We study the question of how to imitate tasks across domains with
discrepancies such as embodiment, viewpoint, and dynamics mismatch. Many prior
works require paired, aligned demonstrations and an additional RL step that
requires environment interactions. However, paired, aligned demonstrations are
seldom obtainable and RL procedures are expensive. We formal
使用 Morphological Adaptation in Imitation Learning (MAIL) 框架,从 3D 带障碍物情况下,带有两个末端执行器的机器人的演示中训练出一个末端执行器的 Franka Panda 机器人的可视化控制策略,比 Learning from Demonstrations 和非 Learning from Demonstrations 基线方法提高了 27% 的成功率,并且在面对不同颜色、厚度、大小和材料等多变性的衣物的姿态(旋转和平移)时展现出很好的通用性。