A Context-sensitive Plan Execution Language for Adaptive Behaviour in Autonomous Systems

The rise of Unmanned Air System in the civil and military aeronautics domain has led to increasing efforts to raise the degree of autonomy in their mission conduct. Especially the dynamic character of aeronautical missions poses a great challenge with respect to autonomy, requiring adaptive behaviour in the face of changing environmental circumstances. The complexity of the environment mandates a stronger focus on the ability to understand the situational developments by drawing connections between observations with the help of context and situation models. Alongside the challenges of autonomy, the way operations are conducted changes as participating units, such as police and firebrigade, share an increasing amount of information and capabilities based on computer networks. Together, these units contribute to a shared operational picture, improving decision making and effectiveness. The general rise in the amount of software in avionics and the increasing distribution aspect of such systems offers new challenges concerning its maintainability and extensibility. Especially in the face of costly certification of safety critical systems, modifications should be isolated as much as possible. An often overlooked source of cascading modifications is the change of exchanged terms in a communication vocabulary of an application-level protocol, such that local representations need to be updated even if such changes are of no concern to all components. The main contribution of this thesis is a framework which integrates the Description Logics knowledge representation formalism in formal syntax and semantics into the existing plan execution language PLEXIL. The introduction of expressive environment models augments the language to a context-sensitive execution framework, tackling the autonomy problem on the layer of an intelligent control system where, at runtime, the pivotal decision are made. Description Logics not only facilitates the modelling and inferencing of environmental circumstances, the context, it also offers an elegant approach to the mentioned problem of cascading modifications in application-level protocol terminologies. Isolation of vocabulary revisions and extensions can be achieved by allowing software components to model their own domain vocabulary using DL and then employ semantic matching at runtime to determine correspondances in vocabularies. In this thesis, a methodology and formal framework is developed which facilitates the ontology engineering process with respect to creating separate domain ontologies, yet, ensuring a large enough overlap to maintain the possibility to communicate. The major and minor contributions are employed in an avionics architecture of an UAS, showing adaptivity to higher- and lower-level context knowledge. Also, the used context models serve as communication vocabularies between components, showing how the described ontology design methodology can be put to use.