1 paper · 1 filter
Yang Tian, Yuyin Yang, Yiman Xie +13
Recent works explore how real and synthetic data contribute to Vision-Language-Action (VLA) models' generalization. While current VLA models have shown the strong effectiveness of…