Implementation of Computer Vision in Low Fidelity Robots: Analysis, Challenges, and Design Implications

We observe the utilization of machine learning in a number of everyday aspects. Robotics is one of the major fields where we celebrate the prowess of machine learning. The application of machine learning and deep learning techniques to robotics has made them elegant, intelligent, and more competent. A subset of robotic gizmos consists of low-fidelity configurations that have abstract shapes and are generally constructed from low-end components. As a result, their capacities are oftentimes limited. Although several prior works sought to enforce machine learning for such contraptions, the lack of high-performance processing often bounds such applications. In this work, we fabricate a low-fidelity robot, named Tokai, implement deep learning-based object detection to perform waste bottle collection, and hence, investigate the feasibility, challenges, and design implications of implementing deep learning for low-fidelity robots.