機器人與自閉兒之無感測器接觸式互動系統

Abstract

自閉症是一種複合行為與認知上的失調，發生率約為千分之二，在社交互動上主要面臨三種障礙：社會化互動、社會化溝通與想像力的欠缺。本研究結合串結式與人型機器人，設計接觸式之互動，並經實測證實可改善自閉症孩童互動之意願與情感表達。 本研究分析機器人內嵌馬達電流，以實現無感測器技術偵測受力事件，提出多重分類器之圖樣辨識架構，藉由肢體之獨立性，馬達運動特性與線性識別分析(linear discriminant analysis; LDA)，將16個馬達電流資訊映射至雙手雙足之4個2維特徵平面進行分類。以橢圓作為子分類器之邊界，由4個子分類器預測各平面之特徵向量所屬的受力事件，交由專家融合步驟進行分類器簡化與調整決策權重後提出最終預測。並與支援向量機(support vector machine; SVM)法比較辨識效果，本系統可將辨識率由75.9%改善至93.6%。 為比較非接觸式(遙控)與接觸式互動之差異，本論文實際應用兩種類型互動於自閉症孩童進行測試。經由量化與質化的評估，接觸式互動系統在對孩童互動意願與情感表達上，的確有較優異的表現。

Autism is a complex symptom in which behavior and cognitive disorder cause the issue cause and 0.2% children present such symptom. Three problems are observed mainly for autistic children as: social interaction, communication, and imagination. In this thesis, the humanoid robot and the chain-type robot is combined to develop the human-robot interactive education system with force contact. This system has been proven to effectively improve interaction and emotional expression for autistic children. In this Thesis, a sensorless technique for the robot to detect the external force is developed by analyzing the current of motors in the robot. The structure of pattern recognition system with multi-classifiers is proposed and the motor current of sixteen motors will be mapped to four two-dimensional planes by the independence of two limbs and two arms through analysis of the kinetic motion and linear discriminant analysis (LDA). Four sub-classifiers are also adopted to predict forced events by analyzing feature vector on each plane with an ellipse boundary. The combination of four sub-classifiers is obtained by applying expert analysis for simplifying the forced event and adjusting the decision weight. Compared the proposed method with the support vector machine (SVM), experimental results indicate that the recognition rate can be improved from 75.9% to 93.6%. Furthermore, both non-contact (remote) and contact interaction approaches have been applied to an autistic child with both quantitative and qualitative analyses. Results indicate that the proposed contact interaction system renders much better improvement in interaction and emotional expression for the child.