Enhancing flight deck decision support with distributed GenAI : a multi-agent approach

Abstract

The integration of Generative Artificial Intelligence (GenAI) into commercial aviation presents transformative opportunities for enhancing flight deck operations, offering an intuitive natural language interface between pilots and automation. This study investigates a novel GenAI-based multi-agent architecture designed to address the unique demands of aviation, such as the need for rapid decision-making and adherence to stringent safety protocols. By employing lightweight, embedded Large Language Models (LLMs), the architecture optimises task allocation among specialised agents, ensuring operational efficiency without reliance on external cloud infrastructure. Preliminary evaluations demonstrate that the proposed architecture achieves performance comparable to systems using larger models, such as GPT-4, while operating locally with lightweight models. This result underscores the feasibility of implementing autonomous, cloud-independent GenAI solutions embedded directly within aircraft systems. Through a comparative analysis of different LLM configurations, the system balances scalability and precision in handling cockpit-specific tasks. Challenges related to explainability, response latency, and integration with broader Human-Machine Interface (HMI) systems are identified as critical areas for future development.