Towards Sonar Based Perception and Modelling for Unmanned Untethered Underwater Vehicles

The ability to automatically generate internal models from sensed data will be of critical importance in the development of flexible, adaptive and useful unmanned underwater vehicles. The subsequent recognition of objects based on these a-priori models, which may be incomplete or partial, will also be an essential component in the success of UUV missions. This paper describes and presents experimental results obtained with our calibrated computer controlled frequency modulated (FM) subbottom penetrating sonar. We present a topographic model of the sea-floor surface and a sequence of sonar “images”, obtained with the same FM sonar. This latter sequence shows a vertical section through the sea-bed itself. Despite the absence of surface features the variations in stratigraphy of this vertical section and the automation of the interpretation of these features opens the possibiliity of using these data to navigate when resurfacing is either not an option or desirable, for example, during under ice missions or from full ocean depth.