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

arXiv:2602.07399v2 Announce Type: replace Abstract: 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 generation--selection perspective and propose a novel framework extbf{VGAS} (extbf{V}alue-extbf{G}uided extbf{A}ction-chunk extbf{S}election). It performs inference-time best-of-N selection to identify action chunks that are both semantically faithful and geometrically precise. Specifically, extbf{VGAS} employs a finetuned VLA as a high-recall proposal generator and introduces the extrm{Q-Chunk-Former}, a geometrically grounded Transformer critic to resolve fine-grained geometric ambiguities. In addition, we propose extit{Explicit Geometric Regularization} (exttt{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 extbf{VGAS} consistently improves success rates and robustness under limited demonstrations and distribution shifts. Our code is available at https://github.com/Jyugo-15/VGAS.

Source: arXiv cs.AI | 2026-05-25