Robot arm training by image processing and reinforcement learning

In this thesis, I describe how to train a robotic arm for tasks such as grasping an object. Firstly, I propose a system that enables a robotic arm to grasp a ball based on its camera image. My algorithm uses OpenCV methods and YOLO to detect the ball in the image and then calculates its position in a fixed global coordinate system of the robotic arm. This information is used to direct the movement of the robotic arm and enable it to grasp the ball. My results show that my system is able to accurately detect and grasp the ball with high success rate. Overall, my system demonstrates the potential of using computer vision techniques to enable robots to perform complex tasks such as object grasping in real-world environments