Skip to main content

Modern ATM via Human/Automation Learning Optimisation

Periodic Reporting for period 1 - MAHALO (Modern ATM via Human/Automation Learning Optimisation)

Reporting period: 2020-06-01 to 2021-05-31

The goal of the MAHALO Project is to answer a specific question: should be de developed an automated system powered by Artificial Intelligence and Machine Learning that matches the human behaviour (i.e. conformal) or an automation that is understandable to the human (i.e. transparent)? Further, what tradeoffs exist, in terms of controller trust, acceptance, and performance? To answer these questions, MAHALO will:

• Develop an individually-tuned hybrid ML system comprised of layered deep learning and reinforcement models, trained on controller performance (context-specific solutions), strategies (eye tracking), and physiological data, which learns to solve ATC conflicts;
• Couple this to an enhanced en-route CD&R prototype display to present machine rationale with regards to ML output;
• Evaluate in realtime simulations the relative impact of ML conformance, transparency, and traffic complexity, on controller understanding, trust, acceptance, workload, and performance; and
• Define a framework to guide design of future AI systems, including guidance on the effects of conformance, transparency, complexity, and non-nominal conditions.
The period covered by the current report is the first year of the MAHALO project. Here the work performed in this cited period for each Work Package.
WP1 activities include both internal coordination (of technical activities, products, and administrative matters) and external communication (with the EU Project Officer and external Advisory Board). It is proceeding regularly and the whole consortium meets regularly on a bi-weekly basis to carry out the work. In addition, it meets when necessary (at least once a week but can be more) to carry out those tasks that require a greater effort or because of some pressing deadlines as for these last few months. The modality remains remote due to restrictions due to the Covid-19 pandemic.
WP1 produced deliverable D1.1 that was correctly submitted and approved.

WP2 Conceptual Definition has firstly developed a State Of The Art Review integrating together Tasks 2.1 and 2.2 on Human Performance and Machine Learning techniques. This State Of The Art Review has been uploaded in the deliverable D2.1 and represents the input of the Concept & Scenario Definition that has correctly been submitted and approved. D2.2 is constructed to be a living document, where new inputs regarding the Concept are included when they arise. In order to track new changes coming from WP5 the deliverable will then be updated again on September 2021 (M17).

In December 2020 the two WP3 and WP4 were kicked off together, as TUD is leading both WPs and the same tasks require joint work. According to the PMP, WP3 and WP4 should have kicked off in September 2020, but the consortium decided to wait for all the inputs coming from D2.2 avoiding to develop multiple ML algorithms not backed by a strong theory. There was not a negative impact on the respective activities and deliverables caused by this delay, as it was foreseen.
According to the PMP, the output of WP3 will be D3.1 Machine Learning Report and D3.2 Machine Learning Demonstrator, both to be submitted by the end of May 2021 (M12). The priority between the two deliverable has been given to the Report in order to not have an impact on WP5 activities and although it is a little bit late, it will be submitted before the intermediate review meeting. The Demonstrator will be submitted just few days later, due to its less important impact on other activities.

The output of WP4 were D4.1. E-UI design document and demonstrator and D4.2. E-UI validation report. Both deliverable have been correctly submitted on STELLAR and SYGMA. The first one include also a video demonstrator which can be reached by accessing a folder linked inside the document.

WP5 Integration and WP6 Simulation are not started yet, according to the PMP. WP5 will be then kicked off shortly after the intermediate review meeting.

In the meantime the consortium as a whole has participated in some important dissemination events, such as DASC 2020 and SIDs 2020, from which it has obtained excellent feedback from the audience. These events have been widely promoted through the use of the website and social channels, LinkedIn and Twitter, of the project. MAHALO took part to an Automation Workshop with other similar projects analysing the Explainability apsects of AI in the ATM industry; this projects have been invited to our Advisory Board.
A YouTube account was also created to collect all the videos of these events and those that will be made in the future. Video demonstrators from D3.2 and D4.1 will be also included soon.

The social channels currently active can be reached at the following links:
the Project website

Further events the project would like to attend or have been already scheduled are:
• First Workshop with the Advisory Board – October 2021
• Second Workshop with the Advisory Board – May 2022
• Human Factors and Ergonomics Society Annual Meeting (HFES) - October 2021
All these events will be collected in the Communication and Dissemination register on Stellar.
MAHALO will create a system that learns from the individual operator, but also provides the operator insight into what the machine has learnt. Several models will be trained and evaluated to reflect a continuum from individually-matched to group-average. Most recent work in areas of automation transparency, Explainable AI (XAI) and ML interpretability will be explored to afford understanding of ML advisories. The user interface will present ML outputs, in terms of: current and future (what-if) traffic patterns; intended resolution maneuvers; and rule-based rationale. The project’s output will add knowledge and design principles on how AI and transparency can be used to improve ATM performance, capacity, and safety.
That said, the knowledge reached within MAHALO could be applied in the future via a transfer learning process also to other transportation domains, such as automotive.