Mutual Prediction Model for Predicting Information for Human Motion Generation

As a model for generating human motion, multiple paired forward and inverse models have been proposed. This model has motor control and motor learning functions by multiple forward/inverse model pairs using responsibility signal obtained from the generation of motion by the forward model. However, this system assumes that the target trajectory of the body is given. In order to learn through interaction with the outside world after the birth of a creature, the generation process of the target trajectory should be modeled. In this study, we propose a hypothesis in which the target trajectory is the sense prediction itself, and propose the mutual prediction model in which forward/inverse model pairs governing the sense and motion intention are mutually coupled as a human motion generation model. We also report some experimental results indicating appropriateness of the hypothesis.