We capture diverse, multimodal human data at scale so robots can learn how to interact with the physical world.