During the second reporting period, several meetings and a workshop were held for the AISA project. The AISA Advisory Board (AB) met on September 20, 2021 and the AB and consortium members participated in a risk assessment workshop on March 8, 2022. Members of the SESAR 3 Joint Undertaking (S3JU) were invited to and participated in a steering board meeting on May 13, 2022 where project results were discussed. The final review meeting (FRM) was held in Brussels on June 30, 2022 and a follow-up meeting via Webex was held on July 19, 2022 to discuss project achievements and Technology Readiness Level criteria fulfillment. The final AISA workshop was held at the 12th European Aeronautics Science Network International Conference in Barcelona on October 20, 2022. The project's close-out meeting on November 25, 2022 marked the end of project activities.
Task 2.1 and D2.1 developed a Concept of Operations (ConOps) for en-route air traffic control (ATC) performed by a human-machine team with shared situational awareness (SA). Task 2.2 analyzed requirements for automation of monitoring tasks via AI SA, showing which tasks can be automated and what requirements are needed. Task 3.1 developed a 4D trajectory prediction module using a neural network and a two-step approach. Task 3.2 developed a Conflict Detection module, which performed well in identifying errors but needed further research. Task 3.3 developed a complexity assessment model to determine air traffic complexity. Task 4.1-4.4 focused on creating a knowledge graph system in Java, connecting it with Prolog, and encoding facts and rules about ATC operations. Task 5.1 evaluated SA among AI and ATCOs through various measurement tools in Experiment 1 and Experiment 2. Task 5.2 conducted a risk assessment of the AI SA system and proposed measures to ensure safety. Task 5.3 evaluated the impact of AI system on human performance in distributed SA, showing high accuracy for most tasks, particularly for conflict detection.
The project has successfully developed a Proof-of-Concept (PoC) knowledge-based system for automating en-route air traffic control (ATC) tasks. Most of the ATCO tasks chosen in the project's ConOps were successfully automated, tested, and applied to traffic data (46 out of 57 tasks), with a focus on monitoring tasks but also including some tasks that involve prediction and decision-making. This has laid the foundation for further automation within the human-machine ATCO team.
The accuracy of the PoC system was deemed very high according to an analysis of its performance, and it was concluded that there is great potential for real-time operation considering it takes about 5 seconds to process a single traffic situation graph, which is similar to the refresh rate of an ATCO's working position. Machine learning (ML) modules were partially integrated into the system, with the conflict detection module showing an accuracy of 70%.
Different assessments of situation awareness (SA) were completed, including assessments of human SA, machine (or artificial) SA, and team SA. The results of these assessments showed that ATCOs are more critical in judging their own SA when they are aware of machine SA, and that inputs from the AI SA system might be beneficial for ATCOs in terms of SA with some minor adjustments, such as providing warnings for important time-critical aspects and providing other information in a more passive manner. Machine SA was assessed through the system's properties and functions and through comparison with human SA, and the results indicated that the ML modules can predict the future state of traffic.
Overall, the PoC system has shown promising results in terms of automating ATC tasks and improving SA within the ATCO team.
