Google has demonstrated its first vision-language-action (VLA) model for robot control that showed improved generalisation capabilities and semantic (the understanding of words and sentences), and visual understanding beyond the robotic data it was exposed to.