Embodied Interpretability: Linking Causal Understanding to Generalization in Vision-Language-Action Models
具身可解释性:将因果理解与视觉-语言-动作模型的泛化联系起来
机构 * University of Science and Technology of China(中国科学技术大学)
AI总结 提出干预显著性评分(ISS)和干扰质量比(NMR),通过干预掩码估计视觉区域对动作预测的因果影响,并量化对任务无关特征的归因,实验表明NMR可预测泛化行为,ISS比现有方法提供更忠实的解释。
Comments Accepted at the 43rd International Conference on Machine Learning (ICML 2026)