階層型モジュラー強化学習による動的環境に適応した学習手法を用いる児童見守りアプリケーションの提案

摘要

Recently, school-age children may meet to dangerous situation such as a kidnapping and a　natural disaster. In order to keep safety children from these accidents, their parents and the surrounding　social community should always monitor or shield from danger or　harm. These countermeasure against　such accidents is required in the current social community, because tight protection becomes hindrance of　child's growth and avoid nurturing his/her talents. School-age children have to judge their own situation inhuman social community. They can　learn‘bushcraft' through experiences in any natural environment. This　paper proposes an Android application which provides kids protect system such as‘ kids keitai'.　Children's　behaviors in playground are defined as the pursuit problem by multi-agent environment. Hierarchical　modular reinforcement learning prorosed by Watanabe, consists of 2 layered learning where Profit-Sharing　works to plan a target position in higher layer and Q-learning trains the state-action pair to the target　in lower layer. In this paper, we developed Android application which the agent can notify that danger　situation close in on the child by the acquired knowledge from the learning result.