로봇 손동작 재현을 위한 학습기반 시스템의 데이터 민감성

Recently, it is not rare that humanoid movement task in programming by demonstration (PbD) makes use of learning algorithms to make robot-learning faster, easy-to-teach, and versatile in a complex environment. However, recent research on PbD make little attention to the possibility of the deterioration of proposed learning-based systems due to the unwise selection of training/testing dataset within the given task. In this paper, we empirically show that the performance of learning systems for PbD hand movement can vary due to the choice of training and testing set configuration.