robot

RDT-1B is an imitation learning diffusion transformer with a parameter size of 1B (currently the largest), pre-trained on more than 1M (currently the largest) multi-robot plot. Given verbal instructions and up to three views of RGB images, RDT can predict the next 64 robot actions. RDT is compatible with nearly all modern mobile manipulators, including single-arm to dual-arm, joint to end-effector, position to velocity, and even wheeled motion. The model was fine-tuned on 6K+ (one of the largest) self-collected two-hand plots and deployed on the ALOHA dual-arm robot. It achieves state-of-the-art performance in terms of dexterity, zero-shot generalization capabilities, and few-shot learning.