Towards Cognitive Coaching in Aircraft Piloting Tasks: Building an ACT-R Synthetic Pilot Integrating an Ontological Reference Model to Assist the Pilot and Manage Deviations

Most aviation assistance systems do not take into account the pilot’s actual cognitive state when providing assistance. Yet, especially in critical situations such as aircraft takeoff, it is important to determine whether the information presented has been correctly processed and understood by the pilot, or whether some has been omitted or misinterpreted. This paper presents a cognitive synthetic pilot based on the ACT-R cognitive architecture and integrating a reference ontology of standard piloting procedures as a knowledge base. The main purpose is to serve as a coaching system for training pilots in simulation environments to perform critical piloting tasks such as takeoff, by exploiting the advantages of a rich semantic representation of the aeronautical context. In this way, the ontology formally models the expert piloting procedures that will be used by the synthetic pilot to advise an trainee pilots during training sessions. For this, we use two types of ontologies to model the pilot’s work: a task ontology describing all the actions the pilot must perform, and a domain ontology containing knowledge about the execution environment. The synthetic pilot uses semantic rules and a reasoner for task automation. The rules define when a task can be executed. The reasoner analyzes these rules and the context to decide which actions to execute. Over time, the actions to be executed are presented in the form of a complex dynamic 3D graph, thus allowing better visualization of the tasks to be performed and intelligent automation of the flight procedures described in the ontological reference model to assist the pilot. This work is an intermediate step towards the implementation of a complete cognitive assistance for novice pilots in a simulation environment. The ultimate goal is to extend the capabilities of the synthetic pilot through machine learning, by analyzing real flight data to extract typical pilot behavioral profiles.