Is anyone feeding models touch data? It seems the main thing we want the robots to do is touch things, but we are just feeding them audio/video/images. The model has to learn how to touch things despite never having touched anything before. Perhaps that's why they all look so hesitant when they touch things?