Intelligent Control of Autonomous Vessels: Bayesian Estimation Instead of Statistical Learning?

Marine vessels have been recently considered for redesign with a view towards autonomous operation. This brings forth a number of safety concerns as regards malware attacks on intra-vehicle communications systems as well as on sensor based communication with their environment. Designing suitable hybrid systems or cyber physical systems as the above, which are data driven, involves a challenge by way of difficulty in abstraction. The current modeling paradigm for cyber physical systems is based upon the abstract idea of a hybrid automaton which involves discrete as well as continuous mathematical models for the physical device (marine vessel/s) Incorporating statistical inference techniques to introduce an element of autonomy in this has been recently proposed in literature. An engineering situation is explored in which a pair of marine vessels is being deployed to navigate avoiding collision with the help of deterministic control as well as with a particle filtering state estimator. A security intrusion is considered to occur in the communication channels and the robustness of the system is studied with the state estimation. Such intrusions can indeed be expected to defeat the collision protection design if sufficiently intense. However, better protection is offered by such Bayesian estimation based intelligent control as compare to statistical learning base control. Our results suggest that the hybrid automaton modeling paradigm with autonomy incorporated needs to be suitably abstracted in order to better design their defence against cyber-attacks