Neural network trajectory tracking of tracked mobile robot

• Main Authors: Asai, Madoka, Chen, Gan, Takami, Isao
• Format: Conference Proceeding
• Language: English
• Subjects: Cost function; Mobile robots; neural network; Neural networks; online learning; Sensors; tracked mobile robot; Trajectory; Trajectory tracking; Wheels
• ISSN: 2474-0446
• DOI: 10.1109/SSD.2019.8893152

In this paper, an online neural network trajectory tracking controller for a tracked mobile robot is proposed. The mobile robot used in this study has only three binary sensors to obtain its position/error from the target trajectory. Due to poor binary sensors which return only 0 or 1, it is impossible to get accurate position/error. Therefore it is not easy to apply ordinary linear control technique directly. For the mobile robot with such poor binary sensors, we propose the online neural network controller. The proposed online neural network controller has the following properties; 1. can train itself and control the mobile robot simultaneously; 2. has feedback structure which can play a role of derivative control like PD controller. Furthermore, we propose using the rough direction of the mobile robot in the input layer. Even though we only have inaccurate position/error, we consider using its direction. The direction is calculated by applying the least square to the inaccurate positions/errors. We expect that the direction by the least square works as a sort of derivatives, even though the positions/errors are inaccurate. Simulation and experimental results are given in order to present the effectiveness of the proposed online neural network controllers for the mobile robot. The comparison of the control performances among three different controllers is also given to show the importance of the derivatives to control the mobile robot and the efficacy of the two redundant derivatives we proposed.