Our Third Newsletter is Out: Key Updates and New Perspectives on CODA
What is CODA?
CODA is a SESAR 3 Exploratory Research project, started in September 2023 and ending in February 2026. CODA involves developing a system in which hybrid human-machine teams collaboratively perform air traffic control tasks. This newsletter summarizes the results from the last issue, delivered in October 2024, and provides some initial insight on them. For more detailed reference on CODA objectives and results, including public deliverables, refer to our webpage, linked bellow.
Available results
From the last newsletter, we completed the technical work regarding ATCO’s task and mental state predictions, as well as the ATCO mental state monitoring. The results of this effort, as the rest of the advances in the project, may be found in the results section of the CODA webpage. More specifically, the following documents are available:
- D3.1 – ATCO Tasks prediction model, summarizing the results on ATCO task short term prediction of ACTO tasks based on data-driven models.
- D3.2 – Cognitive prediction model presents the development and implementation of a psychological model aimed at predicting the mental states (mental workload, fatigue, stress, and attention) of Air Traffic Controllers (ATCOs).
- D4.1 Indexes description and integration with prediction models, describes the neurophysiological sensors and indexes used to assess the mental states of air traffic controllers (ATCOs).
CODA ATCO Tasks prediction advances and results
In CODA we explored the use of artificial intelligence to predict air traffic controller (ATCO) tasks, aiming to improve workload management and support adaptive automation. The research evaluated two different AI-driven approaches, highlighting both the potential and the challenges of task prediction in complex air traffic environments.
The first approach, a machine-learning model trained on historical flight data, radar information, and ATCO task records, tried to provide as output a list of time-ordered tasks. It initially showed promise but struggled with accuracy due to data limitations. To address these shortcomings, researchers developed a second prediction approach using again historical flight data, radar information, and ATCO task records. This approach was based on a probability-driven, divide-and-conquer strategy with simpler and more explainable approaches. This second method calculates the probability of having each type of task in the next few minutes and aggregates the average number of tasks for all flights along time (See Figure for the time evolution of those aggregated number of tasks during a simulated scenario). While this method scaled back on ambition, it proved to be more explainable and feasible for real-world implementation.
Figure 1. Predicted number of tasks along time.
Results, summarized in D3.1, showed that AI can forecast ATCO tasks, but challenges remain, particularly in dealing with real-world variability and the limited availability of labeled data. Among the various models tested, decision trees, random forests, and long short-term memory (LSTM) networks demonstrated the most potential, while simpler algorithms failed to achieve sufficient accuracy. Despite these challenges, the study underscores the potential of AI-driven prediction models to enhance air traffic management by helping distribute tasks more efficiently between controllers and automated systems.
CODA Sensors and Human Machine Performance Envelope description
CODA also explored real-time methods to assess cognitive states such as stress, workload, vigilance, and mental fatigue, providing insights into human-machine collaboration in air traffic management (ATM) contexts. The first phase of WP4 dealt with the selection of sensors and technology able to provide reliable neurophysiological signals. Despite technological advancements, data reliability and user discomfort persist, even with newer sensor technologies. We reviewed the literature and performed internal tests to identify the set of sensors to be used for validation activities able to guarantee accuracy and comfort, reduce invasiveness, and refine the Prediction Model to better anticipate cognitive states and optimize human-machine interaction.
In particular, the integrated CODA system will use data coming from wearable sensors and technology (Mindtooth EEG system, Figure 1 and Shimmer GSR3+, Figure 2).
Figure 1. Mindtooth © Touch EEG headset.
Figure 2. The figure shows the Shimmer3 GSR+ device.
Finally, in WP4 the concept of Human Machine Performance Envelope (HMPE) concept has been defined and proposed. The HMPE defines acceptable performance limits for ATCOs by analysing mental workload, stress, vigilance and mental fatigue. At the same time, the HMPE index also considers the impact of the CODA system in terms of human-machine interaction/teaming and the behaviour of the Controllers. In other words, the HMPE index will be employed to evaluate the benefits of the CODA system by comparing the different experimental conditions and ATC scenarios.
CODA Experimentation Exercise II
In November 2024 CODA performed a series of simulation scenarios to gather information regarding controllers’ metal states (workload, stress, fatigue and vigilance). Four controllers from Italy and Germany executed different exercises along one week designed to trigger different mental state reactions. Data collected included:
- Physiological measures: Electroencephalography (EEG), Hearth Rate (HR), Hearth Variation Rate (HVR), Galvanic Skin Response (GSR), eye blinking, eye saccadic movements, eye fixations, and voice fundamental frequency.
- Subjective measurements: Dundee Stress State Questionnaire (DSSQ), Swedish Occupational Fatigue Inventory (SOFI), Karolinska Sleepiness Scale (KSS), Instantaneous self-assessment (ISA), and debriefings.
- Performance measures: radar traces, occupancy count, control events, and flight plans.
Up to eighteen people from different companies and countries collaborated to the smooth execution of the exercise along the week.
This data is being used to populate the constants and variables of the prediction mental models that have been developed in the project and will be used to provide information to the CODA system. The CODA system will support controllers in managing safely and efficiently their airspace by providing AI support when necessary, considering the future control tasks and current and future mental states.
Next steps
In the coming weeks, we will prepare and execute the final exercise, a human in the loop simulation with ATCOs, in which they will experience the CODA system, so to provide feedback on its impact on performance and safety, as well as its usability and the level of human-AI teaming.
Web page, social media and contact information
If you want more up-to-date news about CODA, please follow us on Linkedin. Also, if you want to join the CODA Advisory Board for future meetings where we will further discuss the project progress and results, you are more than welcome to propose yourself, by contacting us.
Project progress and results may be followed using our communication channels, currently including:
Web page: http://coda-sesar.eu
Linkedin page: https://www.linkedin.com/company/coda-atc
This newsletter
Direct contact, with Project Leader Stefano Bonelli (stefano.bonelli@dblue.it) or Dissemination Leader (juanalberto.besada@upm.es).