Demonstrations

This paper describes a trial of robot learning allowing the evaluation policies based on embodiment in the teach- ing environment with multiple instructors. We have devel- oped a novel method which allows a robot to learn from di- rect demonstrations performed by multiple instructors. This method based on DP matching and clustering. It classi- fies the teachings from multiple instructors, integrates simi- lar teachings into several groups and selects the most suit- able policy of teaching. This paper prepares a group of pet robots AIBO. They have three different types of embodiment. For each embodiment, AIBO require a distinctive teaching policy.