Sonar attentive underwater navigation in structured environment

Abstract

One of the fundamental requirements of a persistently Autonomous Underwater Vehicle (AUV) is a robust navigation system. The success of most complex robotic tasks depends on the accuracy of a vehicle’s navigation system. In a basic form, an AUV estimates its position using an on-board navigation sensors through Dead-Reckoning (DR). However DR navigation systems tends to drift in the long run due to accumulated measurement errors. One way of mitigating this problem require the use of Simultaneous Localization and Mapping (SLAM) by concurrently mapping external environment features. The performance of a SLAM navigation system depends on the availability of enough good features in the environment. On the contrary, a typical underwater structured environment (harbour, pier or oilfield) has a limited amount of sonar features in a limited locations, hence exploitation of good features is a key for effective underwater SLAM. This thesis develops a novel attentive sonar line feature based SLAM framework that improves the performance of a SLAM navigation by steering a multibeam sonar sensor,which is mounted on a pan and tilt unit, towards feature-rich regions of the environment. A sonar salience map is generated at each vehicle pose to identify highly informative and stable regions of the environment. Results from a simulated test and real AUV experiment show an attentive SLAM performs better than a passive counterpart by repeatedly visiting good sonar landmarks.