Copycat Hand - Robot Hand Generating Imitative Behaviour at High Speed and with High Accuracy

Abstract

In recent years, robots demonstrating an aping or imitating function have been proposed. Such functions can estimate the actions of a human being by using a non-contact method and reproduce the same actions. However, very few systems can imitate the behaviour of the hand or the fingers On the other hand, reports in Neuroscience state that mirror neurons Moreover, since human beings perform intelligent actions using their hands, it is important to attempt to artificially imitate the actions of hands and fingers in order to understand the dexterity and intelligence of human hands from an engineering viewpoint. The object actions in the case where one "looks at an action performed by others and imitates it" include not only grasping or manipulating objects but also the actions involving "imitating shapes and postures of hands and fingers" of others such as sign language and dancing motions. The latter two types of actions are often more complicated and require dexterity as compared to the former actions. In the action of imitating "the shape of hands and fingers" such as sign language, it is essential to estimate the shape of the hands and fingers. Furthermore, as compared to the imitation of the actions of the lower limbs, estimating the shape of the hands is significantly more important and difficult. In particular, human hands, which have a multi-joint structure, change their shapes in a complicated manner and often perform actions with self-occlusion in which the portion of one's own body renders other portions invisible. In the case of an artificial system, we can utilize a multi-camera system that records a human hand for imitative behaviours by surrounding it with a large number of cameras. However, all the animals that mimic the motions of others have only two eyes. To realize the reproduction of the actions of hands and fingers by imitating their behaviour, it is desirable to adopt a single-eye or double-eye system construction. To roughly classify conventional hand posture estimation systems, the following two types of approaches can be used. The first approach is a 3D-model-based approach (Rehg &