Stereo-vision graph-SLAM for robust navigation of the AUV SPARUS II

This paper presents the integration of a stereo-vision Graph-SLAM system in the navigation and control architecture of the Autonomous Underwater Vehicle (AUV) SPARUS II. The navigation architecture of SPARUS II is endowed with an Extended Kalman Filter (EKF) that fuses the data provided by a Doppler Velocity Log (DVL), a pressure sensor, a GPS (when the vehicle is in the surface) and an Inertial Measurement Unit (IMU). But due to the nature of the aforementioned sensors, this localization data is prone to drift. Instead, the stereo-vision Graph-SLAM clearly improves the localization data thanks to the additional pose constraints computed from visual (stereo) loop closings. SLAM estimates are thereafter inserted in the control architecture to increase the precision in the navigation and mission tasks. Experiments with SPARUS II in simulated environments show the improvement and benefits in the application of this SLAM approach for localization, navigation and control, with respect to the use of the EKF odometry