How Google’s Robotics ER 1.6 model is trained. Does anyone know what kind of datasets it uses (real-world robot interactions, simulations, human demonstrations, etc.) and how the data is collected or structured? Any insights or resources