以模糊推論做為社交機器人情感行為產生機制之研究

A Study of the Emotional Behavior Generated by Fuzzy Inference for a Social Robot

蔡伯威(Bo-Wei Cai)

指導教授 ： 季永炤

國立虎尾科技大學/工程學院/自動化工程研究所/碩士(2014年)

https://doi.org/10.6827/NFU.2014.00222

摘要

本研究之目的為探討以模糊推論做為對一具有情感表達特性之機器人其情感模型之演算機制，並依此建立一能與環境互動並而能產生成適當情感行為之社交機器人。此機器人具備有差動式雙輪運動機構，其頭部裝設有小型液晶顯示器可以以動畫模擬產生人物臉部之表情，其可運行於一般桌面上與使用者及周邊環境互動。機體所裝設之感測器，包含小型攝相機、紅外線距離感測裝置、與加速度計等，用於偵測外界環境變化與使用者之特定動作。
本文之情感機器人其情感行為推論系統由感知層、心理層、與行為層三部分所構成，機器人對於外界之輸入資訊先進行感知層處理以獲得資訊之特徵，外界資訊則係藉由機器人所裝備之各項感測器來獲取周遭環境參數資訊與具特定性之輸入，後再藉由影像處理、感測融合等處理程序取得感環境感知特徵，特徵將用於各式情感行為之輸出並會影響心理層中依二階動態情感方程式所建立之內部情感狀態。心理層由二階動態情感方程式與情感行為推論機制結合而成，其透過模糊邏輯推論總和合成二者而合成產生情緒行為，同時以模糊演算法將情感動態之輸出平順化，使機器人能夠生成多樣且細膩之情感行為，而使其情感之表達更為接近一般人物之特性。行為之生成與輸出則是在行為層以行為式架構之方式依感知特徵與情感狀態由特定之協調機制對行為輸出進行仲裁而決定，此機制在於整合機器人各個行為輸出之動作反應，以實現合理之行為輸出。
本研究最後以實體機器人進行整體性能之測試，驗證作為一社交機器人其情感互動能力。其中軟體平台部分則是以LabVIEW®進行感測處理與演算推論其情感模型結果，並作為即時監控與狀態回饋之用。

關鍵字

情感模型 ； 模糊推論 ； 行為控制 ； 社交機器人 ； 感測融合

並列摘要

This paper presents a fuzzy based computational algorithm used in the development of an artificial emotion model for the social robot that can deduce the appropriate emotional behavior by interactions with its environment.
The robot is a two-wheel differential drive mobile robot that can interaction with the user and the surrounding environment on the desktop and generate emotional behavior. It has a color liquid crystal display on the head for the facial animation, and equipped with webcam, infrared distance sensors, and accelerometer for detecting changes of the external environment and user’s actions. The artificial emotion inference system of the emotional robot is composed of three functional layers; sense/perception layer, mental layer, and behavior layer. In sense/perception layer, the robot receives inputs from the environment by various types of sensors. The received data will be further processed to get features information. The ambient features are gathered through the fusion of the IR sensor and image processing. Together with other features, they will be used for the emotion inference. Some feature can also generate the direct response of the system. In mental layer, the artificial emotion inference system of the emotional robot generates emotional behavior by fuzzy logic inference which combines the results of a second-order state equation and the output of the OCC based emotional behavior inference mechanism. The robot can generate supple, diverse and delicate emotional behavior by fuzzy algorithm. In behavior layer, the Subsumption structure is used in the behavior generating and producing proper output actions of the robot system. This mechanism integrates action parameters of behaviors to achieve a reasonable reaction of the social robot.
The established emotional behavior inference system is tested with a designed mobile robot system which uses the LabVIEW® programmed data fusion and emotion inference engine. The behavior actions, however, are carried out by the single chip processer on the robot itself. The results show the effectiveness of the proposed inference algorithm.

並列關鍵字

Social Robot ； Emotion model ； Fuzzy inference ； Behavior-based ； Sensing fusion

參考文獻

[1] 戴熒美， "2021年全球機器人市場預測"， 機械工程， vol. 365， pp. 28-37， 2013。.

[2] 林其禹等， 智慧型機器人-原理與應用， 高立， 2013。

[3] E. Ackerman. Japan Earthquake: iRobot Sending Packbots and Warriors to Fukushima Dai-1 Nuclear Plant. 2013, 11/6. http://spectrum.ieee.org/automaton/robotics/industrial-robots/irobot-sending-packbots-and-warriors-to-fukushima

[4] J. Kastrenakes. Google fine tunes its self-driving car for city streets. 2014, 6/2.

http://www.theverge.com/2014/4/28/5660776/google-self-driving-car-city-testing-improvements