Design and Implementation of UFRJ Nautilus’ AUV Lua - A TinyML Approach

The UFRJ Nautilus is a student-driven engineering project team at the Federal University of Rio de Janeiro focused on building and designing AUVs to compete in the AUSVI RoboSub Competition. The team focused on the making of a brand new autonomous underwater vehicle (AUV) called Lua. There are several challenges on building a software system for an AUV. The localization problem is a major challenge. Among the different proposals for localization, the beamforming algorithms have proved to be a reliable option. However, they present some limitations. The beamforming algortihms have two sources of errors: (i) the noise typical of sensors and signals and (ii) the arrangement of the sensors, which is a function of the true azimuth (horizontal angle from an observer - the AUV - to the object of interest) and elevation angles. Therefore, the beamforming has an error factor which is the sum of both errors described. In order to overcome this issue, we propose using a TinyML-based approach - an embedded Machine Learning algorithm that is able to learn about angles combination that reduces the error. To reduce the noise we will combine the TinyML algorithm with a traditional Kalman Filter. Experimental results showed that our proposal has a better performance reducing the errors from previous approaches.