Bonjour à tous,

Vous êtes tous et toutes cordialement invité.es à assister à la présentation de projet du prédoc III de Amin Bonyadkhalaj, le 30 juin à 09:30h .

Titre : A Cognitive Multi-Agent Architecture for Virtual Pilots

Date: mardi 30 juin à 09:30h

Location: Pavillon André-Aisenstadt, 2920 Ch. de la Tour, salle 3195

Jury

Résumé

Modern aviation places pilots in highly complex environments where advanced avionics, automation, and time-critical operational demands converge. Although automation has reduced some aspects of manual workload, it has also introduced cognitive risks such as mode confusion, reduced situational awareness, and overload during critical flight phases. This thesis proposes a Cognitive Multi-Agent Architecture for Virtual Pilot Assistance, developed within the C-Pilot project. The proposed architecture aims to transform processed pilot-state and flight-context information into governed, traceable, and non-intrusive cognitive assistance. It is organized around a central Coordinator and three specialized agents: the Companion Agent, the Recommender Agent, and the Derivation Agent. The Coordinator acts as the sole decision authority, enforcing safety policies, timing constraints, and assistance rules before any output is delivered to the pilot. The research builds on previous simulator-based studies on pilot workload, attention, physiological responses, facial temperature, and experience effects. These studies provide the empirical foundation for the thesis by showing that pilot cognitive state can be measured and used as contextual input. The main contribution of the thesis is therefore not only the estimation of pilot state, but the design and validation of a governed cognitive-agentic architecture capable of using this information safely for virtual pilot assistance.