Path Following of Mobile Robots: Teach and Repeat-based and Learning-based Methods

Navigation of service robots is a relevant research field in mobile robotics. This field opens up the study of robots following paths of autonomous way. This article analyzes path-following techniques. Firstly, a teach and repeat (T\&R) method is proposed, utilizing the representation of the taught paths via Bézier curves. For each trajectory, three experiments were conducted using a real differential mobile robot. More specifically, the user must generate possible paths by teleoperation for the system to be capable of autonomously repeating the trajectories. Experimental results show that the metric of maximum average percentage error (MPE) for the T\&R method reached 1.361% during the teaching process. In the repetition step, the worst MPE was 9.983%. Furthermore, experiments utilizing the Reinforcement Learning (RL) approach for path following enable the robot to learn navigation behaviors directly from experience, achieving 1.821% of MPE. Thus, dual approaches allow an evaluation of geometric-based and learning-based methods, highlighting their strengths and limitations in terms of accuracy and adaptability.