Domain Adaptation for Learning from Demonstration

Learning from Demonstration (LfD) (also known as Imitation Learning) is a paradigm for enabling robots to autonomously perform new tasks by observing humans’ demonstrations. However, imitation learning may be challenging since the human demonstrator and the robot have inherent differences and may lead to systematic domain shifts. For example, there may be a mismatch between the robot's observations and actions and the recorded demonstration from humans, while the actions for human demonstrations are sometimes difficult to obtain. The inherent domain gap may lead to poor performance when using the pre-trained neural network model. To make good use of the human demonstration videos and to enable the robot to perform various tasks by bridging the gap of domain shift between human demonstrations and robot demonstration, transfer learning can be explored.