抄録
This paper presents feature extraction by deep leaning for in-hand manipulation. In the aging society, Robots which have versatile hands to manipulate different sizes and shapes of objects are required. In order to achieve robust in-hand manipulation, the current touch state has to be taken into account. But modeling of the contact states like multiple contacts and complex shapes of the object or the fingertip is difficult. So, machine learning was used for in-hand manipulation. However, information from touch states including redundant information is hard for machine learning to adapt to manipulation with different sizes and shapes of objects. We used feature extraction by deep learning to extract important information for in-hand manipulation.
