Sim-and-Human Co-training for Data-Efficient and Scene-Generalizable Bimanual Manipulation
SimHum is presented, a co-training recipe that extracts kinematic priors from simulation and visual priors from human observations, then fine-tunes on a small real-robot dataset, which exhibits strong scene-generalizable and data-efficient capabilities.
Kaipeng Fang, Wei Liang, Yu-Yang Li et al.
· 0 citations