Teaching grasping to a humanoid hand as a generalization of human grasping data

Humanoid robotics requires new programming tools. Programming by demonstration is good for simple movements, but so far the adaptation for fine movements in grasping is too difficult for it. Grasping of natural objects with a natural hand is known as one of the most difficult problems in robotics. Mathematical models have been developed only for simple hands or for simple objects. In our research we try to use data directly obtained from a human teacher as in imitation learning. To get data from users we built a data glove, we collected data from different experiments, and generalized them through neural networks. Here we discuss the nature of the data collected and their analysis.