研究生(外文):VU TRUNG HIEU
論文名稱:以PyTorch實現基於人工智慧之中華象棋機器人系統
論文名稱(外文):AI-Based Chinese Chess Robotic System by PyTorch
指導教授:王明賢
指導教授(外文):MING, SHYAN-WANG
學位類別:碩士
校院名稱:南臺科技大學
系所名稱:電機工程系
論文出版年:2021
語文別:英文
論文頁數:55
論文摘要
依據報告顯示，全世界有十億多人患有各種類型的疾病。在現實生活中，身障者會遇到很多的困難，身障者及有特殊需要的人必須提高大腦穩定度、增加自信心和專注力等，以解決生活中的所帶來的壓力和焦慮等情緒問題。而研究指出，參與相關心理治療的活動或有創造性的活動將減少身障者心理惡化的風險，身障者必須建立終身學習的觀念來增強活化腦神經網路的聯繫，並額外增強心理的功能。因此，本論文透過下棋提出了一種訓練大腦的方法。
本論文旨在先前的Matlab平台創建一個新的機器人手臂Python平台，以獲得象棋遊戲更高的處理速度和準確性來幫助失智人士或老年人。PyTorch是基於Python的科學計算套裝軟體，用於影像處理和識別中華象棋。棋子的位置將透過棋盤上面的相機識別，之後，數據將被傳送到機械手臂，只要將棋子置於相機和機械手臂的工作範圍，即使棋子是正反顛倒放置，也將受到控制並重新排列到其初始位置。相機照片被傳輸到PyTorch進行影像識別，卷積神經網路（CNN）將確認棋子的字。本論文具有與CNN相同的控制機器手臂的功能、正向和反向運動學用於操縱機械臂，還增加了Faster-RCNN的新技術，專門識別3個棋子：俥，傌，炮。在控制機械手臂的同時，由Matlab/Simulink執行即時模擬將模擬其所有移動。建立Matlab和Python之間的通信後，每個位置的數據將透過Python在螢幕上輸出。
關鍵詞：卷積神經網路、Faster-RCNN、PyTorch。
論文外文摘要
The World report on disability shows that over one billion people around the world are suffering from all types of disability. In fact, people with disabilities are most likely to have more difficulties in daily routines compared to individuals that don't have. People with special needs got to proactively raise brain safety parts and lift confidence, boost their focus, help combat affective disorder, stress and anxiety in their lives. Studies have pointed out that get involved in psychological or inventive tasks that promote mind perform will decrease the danger of making mental deterioration, with a relative threat reduction by half. People with impairments got to establish the behavior of life-long learning to strengthen reliable neural bonds between mind cells and additionally reinforce mind psychological functions. Therefore, this thesis recommends a solution of playing Chinese chess game to train their brains.
This thesis intended to create a new Python computational language platform for the robotic arm based on the previous study about Matlab language platform model, for helping impaired individuals with higher speed and accuracy features. PyTorch, a Python-based scientific computing package, is used for image processing and recognizing the Chinese chess. Positions of the chessmen will be recognized by the attached camera positioned above the chessboard. After that, the data will be transferred to the robotic arm, which will be controlled to rearrange the arbitrary-placed chessmen, even if they were placed up-side-down, to their initial positions, as long as the chessmen are placed in working range of the camera and the robotic arm. The camera photo is transmitted to Pytorch library for image recognition. The character of the chessman is recognized by convolutional neural networks (CNNs). This thesis additionally adds new technique of Faster-RCNN, for specially recognizing 3 chessmen: 俥，傌，and 炮, with same functions of controlling robotic arm as CNN model. The robotic arm is controlled by forward and inverse kinematics algorithm. While the robotic arm is being controlled, the real-time simulation run by Matlab/Simulink simulates all of its movements. The communication between Matlab and Python is established, and the data of each joint’s position will then be outputted on screen by Python.
Keywords: Convolutional neural network, Faster-RCNN, PyTorch.
