github.com/Jyugo-15/VGAS.