Moving mainline train driving from manual operation (Grade of Automation 1, GoA1) to automatic train operation with a monitoring driver on board (GoA2) has been shown in an exploratory simulator research series to be accompanied by directionally lower situation awareness (SA), slower responses to unexpected critical events, reduced workload, accumulating fatigue, and a shift of visual attention from the track ahead to in-cabin displays — but the SA effects did not reach statistical significance, and no validated countermeasure exists. The open problem is to establish, with statistically powered, multi-method comparative evidence using expert train drivers, how GoA2 monitoring and remotely supervised driverless operation (GoA3) actually change operator SA, attention allocation, workload, and fatigue relative to GoA1 baselines, and on that basis to develop and validate information-presentation, training, and workplace interventions — including the remote-operator workplace — that maintain or restore safety-relevant oversight and prevent attentional tunnelling. A 2025 systematic scoping review confirms that the empirical GoA2 human-factors evidence base remains thin (only 4 of 65 reviewed human-in-the-loop studies concerned GoA2) and methodologically unharmonized, so the problem core is open.

Background

Grades of Automation (GoA) classify railway automation by how much of the driving task is delegated to the machine: at GoA1 the driver manually controls speed and monitors the track; at GoA2 automatic train operation (ATO) takes over continuous speed and stopping control while a driver remains on board to supervise and intervene when the automation reaches its limits; GoA3 denotes driverless trains supervised remotely; GoA4 is fully unattended operation. Human-factors research in aviation and road transport has long warned that automation turns active operators into passive monitors, degrading their readiness to take over. A key construct is situation awareness (SA), standardly defined as the perception of environmental elements, comprehension of their meaning, and projection of their near-future state; it is measured objectively with probe techniques such as SPAM (Situation Present Assessment Method, in which the operator is probed with questions about the current traffic situation and response latency and accuracy are scored) and subjectively with rating scales such as SART (Situation Awareness Rating Technique). Attentional tunnelling denotes the maladaptive narrowing of attention onto a single information source, e.g. onto in-cabin displays at the expense of scanning the track ahead. A research series by Brandenburger and Naumann (German Aerospace Center, DLR) — summarized in their 2019 paper in IFAC-PapersOnLine (the conference proceedings series of the International Federation of Automatic Control) — compared GoA1 and GoA2 in high-fidelity simulators with active mainline train drivers. It found: slower operator reaction times to unexpected critical events under GoA2; SA measured with SPAM and SART directionally lower under GoA2 but statistically non-significant (the authors state the tests 'missed the targeted significance'); significantly lower subjective workload under GoA2; rising Karolinska Sleepiness Scale ratings together with electroencephalography (EEG) spectral changes indicating accumulating fatigue (increased low-frequency Delta and decreased high-frequency Beta/Gamma power); and eye-tracking evidence that drivers under GoA2 fixate more on in-cabin displays and less on the outside track view. Because GoA2 keeps a human on board only as a fallback, these effects bear directly on safety: an under-loaded, fatigued monitor with degraded SA may respond late or incorrectly when automation fails. The same authors therefore argued for a GoA3 organization in which remotely placed operators take over trains on request — a concept they implemented as a prototypical remote-control workplace, evaluated only exploratorily and subjectively — and they asserted, without field measurements, that GoA2 yields only limited capacity gains compared with higher automation grades. Their paper and the surrounding research series point to unresolved issues including: statistically powered SA measurement in GoA2; systematic comparative evaluation of the remote-operator workplace against GoA1/GoA2 baselines; validated information-presentation and training interventions to maintain or restore SA and prevent attentional tunnelling; quantification of realised capacity gains for GoA2/GoA3; long-term fatigue, boredom, and vigilance studies beyond short simulator sessions; and organisational and regulatory evaluation of GoA3 adoption. Later literature treats these as an agenda, not as settled questions: a 2025 systematic scoping review of rail human factors and automation (Kusumastuti et al.) found that only 4 of 65 empirical human-in-the-loop studies concerned GoA2, that SA, attention, workload, and fatigue are measured with heterogeneous instruments across small simulator studies with mixed findings, and that a harmonized measurement framework is still missing; a 2021 Cognition, Technology & Work special-issue editorial (Vanderhaegen et al.) frames human factors of future railway systems as an open transverse research challenge.

Problem Statement

The problem, stated from the open items of Brandenburger and Naumann (2019): establish with statistically powered, multi-method empirical evidence how railway automation of intermediate and remote grades changes the human operator's situation awareness, attention allocation, workload, and fatigue — where GoA1 means manual driving, GoA2 means automatic train operation supervised by an on-board driver who intervenes when the automation reaches its limits, and GoA3 means driverless trains supervised from a remote control centre, and situation awareness (SA) means the perception of environmental elements, comprehension of their meaning, and projection of their near-future state. Specifically: (1) determine with adequate statistical power whether SA is lower for train drivers supervising automatic train operation under GoA2 than for drivers under manual GoA1 operation, resolving what the source series could only observe directionally; (2) iterate the prototypical remote-operator workplace for GoA3 and evaluate it systematically against GoA1 and GoA2 baselines on outcomes including operator performance, situation awareness, attention allocation, workload, fatigue, and failure-handling performance; (3) develop and validate information-presentation and training interventions that maintain or restore situation awareness and prevent attentional tunnelling (the maladaptive narrowing of attention onto a single information source, such as in-cabin displays at the expense of scanning the track ahead) in automated railway operations; and (4) determine, through field or long-duration simulated deployments, whether the observed electroencephalographic (EEG) and subjective fatigue trends under automated monitoring generalise beyond short simulator sessions. These four components address complementary facets of one overarching objective: a scientifically grounded determination of whether and how human oversight can be preserved as mainline railway automation moves from GoA1 through GoA2 to GoA3, replacing the current exploratory and underpowered evidence base. Resolving the problem requires all four components: each is a necessary part of the overarching objective, and addressing any single component — for example the statistically powered GoA2-versus-GoA1 situation-awareness comparison of component (1) — resolves that facet alone and does not by itself constitute a resolution of the problem. (Quantification of realised GoA2/GoA3 capacity gains and organisational/regulatory evaluation of GoA3 adoption are additionally open in the source series; they lie outside the scope of this problem.)

References

  1. Niels Brandenburger and Anja Naumann: On Track: A Series of Research about the Effects of Increasing Railway Automation on the Train Driver. IFAC-PapersOnLine 52(19), 288–293 (2019). DOI 10.1016/j.ifacol.2019.12.115, https://doi.org/10.1016/j.ifacol.2019.12.115
  2. Niels Brandenburger and Anja Naumann: Towards remote supervision and recovery of automated railway systems: The staff's changing contribution to system resilience. 2018 International Conference on Intelligent Rail Transportation (ICIRT), pp. 1–5 (2018). DOI 10.1109/ICIRT.2018.8641576, https://doi.org/10.1109/ICIRT.2018.8641576
  3. Niels Brandenburger, Anja Naumann, and Meike Jipp: Task-induced fatigue when implementing high grades of railway automation. Cognition, Technology & Work 23(2), 273–283 (print issue 2021; published online 13 November 2019). DOI 10.1007/s10111-019-00613-z, https://doi.org/10.1007/s10111-019-00613-z
  4. Sarah A. Kusumastuti, Tom H. J. Kolkman, Julia C. Lo, and Simone Borsci: Charting the landscape of rail human factors and automation: A systematic scoping review. Transportation Research Interdisciplinary Perspectives 30, 101350 (2025). DOI 10.1016/j.trip.2025.101350, https://doi.org/10.1016/j.trip.2025.101350
  5. Frédéric Vanderhaegen, Jean-Marie Burkhardt, Weining Fang, Anja Naumann, and Grigore M. Havârneanu: Human factors and automation in future railway systems. Cognition, Technology & Work 23(2), 189–192 (2021). DOI 10.1007/s10111-021-00670-3, https://doi.org/10.1007/s10111-021-00670-3
  6. Jean-Valentin Merlevede, Simon Enjalbert, and Frédéric Vanderhaegen: Grades of operational and tactical automation in rail domain for an artificial intelligence driving assistance system. Cognition, Technology & Work 28(1), 257–267 (2025). DOI 10.1007/s10111-025-00847-0, https://doi.org/10.1007/s10111-025-00847-0
  7. Baris Cogan, Esther Bosch, Birte Thomas-Friedrich, Helena Wasle, David Schackmann, Christian Klotz, and Birgit Milius: Evaluating train drivers' performance to inform the development of automatic train operation. Applied Ergonomics 129, 104622 (2025). DOI 10.1016/j.apergo.2025.104622, https://doi.org/10.1016/j.apergo.2025.104622
  8. Aobo Wang, Beiyuan Guo, Ziwang Yi, and Weining Fang: Research on Enhanced Situation Awareness Model with DMI Visualization Cues for High-Speed Train Driving. International Journal of Human–Computer Interaction 40(20), 6185–6199 (published online 2023; issue 2024). DOI 10.1080/10447318.2023.2247613, https://doi.org/10.1080/10447318.2023.2247613
  9. Emil Jansson, Nils O. E. Olsson, and Oskar Fröidh: Challenges of replacing train drivers in driverless and unattended railway mainline systems — A Swedish case study on delay logs descriptions. Transportation Research Interdisciplinary Perspectives 21, 100875 (2023). DOI 10.1016/j.trip.2023.100875, https://doi.org/10.1016/j.trip.2023.100875
  10. David Schackmann and Esther Bosch: Automated Detection of Train Drivers' Head Movements: A Proof-of-Concept Study. Automation 5(1), 35–48 (2024). DOI 10.3390/automation5010003, https://doi.org/10.3390/automation5010003