申請者らは,社会性を備えたロボット,すなわち,ロボット同士,機械,情報サービス,そして人と能動的に繋がり,協働さらには共発達が可能なロボットをソーシャブルロボットと呼ぶ.本研究課題では,ソーシャブルロボットの論点,特に対人対話戦略における高精度情動認知機能と間合いの制御,共発達する個性の重要性を,上記3つの視点から議論し,それぞれに対して実装モデルを提案する.併せて,提案手法を協力デイケア施設における介護コミュニケーション業務支援に適用し,提案手法の妥当性と有用性を実証的に検証し,協働・共進化型ロボットの構成法を確立する.研究計画2年次となる2020年度は,昨年度の成果を踏まえ,次の2つの課題を並行して実施した.
【課題 1】話者間の社会的関係と対話へのエンゲージメント、メタ対話を考慮した対話戦略の実装と評価
本課題では,対話タスクを中断・再開可能とする対話コンテクストのデータ構造を新たに定義し,対話の構造情報(隣接ペアとその連鎖情報)や対話時間,対話内容の重要度,さらには話者同士の静的な社会的関係を考慮した対話制御フレームワークを提案した.さらに,顔に表出する微表情やしぐさとして表れる身体的微表情,フィラーといった非言語的振る舞い情報,メタ対話情報,話者間の動的な社会的関係性情報を統合的に利用した統合的対話戦略立案手法を議論し,人間と同等の自然な対話制御の予備的実装を行った.
【課題 2】内的情動と対人情動の遷移傾向に基づくパーソナリティ形成モデルの構築と影響調査
本課題では,ヒューマンロボットインタラクションを通した共発達のひとつの形として,対人情動認知能力の獲得とパーソナリティ形成のダイナミクスに着目し,長期的なインタラクションの中でロボットの性格を形成する手法を議論した.具体的には,ロボットの情動を対人情動と内的情動の2つによって定義し,それぞれの情動の遷移傾向(情動傾向)をロボットに対するユーザ行動の履歴によって調整することで,緩やかな性格形成を行うアルゴリズムを提案した.
We call social robots that can actively connect, collaborate and co-develop with machines, information services, and human, as "Social Robots". In this project, the importance of high-precision emotion recognition, spatio-temporal interval control, and co-developed personality in interpersonal interaction strategies is discussed from the above three perspectives, and an implementation model is proposed for each. We will apply the proposed method to support nursing care communication work in a cooperative day care facility, verify the validity and usefulness of the proposed method empirically, and establish a method to construct collaborative and co-evolving robots. In FY 2020, the second year of the research project, we conducted the following two tasks in parallel, based on the results of the previous year.
Research Theme 1: Implementation and evaluation of a conversation strategy that considers social relationships among participants, engagement in dialogue, and meta-dialogue
In this task, we defined a new data structure for dialogue contexts that can suspend and resume dialogue tasks, and proposed a dialogue control framework that takes into account dialogue structure information (neighboring pairs and their chain information), dialogue duration, importance of dialogue content, and even static social relations between participants. In addition, we discussed an integrated dialogue strategy planning method that uses non-verbal behavioral information such as facial expressions, gestures, fillers, meta-dialogue information, and dynamic social relationship information, and made a preliminary implementation of natural dialogue control equivalent to that of humans.
Research Theme 2: Construction of a personality development model based on the transition tendency between internal emotions and interpersonal emotions
In this work, we focused on the dynamics between the acquisition of interpersonal emotional cognition and personality formation as a form of co-development through human-robot interaction, and discussed a method to develop a robot's personality during long-term interaction. Specifically, we defined the robot's emotions as interpersonal emotions and internal emotions, and proposed an algorithm for gradual personality development by adjusting the transition tendency (affective tendency) of each emotion according to the history of user behavior toward the robot.
|