On the development of a framework for underwater localization using all-source data

This paper describes the current state of development of a localization framework with the goal of enabling an autonomous underwater vehicle (AUV) to execute long-range precision-navigation missions where positional resets based on GPS and/or transponders are unfeasible and bathymetric maps of sufficient accuracy are not always available. This framework has three main components: a real-time sonar data processing pipeline providing 3-D point-cloud measurements segmented into objects and underwater terrain; a 3-D synthetic natural environment (SNE) integrating geo-registered data from all available sources; a probabilistic localization method linking the SNE to the sonar processing pipeline and producing a probability map over the space of AUV horizontal positions. The current focus and novel contribution of this work is the specific method of exploitation of FLS 3-D point-cloud and 3-D reference map data, and the design of the corresponding probabilistic measurement model for localization. Initial off-line experiments, using real inwater data for which navigational ground truth is available, show the performance of this measurement model and the promise of the overall approach toward achieving horizontal position resets with a 95% circular error probability (CEP95) of less than 20 m.