A study on efficient adaptation to inexperienced situations by robots with different action control individualities
Yu Yoshida, Yutaro Mikami, Masakazu Suzuki · 2017
In the near future robots will be required to plan and execute complex actions autonomously, and to effectively adapt to situation changes such as obstacle positions and state of the action field. In addition, it is quite difficult for the robots to adequately determine the control parameters for the initial action trial with little action knowledge or without any action knowledge. The easiness criterion was introduced to make it easy, and it was shown that action control individuality is generated by initial teaching based on the easiness criterion[3]. Different individualities lead to different manners of task execution even though they seek the same action result. In this work proposed is a method for efficient adaptation to new situations, especially effective for early stage adaptation when the control knowledge in the new situation is scarce, by utilizing the empirical knowledge of other robots with different action control individualities. The proposed method is applied to hockey shoot actions to study the effectivity.