LATIDIA · Robótica

El análisis causal de los modelos de cimentación de robots permite ganancias de eficiencia

arXiv:2610.20750v1 Anuncio Tipo: nuevo Resumen: Los modelos de visión-lenguaje-acción (VLA) son un enfoque prometedor para la manipulación robótica de propósito general. Si bien el cálculo interno de los modelos de visión-lenguaje, que componen

La noticia

arXiv:2610.20750v1 Announce Type: new Abstract: Vision-language-action (VLA) models are a promising approach to general purpose robot manipulation. While the internal computation of vision-language models, which compose the backbone of VLAs, has been studied extensively, less is known about how their perceptual and semantic representations are transformed into action in VLA models. We study this question mechanistically in three frontier VLA models ($\pi_{0.5}$, MolmoAct2, and MolmoBot) by analyzing the causal pathways through which the input image and the instruction shape actions. Across the three models, activation patching shows that causally relevant information influences the action expert (AE) through a narrow band of mid-to-late cross-attention layers, which we term the action-conditioning interface.