請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/18844
標題: | 移動機器人之認知結構:人類行為與意圖的研究 Cognitive Architecture for Mobile Robots: Studies of Human Behavior and Intention |
作者: | Wei-Zhi Lin 林威志 |
指導教授: | 黃漢邦(Han-Pang Huang) |
關鍵字: | 認知架構,人類行為認識,多模態情緒模型,意圖模型,運動規劃,深度增強學習, Cognitive Architecture,Human Behavior Understanding,Multimodal Emotion Model,Intention Model,Motion Planning,Deep Reinforcement Learning, |
出版年 : | 2020 |
學位: | 博士 |
摘要: | 隨著科技的進步,社交機器人開始逐步進入人們的生活。目前許多研究團隊致力於讓機器人適應環境、了解人類做事方式並且與人們合作做事。為了要讓機器人能夠更進一步讓人們接受,機器人需要具備理解人類行為的能力。 本論文目的旨在發展一套機器人具備人類行為認識與意圖認知的認知架構,並使用上述資訊讓機器人自主行動。在人類行為中,我們提出了多模態情緒辨識系統、人類行為認知地圖。情緒辨識系統中,我們結合臉部與身體資訊並將其結合給予更加穩定的辨識結果,以此理解人類的情緒。在人類行為認知地圖中,記錄且辨識人類在特定地點常有的行為,紀錄空間與人類行為的關係。藉著於意圖偵測中觀察人類注視方向與距離、速度等觀測值來建立對話時候的意圖模型。最後在機器人的行動規劃中,使用深度增強學習演算法學習空間影響與人類行為認知地圖的資訊,並給予機器人自主行走的能力。 最後本論文展示結合上述各功能的認知架構之演算法。使得機器人能夠展現出預測行人意圖並主動給予協助的能力,同時機器人也能表現出符合社會規範的行為。 In recent years, social robotics have developed and entered our lives. In fact, social robotics are acting more and more like humans. Many researchers are focused on developing different behaviors which can let robots adapt to various environments, better understand human decisions, and further collaborate with people. To reach this goal and have the robot be more easily accepted as part of human society, robots must have the ability to understand human behavior. This dissertation attempts to develop a cognitive architecture which combines the understanding of human behaviors and intentions. In the human behavior understanding module, a multimodal emotion recognition system and a human behavior cognitive map are proposed. The multimodal emotion system is built by fusing facial and bodily information to achieve a more stable recognition result. In the human behavior cognitive map, many human actions are recognized and record relationships between space and human actions in a specific environment. Furthermore, the intention model during conversation is established by observing the human gaze direction, distance between human and robot, and walking velocity toward the robot. The motion planning method is built by deep reinforcement understanding, which learns the spatial effective and human behavior cognitive map, giving the robot autonomous walking ability. Finally, the experiment demonstrates the cognitive architecture created by combining the above functions. As a result, the robot exhibits the ability to understand human intentions and offers to actively help with people’s needs, and the robot can also behave in accordance with social norms. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/18844 |
DOI: | 10.6342/NTU202003967 |
全文授權: | 未授權 |
顯示於系所單位: | 機械工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-1808202014115000.pdf 目前未授權公開取用 | 5.23 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。