Phd in INFORMATION AND COMMUNICATION TECHNOLOGY (ICT)

Thesis title: Bio-Inspired Propulsion System and Autonomous Driving Systems Design for a Class of Electric Boats.

The adoption of Autonomous Driving Systems (ADS) is increasingly widespread in land vehicles for private and public transport and more standards have been defined, among which SAE and ADAS are the most complete and have already been qualified and accepted to operate in the real world. In maritime transport, although automation preceded that adopted in land vehicles, there are no standards for complete ADSs. Furthermore, the automation on boats is always partial and refers to only some navigation phases instead of the overall navigation. In this thesis an ADS is presented with the attempt to define a qualified standard for a specific class of electric boats. The design involved both Classical Computation and Computational Intelligence techniques (mainly Fuzzy Logic and Artificial Neural Networks) and several version of the same ADS modular architecture have obtained progressively replacing the classical modules with the intelligent ones. Within the design idea cohabit both the ”divide et impera” paradigm and the holistic point of view, where each module is at the same time an element of a vertical hierarchy but also part of a horizontal organization. This vision leads to specific co-design procedures and precise design choices. The ADS results to be a complex system which interact with another complex system, the environment. Interesting results have been found by adopting the Fuzzy Q-Learning method to make the complex ADS adaptive with the complex environment. An ad hoc simulator is also presented. The proposed simulator, developed in MATLAB environment, is a complete environment for autonomous boat simulation, and eventually training. All the simulation results are also presented giving particular attention to the differences between the performances obtained with a classical versions of the ADS and the ”intelligent" one. It will be clear how computational intelligence adoption allows the ADS to achieve higher probability of success than that obtained with the classical approach on all different scenarios and reinforcement learning allows to specialize the ADS on a singular scenario. Another example of specialized learning is used to control the motion while fuzzy logic is used to re-design some controllers. A performance indicator has been formulated that is inspired by Fish Schooling Behavior, since it has strong similarities with the self-driving boat problem. The same is used as Reward for the Fuzzy Q-Learning method.