GPU-accelerated Monte Carlo Localization for Underwater Robots

Abstract

This paper proposes a navigation system for Remotely Operated Vehicle (ROV) operating around man-made structures such as a tube of a hydroelectric power plant. This kind of environment usually presents a combination of low visibility and ferromagnetic anomalies, so one of the aims of this paper is to propose a navigation method not dependent on cameras neither on magnetometers. In these types of environment, it is customary to have a map of its dimensions known a priori, and with this, it is possible to use an algorithm that is called Monte Carlo Localization. The drawback of using this algorithm is its high computational cost, but as it allows a parallel implementation, this disadvantage can be overcome. Thus, in this work is proposed a parallel implementation using a graphic processing unit (GPU) through CUDA architecture. The navigation system proposed in this work was validated by comparing experimental results with a highly accurate optical reference system. The tests were conducted in a rectangular tank of a known size, and the results show high robustness and real-time capabilities of the proposed navigation system.