Research

Behavior Acquisition by Mobile Robot

We propose multi-layered reinforcement learning by which the hierarchical structure for behavior learning is self-organized. This method enables the behavior learning system to acquire several knowledges/policies, to assign sub-tasks to learning modules by itself, to self-organize its own hierarchical structure, and to simplify the whole system by using only one kind of learning mechanism in all learning modules.

Behavior Acquisition by Teaching

We propose a method for a robot, which has a limited view angle camera with panning facility, to make a decision by efficient observation without explicitly localizing itself. With a limited view camera, a robot can widen the angle by panning, but it takes time. The basic idea of our observation strategy is not for self localization but for decision making, that is, to minimize observation as long as decision making is possible.

Cognitive Developmental Robotics

We are engaged in realizing the mechanizm of imitation learning. Proposed method consits of view transformation mechanizm and adaptive visual servoing. View transformation is based on epipolar geometory which is caused by the similarity of body structure between learner and demonstrator. In this method, the learner can reproduce the demonstration without any explicit 3-D reconstruction.

We study how a robot can acquire an ability of joint attention which is one of communication mechanisms. The proposed learning model has two kinds of developments: a robot's development and a caregiver's one. The experimental results showed that the proposed model can accelerate the learning and improve the final task performance owing to the developments.