Localisation and policy synthesis for underwater swarming autonomous vehicles with probabilistic guarantees about safe exploration and reachability requirements

This project is concerned with the case of an autonomous underwater vehicle (AUV) submerging in deep ocean to locate and inspect underwater infrastructure, i.e. a pipeline. For this task, proper decision-making and positional accuracy are key, both for getting closer to infrastructure whilst keeping safe and for recording data more accurately. The underwater environment raises significant difficulties to these requirements, due to signal attenuation, multipath fading as well as the presence of extreme and unpredictable currents. To address the aforementioned challenges, we estimate that better localisation can be achieved if the receiver nodes (which in this case are also AUVs floating on the surface) stay in an equilateral triangle formation. On the other hand, to synthesise the appropriate policy that satisfies the requirements for safe exploration and reachability of the pipeline, we employ a reinforcement learning (RL) inspired control policy, that satisfies the required properties, which we express as linear temporal logic (LTL) formulae. The performance of our method is satisfactory and evaluated through several test case scenarios, where different parameters involved are taken into account.