You are on page 1of 5

PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 53rd ANNUAL MEETING—2009 1

EFFECTS OF AUTOMATION TYPES ON AIR TRAFFIC CONTROLLER


SITUATION AWARENESS AND PERFORMANCE

Arathi Sethumadhavan
Texas Tech University

The Joint Planning and Development Office has proposed the introduction of automated systems
to help air traffic controllers handle the increasing volume of air traffic in the next two decades
(JPDO, 2007). Because fully automated systems leave operators out of the decision-making loop
(e.g., Billings, 1991), it is important to determine the right level and type of automation that will
keep air traffic controllers in the loop. This study examined the differences in the situation
awareness (SA) and collision detection performance of individuals when they worked with
information acquisition, information analysis, decision and action selection and action
implementation automation to control air traffic (Parasuraman, Sheridan, & Wickens, 2000).
When the automation was unreliable, the time taken to detect an upcoming collision was
significantly longer for all the automation types compared with the information acquisition
automation. This poor performance following automation failure was mediated by SA, with lower
SA yielding poor performance. Thus, the costs associated with automation failure are greater when
automation is applied to higher order stages of information processing. Results have practical
implications for automation design and development of SA training programs.

To meet the increasing demands of air traffic in the when adaptive automation was applied to information
next two decades, the Joint Planning and Development acquisition compared to information analysis, decision
Office has proposed initiatives to modernize the U.S. making, and action implementation. Low LOAs have also
National Airspace System (JPDO, 2007). That proposal been shown to promote faster recovery following
involves introduction of automated systems that would help automation failure (e.g., Kaber, Onal, & Endsley, 2000;
air traffic controllers to cope with increasing traffic Sarter and Schroeder, 2001). For example, using a military
demands. Though fully automated systems offer several decision-making task, Crocoll and Coury (1990) showed
Copyright 2009 by Human Factors and Ergonomics Society, Inc. All rights reserved. 10.1518/107118109X12524440832700

benefits, a potential consequence of such systems is the that when the automation failed, participants who received
out-of-the-loop performance (OOP) problem, which refers only status information (low LOA) performed better than
to the reduced ability of operators working with fully those receiving recommendations (high LOA).
automated systems to perform tasks manually following an
automation failure (e.g., Billings, 1991; Wiener & Curry, EXPERIMENT
1980). Some researchers attribute loss of operator situation
awareness (SA) to be the main factor responsible for the Purpose
OOP problem, bred by overreliance on, and passive
monitoring of automation (e.g., Endsley & Kiris, 1995). The primary purpose of this study was to examine the
Therefore, it is important that automated systems be differences in the SA and collision detection performance
designed that promotes SA and keeps air traffic controllers of individuals when they controlled air traffic using
in control and in the loop. different types of a collision detection system. The PSW
One approach to improve SA and reduce the OOP model of automation (Parasuraman et al., 2000) was
problem is to use varying levels of automation (LOA). The employed in this study. In the information acquisition
central concept of LOA is that automation is not an all or condition, an automated aid supported basic sensory
none phenomenon, but instead can be implemented at processes by cueing the altitude of aircraft using color. In
various levels (e.g., Billings, 1991; Wiener & Curry, 1980). the information analysis condition, the aircraft pairs that
More recently, Parasuraman, Sheridan, and Wickens (2000) were projected to be in conflict were highlighted in the
proposed the PSW model of automation. According to this airspace. In the decision and action selection condition, a
model, automation can be applied to different stages of recommendation to resolve an upcoming conflict between
information processing: information acquisition, aircraft pairs was provided. Finally, in the action
information analysis, decision and action selection, and implementation condition, the automation detected as well
action implementation. In addition, different LOAs can be as resolved upcoming conflicts and provided feedback to
applied to each of these information processing stages. the participants. Though prior research has examined the
Low LOAs have been shown to promote operator SA. extent to which human performance varies under different
For example, using a simulated air traffic control automation conditions in domains such as military (e.g.,
environment, Kaber, Perry, Segall, McClernon, and Prinzel Crocoll & Curry, 1990), robotics (Kaber et al., 2000), and
(2006) showed that individuals were more accurate in driving (e.g., Endsley & Kiris, 1995), there has been
answering perception-based questions about aircraft states limited work (e.g., Kaber et al., 2006) conducted to
Downloaded from pro.sagepub.com at PENNSYLVANIA STATE UNIV on May 12, 2016
PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 53rd ANNUAL MEETING—2009 2

examine the effect of types and levels of automation in the call signs of the aircraft involved in the upcoming collision.
air traffic control context. Examining air traffic controller The advance notification time, which referred to the time a
performance consequences under different automation potential collision was scripted to occur in the scenario
conditions is important to design automated aids for the minus the time the participant detected the potential
future airspace system. collision by pressing the conflict button (Metzger &
This study also examined whether the relationship Parasuraman, 2005) was recorded by the ATST simulator.
between the type of automation and performance following The participants were also instructed to follow the
automation failure was mediated by SA. Though prior rules of the simulation environment. They were told to land
studies (e.g., Endsley & Kiris, 1995, Kaber et al., 2000) aircraft at airports at speed slow and altitude 1 and exit
have attributed the loss of SA to be the central factor aircraft through sector gates at speed fast and level 3. In
responsible for the OOP problem, these studies have failed addition, participants were also instructed to activate the
to demonstrate whether the loss of SA is responsible for aircraft that appeared in their airspace as soon as possible in
performance degradation following automation failure. order to reduce handoff delays.
Secondary Task. In addition to the primary task of
Method controlling air traffic, participants were given an additional
task of monitoring a weather display. They were told that
Participants. Seventy- two individuals (27 men and 45 the weather monitoring task was secondary to the primary
women) ranging in age from 18 to 30 years (M = 21.11, SD task of controlling traffic. Consistent with prior research
= 3.62) participated. (e.g. , Kaber & Endsley, 2004; Metzger & Parasuraman,
Simulation Software. Air traffic scenarios were 2005), the purpose of introducing this task was to induce
simulated using a modification of the Air Traffic Scenarios overreliance on the automated aids provided in the air
Test (ATST), a low fidelity simulation of the radar display traffic control task as well as to be representative of a
(See Figure 1). The airspace consisted of four sector gates to multi-task environment like air traffic control where
neighboring sectors (A, B, C, and D) and two airports (x, y). controllers have to exhibit good performance in multiple,
E, F, G, H, and I were waypoints, through which an aircraft concurrent tasks. This task was presented on a laptop
had to pass to get to its final destination. An aircraft could fly placed adjacent to the radar display. A series of numbers
at three possible speeds (Fast, Medium, or Slow) and at any between 100 and 199 appeared at 3-second intervals.
of the three altitudes (1, 2, or 3). Aircraft were identified by Participants were told that a number that was less than 130
call signs (e.g., 5, 13). or greater than 170 represented a wind shear warning.
Hence, they were told to press the spacebar key on the
laptop when they saw a number less than 130 or greater
than 170. Not more than 3 adjacent numbers were signals
(requiring a key press) or noise. The hit-to-signal ratio (the
number of hits/ (number of hits + misses) and the number
of false alarms were recorded.
Experimental Design. The independent variable in this
study was the automation type, which served as the
between subject grouping variable: information acquisition,
information analysis, decision and action selection, and
action implementation.
In the information acquisition condition, an automated
aid displayed aircraft that were flying at the same altitude
in the same color in the ATC display. Specifically, all the
aircraft flying at altitude 3, 2, and 1 were presented in the
airspace in green, pink, and blue respectively. Research in
visual search tasks have shown that color can be
Figure 1. Snapshot of the ATST airspace. preattentively processed (e.g., Treisman, 1986), thereby
helping controllers to assess altitude, an important piece of
The Air Traffic Control Task. Participants were information that needs to be monitored to project whether
instructed to avoid collisions during the scenarios. collisions would occur, without having to pay focused
Collisions occurred when two aircraft flying at the same attention to datablocks (e.g., Johnston, Horlitz, &
altitude were within 5 miles of each other. They were Edminston, 1993). Thus, a moderate LOA (e.g.,
instructed to avoid collisions by making an altitude Parasuraman et al., 2000) that highlights important
separation. Participants were told to press a button labeled information (i.e., altitude) was applied to the information
‘conflict’ on the ATC display when they detected a acquisition stage of information processing. In the
potential collision. They were also asked to say aloud the information analysis condition, an automated aid
Downloaded from pro.sagepub.com at PENNSYLVANIA STATE UNIV on May 12, 2016
PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 53rd ANNUAL MEETING—2009 3

highlighted the aircraft pairs that were projected to be in them develop appropriate mental models that would help
conflict. Specifically, a high LOA (e.g., Parasuraman et al., them to take over manual control when an automation
2000) that integrates information from multiples sources failure occurs (e.g., Kessel & Wickens, 1982; Moray,
(i.e., altitude, heading, location, and destination) and 1986).
projects future events (i.e., collisions) was applied to the After completing the manual test scenarios,
information analysis stage of information processing. In the participants were randomly assigned to one of the
decision and action selection condition, an automated aid automation types. The participants first received
recommended the course of action (i.e., altitude separation) instructions on the functionalities of the automated aid.
to resolve a potential collision. Specifically, level 4 in the They were informed that the automated aids were highly
Sheridan and Verplank (1978) LOA taxonomy (i.e., reliable, though not 100% reliable and that even if the
‘Computer suggests one alternative’) was applied to the automation fails it was their responsibility to monitor the
decision and action selection stage of information aircraft in their airspace and avoid collisions. They then
processing. In the action implementation condition, an completed several training scenarios in which they were
automated aid detected potential collisions and assisted by the automation. During these scenarios, the
autonomously resolved these collisions by making an automation did not fail so as to build up participants’ trust
altitude separation. Specifically, level 7 in the Sheridan and on the automated aids. This was consistent with the
Verplank (1978) LOA taxonomy (i.e., ‘Computer executes methodology adopted in earlier studies (e.g., Kaber et al.,
selected option and informs the human’) was applied to the 2000; Rovira, McGarry, & Parasuraman, 2007).
action implementation stage of information processing. After completing the training scenarios, participants
The dependent variables were SA, ATC task completed a 15-minute experimental scenario. This
performance, and secondary task performance. SA was scenario involved a total of 12 scripted collisions. The
measured using a modification of the Situation Awareness different automation types were perfectly reliable in
Global Assessment technique (SAGAT; Endsley, 1995). detecting the first 11 upcoming collisions. The automated
The simulation froze twice during the scenario and the aids failed to detect the 12th scripted collision. That is, in
participants were asked to recall the location, call sign, the information acquisition condition, the aid failed to
altitude, heading, and destination of all the aircraft present present the data block of one of the aircraft that would be
in the airspace. Aircraft speed was omitted from the involved in an upcoming collision in the correct color. In
analysis because the aircraft flew at speed fast at virtually the information analysis condition, the aid failed to
all times. The responses during the first and second highlight the aircraft pairs that would be involved in the
SAGAT freezes were averaged to obtain an overall upcoming collision. In the decision and action selection
proportion accuracy score for location, call sign, altitude, condition, the aid failed to provide the recommendation to
heading, and destination. The proportion accuracy scores resolve the collision. Finally, in the action implementation
for all the aircraft attributes were averaged to obtain the condition, the aid failed to detect and resolve the collision.
total SA score. The time taken by participants working with different
The air traffic control performance variable of interest automation types to detect the 12th scripted collision (i.e.,
was the advance notification time (Metzger & advance notification time) following the failure of the
Parasuraman, 2005). Advance notification time was automation was recorded. The scenario ended 8 seconds
computed as the time a planned collision would occur (or after the occurrence of the 12th scripted collision.
would have occurred) in the scenario minus the time the SA of the participants was assessed twice during the
participant reported the collision by pressing the conflict scenario. Both SA assessments were conducted prior to the
button. In addition to performance in the air traffic control occurrence of the automation failure so as to determine the
task, the performance in the secondary task was also extent to which pre-automation failure SA mediated the
assessed. Specifically, the hit-to-signal ratio and the relationship between automation type and collision
number of false alarms were recorded. detection performance following automation failure.
Procedure. The experiment lasted for approximately 3
hours. Participants first watched an instructional video that Results and Discussion
demonstrated the layout of the airspace and then
memorized the map of the airspace. Thereafter, they Air Traffic Control Performance Following
watched instructional videos that demonstrated the rules to Automation Failure. A one-way Analysis of Variance
control traffic in the ATST simulator. They then completed (ANOVA) with the type of automation as the between
manual training in the air traffic control task. During this subject variable was conducted on advance notification
phase, they first learned to control air traffic and following automation failure. The effect of automation type
subsequently to perform the air traffic control task in on advance notification time was significant, F (3, 68) =
conjunction with the secondary task. The purpose of 16.507, p < .001, ηp2= 0.421. Tukey’s HSD analysis
providing manual training to the participants before showed that individuals working with information
assigning them to work with an automated aid was to help acquisition automation were significantly faster in detecting
Downloaded from pro.sagepub.com at PENNSYLVANIA STATE UNIV on May 12, 2016
PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 53rd ANNUAL MEETING—2009 4

the upcoming collision compared to individuals working analysis automation, decision and action selection
with all the other automation types, following the automation, and action implementation automation were
automation failure (See Figure 2). No significant 0.514 (SD =.101), 0.397 (SD =.085), 0.416 (SD =.085), and
differences in the advance notification time were observed 0.383 (SD =.125), respectively.
between the individuals working with the information In order to determine whether the relationship between
analysis automation (95% CI = 18.842, 37.368), decision automation type and the advance notification time
automation (95% CI = 7.173, 25.699) and action following automation failure was mediated by SA, the
implementation automation (95% CI = 4.907, 23.433). methods developed by Baron and Kenny (1986) were used.
However, a less conservative comparison between the Four linear regression analyses were conducted to test for
groups using planned contrasts indicated that individuals in mediation: automation type predicting SA, SA predicting
the information analysis condition were significantly faster advance notification time, automation type predicting
in detecting the collision following the automation failure advance notification time, and finally automation type and
than those in the decision and action implementation SA predicting advance notification time (See Figure 3).
conditions. This suggested that automation of information Sobel test (1982) statistics revealed significant p value for
analysis had some benefits over automation of decision the indirect effect of SA as a mediator between automation
making and action execution. type and advance notification time following automation
Effect of Automation Type on Advance Notification Time
failure (z = -2.287, p < .05). Thus, SA mediated the
(Following automation failure) relationship between automation type and advance
notification time following automation failure, with higher
70
SA before automation failure contributing to earlier
Advance Notification Time (in seconds)

60 detection of the collision following automation failure.


50 SA (Mediator)
40

30 β = -.379 ** β = .488***

β = -.6***
20

10 Without
0
Automation mediation Advance
uisi
tion naly
sis
lect
ion
ntat
ion Type β = -.484*** notification time
A cq atio
nA n Se eme
at ion r m A ctio I mpl
Info
rm Inf o
isio
n an
d
Act
io n
With
D ec
mediation
Automation Type
Figure 3. Mediation model of the association between
automation type and advance notification time following
Figure 2. Effect of automation type on advance notification automation failure as mediated by SA. Standardized
time following automation failure. The ideal advance regression coefficients are presented in the figure. ***p <
notification time is 70 s. Error bars indicate + 1 standard .001, **p < .01.
error of the mean.
Secondary Task Performance. A one-way ANOVA on
In short, when the automation failed, individuals the hit-to-signal ratio was significant, F (3, 68) = 12.087, p <
working with information acquisition automation detected .001, ηp2 = 0.348. Tukey's HSD analysis showed that
the upcoming collision earlier in comparison to the other individuals working with information analysis (M = 0.906,
automation aids. Thus, automation of sensory processing SD = 0.070), decision (M = 0.933, SD = 0.047), and action
had benefits over automation of higher order cognitive implementation automation (M = 0.939, SD = 0.032) had
functions such as prediction generation and decision higher hit-to-signal ratio compared to those working with the
making. This meant that decreasing the level of operator information acquisition automation (M = 0.818, SD = .103).
involvement can prove hazardous if the automation A one-way ANOVA on the number of false alarms
reliability is not 100%.
failed to reach significance, F (3, 68) = 1.167, p > .05, ηp2
SA. A one way ANOVA with type of automation
=0.049. Results of binomial probability calculations showed
as the between subjects variable on the total SA score was
that the number of false alarms for each automation type was
significant, F (3, 68) = 6.304, p < .01, ηp2= 0.218. Tukey’s significantly less than chance probability. These analyses
HSD analysis revealed that individuals working with helped to establish that participants were attending to the
information acquisition automation had higher overall SA secondary task and not merely pressing the spacebar key.
compared to those working with all the other automation The superior performance in the secondary task by
types. The mean proportion accuracy score for total SA for individuals working with information analysis, decision,
the information acquisition automation, information
Downloaded from pro.sagepub.com at PENNSYLVANIA STATE UNIV on May 12, 2016
PROCEEDINGS of the HUMAN FACTORS and ERGONOMICS SOCIETY 53rd ANNUAL MEETING—2009 5

and action implementation automation can be attributed to their valuable comments. Special thanks to Dr. Chris Wickens
the overreliance they exhibited on the collision detection for his comments on this paper.
aids provided to them in the air traffic control task, due to
which they had more cognitive resources or chose to REFERENCES
allocate more cognitive resources to the secondary task.
Baron, R.M., & Kenny, D. A. (1986). The moderator-mediator variable
distinction in social psychological research: Conceptual, strategic,
CONCLUSIONS and statistical considerations. Journal of Personality and Social
Psychology, 51, 1173-1182.
This study provides recommendations for automation Billings, C. E. (1991). Human-centered aircraft automation: A concept and
guidelines (NASA Tech. Memorandum 103885). Moffet Field,
design for the future National Airspace System by applying CA: NASA Ames Research Center.
the PSW model of automation. Based on this model, in Crocoll, W. M., & Coury, B. G. (1990). Status or recommendation: Selecting
order to determine the appropriate degree of automation the type of information for decision aiding. In Proceedings of the
Human Factors and Ergonomics Society 34th Annual Meeting (pp.
that will keep the operator involved, the consequences for 1524-1528). Orlando, FL: Human Factors and Ergonomics
human performance when interacting with automated Society.
systems applied to different stages of information Endsley, M. R. (1995). Measurement of situation awareness in dynamic
systems, Human Factors, 37, 65-84.
processing have to be examined (Parasuraman et al., 2000). Endsley, M. R. & Kiris, E. O. (1995). The out-of-the-loop performance
This study revealed that when 100% reliable automation problem and level of control in automation. Human Factors, 37,
was not provided, automating sensory processing was 381-394.
Johnston, J. C., Horlitz, K. L., & Edmiston, R. W. (1993). Improving situation
beneficial in helping individuals detect an upcoming awareness displays for air traffic controllers. In R. S. Jensen & D.
collision earlier following automation failure in comparison Neumesiter (Eds.), Proceedings of the 7th International Symposium
to automation of information analysis, decision making, on Aviation Psychology (pp. 328-334). Columbus, OH: Ohio State
and action implementation, by improving their SA. In University.
JPDO (2007). Concept of Operation for the Next Generation Air
summary, applying a moderate LOA (i.e., cueing aid) to the Transportation System. Retrieved October 29, 2008, from,
information acquisition stage can be pursued in the future http://www.jpdo.gov/library/nextgen_v2.0.pdf
airspace system as long as the automation reliability is Kaber, D., & Endsley, M. (2004). The effects of level of automation and
adaptive automation on human performance, situation awareness
high. Although applying high LOAs to information and workload in a dynamic control task. Theoretical issues in
analysis, decision and action selection, and action Ergonomic Science, 5, 113–153.
implementation stages can provide benefits such as Kaber, D. B., Onal, E., & Endsley, M. R. (2000). Design of automation for
telerobots and the effect on performance, operator situation
enhancing performance in secondary tasks that controllers awareness, and subjective workload. Human Factors and
have to perform, the risk associated with their failure is Ergonomics in Manufacturing, 10, 409-430.
high. Kaber, D.B., Perry, C. M., Segall, N., McClernon, C. K., & Prinzel, L.J.
(2006). Situation awareness implications of adaptive automation
for information processing in an air traffic control-related task.
LIMITATIONS AND FUTURE DIRECTIONS International Journal of Industrial Ergonomics, 36, 447-462.
Kessel, C. J., & Wickens, C. D. (1982). The transfer of failure-detection skills
between monitoring and controlling dynamic systems. Human
Caution should be exercised in applying the results of Factors, 24, 49-60.
this study to the ATC domain. The background of the Lee, J., & See, J. (2004). Trust in automation: Designing for appropriate
participants and the fidelity of the simulation environment reliance. Human Factors, 46, 50-80.
limit my ability to generalize the findings of this study to Metzger, U., & Parasuraman, R. (2005). Automation in future air traffic
management: Effects of decision aid reliability on controller
the air traffic control domain. Future work should thus be performance and mental workload. Human Factors, 47, 35-49.
directed towards examining the costs associated with Moray, N. (1986). Monitoring behavior and supervisory control. In K. Boff
automation failure with professional controllers who have (Ed.), Handbook of perception and human performance (pp. 40/1-
40/51). New York: Wiley.
had extensive manual experience in the air traffic control Parasuraman, R., Sheridan, T. B., & Wickens, C. (2000). A model for types
domain. Future work should also examine whether and levels of human interaction with automation. IEEE
explaining the situations under which the automation Transactions on Systems, Man, and Cybernetics, 30, 286-297.
Rovira, E., McGarry, K., & Parasuraman, R. (2007). Effects of imperfect
algorithm can fail and making automation failures more automation on decision making in a simulated command and
salient by providing the intermediate results of the control task. Human Factors, 49, 76-87.
automation algorithm and designing more intuitive displays Sarter, N. B., & Schroeder, B. (2001). Supporting decision making and action
selection under time pressure and uncertainty: The case of in-flight
(e.g., Lee & See, 2004; Metzger & Parasuraman, 2005) can icing. Human Factors, 43, 573-583.
help to reduce the misuse of information analysis, decision, Sheridan, T. B., & Verplank, W. L. (1978). Human and computer control of
and action implementation automation. undersea teleoperators. Cambridge, MA: MIT Man-Machine
Laboratory.
Sobel, M. E. (1982). Asymptotic confidence intervals for indirect effects in
ACKNOWLEDGMENT structural equations models. In S. Leinhart (Ed.), Sociological
methodology (pp. 290-312). San Francisco: Jossey-Bass.
This work was part of the author’s dissertation. The Treisman, A. (1986). Features and objects in visual processing. Scientific
American, 255, 114–125.
author would like to thank her academic advisor, Dr. Frank Wiener, E. L., & Curry, R. E. (1980). Flight deck automation: Promises and
Durso and the other members in her dissertation committee for problems. Ergonomics, 23, 995-1011.

Downloaded from pro.sagepub.com at PENNSYLVANIA STATE UNIV on May 12, 2016

You might also like