https://scholars.lib.ntu.edu.tw/handle/123456789/119551
Title: | 從人機互動中觀察人員的注意力反應以達成機器人提供服務行為之調適 Adapting Robot Behaviors for Providing Services through Observing Human's Attention Responses from Human-Robot Interactions |
Authors: | 蔣亦修 Chiang, Yi-Shiu |
Keywords: | 機器人行為調適;互動學習;動態貝氏網路;人機互動;日常生活輔助;Robot Behavior Adaptation;Interactively Learning;Dynamic Bayesian Network;Human-robot Interaction;Robots in Daily Life | Issue Date: | 2015 | Abstract: | 當機器人服務於家庭、安養中心等私人環境下,如何讓機器人考慮個人之喜好以維持人與機器人間的社交規範將成為一個重要的議題。在人與人的互動之中,人們總能不自覺地學習每個人潛在的社交規則;相對的,機器人卻難以在互動的過程中察言觀色,以致於打擾到當下人們的活動,進而造成不愉快的互動體驗。為了賦予機器人如此的社交能力,在本篇碩士論文提出了一個人員感知之互動學習架構,以此來建構機器人了解自身行為對於使用者的干擾程度,並在推論使用者社交注意力之同時最佳化自身的服務提供行為。以此為目標,我們提出了一個人員感知之馬可夫決策過程 (Human-Aware Markov Decision Process) 來描述此類需同時規劃機器人之行為並推論使用者之社交注意力之問題。對於社交注意力模型,我們採用了動態貝氏網路 (Dynamic Bayesian Network) 來推論使用者察覺機器人存在之機率。此外,使用者之社交注意力,機器人自身對於使用者察覺之揣測,以及機器人行為動作此三項相互間的關聯性則由增強式學習法 (Reinforcement Learning) 來進行探索發掘。同時地,為了達成更自然的機器人互動,使用者當下之心情回饋則由以身體姿態為基礎之情感辨識來粗略地估計出來。本篇論文之最後則進行了數個以實際社交情境為基礎之實驗,以驗證我們所提出之人員感知互動學習架構。 Robots that service humans in private places, such as homes or senior centers, must consider humans'' preferences to behave in a socially acceptable manner. Human beings subconsciously adapt their actions to start a conversation according to the historical interaction experiences, but robots often fail to do this and result in disrupting their users. To endow service robots with such socially acceptable ability, this thesis proposes an online human-aware interactive learning framework, under which the robot behaves so as to optimize its service providing behavior while inferring user''s awareness of robot itself. To this purpose, a human-aware Markov decision process (HAMDP) is proposed to model this kind of problem, which requires planning of robot actions and inference on user''s social attention concurrently. For social attention inference model, it is based on a Dynamic Bayesian Network (DBN), which is also employed to infer the possibility of user''s awareness of the robot, extit{i.e.} the robot''s theory of awareness. The correlation between the robot''s theory of awareness, the user''s social attention, and the robot behavior are explored through reinforcement learning. Besides, to let the robot behave more naturally, the mood of the user is estimated by recognizing his/her body gesture based gross affective state. In order to verify the effectiveness of our proposed framework, experiments with real social scenarios have been conducted. |
URI: | http://ntur.lib.ntu.edu.tw//handle/246246/275620 | Rights: | 論文公開時間: 2020/8/11 論文使用權限: 同意有償授權(權利金給回饋學校) |
Appears in Collections: | 資訊工程學系 |
File | Description | Size | Format | |
---|---|---|---|---|
ntu-104-R01922047-1.pdf | 23.32 kB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.