Single Board Computer ROS-Based Tennis Balls Collecting Mobile Robot

Abstract

With the recent increases of mobile robot deployment that rely on robot operating system (ROS), new challenges have emerged as a result of the hardware requirements imposed by ROS on the host computer. Installing ROS on a mobile robot requires the target robot to be equipped with a full personal computer (PC) with specific specifications. However, deploying ROS on such PC’s will introduce new issues such as increased size, weight, cost, and power consumption. This research presents the development and implementation of a fully integrated standalone tennis balls collecting mobile robot using ROS. The operating system is deployed on a compact, low-cost, low power consumption, light weight and embedded single board computer (Raspberry Pi 4). The robot goal is to assist playground attendees by collecting scattered tennis balls. This is accomplished by integrating and implementing a miniature series of algorithms that construct the robot tasks. These algorithms are used to detect objects, classify them, plan optimal paths, and avoid obstacles. During the implementation process, a significant challenge arose in the form of a high computational load on the main processing unit (CPU). The vision detection algorithm is to blame for this. This was resolved by using a lighter version of the algorithm, which reduced the computational load. The proposed method was investigated in this work. The results show that a single board computer (Raspberry Pi 4) can complete the required objectives and run the algorithms within acceptable constraints. The vision algorithms performed as expected, detecting all of the objects in the robot workspace. However, the Raspberry Pi requires a longer execution time than a standard PC to perform vision tasks. The extra time is due to the Raspberry Pi's hardware resource limitations, as well as the limitation on utilizing hardware acceleration abilities. Keep in mind that hardware acceleration employs the graphical processing unit to address vision algorithms in order to shorten execution time. Furthermore, the A* algorithm was used to help the robot find the shortest obstacle-free path. Other algorithms are in charge of formulating the wheel's trajectory and control law. All of the robot algorithms were coded to use fewer computational resources, resulting in less extra execution time. As a result, the robot is able to complete the tasks in a reasonable amount of time. Finally, the proposed low-cost solution was shown to be capable of running ROS-based mobile robot algorithms.