HYDRO 2016 Paper 9A3
IMGAM is an autonomous underwater vehicle designed for autonomous gas flare detection, localisation, and sampling. It employs sensor reactive behaviour throughout the various autonomous phases of a mission. Such behaviour has an inherent risk of "wrong" decisions with potential catastrophic consequences. Therefore, care must be taken to ensure a thorough understanding of how the vehicle reacts to external stimuli and to itself. For highly complex systems, such understanding can best be achieved by broadly measuring or estimating performance of all known subsystems and by combining this knowledge into an overall simulation model. This model can then be used for development and testing of control algorithms and autonomy behaviour. In IMGAM, Matlab/Simulink® has been used as tool for such a model, which now features both virtual reality visualisation and real-time manual user interfaces for testing overall pilot-in-the-loop behaviour during hybrid AUV/ROV-mode operation.
Using highly detailed models has an inherent risk that sub-model inaccuracies distort overall model performance. This is best mitigated by identification and validation against measured real vehicle performance. Unfortunately, optimisation of models with high level of detail is often too complex for trial-and-error methods. Development work in IMGAM therefore employs system identification procedures and software that were originally developed for identification of highly complex systems in aviation.
By maintaining and developing a model in parallel with system design, it has been possible to perform extensive tests of vehicle, controller, and autonomy performance. This has allowed ATLAS ELEKTRONIK GmbH to identify risks and necessary system improvements long before the first metal was cut. Furthermore, automatic code generation of controller and autonomy algorithms has facilitated easy and trouble free integration into ATLAS MARIDAN's highly proven AUV-core system that forms the vehicle's back-bone.
The paper will discuss the subjects mentioned above in greater detail. It will also discuss test results and lessons learned during trials (planned for late summer 2016).