人型機器人情緒動作設計與即時性音樂情緒展演整合系統

Abstract

本研究建立一套機器人情緒表達展演系統，由即時性音樂情緒辨識分類演算結果，並將此即時音樂情緒量化分級，利用人形機器人演示音樂情緒的動作表達。人形機器人KHR-3HV 由日本KONDO公司所製造，其機構由雙腳、雙臂和頭部所構成，分別擁有十個足部自由度、六個手部自由度和一個頭部自由度。人型機器人程式(HEARTTOHEART)編輯也以圖形化介面為主，其情緒動作的設計以拉邦動作分析(Laban Movement Analysis, LMA)推導，建構出可量化之動作描述系統。即時性音樂情緒系統引用Thayer提出的情緒模型，由200首已標記情緒的音樂片段之WAVE檔中萃取五種特徵包括音量、音樂事件密集度、音色、調性與音程的和諧度，經過情緒計分以界定在情緒平面上的分布，主要分為”平和”；”興奮”；”焦慮”；”消沉”四種情緒，彼此間的邊界以高斯混和模型(GMM)來標定。本研究將人型機器人情緒動作展演描述系統與音樂播放時之即時情緒軌跡整合，以實況動作表達當下音樂所呈現之情緒。 本研究為了能驗證機器人情緒動作展演描述系統與觀眾感受到情緒的相似程度，安排53名受測者進行共十二組分別由數據分析探討以拉邦勁力圖為基底設計之情緒動作(”平和”；”興奮”；”焦慮”；”消沉”)的問卷調查，作為情緒動作展演的客觀結果驗證。調查結果顯示，其動作的設計可以有效地使人感受到當下情緒，由機器人作情緒動作展示可延伸應用在3D動畫、居家照護機器人、電腦遊戲人物動作設計、人機互動軟體。

This study proposes an integrated system to demonstrate the emotional movements of humanoid robot inspired by the real-time music emotions. The music emotions tracking system progressively extracts the features of music and characterizes music-induced emotions in an emotion plane to trace the real-time emotion locus of music. Thayer’s model of mood is adopted as the emotion plane of classification and detection. The model is consisted of four quadrants: (i)Contentment, (ii)Depression, (iii)Anxious, (iv)Exuberance. Each emotion is quantized to three levels to express the degree of mood. A humanoid robot, Kondo KHR-3HV, is used as a base robot which includes 17 servomotors that allow for 17 Degree-Of-Freedom (DOF) movement. The motions designs for emotional expression are based on Laban Movement Analysis (LMA) to construct a quantifiable action description system. The system is capable of describing and interpreting nearly all varieties of human movements. Furthermore, a questionnaire survey is conducted to evaluate the results of proposed emotions rendering system judged by the participants’ experience. For the comparative emotion model checklist, 53 participants had to rate their felt emotional reaction to emotions movement on all three motions levels checklists. Preliminary evaluations indicate that the proposed algorithms produce results agreed approximately.