Resumen de Online acoustic localization methods for autonomous underwater vehicles
Guillem Vallicrosa Massaguer
Autonomous Underwater Vehicles (AUVs) true autonomy capabilities in complex potentially unknown environments, have not yet been fully achieved because of the lack of online algorithms that can solve fundamental problems such as localization, mapping and path-planning on-board the AUV and consequently react according to their outputs. These algorithms can empower them for new capabilities such as long-term deployments, autonomous exploration and autonomous intervention.
This thesis presents the development of two online localization algorithms for AUVs. The first algorithm is based on a Sum of Gaussian (SOG) filter for online range-only localization of an acoustic beacon, e.g. localization of a Docking Station (DS) for battery recharging and data uploading. Two different versions of the algorithm are developed, one based on Dead Reckoning (DR) navigation and one based on a full Simultaneous Localization and Mapping (SLAM) solution. Moreover, an Active Localization (AL) algorithm is also developed to autonomously select the best actions that minimize the range-only localization uncertainty. This algorithm is also tested as part of a wider project where it is combined with other algorithms to produce a complete homing and docking strategy. Consequently, this algorithm can help long-term deployed AUVs being able to always return to their base DS for battery recharging.
The second algorithm proposes a new online SLAM framework for continuous occupancy mapping named H-SLAM. It uses a Rao-Blackwellized Particle Filter (RBPF) where each particle carries a HM representation of the environment. HMs offer a low memory footprint and constant computational complexity O(1) for insertion and query, suitable for online processing. This algorithm is tested on two real-world datasets offering a significantly better reconstruction of the environment than using DR navigation. Producing correct continuous occupancy maps and trajectories, opens plenty of possibilities for future combination with online path-planning algorithms.
