Abstract
In general, the most difficult aspect of realization of cooperative behaviors between robots, as well as between a robot and human, stands on how one acquires another dynamical, mechanical, and control model. For learning the other models, it is necessary to dissociate whether the resulting motion are by self or other operations, that is to say, self-other distinction in learning. To discuss the mechanisms of self-other distinction, we consider the cooperative manipulation task by two arm-robots, in which robots are physically interact with each others, and are equipped for independent controllers. In the simulation the feedback-error learning is used for acquisition of required motions. In this study we show that the minimization of various parameters concerned with the controlling arms, along with typical learning, can lead the feedforward controller which is nearly independent of the other motions.
