Content uploaded by Gijs de Rooij
Author content
All content in this area was uploaded by Gijs de Rooij on Jun 03, 2021
Content may be subject to copyright.
FLIGHT ALLOCATION IN SHARED HUMAN-AUTOMATION
EN-ROUTE AIR TRAFFIC CONTROL
G. de Rooij, C. Borst, M.M. van Paassen, M. Mulder
Delft University of Technology
Delft, the Netherlands
Allocation is a challenge for higher levels of automation in air traffic control,
where flights can be dynamically assigned to either a human or an automated
agent. Through an exploratory experiment with six professional air traffic
controllers, insight was gained into the possibilities and challenges of human-
automation teamwork in an en-route environment. Participants showed high levels
of automation trust, but mostly ignored automation-suggested allocations,
preferring a highly automated sector instead. Most flights were delegated to
automation, after they were given a direct and conflict-free path. Flights handled
manually were those requiring level changes or non-standard routing. Future
research should focus on establishing specifically which flights can be automated.
Air traffic controllers (ATCOs) work in a challenging and demanding environment. The
continuous quest for more efficient and safer air travel, drives the development of more advanced
automation. Both Europe and the United States aim for higher levels of automation in the com-
ing decades with a more supervisory/strategic role for humans (Prevot, Homola, Martin, Mercer,
and Cabrall (2012); SESAR Joint Undertaking (2019)). In such an environment, less people can
handle more traffic in larger sectors. Despite high levels of automation, humans are expected to
play an important role in supervising these future systems and to intervene when automation falls
short (Metzger and Parasuraman (2005)); people will ultimately remain responsible.
To be able to intervene, it is essential that ATCOs maintain vigilance, situation awareness
and a sufficient skill level to perform tasks manually (Bainbridge (1983)). This could be achieved
by not making the human a supervising bystander, but have him/her work side-by-side with au-
tomation in a team, both able to perform and share tasks. This sparks the question of what such
co-operation should look like, and what impact it will have on human-automation performance.
Currently, airspace is divided into sectors, each under the responsibility of a different
ATCO. This requires considerable coordination between adjacent sectors and may lead to an
imbalance in traffic load (and thus workload). To mitigate these issues, Birkmeier, Tittel, and
Korn (2016), among others, considered so-called flight centric or sectorless operations. Instead of
coupling controllers to geographic areas, a single controller would be assigned to several flights,
from departure to arrival, reducing the number of handoffs and possibly providing a better work-
load balance. This, however, also introduces new challenges. Consider, for example, when two
flights under control by different ATCOs are in conflict. Who should then solve the conflict?
What if that other controller is not another human, but an automated system? How are
flights then assigned to either a controller or automation? Should all aircraft involved in a conflict
be controlled by either the ATCO or the automation, so as to mitigate additional workload related
to coordination? If not, who solves a conflict? In addition, with an automated agent, it becomes
possible to share (sub)tasks dynamically, back and forth, between human and automation. This
could establish true teamwork, but only if above-mentioned questions have been answered first.
This paper discusses an exploratory experiment on the allocation of flights in a shared
human-automation en-route airspace. Control over which flights were automated was given to the
ATCOs themselves, although initial automation-based suggestions were given for each flight.
Method
Participants and Apparatus
Six professional ATCOs (age M = 38.3, SD = 10.0, years of experience M = 14.8,
SD = 8.7), from Maastricht Upper Area Control (MUAC) participated in a real-time simulator
experiment. A TU Delft-built Java-based simulator (Fig. 1) was designed to mimic the MUAC
interface, to ensure that participants could focus on working with the experimental automation.
A 1920 x 1920 pixels 27" display was used with a standard computer mouse for control inputs.
Figure 1. Simulator interface, with blue aircraft allocated to au-
tomation and green aircraft to the human ATCO. Background col-
ors have been inverted here for clarity.
Figure 2. Callsign menu,
as shown when clicking the
callsign in an aircraft label.
The ATCO could delegate
a flight to automation by
pressing "ASSUME TO
AUTO". Once the flight
was assumed, a "TRANS-
FER" button was added to
the menu.
Airspace and Traffic Scenario
Participants were responsible for traffic above FL245 in the combined DELTA and
JEVER sectors, above the Netherlands and part of Germany. Each ATCO experienced the same
traffic scenario, resembling an average day in February 2020 (prior to the COVID-19 pandemic).
There were between 15 and 30 flights in the sector at any time (M = 21, SD = 4). Flights followed
standard routing or directs to their designated exit points. Besides overflying traffic, arrivals and
departures to several airports, within or close to the sector, were included. There was no wind.
Automation
During the exercise, the ATCOs were accompanied by an automated “colleague". When
flights entered their sector, the ATCOs had to decide whether to manually assume the flight or
delegate it to automation (Fig. 2). This allocation remained flexible, such that they could re-
assume manual control or delegate flights to automation at any time, anywhere in the sector. All
flights had to be manually transferred to the next sector, including those delegated to automation.
Automation was capable of performing the following tasks:
•Ensure sufficient separation between automated aircraft (5 NM, 1000 ft),
•Deliver aircraft at their exit point and transfer level, descending as late as possible, and
•Descend arrivals to FL260 to be transferred to lower area control.
When two automated aircraft encountered a conflict, it was always solved in the vertical
plane. Automation would never issue any heading commands or direct-to’s. In case of a human-
automation conflict, it was up to the ATCO to solve it, under the presumption that automation
would not know the ATCO’s intents. Apart from showing the clearances in the aircraft labels,
automation did not provide any feedback on its intentions.
Procedure
After signing a consent form, each participant received a ten-minute training, during
which the automation was introduced and participants familiarized themselves with the interface.
Both a human-automation and automation-automation conflict were shown to demonstrate how
automation would handle both situations. The training was concluded with a short questionnaire.
Next, the measurement run started with a five-minute take-over period, during which no
commands could be issued, followed by 90 minutes of real-time simulation. Each ATCO was
subjected to one allocation suggestion scheme (Table 1), based on flight type or entry sector. The
suggestions were shown by the label color upon sector entry (green = manual, blue = automated),
but the ATCOs were not told which scheme was applied to them. In all cases, they could ignore
the suggestions and re-allocate each flight at any time, even after delegating it to automation.
Throughout the run, an observer asked the ATCOs to explain their actions and what they
were taking into consideration. Every three minutes, the ATCOs rated their instantaneous work-
load by clicking on an on-screen 0-100 scale. After the experiment, they completed an extensive
questionnaire, followed by a radar replay allowing specific situations to be reviewed.
Table 1. Suggested human-automation flight allocation strategies.
ATCO Basic traffic Complex traffic DELTA JEVER
1 Human Automation - -
2 Automation Human - -
3 - - Human Automation
4 - - Automation Human
5 Human Human Human Human
6 Automation Automation Automation Automation
Note. Basic traffic has to descend/climb 2000ft or less in the sector. All other traffic is labelled as complex.
Results and Discussion
Allocation Strategies
All ATCOs delegated at least 50% and up to 100% of traffic to automation, regardless of
the suggested allocation (Fig. 3). Whereas most ATCOs largely ignored the suggested allocation,
ATCO-3 tried to follow it when he realized that one of the sectors was completely handled by
automation. He even delegated/assumed flights as they crossed the border between the two sec-
tors, commenting that solitary manual flights in a predominantly automated area were difficult to
handle. The big drop in automation observed for ATCO-4 around 50 minutes, was caused by him
purposely re-directing flights manually to “test automation” with a more complex scenario. He
stated that he would have been okay with purely monitoring a completely automated scenario.
0 90
0
1
Time, min
Automation fraction
ATCO-1
0 90
Time, min
ATCO-3
0 90
Time, min
ATCO-5
0 90
0
1
Time, min
Automation fraction
ATCO-2
0 90
Time, min
ATCO-4
0 90
Time, min
ATCO-6
Figure 3. Time traces of the fraction of flights allo-
cated to automation (red). The blue line shows the
fraction, if the ATCOs would have followed the sug-
gested allocation on airspace entry (see Table 1).
0.01 0.5 0.75 1
0
1
Min. fraction of flight automated
Share of flights
Threshold Basic Complex
0 ft
2,000 ft
5,000 ft
Figure 4. Cumulative share of flights
that was delegated to automation for a
minimum fraction of their duration, as
a function of level change threshold.
When asked about all allocation strategies from Table 1, the ATCOs unanimously agreed
that complex flights, here defined as requiring more than 2,000 ft level change, need to be han-
dled manually (potentially with support tools). They indicated a strong preference for delegating
basic flights to automation, which is for most ATCOs also reflected in the time that they delegate
such flights to automation (Fig. 4). Although some ATCOs said 5,000 ft would have been a more
appropriate choice of level change threshold, at which traffic was divided in basic and complex,
this is not directly reflected in the figure. All traffic that had to change levels has evoked more
manual control than overflights with zero level change and could thus be considered “complex”.
Apart from this division into basic and complex traffic, the questionnaire provided more
insight into how ATCOs determined whether flights should be delegated or not (Fig. 5). Traffic
directly around the flight was especially important when there were many manual flights and del-
egating a single flight to automation would have added (too) much uncertainty. The suggested
allocation was given low priority, or ignored by most ATCOs (except ATCO-3), as confirmed by
Fig. 3. In general, flights were assumed manually, sent on a direct to their exit point and only del-
egated to automation when clear of conflicts, irrespective of the suggested allocation. If automa-
tion would have been capable of giving directs, the ATCOs would have delegated more flights.
Traffic directly around the aircraft
The type of flight
Traffic along the route of the aircraft
Capabilities of automation
The suggested allocation
The route of the aircraft
My workload
Flight level of the aircraft
The distance of the aircraft to the next sector
1
1
2
3
3
4
5
6
1
1
1
1
1
1
1
1
1
1
1
1
3
1
1
4
22
1
2
1
Most important
Least important
Not considered
Figure 5. Driving factors that made ATCOs decide to delegate flights to automation, or not.
Trust in Automation
At the start of the experiment, all ATCOs reported to have a high level of trust in automa-
tion in general (Fig. 6). Nonetheless, they were suspicious of the experimental automation after
the (short) training. Throughout the 90-minute run their trust increased considerably, according
to the ATCOs mainly due to seeing the automation perform well. The rule-based form of au-
tomation (programmed to be “perfect”), clear separation of responsibilities and absence of un-
certainties, such as wind and pilot behaviour, further contributed to this. ATCOs did, however,
not like the lack of feedback, a common pitfall in automation design hindering the establishment
of human-automation teamwork (Norman (1990)). As automation did not indicate where or when
it would descend aircraft, ATCOs sometimes assumed aircraft manually, solely to prevent them
from descending unexpectedly. All ATCOs would have liked automation to at least show its in-
tentions about where on the trajectory it would start and end a climb or descent.
I trust automation (pre-experiment)
I trust the automation (post-training)
I trusted the automation (post-experiment)
1
2
5
4
12
3
Strongly disagree
Somewhat disagree
Neither disagree nor agree
Somewhat agree
Strongly agree
Figure 6. Trust in automation as reported by the ATCOs.
Task Allocation
While this experiment focused on aircraft allocation, a human-automation team may also
be created by sharing tasks. Four out of six ATCOs included the capabilities of automation in
their allocation strategy (Fig. 5). We replicated part of the study from Prevot et al. (2012), to
see what kind of tasks the ATCOs would like to do themselves, share with automation or com-
pletely delegate to automation. In line with the findings of Prevet et al., the ATCOs indicated
that a considerable number of tasks can be either shared with or completely delegated to automa-
tion (Fig. 7). Transfer of control can be automated as a first step towards more automation, but
ATCOs should be able to reject auto-transfers as well as to initiate early transfers. The ATCOs
prefer to keep short-term, tactical actions manual, while more strategic long-term planning and
routine tasks can be (partially) delegated to automation. Presumably this is because automation
can introduce too much uncertainty in critical short-term situations.
Changing display range (zoom)
Approving weather reroutes
Solving short-term conflicts
Issuing climbs and descends
Coordinating with neighbouring controllers
Solving medium-term conflicts
Detecting short-term conflicts
Transfers of control
Detecting medium-term conflicts
Moving labels (decluttering)
1
1
2
2
2
3
4
4
2
1
4
3
3
4
2
2
2
6
4
4
1
1
1
1
Human
Shared
Automation
Unsure
Figure 7. Allocation of tasks between human and automation as desired by the ATCOs.
Situation Awareness
All ATCOs classified their situation awareness as “okay”, the middle score on a five-point
Likert scale from “poor” to “very good”. Several mentioned that they paid less attention to the
blue automated aircraft, akin to transferred flights, even though they were still responsible for
these flights. At the only (not explicitly programmed) occurrence of a human-automation con-
flict in the experiment, the involved ATCO was surprised by the short-term collision alert and ex-
plained that he had not seen the automated aircraft as it was emerging from, in his words, “a sea
of blue aircraft”. Future experiments with eye trackers could give insight in changing scanning
patterns when aircraft are delegated.
Conclusion
This exploratory study gained useful insights into human-automation teaming in a real-
istic ATC setting. We showed that professional en-route ATCOs are not averse to sharing their
work in a sector with automation. In a simplified situation, lacking uncertainties by wind, emer-
gencies and pilot requests, a high level of delegation to automation was reached, under the con-
dition that flights were on direct routes and free of conflicts. ATCOs generally ignored the sug-
gested allocation, suggesting a need for a different allocation scheme that may be more accepted.
Future research should take a closer look at determining specifically which flights should
be considered “basic” or “complex”, such that a fitting allocation scheme can be applied. Addi-
tionally, the influence of environmental uncertainty (e.g., wind and pilot delays) and automation
capabilities should be researched. Together with empirical studies on the various forms of task
sharing and distribution, this can help establish human-automation teamwork in a shared ATC
environment.
Acknowledgements
The authors would like to express their gratitude to all participating ATCOs, as well as to
MUAC for facilitating an experiment in these challenging times of COVID-19.
References
Bainbridge, L. (1983). Ironies of Automation. Automatica,19(6), 775–779. doi: 10.1016/0005-1098(83)90046-8
Birkmeier, B., Tittel, S., & Korn, B. (2016). Controller Team Possibilities for Sectorless Air Traffic Management. In
ICNS 2016: Securing an Integrated CNS System to Meet Future Challenges (pp. 3–1). Herndon, VA, USA:
IEEE. doi: 10.1109/ICNSURV.2016.7486362
Metzger, U., & Parasuraman, R. (2005). Automation in Future Air Traffic Management: Effects of Decision
Aid Reliability on Controller Performance and Mental Workload. Human Factors,47(1), 35–49. doi:
10.1518/0018720053653802
Norman, D. A. (1990). The ’Problem’ with Automation: Inappropriate Feedback and Interaction, not ’Over-
Automation’. Philosophical transactions of the Royal Society of London. Series B, Biological sciences,
327(1241), 585–593. doi: 10.1098/rstb.1990.0101
Prevot, T., Homola, J. R., Martin, L. H., Mercer, J. S., & Cabrall, C. D. (2012, 2). Toward Automated Air Traffic
Control-Investigating a Fundamental Paradigm Shift in Human/Systems Interaction. International Journal of
Human-Computer Interaction,28(2), 77–98. doi: 10.1080/10447318.2012.634756
SESAR Joint Undertaking. (2019). European ATM Master Plan - Digitalising Europe’s aviation Infrastructure. doi:
10.2829/10044
