A Lightweight Target Following Architecture for Indoor Mobile Robot

Manufacturing efficiency and transport operations are being significantly improved by mobile robots. As the implementation of a configurable, lightweight, and stateof-the-art robotic system is required for current manufacturing sectors to maximize the use of mobile robots, this paper presents a software architecture comprised of perception and action systems for human following in real time. The perception system uses the YOLOv8 computer vision model to identify and estimate human pose. Additionally, an algorithm is developed using the 3D information from a stereo camera to determine which target and whether to follow. Once the perception system perceives the desired target information, the action system can control and coordinate the robot using the ROS. Experimental results from the physical robot demonstrate the feasibility of the system architecture.