2A1-E19 注意に基づく物理的因果性の獲得に向けた状況依存型予測器の学習

概要

論文の詳細を見る
Physical causality, like object permanence, is one of the most important concepts that robots working in our daily lives should have. The fundamental faculty enabling physical causality is prediction of objects' movements. If an object or the environment changes, the prediction should cope with this change. We propose a situation-dependent predictor which can change prediction depending on the object's shape and the environment. In order for a robot to obtain the concepts by itself, we propose an attention model with which a robot decides what to attend and when to serch a new object. This paper presents this situation-dependent predictor and shows that the system can predict object's flow depending on situation, and consequently robots acquires the physical causality with the attention model.
一般社団法人日本機械学会の論文
2009-05-25