Development of a Basic Educational Kit for Robot Development Using Deep Neural Networks

In this paper, we propose a basic educational kit for robotic system development using deep neural networks (DNNs). To develop systems robust to changes in dynamic environments, much research is focusing on learning-based recognition and robotic manipulation systems. However, existing robotic techniques and DNNs are not systematically integrated and packages for beginners have not yet been developed. Therefore, we developed a robotic system using DNNs and a system manual to serve as a basic educational kit that can be easily used by anyone. Our goal was to educate beginners in both robotics and machine learning, especially when DNNs are used. Initially, we set the following requirements for the kit: (1) easy to understand; (2) experience-based; and (3) applicable in many areas. We analyzed the research and development of DNNs and divided the process into Date Collecting (DC), Machine Learning (ML), and Task Execution (TE) steps. Finally, we applied our hierarchical system architecture to a physical robotic grasping system. We implemented the DC and TE steps in the physical robotic system by using robotics middleware and in the ML step, we prepared a sample script.