UniVLA: Learning to Act Anywhere with Task-centric Latent Actions

具身智能