Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 05, 2019 - June 08, 2019
There are various kinds of information in human-human interaction, but it is difficult to process all of them. This research aims at understanding mechanism of information selection and proposes a computational model based on prediction error minimization (PEM) mechanism. The PEM mechanism is a hypothesis that explains brain function as a minimization process of prediction error. Although it has been demonstrated that action generation by the PEM mechanism is important in interactions, the situation dealing with multiple information has not been considered. This research focused on uncertainty of information and used a recurrent neural network model that learns to predict the next state of information and uncertainty. In the experiment, ball-rolling interaction between a robot with the model and that operated by an experimenter was conducted. The robot switched its action in accordance with only the information with small variance. This result suggests that uncertainty is important for information selection.
