How to efficiently finetune robot policies to learn new tasks on the fly? State of the art robotic manipulation policies are based on behaviour cloning of la…
机构:DeepMind
来源:arXiv 2608.19891 | AI4Papers 论文推荐平台