On 11 November 2024, the SESAR AI Flagship workshop brought together researchers, industry stakeholders and regulatory representatives at Spazio Europa in Rome. This second session of the flagship series followed the inaugural online workshop held on 20 September and coincided with the SESAR Innovation Days, which kicked off the next day. With 56 participants from 23 SESAR projects, the workshop served as a strong continuation of discussions on aligning AI innovation with aviation certification requirements.

A key highlight of the event was the active participation of representatives from the European Union Aviation Safety Agency (EASA). Their involvement deepened the connection between SESAR projects and the institution driving the AI Roadmap 2.0, fostering a shared understanding of the roadmap’s goals and bridging the gap between research outputs and regulatory needs.

Bridging research and regulation: insights from EASA

Central to the workshop was EASA’s detailed presentation on the AI Roadmap 2.0, a strategic framework guiding the development and deployment of AI systems in aviation. Launched in 2019, the roadmap has evolved into its second phase, focusing on Level 3 AI capabilities and the consolidation of an AI/ML framework. EASA’s direct engagement with the workshop bridged the divide between research and regulation, enabling SESAR projects to align their innovations with the priorities and expectations of Europe’s leading aviation safety authority.

Guillaume Soudain, joined by colleagues François Triboulet and Andrew Kliner, elaborated on how the roadmap integrates existing safety and certification standards with the unique demands of AI systems. The session highlighted the importance of defining a robust safety assurance process that accommodates the non-deterministic nature of AI. Unlike traditional deterministic automation, AI systems require novel approaches to data management, security threat modeling, and system monitoring.

The discussion on EASA’s “W-model” for learning assurance exemplified these challenges. Designed to address issues such as bias-variance trade-offs and model generalization, the W-model introduces iterative steps for identifying edge cases and ensuring operational robustness. EASA emphasized the criticality of understanding and managing the influence of AI components within hybrid systems, where deterministic and non-deterministic elements coexist.

The roadmap also underscored the significance of operational explainability, tailored to the specific needs of stakeholders such as air traffic controllers and system operators. This focus on user-centric design resonated with participants, especially as the SESAR projects continue to explore how AI can be effectively integrated into real-world ATM operations.

SESAR projects showcase collaborative innovation

The workshop spotlighted five SESAR projects, each tackling unique challenges in applying AI to air traffic management.

JARVIS The JARVIS project introduced its human-centric design approach for digital assistants, emphasizing the critical role of trust, workload-adaptive interfaces, and robust interaction mechanisms. The project highlighted the risks of over-reliance on AI and the potential for de-skilling, offering a thought-provoking perspective on human-AI teaming.

CODA and TRUSTY The CODA and TRUSTY projects shared their advances in digital systems for air traffic controllers, exploring adaptive task distribution and the psychological nuances of trust in AI-powered digital towers. These presentations illuminated the transformative potential of AI in redefining ATM roles, while also addressing the philosophical and practical challenges of integrating AI into high-stakes environments.

AI4HyDrop AI4HyDrop project explored how AI can optimize airspace management, particularly with the growing presence of drones. By leveraging computer vision and other robust technologies, the project demonstrated its commitment to achieving technical resilience and operational efficiency.

HUCAN The HUCAN project showcased its work on a certification roadmap for advanced ATM systems. With EASA now leading this regulatory effort, HUCAN has shifted focus to clarifying critical definitions and approaches to AI, as evidenced by its paper presented at the SESAR Innovation Days.

The in-person gathering in Rome provided a dynamic forum for meaningful dialogue, building on the foundational discussions of the September online session. Roundtable discussions fostered rich exchanges on topics ranging from technical robustness and AI explainability to human-AI teaming and certification challenges. Participants collaborated on identifying shared terminology, exploring visualization tools, and addressing liability concerns.

As the day concluded, Alessandro Prister from SESAR thanked the participants for their active engagement and underscored the importance of such collaborative efforts. The SESAR AI Flagship Workshop exemplified the critical role of community and shared expertise in advancing AI for aviation, setting the stage for future innovations and robust frameworks that will shape the industry's evolution.