具自主學習與修正之中文書法機器手臂的設計與實現

Abstract

本論文的主要研究目的為利用Delta Robot並聯式機器手臂簡單的機械結構和可靠的定位座標來實作需要精確參數才能寫字的中文書法寫字系統。設計好的學習模型也需要一個穩定的實作系統來確認，經過驗證假設生成網路(HGN)學習模型的可靠性，可讓我們的Delta Robot並聯式機器手臂獲得具有自主學習書法的能力。從人類的筆畫輸入經由一連串影像轉換成動作的處理，如提取筆畫骨架、筆畫拆解、筆畫座標排序、筆畫辨識、座標轉換再到HGN學習系統回傳動作命令，再經由逆向運動學得到馬達輸出轉角，最後由機器手臂完成毛筆五個自由度的寫字動作。寫字的結果再透過網路攝影機回傳給系統做誤差計算與座標修正，經過幾次的循環與修正，最後可以得到趨近目標樣本的輸出結果。透過Delta Robot並聯式機器手臂的實作結果可以證明，此中文書法寫字系統具有自主學習與修正的能力。

The main goal of the thesis is to use a simple mechanical structure and reliable coordinates to develop a Chinese calligraphy-writing system that requires precise control. A well-designed learning model also needs to be verified by a real robotic system. Using the reliability of the hypothesis generation net (HGN) learning model makes the delta robot with the ability to learn Chinese calligraphy writing autonomously. After input samples of strokes by human writing through a webcam, the image-to-action processes include extracting stroke skeletons, disassembling strokes, sorting skeleton coordinates, recognizing strokes, converting coordinates, receiving instructions from the HGN, and getting the angle instructions of motors by inverse kinematics. Then the delta robot completes the writing with 5 degrees of freedom. The other webcam provides the writing results of the delta robot to the writing system for error calculation and coordinate correction. After several cycles and corrections, the writing results could approach the target sample finally. Through the implementation of the Chinese calligraphy-writing system, it can be proved this writing system with the ability of autonomous learning and correction.