Enhancing maritime defence and security through persistently autonomous operations and situation awareness systems
MetadataShow full item record
This thesis is concerned with autonomous operations with Autonomous Underwater Vehicles(AUVs) and maritime situation awareness in the context of enhancing maritime defence and security. The problem of autonomous operations with AUVs is one of persistence. That is, AUVs get stuck due to a lack of cognitive ability to deal with a situation and require intervention from a human operator. This thesis focuses on addressing vehicle subsystem failures and changes in high level mission priorities in a manner that preserves autonomy during Mine Counter measures (MCM) operations in unknown environments. This is not a trivial task. The approach followed utilizes ontologies for representing knowledge about the operational environment, the vehicle as well as mission planning and execution. Reasoning about the vehicle capabilities and consequently the actions it can execute is continuous and occurs in real time. Vehicle component faults are incorporated into the reasoning process as a means of driving adaptive planning and execution. Adaptive planning is based on a Planning Domain Deﬁnition Language (PDDL) planner. Adaptive execution is prioritized over adaptive planning as mission planning can be very demanding in terms of computational resources. Changes in high level mission priorities are also addressed as part of the adaptive planning behaviour of the system. The main contribution of this thesis regarding persistently autonomous operations is an ontological framework that drives an adaptive behaviour for increasing persistent autonomy of AUVs in unexpected situations. That is, when vehicle component faults threaten to put the mission at risk and changes in high level mission priorities should be incorporated as part of decision making. Building maritime situation awareness for maritime security is a very difﬁcult task. High volumes of information gathered from various sources as well as their efﬁcient fusion taking into consideration any contradictions and the requirement for reliable decision making and (re)action under potentially multiple interpretations of a situation are the most prominent challenges. To address those challenges and help alleviate the burden from humans which usually undertake such tasks, this thesis is concerned with maritime situation awareness built with Markov Logic Networks(MLNs) that support humans in their decision making. However, commonly maritime situation awareness systems rely on human experts to transfer their knowledge into the system before it can be deployed. In that respect, a promising alternative for training MLNs with data is presented. In addition, an in depth evaluation of their performance is provided during which the signiﬁcance of interpreting an unfolding situation in context is demonstrated. To the best of the author’s knowledge, it is the ﬁrst time that MLNs are trained with data and evaluated using cross validation in the context of building maritime situation awareness for maritime security.