機器學習型影像辨識引導機械手臂

本專題提出利用機器學習技術進行影像辨識，並以此作引導機器手臂的依據。在硬體方面，我們以3D列印來自製機械手臂模型，搭配控制模組以驅動各軸步進馬達，並使用CCD進行影像擷取，在軟體方面則進行影像分析與處理，以類神經網路進行特定圖案的學習，使機械手臂記住每一張圖樣。執行時機械手臂能透過CCD影像分析與處理，找出目標圖樣在影像中的位置，並以類神經網路辨識該圖樣，使用者能夠指定要抓取哪一個圖樣物件，控制手臂到達指定位置並進行擷取。本專題目前以麻將為辨識標的，場景設定為老人陪伴、復健治療等智慧型機械手臂的應用。 This project employs machine learning techniques for image pattern recognition that is used for the guidance of robotic arm manipulation. For this purpose, a 6-DoF robotic arm is built via 3D printing. The robot is equipped with stepper motors, CCD cameras, and pressure sensors. Target images are captured and analyzed by Artificial Neural Network. The position of the target to be grabbed is computed and feed into a PID controller. Currently, this project is applied for playing Mahjong for companion of the elderly. The robot senses a specific Mahjong tile according to the ANN models and drives the PID controller to manipulate the robotic arm and to play the game.