causse thesis summary - free
TRANSCRIPT
PhD Thesis
INSTITUT SUPÉRIEUR DE l’AÉRONAUTIQUE ET DE L’ESPACE
Mickaël Causse
Influence of age and reward on piloting performance:
a neuroergonomics contribution to aeronautic safety
Doctoral School : Aéronautique Astronautique
Thesis advisor:
Josette Pastor
Frédéric Dehais
General abstract
Traditionally, the human-system interactions (HIS) have been studied through the modification of
the human behavior in response to system variations. Although this approach has opened the way to
great progresses in HIS framework, especially when observations led to descriptive modeling, an
important part of the interaction remains unknown. Indeed, while, for instance in military
aeronautics, the physiological constraints were taken into account (eg. G-lock), the cognitive
constraints inherent to the brain functioning were largely unknown. In this perspective, human error
becomes a symptom of the system deficiency. An innovative and original approach to cope with the
problematic of the HIS is to use the tools and concepts of the integrative neurosciences and of the
neuropsychology. These researches, based on the study of the neural substrate of cognitive
mechanisms, have made the knowledge related to the man-environment interaction grow. These
scientific results arouse human factor interest and generated a new discipline in the USA:
neuroergonomics.
In this work, we aimed at applying this approach to better understand the pilot performance,
especially its degradation. We tried to comprehend the similarity of executive functions (EFs)
impairments that appear in the diminished subject (neurological pathology, sleep deprivation, effects
of some medications etc.) and in the healthy pilot facing degraded conditions. Indeed, piloting errors
may be, in some cases, attributed to temporary perturbations of the EFs. These findings plead in
favor of the hypothesis of common cerebral mechanisms and suggest the existence of a cognitive
continuum, spanning from optimal intellectual performances to a degraded functioning, as found in
pathological subjects. Therefore, the hypothesis of the cognitive continuum leads us to focus on the
deleterious impact of aging and emotion on the pilot’s performance.
We have performed a first study aiming at understanding the cross influences of the risk and the
reward on decision making in aeronautics. A first experiment has used neuroimaging tools (fMRI) in a
neuroeconomics experimental protocol (See part 1). The behavioral results showed an increase of
risky decisions under the influence of the reward. The neuroimaging preliminary results point out
that the shift from “cold” reasoning to “hot" reasoning has resulted in a transfer from dorsolateral
prefrontal cortex (main center of the EFs) activations toward ventromedial prefrontal cortex
(strongly connected to emotional brains areas) activations. A separate psychophysiological
experiment was conducted with the same neuroeconomics experimental protocol (see part 2). The
behavioral results confirmed the increased risky decision under the financial incentive. In addition,
this behavioral shift was associated with an increased heart rate, sign of the emotional impact of the
reward.
We have also conducted a research to assess the subtle perturbations of the EFs during normal
aging so as to identify predictive clues of the ability to pilot in general aviation, where no age limit is
defined (see part 3). The results showed that reasoning, working memory abilities and flight
experience were predictive of the piloting performance. The most the reasoning and the working
memory abilities were efficient, the smaller the flight path deviation (FPD), amount of angular
deviation in X axes from the ideal flight path, was. In the same way, the most the pilots were
experienced, the smaller the FPD was. The adjusted r² showed that our model explained 44.51% of
the variation of the FPD. Besides, the aging effects on cognitive functioning were examined and the
predictibility of this latter, plus age and flight experience, on the relevance of a decision-making
performed during landing was assessed. The results confirmed the strong vulnerability of EFs to the
aging effects, except the reasoning abilities that were somewhat preserved in our sample of pilot. In
addition, the working memory, flight experience and trait impulsivity were predictive of poor
decision-making. The pilots who made the good decision to go-around demonstrated a better
percentage of correct answers in the 2-back task, a larger total flight experience and a lower motor
impulsiveness compared to pilots who made a poor decision.
Finally, we have operationalized our results by experimentations carried out in real flight with
oculometric means (see part 4). An eye tracker was embedded in a light aircraft and a flight scenario
was defined. In addition, two different landing were performed, a nominal one and a degraded one
(with a simulated engine failure). The study revealed the feasibility of collecting visual information
taking of pilots in real flight, and the pupil dilation indicated stress and mental load effects. The pilots
spent less time glancing at the instruments, and focused on fewer instruments in the degraded
condition. Moreover, the pupil size varied with the flight phases in the degraded condition, which
reflected the variations of stress and attention levels. These encouraging results open two tracks: the
development of new eye trackers able to overcome current technical limitations, and
neuroergonomics researches providing guidelines for new man-machine interfaces integrating both
flight and crew state vectors.
All theses researches illustrate the capacity of neuroergonomics to grow beyond the disciplines
from which it emerges, in particular human factors.
Part 1
Economic issues provokes hazardous landing decision-making by enhancing the activity of
"emotional" neural pathways
Submitted to ICRAT 10
Mickaël Causse
Centre Aéronautique et Spatial ISAE-SUPAERO; Inserm; Imagerie cérébrale et handicaps neurologiques UMR 825; F-31059 Toulouse, France
Université de Toulouse; UPS; Imagerie cérébrale et handicaps neurologiques UMR 825; CHU Purpan, Place du Dr Baylac, F-31059 Toulouse Cedex 9, France
Frédéric Dehais Centre Aéronautique et Spatial ISAE-SUPAERO; Université de Toulouse, France
Patrice Péran Laboratoire de Neuroimagerie, IRCCS Fondation Santa Lucia, Rome, 00149, Italie
Umberto Sabatini Laboratoire de Neuroimagerie, IRCCS Fondation Santa Lucia, Rome, 00149, Italie
Josette Pastor Inserm; Imagerie cérébrale et handicaps neurologiques UMR 825; F-31059 Toulouse, France
Université de Toulouse; UPS; Imagerie cérébrale et handicaps neurologiques UMR 825; CHU Purpan, Place du Dr Baylac, F-31059 Toulouse Cedex 9, France
Abstract— The analysis of aeronautical accidents highlights the fact that some airline pilots demonstrate a
trend to land whereas the approach is not well stabilized. This behavior seems to be the consequence of
various factors, including financial issues. Our hypothesis is that financial constraints modulate the brain
circuitry of emotion and reward, in particular via the interactions between two prefrontal structures: the
dorsolateral prefrontal cortex (DLPFC), main center of the executive functions (EFs), high level cognitive
abilities, and the ventromedial prefrontal cortex (VMPFC), structure linked with the limbic system, major
substratum of emotional processes. In our experiment, participants performed a simplified task of landing
in which the level of uncertainty and the financial incentive was manipulated. A preliminary behavioral
experiment (n = 12) was conducted. A second experiment using fMRI is in progress and a case study only
is reported here. The behavioral data showed that the participants made more risky decision to land in the
financial incentive condition in comparison to the neutral condition, where no financial incentive was
delivered. This was particularly true when the uncertainty was high. The functional neuroimaging results
showed that the reasoning performed in neutral condition resulted in enhanced activity in DLPFC. On the
contrary, under the influence of the financial incentive, VMPFC activity was increased. These results
showed the effectiveness of the financial incentive to bias decision-making toward a more risky and less
rational behavior from a safety point of view. Functional neuroimaging data showed a shift from cold to
hot reasoning in presence of the financial incentive, suggesting that pilot erroneous trend to land could be
explained by a temporary perturbation of the decision-making process due to the negative emotional
consequences associated with the go-around.
Keywords: decision making, emotion, reward, piloting performance.
I. INTRODUCTION
Approach and landing are critical flight phases. They require formalised sequences of actions
(e.g. to put and lock the gear down, to extend the flaps) and to follow an arrival procedure through
several waypoints. They also require decision-making processes based upon rational elements like
the maximum crosswind speed for a given aircraft. Uncertainty, a worsening factor since it generates
psychological stress, can be high during landing. Moreover, several psychosocial factors lead some
pilots to irrational decision-making, such as keeping landing on whereas all safety parameters are not
respected [1]. According to the legislation, such hazardous conditions (e.g. thunderstorm, heavy rain,
strong crosswind or windshear) require to go-around and to perform a new attempt to land more
securely or a diversion to another airport. A study conducted by the MIT [2] demonstrates that in
2000 cases of approaches under thunderstorm conditions, two aircrews out of three keep on landing
in spite of incompatible meteorological conditions. This phenomenon, called plan continuation error
[3], also exists in general aviation. Indeed, the BEA (the French national institute for air accident
analysis) revealed that this pilots’ trend to land (the get-home-itis syndrome) have been responsible
for more than 41.5 percent of casualties in light aircrafts [4].
II. ECONOMIC PRESSURES AND LANDING DECISION
Many experiments have addressed the difficulty for pilots to revise their flight plan and several
cognitive and psychosocial explanatory hypotheses have been put forward [5] [6] [7] [8]. Another
explanation to this trend to land in spite of bad meteorological conditions or an unstabilized approach
may resides in the impact of a large range of aversive consequences associated with the decision to
go-around. Indeed, a go-around generates a stress in the crew and the passengers, the pilot can feel it
like a failure and it may lead to difficulties to reinsert the aircraft in the traffic. Moreover, a go-around
has negative financial consequences for the company (fuel consumption in particular). An
organisation’s emphasis on productivity may unconsciously set up goal conflicts with safety. The
culture of the company weighs on security: if it attaches a negative connotation to a go-around, it is
an excellent candidate for landing accidents. One now-defunct airline used to pay passengers one
dollar for each minute their flight was late until a crew attempted to land through a thunderstorm and
crashed [9]. According to Orasanu [8], companies also emphasize fuel economy and getting
passengers to their destinations rather diverting the flight, perhaps inadvertently sending mixed
messages to their pilots concerning safety versus productivity. Those blurred messages create
conflicting motives, which can affect unconsciously pilots’ risk assessments and the course of action
they choose. All these emotional pressures could alter the rational reasoning by shifting decision-
making constraints from safety rules to economic optimization.
III. FROM COLD TO HOT DECISION MAKING
Neglected during the first half of the 20th century, the role of emotion on cognitive functioning has
been recently fully established in the cognitive neurosciences. According to Koenig and Sander [10],
this historical neglect of emotion is explained by the difficulty inherent to its investigation and by the
influence of a scientifically-correct Cartesianism that considered the cognitive system as the
“incarnation of reason”. Today, it is commonly admitted that experiencing an emotion can trigger
unconscious processes useful to decision-making, in particular when the uncertainty is high [11]. Many
experiments put forward evidence of a strong interaction between the limbic system, “emotional
brain”, and other structures like the prefrontal cortex, the “rational brain”. For instance, Drevet &
Raichle [12] have shown the existence of a dynamic balance between regions of the limbic system
(amygdala, posteromedial cortex, ventral anterior cingulate cortex) and regions more associated to
EFs (dorsal anterior cingulate cortex, DLPFC). Similarly, Mayberg and colleagues [13] have put in
evidence that an increased activity of limbic and paralimbic regions (subgenual cingulate, anterior
insula) was proportional to the decrease in activity of neocortical regions (right prefrontal cortex,
inferior parietal cortex) during the experience of sadness. These types of observations are supported
by a study of Mitchell [14], which demonstrated that the activity of the DLPFC was inversely
proportional to the VMPFC. A previous study of Goel and Dolan [15] have also highlighted this type of
emotional and cognitive subdivision in the prefrontal cortex (PFC) in a reasoning task. In this study,
when the reasoning task was performed without emotional induction, cold reasoning, DLPFC
activations were found. When the same task was performed with emotional induction, hot reasoning,
VMPFC activations were observed. All these studies allow to understand how emotion or stress are in
relation to cognitive functions and how they can modulate the cognitive performance, in particular
the EFs [16], mainly implemented within the PFC.
Our hypothesis is that plan continuation error may be, at least in part, related to a shift from cold
reasoning to hot reasoning, in result of the different negative emotional consequences associated
with the decision to go-around. Cold reasoning may be supported by EFs whereas hot reasoning may
be less rational from a safety point of view and more oriented toward company’s financial interest. In
a fully neuroergonomics approach, we propose to investigate this hypothesis by reproducing a
simplified landing task performed in an fMRI.
IV. METHODS
A. Subjects
Two separate experiments were conducted. 12 physically and psychiatrically healthy volunteers
were recruited from the local population to participate to the behavioral experiment (age: mean = 28,
SD = 3.69). 1 participant (age 28) was scanned in the fMRI. All subjects were right handed as measured
by the Edinburgh handedness inventory [17]. The experiment was approved by the local ethic
committee and an informed consent was obtained before participation.
B. From aicraft to fMRI
The task was based on a simplified reproduction of a real flight instrument, the ILS (Instrument
Landing System). The participants were instructed that they were flying a plane during the landing
phase and that like pilots, they were allowed to avoid landing if they believed that landing was unsafe
(Figure 1). Decisions were based on two elements of the ILS: the localizer and the glide path, which
provide lateral and vertical guidance to adjust the trajectory of the aircraft to the runway. This
information was given by two rhombuses, like in real life, displayed below and on the right of the PFD
(Primary flight Display). It was explained to participants that the landing was safe when both
rhombuses were close to the center of their axes, the farthest from the center the rhombuses were,
the higher was the risk of crash. For each trial, the participants indicated their responses by pressing a
button on the response pad. A first independent variable with two modalities was the degree of
uncertainty, high or low, linked with the rhombuses position. The second independent variable was
the type of incentive, neutral or financial. During the neutral condition, the incentive was only based
on a feedback that gave information on the accuracy of the responses. During the financial condition,
a financial incentive was added to the one that gave feedback on the accuracy of the responses. The
task consisted of a set of 4 runs, 2 neutral, and 2 financial in which the level of uncertainty was
manipulated according to the two modalities (high and low).
Figure 1. Simplified reproduction of the decision-making performed during the landing phase. In the upper part, the real environment. From left to right: the PFD within the cockpit, a zoom on the PFD and the throttle. In the bottom part, the experimental environment. From left to
right: the fMRI, the simplified PFD (with only the two rhombuses of the ILS) and the response pad that allowed to decide to accept to land or to go-around.
C. Stimuli
1) The ILS
The stimuli was based on a 480x480 pixels simplified PFD with ILS and they reproduced a landing
situation without external visibility. During the scan, they are displayed via back projection and an
angled mirror in the head coil housing. Two different levels of uncertainty, depending of the positions
of the two rhombuses, are randomly sorted within the 4 runs. In landings with low uncertainty, the
decision-making was straightforward: either the rhombuses were very far from their respective
center, requiring a go-around (likelihood of successful landing: 0%), either they were very close,
requiring a landing acceptance (likelihood of successful landing: 100%). In landings with high
uncertainty, rhombuses were borderline (not very far, not very close from the center) and the
likelihoods (unknown by the subjects) of a successful landing or a crash is 50%. The positions of the
two rhombuses were related to a score. Each axis was graduated with a 16 points scale, the most the
rhombuses were far from the center, the higher was the score and the weaker was the likelihood to
land securely (Figure 2).
Figure 2. Categorization of the level of uncertainty according to the rhombuses positions. A score was computed according to the position of
both rhombuses (the zoom gives information on only one rhombuse, the graduation was not displayed during the experimentation). The rhombuses position were counterbalanced to avoid laterality effects. The order of presentation of the stimuli was randomized.
2) Feedback
After each response, the participants received a feedback that informed on the response accuracy
(OK, for a successful landing or a justified go-around; NO, for an erroneous decision to land or an
unjustified go-around). During the condition with financial incentive, a second feedback gave
information about the financial consequences of the decision (Figure 3). At the end of each run, a
global feedback indicated the percentage of correct responses (safety score). Concerning the run with
financial incentive, another feedback indicated the cumulative amount of money won or loss (financial
score).
Figure 3. The various feedbacks displayed after each decision making. Without incentive, only the accuracy feedback was delivered (ok/no),
with financial incentive, the monetary consequences are also displayed after the accuracy feedback.
D. The payoff matrix
During the financial incentive condition, negative emotional consequences associated with a go-
around have been reproduced by a payoff matrix. This matrix was set up to bias responses in favor of
landing acceptance. A go-around was systematically punished by a financial penalty. The penalty was
less important (-2€) when the go-around was justified (in the case where rhombuses were very far
from their center) than when it was unjustified (-5€). This systematic punishment of the decision to
go-around reproduced the systematic negative consequences associated with this latter in real life. A
successful landing was rewarded (+5€) whereas an erroneous decision to land was punished (+5€).
The fact that the erroneous decision to go-around was less punished that the erroneous decision to
land may appears counterintuitive but the matrix was set up in this way for two reasons. Firstly, in real
life, the pilots know that crash and overrun are rather unlikely events whereas the negative
consequences associated with a go-around are systematic. The analysis of unstabilized approach
confirms that accidents are rather rare in spite of frequent risk taking [2]. Secondly, we could not
reproduce the low frequency of accidents in an fMRI experiment because the cerebral signal
associated with rare events could not emerge from a statistical point of view. For this reason, we were
compelled to modulate the weight of the punishment rather than the frequency (Figure 4).
Case GO GO-A
Choice
GO +5€ -2€
GO-A -5€ -2€
Figure 4. Payoff matrix biased in favor of landing acceptance. A successful landing pays 5€, an erroneous decision to land costs 2€, a justified go-around costs 2€ and an unjustified one costs 5€.
E. Experimental design for the behavioral study
We used a 2x2 factorial design crossing the financial incentive (neutral and financial) and the
uncertainty (high, 50% chance of crash, or low, 100% or 0% chance of crash). Stimulus display and
data acquisition were done with Cogent 2000 v125 running under Matlab environment (Matlab
7.2.0.232, R2006a, The MathWorks, USA). Two types of runs were presented during the experiment:
neutral and financial ones. There were three likelihoods (0%, 100%, 50%) of successfully landing (40
trials for 0%; 40 trials for 100% and 80 trials for 50%), depending on both positions of the rhombuses
displayed on the PFD. These likelihoods were unknown by the subjects. Each trial consisted in a
presentation of the stimulus (3 s) during which the participant performed his decision thanks to a
response pad, followed after a delay (5.5 s) by the feedback informing of the accuracy of the response
(2 s). During the incentive condition the financial outcome was also displayed ({+5€}, {-5€} or {–2€}.
Finally, an inter trial interval (2 s) was introduced. Response and financial feedback delivery was
contingent upon the subject’s response
F. Experimental design for the fMRI study
The fMRI design was identical to that of the behavioral study excepted that the stimulus display
duration was shorter (2.5 s) and that the delay duration (6-10 s) and the inter trial interval duration (3-
9 s) were variables for neuroimaging technical issues. The long variable delay before the feedback
allowed us to distinguish the hemodynamic signal associated with the decision taking during the
stimulus presentation from the sustained signal associated with the reward uncertainty during the
delay (Figure 5).
Figure 5. Experimental design. Four runs were performed by the participants (160 trials). The analysis of neuroimaging data was performed during the stimulus presentation (at choice) and during the delay. The order of presentation of the run was fully counterbalanced to avoid
order effects.
Before the experiment, participants performed two runs (neutral and financial) to become familiar
with the task and the payoff matrix. The training is identical to the behavioral task.
G. fMRI image acquisition
The experiment was conducted at the Fondazione Santa Lucia (Rome). All the data were acquired
in a single session on a 3 T Allegra scanner (Siemens Medical Solutions, Erlangen, Germany) with a
maximum gradient strength of 40 mT/m, using a standard quadrature birdcage head coil for both RF
transmission and RF reception. The fMRI data were acquired using a gradient echo-EPI, with 38 axial
slices with a voxel size of 3 × 3 × 3.75 mm3 (matrix size 64 × 64; FOV 192 × 192 mm2) in ascending
order. The acquisition time was 2.47 s / 65 ms/ slice.
H. Analysis of fMRI data
Data analysis was performed within the Statistical Parametric Mapping analytic package (SPM5,
Wellcome Department of Cognitive Neurology, London, UK). The data were sinc-interpolated in time.
All images were re-aligned to the first acquired volume to correct head motion. Image was then
spatially normalized and the transformation parameters were then applied to the functional volumes,
smoothed with a (6*6*6*8 mm) isotropic Gaussian smoothing kernel. The preliminary analysis
focused on non-specific effect of the financial incentive by collapsing reward regressor for the period
of the stimulus and the delay and for every level of uncertainty. Thus two regressors were used: [low /
high uncertainty, neutral], [low / high uncertainty, financial].
V. RESULTS
A. Statistical analysis
All behavioral data were analyzed with Statistica 7.1 (© StatSoft). The Kolmogorov-Smirnov
goodness-of-fit test shown that data distribution was not normal, therefore, the effects of the
financial incentive (neutral vs. financial), of the level of uncertainty (low vs. high) and their interactions
on our dependant variables, the reaction times (RT) and the percentage of landing, were examined
thanks to Friedman’s ANOVA for overall effects and Sign test for paired analyses. The same type of
analysis was also used to examine the effects of the type of stimulus (0%, 100% and 50%) on the same
dependant variables.
B. Behavioral results
1) Effect of Uncertainty on reaction times
The Sign test revealed an effect of uncertainty on RT (p < .023). Higher uncertainty generated
longer mean RT than low uncertainty (see Figure 6).
Figure 6. Reaction times (in ms) according to the two level of uncertainty (low and high).
2) Cross-effect of incentive and uncertainty on reaction times
The Friedman’s ANOVA did not revealed an overall effect of the type of incentive on the RT.
However, the Sign test revealed an effect of the financial incentive on RT (p < .023) for the stimuli
where the landing was obviously possible (100% vs. 100%*). During the financial condition, the
subjects RT were shorter than during the neutral condition, see Figure 7.
Figure 7. Reaction times (in ms) according to the 3 type of stimulus and for the two types of incentive. The asterisk indicates the presence of the financial incentive.
3) Cross-effect of incentive and uncertainty on decision making
In response to the asymmetric payoff matrix, subjects exhibited a significant shift in the likelihood
of accepting landings. More precisely, the Sign test showed that under uncertainty (50% vs. 50%*), the
percentage of landing acceptance increased (p = .026), from 32.093% (SD = 12.06) to 74.03% (SD =
27.85), see Figure 8.
Figure 8. Percentage of landing acceptance according to the level of uncertainty and the type of financial incentive. Concerning the high
uncertainty (50%), results showed a conservative behavior without incentive (under 50% of acceptance) and a risky behavior with incentive (beyond 50% of acceptance). The asterisk indicates the presence of the financial incentive.
C. Functional neuroimaging case study
The behavioral results of the subject that performed the task in the scanner were coherent with
the behavioral group. The RT decreased in the financial incentive condition. Moreover the financial
incentive generated a shift in the likelihood of accepting landings under high uncertainty, from 50% in
the neutral condition to 85% of landing acceptance in the financial condition.
We investigated which brain regions were differently involved in decision-making under monetary
incentive and in the neutral condition by performing overall contrasts that collapsed the time of
choice and the time of the delay. The cold reasoning, performed during the neutral condition was
associated with an increased activity in right DLPFC and occipital cortex. On the contrary, the hot
reasoning, performed under financial incentive was related with an increased activity in bilateral
VMPFC (Figure 9, TABLE I. ).
Figure 9. (A) Increased activation of the bilateral VMPC (BA11) during stimulus and delay for monetary incentive vs. neutral. (B)
Increased activation of the right DLFPC (BA46) during stimulus and delay for neutral vs. monetary incentive (p<0.01; cluster > 15). (C) Global patterns for monetary incentive (in green) and neutral (in red) conditions.
TABLE I. TALAIRACH COORDONINATES, Z-VALUE AND CLUSTER SIZES (K) OF THE ACTIVATED ANATOMICAL STRUCTURES FOR THE CONTRASTS NEUTRAL VS. FINANCIAL AND FINANCIAL VS. NEUTRAL. ALL AREAS WERE SIGNIFICANT AT P < . 01 UNCORRECTED.
Anatomical structures
(Broadman’s area)
Neutral - financial Financial - neutral
Talairach coordinates
Talairach coordinates
x y z z-value k x y z z-value k
Frontal
CPFDLD (BA46) 53 36 19 2.35 22
CPFVMBL (BA11) -12 42 -24 2.33 21
Occipital
V2 (BA18) -33 -85 -12 2.30 15
VI. CONCLUSION
We used an approach borrowed from neuroeconomics to investigate the impact of an economic
pressure, namely the cost of a go-around, on landing decision-making. This preliminary work reports
behavioral results and a case study in fMRI. The behavioral data confirmed the impact of the financial
incentive. Firstly, subjects responded with a slightly faster response time for the financial incentive
condition when the decision to land was obvious (100%), showing more precipitate responses. The
decision to land in this context is rewarded by 5€ and this reduction of the RT may be interpreted as a
search of the reward at the expense of a detailed analysis of the rhombuses positions. Secondly, the
financial incentive has biased responses toward more risky decision-making. Whereas, under
uncertainty, participants are rather conservative during the neutral condition (32.093% of landing),
they took more risky decisions under the influence of the biased payoff matrix. The payoff matrix has
associated the go-around with immediate negative consequences and participants became reluctant
to do it.
The preliminary neuroimaging results confirmed that the change in decision-making entailed by
the financial incentive was subserved by a shift from a cerebral region dedicated to reasoning (DLPFC)
to a region involved in emotion processing (VMPFC). The modulation of occipital cortex activation
gives clues on the possible top-down perturbing effect of monetary incentive on visual information
processing The behavioral data associated to these neuroimaging results are in favor of a shift from a
cold reasoning under the neutral condition to a hot reasoning in presence of the financial incentive.
According to us, this shift can be generalized to pilots and demonstrates that the erroneous trend to
land whereas the approach is not stabilized is the result, at least for a part, of the different aversive
negative consequences associated with the go-around decision, in particular the financial cost for the
company. This is suggesting that this phenomenon may be explained by a temporary perturbation of
decision-making process under an emotional factor. Data from fMRI sessions are currently analyzed
and include a total of 16 participants.
REFERENCES
[1] Dehais, F., Tessier, C., Christophe, L., & Reuzeau, F. (2009). The perseveration syndrome in the pilot’s activity: guidelines and cognitive countermeasures.
[2] Rhoda, D., & Pawlak, M. (1999). An assessment of thunderstorm penetrations and deviations by commercial aircraft in the terminal area. Massachusetts Institute of Technology, Lincoln Laboratory, Project Report NASA/A-2, 3.
[3] Orasanu, J., Martin, L., & Davison, J. (1998). Errors in Aviation Decision Making: Bad Decisions or Bad Luck? RECON(20020063485).
[4] BEA. (2000). Objectif : destination. Technical report, Bureau Enquête Analyse.
[5] Goh, J., & Wiegmann, D. (2002). Human factors analysis of accidents involving visual flight rules flight into adverse weather. Aviation, space, and environmental medicine, 73(8), 817.
[6] Muthard, E., & Wickens, C. (2002). Change detection after preliminary flight decisions: Linking planning errors to biases in plan monitoring.
[7] O'Hare, D., & Smitheram, T. (1995). "Pressing on" into deteriorating conditions: An application of behavioral decision theory to pilot decision making. The International Journal of Aviation Psychology, 5(4), 351-370.
[8] Orasanu, J., Ames, N., Martin, L., & Davison, J. (2001). Factors in Aviation Accidents: Decision Errors. Linking expertise and naturalistic decision making, 209.
[9] Nance, J. (1986). Blind trust: William Morrow & Co.
[10] Sander, D., & Koenig, O. (2002). No inferiority complex in the study of emotion complexity: A cognitive neuroscience computational architecture of emotion. Cognitive Science Quarterly, 2, 249-272.
[11] Damasio, A. (1995). L'erreur de Descartes Paris. Éditions Odile Jacob.
[12] Drevets, W., & Raichle, M. (1998). Reciprocal suppression of regional blood flow during emotional versus higher cognitive processes: Implications for interactions between emotion and cognition. Cognition and Emotion, 12(3), 353-385.
[13] Mayberg, H., Liotti, M., Brannan, S., McGinnis, S., Mahurin, R., Jerabek, P., et al. (1999). Reciprocal limbic-cortical function and negative mood: converging PET findings in depression and normal sadness. American Journal of Psychiatry, 156(5), 675.
[14] Mitchell, R., & Phillips, L. (2007). The psychological, neurochemical and functional neuroanatomical mediators of the effects of positive and negative mood on executive functions. Neuropsychologia, 45(4), 617-629.
[15] Goel, V., & Dolan, R. (2003). Reciprocal neural response within lateral and ventral medial prefrontal cortex during hot and cold reasoning. Neuroimage, 20(4), 2314-2321.
[16] Schoofs, D., Wolf, O., & Smeets, T. (2009). Cold pressor stress impairs performance on working memory tasks requiring executive functions in healthy young men. Behavioral Neuroscience, 123(5), 1066-1075.
[17] Oldfield, R. (1971). The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia, 9(1), 97.
Part 2
Economic issues provokes hazardous landing by enhancing emotional decision-making:
evidence from psychophysiological measures
I. METHODS
A. Psychophysiological experimental design
Like in the fMRI study, we used a 2x2 factorial design crossing the type of financial incentive,
neutral or financial, and the level of uncertainty, high, 50% chance of crash, or low, 100% or 0%
chance of crash, depending on both positions of the rhombuses displayed on the PFD. These
likelihoods were unknown by the subjects. The two types of incentive were introduced in two
separate runs and the two level of uncertainty were manipulated within each run. Each trial consisted
in a presentation of the stimulus (2.5 s), during which the participant made his choice thanks to a
response pad, followed after a variable delay (6-10 s) by the feedback informing of the accuracy of the
response (2 s). During the financial condition, the financial outcome {+5€}, {-5€} or {–2€}) was
displayed (1.5 s) after the accuracy feedback (0.5 s). Finally, a variable inter trial interval (3-9 s) was
introduced. Response and financial feedback delivery was contingent upon the subject’s response.
Before the experiment, participants performed two runs (neutral and financial) to become familiar
with the task and the payoff matrix. The training is identical to the behavioral task. During the
experience, after the run with financial motivation, the participants rated how much the payoff matrix
had influenced their decisions on a 1-9 scale.
B. Data acquisition and analysis
Stimulus display and data acquisition were done with Cogent 2000 v125 running under Matlab
environment (Matlab 7.2.0.232, R2006a, The MathWorks, USA). All behavioral data were analyzed
with Statistica 7.1 (© StatSoft).
1) Behavioral data Mean response times (RT) and percentages of landing acceptance were calculated for each level of
uncertainty (low and high) and for each type of stimulus (0%, 100% and 50%). The effects of the type
of incentive (neutral vs. financial), the level of uncertainty and their interactions on our dependant
variables, RT and the percentage of landing, were examined thanks to two-way repeated measures
ANOVA. The same ANOVA were also used to examine the effects of the type of stimulus (0%, 100%
and 50%) on the same dependant variables. The analysis of the effect of the type of stimulus required
separate two-way repeated measures ANOVA because this independent variable was nested within
the level of uncertainty. Fisher's least significant difference post hoc test was used to examine paired
comparisons. Bravais-Pearson correlation was used to estimate the matching between self-perceived
level of influence of the financial incentive and its effective impact on the increase acceptance of
landing between the neutral incentive and the financial incentive.
2) Psychophysiological data Participants were comfortably installed and tested in a moderately lit room, in which the
illumination was held constant (background luminance: about 450 lux). The ProComp Infiniti
(©Thought Technology Ltd.) was used to record continuously the two physiological measures, the
heart rate (HR) and the skin conductance responses (SCRs). Data were digitized at a sampling rate of
256 Hz. Before the experiment, they were asked to relax during 5 minutes to collect the rest state
(baseline).
Skin conductance was recorded using the SC-Flex/Pro sensor, a pair of electrodes placed on the
distal phalanges of digits II and III of the left hand. The SCRs were measured in micro Siemens (μS) and
analyzed off-line. Responses were computed as a change in conductance from the pre-stimulus level
to the peak of the response. Following information provided by Dawson [45], the SCRs were
computed as follow (where G is the skin conductance value and Ts the stimulus onset): Gmax [ts + 3;
ts + 7] - Gmin [ts - 3; ts - 1]. The minimum level occurring within 1-3 s from stimulus presentation was
subtracted from the peak value occurring within a 3-7 s window, an absence of response was
computed as 0. The time course of phasic SCRs being 4–6 s, our long inter-stimulus interval paradigm
allowed avoiding SCRs overlapping.
The heart rate was measured using the EKG-Flex/Pro sensor. Three electrodes connected to an
extender cable were applied on the participant’s chest. We used and Uni-Gel electrodes to enhance
the quality of the signal. This latter was measured in microvolts (μV) and inter-beat-intervals were
converted to beats per minute (bpm).
Mean values were calculated for the heart rate parameter during the complete duration of three
periods of interest: the baseline and the two types of runs, neutral and financial. Two-way repeated-
measures ANOVA were performed to assess the significance of heart rate changes across these three
periods and Fisher's least significant difference post hoc test was used to examine the impact of each
task in comparison to baseline. Separate one-way repeated-measures ANOVA were performed on
delta values (i.e. run – baseline) to compare the significance of physiological changes across the two
types of runs. Then, the effect of the type of incentive, the level of uncertainty and their interactions
on the heart rate was examined during shorter time windows. The mean values were calculated in the
time window between the stimulus onset and 10 s post-stimulus onset. The mean values per
condition of these stimulus locked data were submitted to two-way repeated-measures ANOVA to
examine the effect of the type of incentive, of the level of uncertainty, and their interactions on the
HR. As in behavioral analysis, the effect of the type of stimulus was examined with separate two-way
repeated-measures ANOVA. Fisher's least significant difference post hoc test was used to examine
paired comparisons. In addition, similar separate analyses were conducted for SCRs to assess the
effect of the same independent variables.
II. RESULTS
A. Behavioral results
1) Effect of Uncertainty and incentive on reaction times The ANOVA revealed a main effect (for both type of incentive) of uncertainty on RT (p<.001,
F(1,15) = 45.539). High uncertainty generated longer mean RT than low uncertainty, see Figure 10.
Figure 10. Main effect of the level of uncertainty on reactions times (for both type of incentive, neutral and financial).
The ANOVA revealed a main effect (for both level of uncertainty) of the type of incentive on the RT
(p = .001, F(1,15) = 16.143). During the financial incentive conditions, RT were shorter than during the
neutral condition, see Figure 11. The analysis of the interactions between the level of uncertainty and
the type of incentive did not revealed significant results (p = .006, F(1,15) = 3.930). However, a
separate analysis examining the effect of the type of incentive according to the different type of
stimuli showed a significant interaction (p = .001, F(2,30) = 6.826), see Figure 12. The Fisher's LSD post
hoc test revealed that the decision-making was performed faster under the influence of the financial
incentive than the neutral incentive during the 100% (p = .001) and 50% (p = .004) stimuli. That was
not the case concerning the 0% stimuli (p = .263). These result explain the non significance of the
interaction between incentive and uncertainty and suggests that the RT reduction under the financial
incentive influence is associated to the landing decision but not with the decision to go-around (i.e.
0%).
Figure 11. . Main effect of the type of incentive on reactions times (for both level of uncertainty, low and high).
Figure 12. Reaction times according to the three types of stimuli for * the two type of incentive. There was a main effect of the type of
incentive. More precisely, during the financially motivated condition, the reaction times of the stimuli 100% and 50% were significantly shorter under the financial incentive.
2) Effect of uncertainty and incentive on decision-making The magnitude of the mean total outcome was important (39.31€, SD = 11.42), beyond the
maximum outcome of 30€ that the participants thought they could win. Its demonstrated that the
decision-making were oriented toward economic optimization. Indeed, in response to the asymmetric
payoff matrix, subjects exhibited a significant shift in the likelihood of accepting landings. We
calculated a response bias score: a negative response bias would correspond to a conservative
behavior (less landing acceptance that expected) whereas a positive response bias would correspond
to a risky behavior (more landing acceptance than expected). The ANOVA showed that there was a
main effect of the uncertainty (p < .001, F(1,16) = 121.165) and a main effect of the incentive (p < .001,
F(1,16) = 16.211) on the magnitude of the response bias: there was an overall increased likelihood of
making a landing acceptance when the uncertainty was high and under the financial incentive. In
addition, the ANOVA also revealed an interaction effect between these two variables (p = .021, F(1,15)
= 6.530): the financial incentive increased the likelihood to accept a landing criterion, in particular
when the uncertainty was high (Figure 13).
Figure 13. Response bias according to the level of uncertainty and the type of incentive. A positive response bias demonstrated an increase likelihood of accepting landing.
The separate ANOVA examining stimulus effect showed that there was an overall effect of the type
of incentive on the decision-making performed during the three types of stimuli (p < .001, F(1,15) =
20.242). The Fisher's LSD post hoc test revealed that the landing acceptance rate was increased with
the financial incentive in comparison to the neutral incentive during the three different types of
stimuli: 0% (p=.048), 100% (p=.008) and 50% (p=.001), see Figure 14, TABLE II. And TABLE III. ).
Figure 14. Percentage of landing acceptances according to the three types of stimuli and the type of financial incentive. With the financial
incentive, the landing acceptance rate is higher across the three types of stimuli.
Finally, the Bravais-Pearson correlation showed that there was no significant relationship between
the self-perceived level of influence of the financial incentive and its effective impact on the increase
acceptance of landing between the neutral incentive and the financial incentive (p = .578, r = +.17).
The conscious perception of change in their decision criterion was not consistently related to their
objective behavior.
TABLE II. MEAN REACTION TIMES (IN MS), MEAN PERCENTAGE OF LANDING ACCEPTANCE AND STANDARD DEVIATION ACCORDING TO THE THREE TYPES OF STIMULI AND THE TWO TYPES OF INCENTIVE.
RT (ms)
Low uncertainty High uncertainty
0% 100% 50%
Neutral incentive 1351.06 (±184.85) 1385.63 (±221.59) 1592.43 (±174.45)
Financial incentive 1415.03 (±231.92) 1186.56 (±160.46) 1421.34 (±158.44)
% of landing acceptance
Neutral incentive 2.66 (±7.98) 87.33 (±7.03) 63.66 (±19.12)
Financial incentive 10 (±8.94) 97.5 (±4.47) 85.9375 (±10.03)
TABLE III. MAIN ANOVA RESULTS (P AND F VALUES) USING THE THREE DIFFERENT PREDICTORS AND THEIR INTERACTIONS TO ASSESS THEIR EFFECTS ON TWO BEHAVIORAL MEASURES, REACTION TIMES AND PERCENTAGE OF LANDING (** ≤.01 ; *** ≤.001).
Predictors Level of
uncertainty
Stimulus type Type of
incentive
Level of
uncertainty *
Type of incentive
Stimulus type *
Type of
incentive
RT p <.001*** <.001*** .001*** .067 <.001***
% of landing p
F
/
/
<.001***
802.109
<.001**
20.242
/
/
.013**
5.079
B. Physiological results
Since informatics issue, physiological data were available for only 15 participants.
1) Skin conductance response The ANOVA did not reveal main effect neither of the uncertainty (p = .560, F(1,14) = 0.356) nor the
type of incentive (p < .451, F(1,14) = 0.603). However, the separate ANOVA showed that there was an
effect of the type of stimuli (p < .035, F(2.28) = 3.810). In particular, the Fisher's LSD post hoc test
revealed that 0% stimuli type elicited more important SCRs than 100% ones (p = .011). See Figure 17
for illustration.
2) Heart rate The ANOVA performed on the three periods of interest (baseline, run neutral and run money)
revealed that the HR was significantly different across these three periods (p < .001, F(2.28) = 11.233),
see Figure 15. The LSD post hoc test demonstrated that the mean HR was higher during the two runs
than during the baseline (p < .001 in both comparisons). In addition, the comparison performed on
delta values between the two runs showed that the heart rate was more elevated during the financial
run than the neutral one (p < .009, F(1.14) = 9.570).
Figure 15. Mean heart rate across three periods of interest: at rest (baseline), the neutral run and the financial run. The HR was significantly
lower during the baseline than the two runs and was more elevated during the financial run than the neutral one.
The stimulus locked analyzes did not demonstrated any effect of the uncertainty (p < .102, F(1.14)
= 3.126). However, a main effect of the type of incentive was also found (p = .030, F(1.14) = 6.023),
which is consistent with the previous results found by performing whole run comparisons. The ANOVA
did not revealed main effects of the type of stimuli. The LSD post hoc test showed that (asterisk
indicates the financial incentive) 50%* elicited a higher HR that 0% (p < .001), 100% (p < .001) and 50%
(p = .003); that 100%* elicited a higher HR than 0% (p = .005), 100% (p = .003) and 50% (p = .025) and
finally that HR was more elevated during 0%* stimuli that 0% (p = .006), 100% (p = .004) and 50% (p =
.032). The highest mean heart rate was recorded during 50%* stimuli (74.90 bpm), when the
uncertainty was high and the financial motivation was administered, see Figure 16, TABLE IV. and
TABLE V. ).
Figure 16. Mean heart rate across the three types of stimuli * the two types of incentive. The highest mean heart rate was recorded during
the 50%* stimuli. The asterisk indicates the financial incentive.
Figure 17. Illustration of the skin conductance measure for one subject during a whole run. The last rise corresponds to the display of the global outcome.
TABLE IV. MEAN STIMULUS LOCKED VALUES ACCORDING TO THE THREE TYPES OF STIMULI AND THE TWO TYPES OF INCENTIVE.
TABLE V. MAIN ANOVA RESULTS (P AND F VALUES) USING THE THREE DIFFERENT PREDICTORS AND THEIR INTERACTIONS TO ASSESS THEIR EFFECTS ON TWO BEHAVIORAL MEASURES, REACTION TIMES AND PERCENTAGE OF LANDING (* ≤.05).
Predictors Level of
uncertainty
Stimulus type Type of
incentive
Level of
uncertainty *
Type of incentive
Stimulus type *
Type of
incentive
HR p .102 .177 .030* .083 .096
SCRs p
F
.560
0.356
.035*
3.810
.451
0.603
.788
0.075
.268
1.381
III. DISCUSSION
We used an approach borrowed from neuroeconomics to investigate the impact of an economic
pressure, namely the cost of a go-around, on landing decision-making. In this experiment, both
uncertainty and type of incentive were manipulated. As expected, these variables have significantly
impacted the behavior of participants. The uncertainty generated two types of reasoning: when the
rhombuses positions were non-ambiguous, all participants reported that the decision was
straightforward; on the contrary, when they tried hard to find a rule when the rhombuses positions
were ambiguous. This latter subjective result is supported by the increase of the reaction when the
ambiguity was high. Nevertheless, these results were expected and suggest that low uncertainty
stimuli were easily categorized and high uncertainty stimuli engendered deepest reasoning.
The payoff matrix was designed to incite participants to maximize their monetary reward by
biasing their response criterion (i.e. a bias to indicate that the landing was possible). All subjects
HR
Low uncertainty High uncertainty
0% 100% 50%
Neutral incentive 72,47 (±5.14) 72.38 (±5.67) 72.84 (±6.05)
Financial incentive 74.48 (±7.38) 74.63 (±7.07) 74.90 (±6.21)
SCRs
Neutral incentive 0.62 (±0.58) 0.35 (±0.27) 0.47 (±0.43)
Financial incentive 0.42 (±0.38) 0.36 (±0.33) 0.37 (±0.03)
understood that the payoff matrix biased decisions toward the landing acceptance. They showed a
significant shift in the response criterion, and trend to increase their landing acceptance rate with the
financial incentive, in particular when the uncertainty was high. Whereas their behaviors were rather
conservative in the neutral condition, they made more risky decisions under the influence of the
financial incentive, to avoid the risk of a penalty in case of go-around. In addition, reaction times were
longer in the neutral condition, showing more reasoned decision making. Cold reasoning appeared to
be more analytic, more objective from a safety point of view whereas hot reasoning was associates
with a search of the reward at the expense of a detailed analysis of the rhombuses positions.
The comparison of the mean HR during the baseline with the period of each run has confirm that
this measure is sensitive to the increased activity of SNA in response to the task energy mobilization
and the investment of mental effort to deal with the task [1] [1]. Indeed, the mean HR was
significantly higher during both tasks compared with the resting state.
In addition, the heart rate measure showed that the financial incentive elicited ANS arousal. Both
stimulus locked and whole runs analyzes showed that the mean HR was higher in the financially
motivated condition. This result demonstrated that the financial incentive was linked with a
modulation of the emotional state of the participants. Although the magnitude of the change between
the two runs was small (2.21 bpm) it was nevertheless very significant and persistent on all
participants. It should be noted that the magnitude of this increase is consistent with the literature
related to emotional induction. For instance, Brosschot [3] showed that negative and positive emotion
induction elicited respectively an HR rise of respectively 2.14 and 1.06 bpm. This increase of the HR
between the neutral and the financial incentive cannot be attributable to a significant improvement of
the performances or of the difficulty, in particular in the ambiguous condition. On the contrary, in this
latter condition, the false alarms increased drastically in presence of the financial motivation. Yet,
since the work of Wright, making the link between motivational theory and cardiovascular activity,
many studies shown that cardiac activity is related to the increased difficulty of a task as long as the
success is important and the improvement of the performance is possible. Indeed, the organism
manages the investment of its resources so as not to waste them. In our experiment, there were no
clear limits beyond which a landing could be categorized as possible or not, the visual cues available
on the real PFD has been removed for this purpose. In addition, in ambiguous situation, the
improvement of the performance was not possible because the chance of landing successfully was
50/50. These considerations lead us to think that, beyond an emotional effect, the increase HR reflects
the motivation to maximize financial gains. In this perspective, the payoff matrix leads participants to
waste energy for a purpose that is in total discrepancy with the safety. Indeed, the increase of landing
acceptance was directly linked with increased risk taking. The fact that O% stimuli type elicited more
important SCRs than 100% stimuli suggested that the decision to go-around generated an ANS arousal
than the decision to land, in response to the overall negative context recreated around the go-around
in this experiment. It may argue that participants felt frustration when they were constrained to go-
around. However, no significant effects neither of the uncertainty not of the type of incentive was
found on SCRs. Several explanations main be put forward. Each type of stimuli was repeated a large
number of times to get reliable statistics based. The SCRs are extremely sensitive to novelty, and the
repetition may have decreased progressively the SCRs response magnitudes. Moreover, the
systematic feedback displayed after each response may have also contributed to lower the SNA
responses. For instance, Van der Veen [3] has shown that the display of the feedback during task
performance reduced cardiovascular reactivity.
Our experiment was originally designed to understand pilot trend to land despite wrong
meteorological conditions. Our assumption was that pilots frame their decision to keep on landing in
terms of potential losses (such as money spent of fuel used up). Our behavioral and physiological
results showed that the risky decision to land in pilots may be, at least for a part, explained by the
immediate negative consequences associated with the decision to go-around. An organization that
emphasizes productivity (e.g. on time arrivals or saving fuel) may unconsciously set up goal conflicts
with safety. Pilots may be willing to take a risk with safety (a possible loss) to arrive on time (a sure
benefit). It has been generally assumed that people are risk aversive in their choice. Most people will
be more likely to choose 80 dollars for certain than 85% chance of winning 100 dollars. Kahnenman &
tversky [5] demonstrated those choice criterions are reversed when the sign of outcome is reversed:
faced with a choice between a certain loss of 80 dollars and a gamble, 85% chance of losing 100
dollars, most of people will choose the gamble. Thus, people are only risk averse when considering
possible gains; when faced with the prospect of losses, people suddenly become risk seeking. Our
payoff matrix clearly associates the go-around with expected losses: the mathematical expectation
linked with this decision was either null (0€), when a go-around was necessary, or negative, (-3.5€),
when a landing was possible.
In this experiment, the trial-related design sought to break up the task into three stages: decision-
making, feedback expectancy, and feedback processing. We plan to analyze the physiological data
during these three stages.
REFERENCES
[1] Fairclough, S., Venables, L., & Tattersall, A. (2005). The influence of task demand and learning on the psychophysiological response. International Journal of Psychophysiology, 56(2), 171-184.
[2] Gaillard, A. (2001). Stress, workload, and fatigue as three biobehavioral states: A general overview. Stress, workload, and fatigue. Mahwah, NJ: L. Erlbaum.
[3] Brosschot, J., & Thayer, J. (2003). Heart rate response is longer after negative emotions than after positive emotions. International Journal of Psychophysiology, 50(3), 181-188.
[4] Van der Veen, F., Van der Molen, M., Crone, E., & Jennings, J. (2004). Phasic heart rate responses to performance feedback in a time production task: effects of information versus valence. Biological Psychology, 65(2), 147-161.
[5] Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263-291.
Part 3
Flight experience and executive functions predict flight simulator performance and poor
decision-making in general aviation pilots
Submitted to ICRAT 10
Mickaël Causse Centre Aéronautique et Spatial ISAE-SUPAERO ; Université de Toulouse, France
Inserm ; Imagerie cérébrale et handicaps neurologiques UMR 825; F-31059 Toulouse, France Université de Toulouse ; UPS ; Imagerie cérébrale et handicaps neurologiques UMR 825; CHU Purpan, Place du Dr Baylac, F-
31059 Toulouse Cedex 9, France [email protected]
Frédéric Dehais
Centre Aéronautique et Spatial ISAE-SUPAERO ; Université de Toulouse, France [email protected]
Josette Pastor
Inserm ; Imagerie cérébrale et handicaps neurologiques UMR 825; F-31059 Toulouse, France Université de Toulouse ; UPS ; Imagerie cérébrale et handicaps neurologiques UMR 825; CHU Purpan, Place du Dr Baylac, F-
31059 Toulouse Cedex 9, France [email protected]
Abstract— Unlike professional pilots who are limited by the FAA's age rule, no age limit is defined in
general aviation (GA). Some studies revealed significant aging issues on accident rates but these results
are criticized. Our overall goal is to study how the effect of age on executive functions (EFs), high level
cognitive abilities, impacts on the flying performance in GA pilots. This study relies on four components:
EFs assessment, pilot characteristics (age, flight experience, level of impulsiveness), the navigation
performance on a flight simulator and a decision-making performed during the approach. The results
showed that contrary to age, reasoning, working memory (WM) and total flight experience were
predictive of the flight performance. The WM, the flight experience and the level of impulsiveness were
predictive of poor decision-making. These results suggest that “cognitive age” is a better mean than
“chronological age” consideration to predict the ability to pilot, in particular because of the inter-
individual variability of aging impact and the beneficial effect of the flight experience.
Keywords: piloting performance, executive functions, flight experience, decision making, normal aging.
I. INTRODUCTION
The population of GA pilots is getting older in the USA [1] and in European countries like France
where forty one percent of private pilots are more than fifty (BEA1, 2008). Unlike professional pilots
who are limited by the FAA's age rule, no such restriction exists for GA pilots. Moreover, contrary to
commercial aviation (CA) pilots, GA pilots have not necessarily experienced a professional training, fly
mostly on their own, without any co-pilot and very few assistance systems, have less support from the
air traffic control and are more affected by weather conditions. Not surprisingly, in GA, the accident
rate is considerably higher than in CA [2].
Several studies have revealed significant aging issues on accident rates in GA [3] [4] [5], though
these results are called into question [6] [7]. The assessment of the cognitive functioning is a key issue
in pilot’s aging as long as its decline represents a much higher risk of accidents than sudden physical
incapacitation [8]. A substantial literature focuses on the evaluation of the cognitive state of pilots but
its conclusions remain contradictory. Several reasons may explain the difficulty to draw a definitive
conclusion on the effects of aging on flight performance in GA pilots. There is a great inter-individual
variability in the deleterious effects of aging on cognition [9]; the evaluations performed in classical
human factors studies are rather nonspecific in terms of explored cognitive functions and do not
necessarily focus on the ones that are the most impacted by aging; very few researches attempt to
link, in the same population, the cognitive performances to the flight abilities; the greatest part of the
studies is interested on safety aspects like communications [10], or decision making during landing
[11]; few researches are exclusively related to the GA population; finally, another source of complexity
arises from the suspected compensative role on aging effects of the flight experience [12].
II. COGNITIVE FUNCTIONS AND PILOTING
Numerous studies have been conducted to attempt to link the cognitive functioning with the flight
performance. Different measurements of cognitive efficiency have been identified as crucial to the
piloting ability, for example: time-sharing [13], speed of processing [14], attention [15] or problem
solving [16]. Cogscreen-AE [17] is among the most widely used cognitive tests batteries in pilots aging
studies. It consists of a series of computerized cognitive tasks that evaluate a large set of cognitive
functions. This battery has been shown to be able to successfully discriminate between neurologically
impaired and cognitively intact pilots [18]. Some Cogscreen-AE variables were predictive of flight
parameter violation in Russian CA pilots [19]. Furthermore, Taylor and colleagues [20] were able to
1 French Accident Investigation Bureau.
predict 45% of the variance of the flight simulator performance with four Cogscreen-AE predictors
(speed/WM, visual associative memory, motor coordination and tracking) in a cohort of 100 aviators
(aged 50-69). Contrary to this latter study that involved Cogscreen-AE, a rather generalist battery in
terms of explored cognitive functions, we propose to focus specifically on EFs. Indeed, these functions
are the earliest ones to be impacted by aging [21] and represent excellent clues of aging effects on the
cognitive performance. The study of EFs has appeared recently in aeronautics, for instance, Hardy [22]
found significant age-related differences in pilots’ executive functioning (e.g. inhibition, set-shifting)
and Taylor [23] established a relationship between interference control and the ability to follow air
traffic instructions.
III. EXECUTIVE FUNCTIONS AND PILOTING
EFs underlie goal-directed behavior and adaptation to novel and complex situations [24]. They
allow the inhibition of automatic responses in favor of controlled and regulated behavior, in particular
when automatic responses are no more adapted to the environment. Three major low level EFs are
moderately correlated with one another, but clearly separable [25]: set-shifting between tasks or
mental sets (“shifting”), inhibition of dominant or prepotent responses (“inhibition”), and updating
and monitoring of WM representations (“updating”). The prefrontal cortex (PFC) plays a dominant
role in the implementation of EFs that also encompass decision-making [26] or reasoning abilities [27].
According to our hypotheses, EFs are crucial to piloting. Indeed, this activity takes place in a dynamical
and changing context where new information must be integrated and updated continuously. We
assume that flying light aircraft with no autopilot and very few assistant systems (like the TCAS2 or
weather radar) presupposes a strong involvement of the EFs for handling the flight, to monitor the
engine parameters, to plan the navigation, to maintain and update situation awareness and to
correctly adapt to traffic and environmental changes and perform accurate decision-making by
inhibiting wrong behavioral responses.
IV. EXECUTIVE FUNCTIONS AND NORMAL AGING
Functional neuroimaging brings evidence that the brain is subject to anatomical and physiological
modifications in normal aging [28]. The prefrontal lobes appear to be the earliest cerebral regions to
be affected [29] and may account for a great part to age-related cognitive changes [31]. Because the
prefrontal lobes mainly implement EFs, aging is suspected to provokes a selective alteration of these
latter, for example the reasoning [30], inhibition [31] or updating [32] abilities. However, the executive
changes vary considerably across people. The complex interactions between the cerebral structures
2 Traffic Collision Avoidance System.
underlying EFs [9], sociocultural factors and genetic factors [33] may explain the heterogeneity of this
decline.
In this experiment, we proposed to evaluate specifically the EFs, high level cognitive abilities that
present a strong vulnerability to aging effects [21]. More precisely, we assessed three low levels EFs
(shifting, inhibition and updating) and a more established general ability: the reasoning. The reasoning
performance reflects fluid intelligence, that support processes relevant for many kinds of abilities
(verbal, spatial, mathematical, problem solving etc.) and adaptation to novelty. It is a concept very
close to the executive functioning [34] [35]. The speed of processing was also collected because it
represents a reliable measure of general cognitive decline during aging. Finally, we have also taken
into account age and the total flight experience to assess their respective participation to the flight
performance variation. Our hypothesis is that the “chronological age” is not a sufficient criterion to
predict the piloting performance and that the “cognitive age”, evaluated by the cognitive functioning,
is a more relevant criterion.
V. IMPULSIVITY
According to Sicard [36], flight safety is dependent on the quality of the decision-making process,
which is closely related to risk taking. Impulsive individuals are more likely to make risky decisions,
choosing immediate rewards despite potential long-term negative consequences [37]. Impulsivity is a
personality characteristic described as ‘‘acting without thinking”. Impulsive individuals make risky
decisions, motivated more by immediate reward rather than by the potential long-term negative
consequences of their choices, suggesting heightened sensitivity to reward and/or reduced sensitivity
to negative outcomes [38]. Several studies have highlighted the close relationship between trait
impulsiveness and decision-making. For instance, Martin [39] shown that high impulsive people do not
present error related negativity (an event related potential obtain with an electroencephalogram)
contrary to low impulsive people during risky choice, suggesting that low impulsive individual
evaluated risky choice as a poor decision whereas high impulsive individuals are biased towards
immediate reward and are less sensitive to the negative consequences associated with their choice.
Interestingly, some executive functions, like planning and impulse control, has shown to be predictive
of the level of trait impulsiveness [40], showing that this psychological trait is sustained by
neuropsychological functions. In addition, executive functions may modulate the decision making
style. For instance, Hinson [41] showed that limits on working memory (WM) function are predictive
of a more impulsive decision-making style. In this way, Keilp [42] has shown that the Go–No Go task,
verbal fluency, executive function measures and tasks requiring decision-making against time were
the strongest correlate of self-rated impulsiveness. These measures remained linked with the level of
impulsiveness after adjustment for age and education. According to the authors, these findings
suggest that trait impulsiveness is effectively assessed by tasks that require decision-making and
response organization under time pressure. The piloting activity takes place in a dynamic environment
where decision-making are often performed under time pressure. Our hypothesis is that trait
impulsiveness is a psychological trait that may strongly modulates the relevance of decision-making, in
particular during approach and landing where the time pressure is important.
VI. METHODS
A. Participants
The participants were 24 private licensed pilots (age 43.3, SD = 13.6) rated for visual flight
conditions. The pilots that had no longer flown during the past two years were excluded because of
the potential impact on flight simulator performance. Inclusion criteria were male, right handed, as
evaluated by the Edinburgh handedness inventory [43], native French speakers, under or
postgraduate. Non-inclusion criteria were expertise in logics, airline pilots and sensorial deficits,
neurological, psychiatric or emotional disorders and/or being under the influence of any substance
capable of affecting the central nervous system. All subjects received complete information on the
study’s goal and experimental conditions and gave their informed consent. Given that flight
experience may moderate age related deficits in the performance of domain relevant task [12], we
attempted to homogenize the flight experience distribution across the life span of our sample.
B. Flight performance
1) Navigation The flight scenario has been setup in collaboration with flight instructors to reach a satisfying level
of difficulty and realism. To familiarize the participants with the PC-based flight simulator and
minimize learning effects in order to obtain reliable flight simulator performances, each volunteer
underwent a training session. Before the navigation, they received the instructions, a flight plan and
various technical information related to the aircraft (e.g. aircraft's crosswind limit). Basically, the
scenario implied to take off, reached a waypoint with the help of the aircraft radio navigation system
and finally, land on a given airport. The pilots were instructed that they were in charge of all the
decisions and that they could only received an informative weather report before landing. In order to
increase the subject’s workload, the pilots had to perform a mental arithmetic calculation of the
ground speed (thanks to the embedded chronometer). Moreover, a failure of the compass was
scheduled. After this failure, the pilots had to navigate thanks to the magnetic compass, which
presents the particularity to be difficult to use as it is anti-directional. The flight scenario lasted
approximately 45 min. The performance assessment was exclusively founded on the flight path
deviations (FPD), expressed in terms of amount of angular deviation in X axes from the ideal flight
path.
2) Crosswind landing decision After the waypoint and before reaching the runway threshold, the pilots must state if the
meteorological conditions, as provided by the automatic information system of airport arrival, were
compatible with a landing or necessitate a go-around and a diversion. In this purpose, the pilots
should assess the crosswind component using a commonly used formula. The calculation result
exceeded of 6 knots the aircraft's maximum crosswind limit, as specified in the documentation
provided to the pilots at the time of the flight preparation. The measured dependant variable was
binary: correct when the pilot decided to divert before the runway threshold, incorrect when the
pilots continued their landing beyond the runway threshold.
C. Pilots impulsiveness measurement
The level of impulsivity of the pilots was measured by the French version of the Barratt
Impulsiveness Scale [44]. This test includes 34 items wich may be scored to three first-order factors:
cognitive (quick decision, 11 items), motor (acting without thinking, 11 items) and non-planning
impulsiveness (present orientation, 12 items).
D. Neuropsychological battery
1) Target hitting This test provides a basic psychomotor reaction time [45]. The instruction is to click as fast as
possible on each target. The performance is measured by a velocity index inspired by the Fitts’ law
[46]. The index is the average ratio of the base 10 logarithm of the distance in pixels between two
targets, divided by the time in seconds to go from the first target to the second.
2) The 2-back test. The 2-back test aims at assessing working memory (WM), in particular maintenance and updating
abilities [47]. Subjects view a continuous stream of stimuli and have to determine whether the current
stimulus matches in a specific dimension (shape for our test) the stimulus 2-back in the sequence
(Figure 18). For each condition, the percentage of correct responses was collected.
Figure 18. The 2-back test. The participant stated if the current shape match to the 2-back shape in the sequence thanks to the response box.
3) Deductive reasoning The logical reasoning test has been used in a previous study to assess executive functioning [48].
The goal of the task is to solve syllogisms by choosing, among three suggested solutions, the one that
allows concluding logically. Syllogisms are based on a logical argument in which one proposition (the
conclusion) is inferred from a rule and another proposition (the premise). We used four existing forms
of syllogisms: modus ponendo ponens, modus tollendo tollens, setting the consequent to true and
denying the antecedent. Each participant had to solve 24 randomly displayed syllogisms. The
measurement was the percentage of correct responses.
4) The computerized Wisconsin Card Sorting test The Wisconsin Card Sorting test (WCST) [49] gives information on the subject’s abstract reasoning,
discrimination learning and shifting abilities [50]. The test version here was a computer
implementation very similar to the clinical version of the WCST [51]. The participant must sort cards
according to three different unknown categories (color, shape, number); an audio feedback indicated
whether the response is correct or not (yes/no). When the participant categorized successfully ten
cards, the target category was automatically changed. The task ended when six categories was
achieved (color, shape, number, color, shape, number) or when the deck of 128 cards was used. The
total numbers of perseverative errors (at least two unsuccessful sorting on the same category) was
derived from the individual cards’ records (Figure 19).
Figure 19. The Wisconsin card sorting test. The participant sorted the cards according to a specific dimension. An audio feedback informed if
the sorting was correct or no.
5) Spatial stroop Spatial Stroop tests generally assess the conflict between the meaning of a word naming a location
(e.g. “below”) and the location where the word is displayed. The ability to restrain a response
according to the localization of the word gives information on inhibition efficiency. This conflict
appears to be provoked by the simultaneous activation of both motor cortices [52]. Our test
encompasses four control conditions (Figure 20). “Stroop neutral meaning” (SNM): a motor answer is
given with the appropriate hand according to the word meaning; “Stroop neutral position” (SNP): the
response is given according to the location of a string of XXXXX, displayed at the left or the right of the
screen; “Stroop meaning incompatible/compatible” (SMI/SMC): the response is given according to the
meaning of the word, compatible or incompatible with is location at the screen. In order to get the
pure effects of inhibition, the interference score was calculated to control reading and localization
effects by:
Figure 20. The four conditions of the spatial stroop. On the left: SNM, the participant pressed on the left/right button according to the
meaning of the word; SNP, the participant pressed the left/right button of the response box according to the location at the screen of the pattern of XXXXX. On the right: SMC/SMI, the participant pressed the left/right button according to the meaning of the word, congruent or
incongruent with its location at the screen.
E. Pilots caracteristics
Age and total flight experience in hours were collected to assess their effects on the flight
performance. We attempted to homogenize the flight experience distribution across the life span of
our sample in order to minimize the perturbation of this parameter on the flight performance
measurement.
VII. RESULTS
A. Statistical analysis
All data were analyzed with Statistica 7.1 (© StatSoft). The relationship between age and the total
flight experience was examined thanks to Bravais-Pearson correlation. The ability of our control
variables to predict the piloting performance was tested using exhaustive regression (ER) that
searches for the best possible fit between a dependent variable and a set of potential explanatory
variables. ER searches the entire space of potential models and returns those for which all parameter
estimates are statistically significant. The goodness of fit of the models was evaluated by the adjusted
coefficient of determination r². The relationship between age and neuropsychological variables was
examined thanks to Bravais-Pearson correlation. The Bonferroni-Holm [53] correction was applied to
control the familywise error rate. The landing decision being a Boolean decision, we have performed
discriminant analysis to examine which variables discriminate between the pilots that had erroneously
land and the pilot that goes-around.
B. Pilots characteristics
The mean level of education of our sample was high (15.45 years, SD = 2.06) and did not
significantly correlated with age (p =. 110, r = -.32). The mean Barrat total score was 41.85 (SD = 8.89)
and did not evolve significantly with age either (p =. 129, r = -.35). This impulsiveness mean score is
remarkably low compared to those of the general male population which is of 54 (SD = 17) [44].
C. Aging and executive functions
A first objective of this study was to evaluate the effects of normal aging on executive functioning,
domain independent skills. This type of analysis on general aviation pilots is relatively uncommon and
is interesting because general aviation pilots is usually a population that presents a high level of
education, supposed to be protective from aging effects. In spite of this high level of education, in our
experiment, the pilots demonstrate a significant decrease in executive functioning in nearly every
neuropsychological measure. Consistently with literature (ex: [54] [55] [56] [57] [58] [59]), the
updating in WM performances was strongly impacted by age, as shown by the decrease of the
percentage of correct answers in the 2-back test with aging. This decline is particularly obvious from
55, where six out of seven participants had a correct response rate below 70%, which is not the case
for any other of the younger participants. This is consistent with Hardy results which showed that
despite of the observation of cognitively preserved people, the number of outliers in WM
performances terms increases significantly after 50. This 2-back test performances decline highlights
an increase difficulty to update the WM content but also in maintenance abilities.
This cognitive decline, despite a high level of education may appears counterintuitive. However,
Tucker-Drob [60] has shown that a high level of education has no neuroprotective effect. According to
the author, education increases cognitive skills in advanced age, but do not influences the rate of the
cognitive decline. In other words, the best cognitive performances bring by a high level of education
are the manifestation of the persistence of higher cognitive abilities already present in the past. It is
not surprising that linear regression were able to find a significant decline with age in our population.
The speed of processing was also reduced with age. This is also a classic effect of normal aging [61]
[62] [63] and it is considered as a factor of cognitive deterioration, and not as a symptom of this
degradation. According to Salthouse [64], the speed at which the neural connections are performed
determines the number of nervous centers that cooperate at the same time, and therefore,
modulates the intellectual performance.
The inhibition performances were also found to be affected by age, as demonstrated by the
increase of the interference score. This decline occurs strongly after 55, where five pilots had a higher
interference rate than any other pilots. Various studies have shown that older adults demonstrate a
greater sensitivity to interference compared to young adults [65] [66]. The lack of inhibition is
considered by some authors as a major cause of decline in cognitive performance. Finally, there was
also a significant increase of perseverative errors, linked with set-shifting performances. This result
must be considered with caution: the correlation was significant mainly in reason of four outliers of
more than 50 that made a high number of perseverative errors. The reasoning abilities seemed to be
relatively preserved, the decline of correct response during the syllogism resolution task was only near
significant (p = .066). This may be explained by the relative low mean age of our sample (43.3, SD =
13.6). Indeed, De Neys [67] has shown that the reasoning performance began to decline drastically
after 65, which is precisely the age of our oldest pilot.
D. Age and experience relationships
The mean total experience of our sample was of 1676 hours of flight (Range = 57-13000). The
Bravais-Pearson correlation revealed that there was no relationship between age and total flight
experience. However, in particular because of three aged pilots that owned a large total flight
experience (respectively 61 and 13000 hours; 61 and 5000 hours; 58 and 6700 hours), the correlation
was close to reach the significance (p = .0561, r = +.39).
E. Correlation of aging and executive functions
With the exception of the reasoning performances, all the neuropsychological variables were
significantly correlated with the age. Bravais-Pearson correlation shown that the three low level
executive functions performances ― updating in WM (p < .001, r = -.73), inhibition (p = .011, r = +57)
and set-shifting (p = .034, r = +.48) ― decreased with age. The speed of processing was also
significantly reduced with age (p < .001, r = +71), see Figure 21, Figure 22, Figure 23 and Figure 24 and
TABLE VI. That was not the case of the reasoning performance that solely showed a trend to decline
(p = .066).
Figure 21. Updating working memory performances as a function of age.
Figure 22. Psychomotor velocity index as a function of age.
Figure 23. Interference score as a function of age.
Figure 24. Number of perseverative errors as a function of age
TABLE VI. CORRELATIONS OF CHRONOLOGICAL AGE WITH THE FIVE NEUROPSYCHOLOGICAL INDICES OF PERFORMANCES (* ≤.05 ; *** ≤.001).
Variables Corrected p-value r
UP in WM < .000*** - .73
Speed of processing < .000*** - .71
Interference = .011* + .57
Set-shifting = .034* + .48
Reasoning = .066 - .38
F. Explanatory variables of the piloting performance
The mean FPD amplitude was 27.69 (SD = 10.38). The ER revealed that the performances of two
cognitive abilities were predictive of the FPD: the reasoning and the WM (respectively, p = .0083,
F(1,15) = 9.20, t = -3.03; p = .0395, F(1.15) = 5.08, t = -2.25). Moreover, the total flight experience was
also a significant explanatory variable (p = .0275, F(1,15) = 5.95, t = -2.44), see Figure 25.
The most the reasoning (see Figure 26 and Figure 27) and the WM abilities were efficient, the
smaller was the FPD. In the same way, the most the pilots were experienced, the smaller was the FPD.
The adjusted r² showed that this model explained 44.51% of the FPD.
As expected, the ER did not revealed any significant effect of age on the piloting performance (p =
.2488, F(1,15) = 5.95, t = -1.19). In the same way, the speed of processing and the two others low level
EFs, set-shifting and inhibition, were not predictive of the flight performance (respectively, p = .5603,
F(1,15) = 0.35, t = -0.59; p = .8979, F(1,15) = 0.17, t = -0.13; p = .9008, F(1,15) = 0.16, t = -0.12), see
Figure 25.
It is interesting to note that the worst piloting performance (FPD = 52.01) has been done by a
rather old pilot (62) with a very small total flight experience (90 hours) whereas two others aged pilots
(both 61) with a high experience (13000 and 5000 hours) demonstrated correct flight performances
(respectively FPD = 21.08 and 32.30).
Figure 25. Synthesis of the ER. The Pareto diagram shows the three predictive variables of the flight performance: the reasoning abilities, the updating and the total flight performance.
Figure 26. FPD as a function of the reasoning performances. The ER revealed that the reasoning performance predicts significantly the FPD.
Figure 27. Flight path of two pilots and their respective reasoning performances. In blue, the pilot had a small flight path deviation and a good reasoning performance (83.3% of correct answers). In pink, the pilot had a large flight path deviation, he lost himself and flew by
mistake above the Blagnac airport. His flight path deviation was very important and his reasoning performances were very low (41.6% of correct answers). Flight path are rendered with FromDady [68], the width of the line codes the altitude.
G. Discriminative variable of the crosswind landing decsion
The discriminant analysis revealed that three variables were predictive of the correct decision do
go-around: updating in WM, the flight experience and the level of motor impulsiveness. The pilots
who made the good decision to go-around demonstrated a better percentage of correct answers in
the 2-back task, a larger total flight experience and a lower motor impulsiveness compared to pilots
who made a poor decision (respectively p < .001, F(1,10) = 21.27, t = + 4.61 ; p = .003, F(1,10) = 14.77,
t = + 3.84 ; p = .037, F(1,10) = 5.73, t = -2.39) (see TABLE VII. The non-planning impulsiveness was
nearly significantly predictive of the decision (p = .059). Neither the age nor the reasoning
performance were predictive of the relevance of the decision making (respectively, p = .534; p = .728).
The classification matrix show that this model correctly discriminate 100% of the pilots who made a
poor decision and 91.6% of the pilot who choose to go-around. It is interesting to note that a separate
analysis showed that the level of education was negatively correlated to the total score of
impulsiveness (p = .043, r = -.456).
TABLE VII. SUMMARY OF DISCRIMINANT ANALYSIS BY EXHAUSTIVE SEARCH: NEUROPSYCHOLOGICAL INDICES OF PERFORMANCES THAT PREDICT CROSSWIND LANDING DECISION (* ≤.05 ; ** ≤.01 ; *** ≤.001).
Variables β Standard error F(1,14) t p
Age .224 .172 .417 .646 .534
TFE .925 .113 14.263 3.776 .004**
Motor Imp. -.627 .123 6.528 -2.555 .030*
Cognitive Imp. .041 .102 .042 .205 .841
Non-planning Imp. -.475 .110 4.630 -2.151 .059
Speed of processing .268 .132 .928 .963 .360
Reasoning -.144 .116 .486 -.697 .503
Up in WM 1.551 .162 20.676 4.547 .001***
Set-shifting -.379 .112 2.584 -1.607 .142
Inhibition .264 .130 1.072 1.035 .327
VIII. DISCUSSION
A. Aging and piloting performance
According to our hypotheses and other authors [6] [7], the chronological age was not a relevant
variable to predicts the piloting performance. However, although the total flight experience was not
correlated with age, it may have played a beneficial effect on some aged pilots. It is interesting to note
that the worst piloting performance has been performed by a rather old pilot with a weak experience,
whereas two others aged pilots, with a high experience, demonstrated quite good flight
performances. In spite of these observations, our results raised the limitation of using the
chronological age as a single criterion to decide if a given pilot is able to fly or not. In accordance with
such statement, Schroeder [69] have pointed out the necessity to use neuropsychological tests rather
than relying on chronological age.
B. Neuropsychological tests and piloting performance
The pilots performed a neuropsychological battery that taped three crucial low-level executive
functions [25] plus reasoning and speed of processing. Finally, as revealed by the ER, reasoning
performance was the variable the most predictive of the ability to pilot in our study. This result is not
surprising, the reasoning abilities were strongly involved in our scenario. The pilots ought to perform
numerous observations during the navigation to estimate their position and they had to use the radio
navigation systems to reach a waypoint. Moreover, the scheduled compass failure required pilots to
use the anti-directional magnetic compass as a backup. The utilization of this instrument is complex
and could be a source of difficulty. Although we did not assess precisely the errors associated with the
use of this instrument, it seems likely that it has participated to increase the path deviation of some
pilots. These results concerning the reasoning are in line with Wiggins and O’Hare [70] that have
highlighted the links between reasoning performance, evaluated by a syllogism resolution (duncker’s
candle problem), and piloting performance. The reasoning performances reflect fluid reasoning,
central cognitive ability linked with various types of mental activity (mental calculation, problem
solving etc.) and essential to the adaptation to novel problems. Complex and novel problems cannot
be solved directly by referring to a store of long-term knowledge but require analytic or fluid
reasoning. The complexity of our scenario with unexpected event like the compass failure appears to
have contributed to a strong involvement of reasoning abilities.
The total flight experience was also predictive of the FPD. In accordance with other studies [12],
this data has confirmed the beneficial impact of experience on flight performance. This is coherent
with Taylor’s results [5] that showed that more expert pilots demonstrated better flight summary
scores, especially in the communication and approach-to-landing. Moreover, this 3- year longitudinal
study showed that aviation expertise was associated with less declines in flight simulator performance
over time.
Finally, updating ability was also linked with the pilot’s performances. This is coherent with our
expectation. Indeed, the pilot’s activity takes place in a dynamical and changing context where new
information must be integrated and updated continuously. The updating performances are crucial in
this context. Another study of Taylor et al. [20] found that the WM and the speed of processing were
predictive of the piloting performance. We are partially in line with these results. We did not retrieve a
significant effect of the speed of processing. The mean age of our sample was relatively low (43.3, SD
= 13.6) and only seven participants of more than fifty were involved in the experiment. We may argue
that more severe aging effects on speed of processing occur later in life, the sample of Taylor was
more extreme and included participants from 50 to 69, these latter probably demonstrated more
pronounced variations of speed of processing. Moreover, the task that we used to assess the speed of
processing had a strong motor component that could have been less relevant to flight performance
assessment.
Our overall results suggest that “cognitive age” is a better criterion than “chronological age” to
predict the ability to fly and that reasoning and updating are good candidate to assess the cognitive
age. The design of such neuropsychological batteries of tests that could be administrated during the
pilot’s periodic physical examinations could help to detect cognitive impairment associated with
increased risk of accidents.
C. The landing decision relevance
Many studies have tried to understand the difficulty of pilots to review their flight plan [71] [72]
[73] [74], especially during the final approach [75]. The updating in WM, the flight experience and the
motor impulsiveness were predictive of the landing decision relevance. Muthard [73] had already
mentioned the great difficulties encountered by some pilots to integrate critical contextual changes
such as deteriorating weather. The fact that WM performances were linked with the erroneous
decision to continue the landing is in line with this author. Indeed, updating in WM performances may
modulate the ability to integrate the meteorology evolution during the piloting scenario course.
Moreover, the aircraft's maximum crosswind limit was not recollected in WM during the critical time
of the approach. This inability to recollect critical information and to update their situation awareness
seems to lead pilots to erroneously continue the landing. The mental calculation of the crosswind is
not to incriminate, no error of that type was observed in our pilot, this is consistent with the fact that
reasoning performance was not a predictive variable. This result provides new insight into the get-
home-itis syndrome, particularly hazardous in commercial aviation [75] and general aviation, where it
is accounting for over 41.5% of casualties [76]. Assessing the integrity of the pilot’s WM seem to be a
major challenge for aviation safety since many studies have also demonstrated that its predicts the
flight performance [71] [72] [73] [74], or the ability to perform radio communications [77][78][79].
The total flight experience was also predictive of the landing decision relevance. Indeed, that was
the least experienced pilots who were more likely to make poor decision. These results are quite
consistent with those of Wiegman [80] who has shown that the time and distance traveled into
adverse weather prior to diverting were negatively correlated with pilots' flight experience. These
findings and our results suggest that landing continuation may be attributable, at least in part, to poor
situation assessment, consequence of a lack of experience.
Finally, the level of motor impulsiveness, habitual propensity of a person to act without fully
considering the consequences of his actions, was also a relevant predictor of the erroneous decision
to land. To our knowledge, no study has linked the level of impulsiveness with the plan continuation
error. According to Sicard [36], increased impulsivity is deleterious to the relevance of decision-making
because of increased risk-taking that it generates. Goh [71] highlighted the fact that overconfidence is
seven times more frequently cited as the main cause of accidents in those of the VFR-IMC (Visual
Flight Rules - Into instrument meteorological conditions) category. Laboratory experiments have
shown that pilots that present a trend to fly in VFR-IMC evaluate themselves as more skilled and think
that they make more relevant decision than those who does not demonstrate a trend to fly in VFR-
IMC. O’Hare and Smitheram [81] support the hypothesis that the accidents that occur in degraded
condition type VFR-IMC are largely due to poor risk assessments.
The identification of the brain mechanisms related to pilot error requires much progress. However,
the definition of such batteries of neuropsychological tests, administered to pilots during their medical
examination, would be a significant step forward in terms of aviation safety, particularly when obvious
executive decline are observed. In the same way, psychological evaluation like the impulsiveness level
seems to be relevant. A future protocol will include a larger sample of pilot, including older ones.
ACKNOWLEDGMENT
The authors wish to thank all the pilots who participate to this experiment. The study was
supported by a “Gis Longévité” grant, by the DGA grant 0434019004707565 and the Midi-Pyrenees
Regional Council grants 03012000 and 05006110.
REFERENCES
[1] Hardy, D. J., & Parasuraman, R. (1997). Cognition and flight performance in older pilots. Journal of experimental psychology applied,
3, 313-348.
[2] Li, G., & Baker, S. (2007). Crash risk in general aviation. Jama, 297(14), 1596.
[3] Harkey, J. A. Y. (1996). Age-related changes in selected status variables in general aviation pilots. Transportation Research Record:
Journal of the Transportation Research Board, 1517(-1), 37-43.
[4] Kay, G. (2001). Effects of aging on aviation-related cognitive function in commercial airline pilots. Paper presented at the In
Proceedings of the 72nd Meeting of the Aerospace Medical Association, Aviat Space Environ Med 2001; 72 :241.
[5] Taylor, J., Kennedy, Q., Noda, A., & Yesavage, J. (2007). Pilot age and expertise predict flight simulator performance: A 3-year
longitudinal study. Neurology, 68(9), 648.
[6] Li, G., Baker, S., Grabowski, J., Qiang, Y., McCarthy, M., & Rebok, G. (2003). Age, flight experience, and risk of crash involvement
in a cohort of professional pilots. American journal of epidemiology, 157(10), 874-880.
[7] Li, G., Baker, S., Lamb, M., Grabowski, J., & Rebok, G. (2002). Human factors in aviation crashes involving older pilots. Aviation,
space, and environmental medicine, 73(2), 134-138.
[8] Schroeder, D., Harris, H., & Broach, D. (2000). Pilot age and performance: an annotated bibliography. Oklahoma City, OK: Federal
Aviation Administration, Civil Aeromedical Institute. Human Resources Research Division.
[9] Buckner, R. L. (2004). Memory and Executive Function in Aging and AD Multiple Factors that Cause Decline and Reserve Factors
that Compensate. Neuron, 44(1), 195-208.
[10] Morrow, D., Menard, W., Ridolfo, H., Stine-Morrow, E., Teller, T., & Bryant, D. (2003). Expertise, cognitive ability, and age effects
on pilot communication. The International Journal of Aviation Psychology, 13(4), 345-371.
[11] Wiggins, M., & O’Hare, D. (1995). Expertise in aeronautical weather-related decision making: A cross-sectional analysis of general
aviation pilots. Journal of Experimental Psychology: Applied, 1(4), 305-320.
[12] Morrow, D., Menard, W., Stine-Morrow, E., Teller, T., & Bryant, D. (2001). The influence of expertise and task factors on age
differences in pilot communication. Psychology and Aging, 16(1), 31-46.
[13] Tsang, P., & Shaner, T. (1998). Age, attention, expertise, and time-sharing performance. Psychology and Aging, 13, 323-347.
[14] Taylor, J., Yesavage, J., Morrow, D., Dolhert, N., Brooks III, J., & Poon, L. (1994). The effects of information load and speech rate on
younger and older aircraft pilots' ability to execute simulated air-traffic controller instructions. The Journal of Gerontology, 49(5), P191.
[15] Knapp, C., & Johnson, R. (1996). F-16 Class A mishaps in the US Air Force, 1975-93. Aviation, space, and environmental medicine,
67(8), 777.
[16] O'Hare, D., Wiggins, M., Batt, R., & Morrison, D. (1994). Cognitive failure analysis for aircraft accident investigation. Ergonomics,
37(11), 1855-1869.
[17] Horst, R., & Kay, G. (1991). COGSCREEN- Personal computer-based tests of cognitive function for occupational medical
certification.
[18] Kay, G. (1995). CogScreen: Aeromedical Edition. Odessa, Florida: Psychological Assessment Resources: Inc.
[19] Yakimovitch, N., Strongin, G., Go'orushenko, V., Schroeder, D., & Kay, G. (1994). Flight performance and CogScreen test battery in
Russian pilots.
[20] Taylor, J., O'Hara, R., Mumenthaler, M., & Yesavage, J. (2000). Relationship of CogScreen-AE to flight simulator performance and
pilot age. Aviation, space, and environmental medicine, 71(4), 373.
[21] Raz, N. (2000). Aging of the Brain and Its Impact on Cognitive Performance: Integration of Structural and Functional Findings. The
handbook of aging and cognition.
[22] Hardy, D., Satz, P., D Elia, L., & Uchiyama, C. (2007). Age-Related Group and Individual Differences in Aircraft Pilot Cognition. The
International Journal of Aviation Psychology, 17(1), 77-90.
[23] Taylor, J., O’Hara, R., Mumenthaler, M., Rosen, A., & Yesavage, J. (2005). Cognitive ability, expertise, and age differences in
following air-traffic control instructions. Psychology and Aging, 20(1), 117-133.
[24] Royall, D. R., Lauterbach, E. C., Cummings, J. L., Reeve, A., Rummans, T. A., Kaufer, D. I., et al. (2002). Executive Control Function
A Review of Its Promise and Challenges for Clinical Research. A Report From the Committee on Research of the American
Neuropsychiatric Association (Vol. 14, pp. 377-405): Am Neuropsych Assoc.
[25] Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D. (2000). The unity and diversity of executive
functions and their contributions to complex “frontal lobe” tasks: A latent variable analysis. Cognitive Psychology, 41(1), 49-100.
[26] Sanfey, A. G., Hastie, R., Colvin, M. K., & Grafman, J. (2003). Phineas gauged: decision-making and the human prefrontal cortex.
Neuropsychologia, 41(9), 1218-1229.
[27] Colom, R., Rubio, V., Shih, P., & Santacreu, J. (2006). Fluid intelligence, working memory and executive functioning. PSICOTHEMA-
OVIEDO-, 18(4), 816.
[28] Cabeza, R., Anderson, N. D., Locantore, J. K., & McIntosh, A. R. (2002). Aging gracefully: compensatory brain activity in high-
performing older adults. Neuroimage, 17(3), 1394-1402.
[29] Tisserand, D. J., & Jolles, J. (2003). Special issue on the involvement of prefrontal networks in cognitive ageing. Cortex, 39, 1107-
1128.
[30] De Neys, W., & Van Gelder, E. (2009). Logic and belief across the lifespan: the rise and fall of belief inhibition during syllogistic
reasoning. Developmental Science, 12(1), 123-130.
[31] West, R., & Baylis, G. C. (1998). Effects of increased response dominance and contextual disintegration on the Stroop interference
effect in older adults. Psychology and Aging, 13, 206-217.
[32] Van der Linden, M., Bredart, S., & Beerten, A. (1994). Age-related differences in updating working memory. British journal of
psychology(1953), 85(1), 145-152.
[33] Nagel, I. E., Chicherio, C., Li, S. C., von Oertzen, T., Sander, T., Villringer, A., et al. (2008). Human aging magnifies genetic effects on
executive functioning and working memory. Frontiers in Human Neuroscience, 2.
[34] Decker, S. L., Hill, S. K., & Dean, R. S. (2007). Evidence of construct similarity in executive functions and fluid reasoning abilities.
International Journal of Neuroscience, 117(6), 735-748.
[35] Roca, M., Parr, A., Thompson, R., Woolgar, A., Torralva, T., Antoun, N., et al. (2009). Executive function and fluid intelligence after
frontal lobe lesions. Brain.
[36] Sicard, B., Taillemite, J., Jouve, E., & Blin, O. (2003). Risk propensity in commercial and military pilots. Aviation, space, and
environmental medicine, 74(8), 879-881.
[37] Moeller, F., Barratt, E., Dougherty, D., Schmitz, J., & Swann, A. (2001). Psychiatric aspects of impulsivity. American Journal of
Psychiatry, 158(11), 1783.
[38] Ainslie, G. (1975). Specious reward: A behavioral theory of impulsiveness and impulse control. Psychological bulletin, 82(4), 463-496.
[39] Martin, L., & Potts, G. (2009). Impulsivity in decision-making: An event-related potential investigation. Personality and Individual
Differences, 46(3), 303-308.
[40] Pietrzak, R., Sprague, A., & Snyder, P. (2008). Trait impulsiveness and executive function in healthy young adults. Journal of
Research in Personality, 42(5), 1347-1351.
[41] Hinson, J., Jameson, T., & Whitney, P. (2003). Impulsive decision making and working memory. Journal of Experimental Psychology-
Learning Memory and Cognition, 29(2), 298-306.
[42] Keilp, J., Sackeim, H., & Mann, J. (2005). Correlates of trait impulsiveness in performance measures and neuropsychological tests.
Psychiatry research, 135(3), 191-201.
[43] Oldfield, R. (1971). The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia, 9(1), 97.
[44] Bayle, F., Bourdel, M., Caci, H., Gorwood, P., Chignon, J., Ades, J., et al. (2000). Factor analysis of french translation of the Barratt
impulsivity scale (BIS-10). Canadian journal of psychiatry. Revue canadienne de psychiatrie, 45(2), 156.
[45] Loubinoux, I., Tombari, D., Pariente, J., Gerdelat-Mas, A., Franceries, X., Cassol, E., et al. (2005). Modulation of behavior and cortical
motor activity in healthy subjects by a chronic administration of a serotonin enhancer. Neuroimage, 27(2), 299-313.
[46] Fitts, P. (1954). The information capacity of the human motor system in controlling the amplitude of movement. Journal of
Experimental Psychology, 47, 381–391.
[47] Chen, Y., Mitra, S., & Schlaghecken, F. (2008). Sub-processes of working memory in the N-back task: an investigation using ERPs.
Clinical Neurophysiology, 119(7), 1546-1559.
[48] Causse, M., Sénard, J.-M., & Démonet, J.-F., Pastor, J. (2009). Monitoring cognitive and emotional processes through pupil and cardiac
response during dynamic vs. logical task. Applied Psychophysiology and Biofeedback. DOI 10.1007/s10484-009-9115-0.
[49] Berg, E. A. (1948). A simple objective technique for measuring flexibility in thinking. The Journal of general psychology, 39, 15.
[50] Eling, P., Derckx, K., & Maes, R. (2008). On the historical and conceptual background of the Wisconsin Card Sorting Test. Brain and
Cognition.
[51] Heaton, R. K. (1981). A MANUAL for the Wisconsin Card Sorting Test. Odessa, Florida 33556: Psychological Assessment Resources,
Inc.
[52] DeSoto, M., Fabiani, M., Geary, D., & Gratton, G. (2001). When in doubt, do it both ways: Brain evidence of the simultaneous
activation of conflicting motor responses in a spatial Stroop task. Journal of Cognitive Neuroscience, 13(4), 523-536.
[53] Holm, S. (1979). A simple sequentially rejective multiple test procedure. Scandinavian Journal of Statistics, 6(2), 65-70.
[54] Gregoire, J., & Van Der Linden, M. (1997). Effect of age on forward and backward digit spans. Aging, Neuropsychology, and
Cognition, 4(2), 140-149.
[55] Orsini, A., Chiacchio, L., Cinque, M., Cocchiaro, C., Schiappa, O., & Grossi, D. (1986). Effects of age, education and sex on two tests
of immediate memory: a study of normal subjects from 20 to 99 years of age. Perceptual and motor skills, 63(2 Pt 2), 727.
[56] Taylor, J., Kennedy, Q., Noda, A., & Yesavage, J. (2007). Pilot age and expertise predict flight simulator performance: A 3-year
longitudinal study. Neurology, 68(9), 648.
[57] Taylor, J., O'Hara, R., Mumenthaler, M., & Yesavage, J. (2000). Relationship of CogScreen-AE to flight simulator performance and
pilot age. Aviation, space, and environmental medicine, 71(4), 373.
[58] Verhaeghen, P. (1993). Marcoen, A, & Goossens, L. Facts an& fiction about memory aging: A quantitative integration of research
findings. Journal of Gerontology: Psychological Sciences, 48, 157-171.
[59] Yesavage, J., Dolhert, N., & Taylor, J. (1994). Flight simulator performance of younger and older aircraft pilots: effects of age and
alcohol. Journal of the American Geriatrics Society, 42(6), 577.
[60] Tucker-Drob, E. M., Johnson, K. E., & Jones, R. N. (2009). The cognitive reserve hypothesis: A longitudinal examination of age-
associated declines in reasoning and processing speed. Developmental psychology, 45(2), 431.
[61] Birren, J. E., & Botwinick, J. (1955). Speed of response as a function of perceptual difficulty and age. Journal of Gerontology, 10(4),
433.
[62] Houx, P. J., Jolles, J., & Vreeling, F. W. (1993). Stroop interference: aging effects assessed with the Stroop Color-Word Test.
Experimental Aging Research, 19(3), 209-224.
[63] Salthouse, T. A. (1996). The processing-speed theory of adult age differences in cognition. Psychological review, 103(3), 403-427.
[64] Salthouse, T. A. (1991). A theory of cognitive aging: Elsevier.
[65] Bruyer, R., Van der Linden, M., Rectem, D., & Galvez, C. (1995). Effects of age and education on the Stroop interference. Archives de
psychologie(Genève), 63(247), 257-267.
[66] West, R., & Baylis, G. C. (1998). Effects of increased response dominance and contextual disintegration on the Stroop interference
effect in older adults. Psychology and Aging, 13, 206-217.
[67] De Neys, W., & Van Gelder, E. (2009). Logic and belief across the lifespan: the rise and fall of belief inhibition during syllogistic
reasoning. Developmental Science, 12(1), 123-130.
[68] Hurter, C., Tissoires, B., & Conversy, S. (2009). FromDaDy: spreading data across views to support iterative exploration of aircraft
trajectories. IEEE Transactions on Visualization and Computer Graphics.
[69] Schroeder, D., Harris, H., & Broach, D. (1999). Pilot age and performance: An annotated bibliography (1990-1999).
[70] Wiggins, M., & O’Hare, D. (1995). Expertise in aeronautical weather-related decision making: A cross-sectional analysis of general
aviation pilots. Journal of Experimental Psychology: Applied, 1(4), 305-320.
[71] Goh, J., & Wiegmann, D. (2001). Visual flight rules flight into instrument meteorological conditions: An empirical investigation of the
possible causes. The International Journal of Aviation Psychology, 11(4), 359-379.
[72] Goh, J., & Wiegmann, D. (2002). Human factors analysis of accidents involving visual flight rules flight into adverse weather.
Aviation, space, and environmental medicine, 73(8), 817.
[73] Muthard, E., & Wickens, C. (2003). Factors that mediate flight plan monitoring and errors in plan revision: Planning under automated
and high workload conditions.
[74] Orasanu, J., Ames, N., Martin, L., & Davison, J. (2001). Factors in Aviation Accidents: Decision Errors. Linking expertise and
naturalistic decision making, 209.
[75] Rhoda, D., & Pawlak, M. (1999). An assessment of thunderstorm penetrations and deviations by commercial aircraft in the terminal
area. Massachusetts Institute of Technology, Lincoln Laboratory, Project Report NASA/A-2, 3.
[76] BEA. (2000). Objectif : destination. Technical report, Bureau Enquête Analyse.
[77] Morrow, D., Menard, W., Stine-Morrow, E., Teller, T., & Bryant, D. (2001). The influence of expertise and task factors on age
differences in pilot communication. Psychology and Aging, 16(1), 31-46.
[78] Morrow, D., Ridolfo, H., Menard, W., Sanborn, A., Stine-Morrow, E., Magnor, C., et al. (2003). Environmental support promotes
expertise-based mitigation of age differences on pilot communication tasks. Psychology and Aging, 18(2), 268-284.
[79] Taylor, J., O’Hara, R., Mumenthaler, M., Rosen, A., & Yesavage, J. (2005). Cognitive ability, expertise, and age differences in
following air-traffic control instructions. Psychology and Aging, 20(1), 117-133.
[80] Wiegmann, D., Goh, J., & O'Hare, D. (2002). The role of situation assessment and flight experience in pilots' decisions to continue
visual flight rules flight into adverse weather. Human factors, 44(2), 189.
[81] O'Hare, D., & Smitheram, T. (1995). "Pressing on" into deteriorating conditions: An application of behavioral decision theory to pilot
decision making. The International Journal of Aviation Psychology, 5(4), 351-370.
Part 4
Embedded eye tracker in a real aircraft: new perspectives on pilot/aircraft interaction monitoring
Presented at ICRAT 08
Frédéric Dehais Centre Aéronautique et Spatial ISAE-SUPAERO, University of Toulouse
Toulouse, France [email protected]
Mickaël Causse Centre Aéronautique et Spatial ISAE-SUPAERO; INSERM, U825, University of Toulouse. Toulouse, F-31000 France
Josette Pastor INSERM, U825, Toulouse, F-31000 France;
University Paul Sabatier, University of Toulouse, Toulouse, F-31000 France. [email protected]
Abstract—Currently, online assessment of the aircrew performance focuses on behavioural data (flight
data and pilot’s actions) and the detection may intervene too late for coping with the situation
degradation. An early assessment of the pilot’s “internal state”, based on physiological data collected from
his autonomous nervous system (ANS) and predictive of his behaviour, is necessary. These data give clues
both on the cognitive activity and on the emotional states and stress. The integration of ANS devices in a
cockpit presents practical drawbacks and their use is often limited to simulators. In this preliminay study,
the pros and cons of the adaptation of a standalone eye tracker in a light aircraft are presented. In spite of
a sensitivity to light conditions and a definition of areas of interest limited to a part of the cockpit, the eye
tracker has provided interesting behavioural (fixations) and physiological (pupillometry) measures in
nominal (from take-off to landing) and degraded (provoke a simulated engine failure and plane down
toward the airfield) conditions. The pilots spent less time glancing at the instruments, and focused on less
instruments in the degraded condition. Moreover, the pupil size varied with the flight phases in the
degraded condition, which reflected the variations of stress and attention levels. These encouraging results
open two tracks: the development of new eye trackers able to overcome current technical limitations, and
neuroergonomics researches providing guidelines for new man-machine interfaces integrating both flight
and crew state vectors.
Keywords: Neuroergonomics, Eye Tracker, pilot activity, pupillometry, human factors
I. INTRODUCTION
As long as the human operator (i. e. the pilot) is a key agent in charge of the flight, the definition of
metrics able to predict his performance is a great challenge. Currently, online assessment of the
aircrew performance focuses on behavioural data measured from the pilot/aircraft interactions. In
particular, formal methods are developed to detect human errors thanks to aircraft/pilot behaviour
monitoring [1, 2]. A predictive approach based on particle Petri nets [3] is also proposed to anticipate
possible pilot-systems conflicts [4] that are accurate precursors of the pilot’s loss of situation
awareness [5]. However, all these methods, which rely on the operator’s actions, may intervene too
late for coping with the situation degradation. An early assessment of the pilot’s “internal state”,
predictive of his behaviour, should tackle the problem. Physiological data collected from the pilot’s
autonomous nervous system (ANS) are good candidates since they give clues both on the cognitive
activity and on emotional states and stress [6] [7]. Arousal, vigilance, emotional states, attentional
demand can be derived from heart rate and blood pressure, theta and alpha brain waves,
temperature variations, respiration, skin conductance and oculometry [8] [9].
All ANS devices though present practical drawbacks. They are sensitive to the operator’s physical
state (e.g. sweating perturbs skin conductance, fever changes temperature and heart responses ...) or
to the environment (e.g. magnetic and electric fields may create artefacts on electroencephalograph
responses), and they are too cumbrous to be easily adapted to a cockpit. For example, oculometry
suffers the following limitations:
• As eye fixations “fill up” the total time, not all fixations are relevant to assess the pilot’s visual
demand. Moreover, a fixation does not necessarily imply perception [10];
• As the pupil diameter varies in function of light intensity to maximise visual capacity,
pupillometry cannot be used to assess the level of stress of the pilot under changing light
environments ;
• Electro-oculograms and most eye trackers are cumbersome devices and may disrupt pilots’
activity since pilots have to wear an electrode close to the eye or special equipment like
helmets.
These considerations tend to restrict the utilisation of ANS devices to controlled studies in
laboratory (i.e. flight simulator) [11] [12], although they have already been used in real flight
conditions [13]. Eye tracking offers a fruitful perspective since visual perception is a key for pilot to
control the flight and oculometry may provide both behavioural and cognitive/emotional physiological
measures [7] to assess the pilot’s performance:
• The visual-search strategy, or the selective attention to relevant visual stimuli is an index of
information needs [14];
• The eye-scanning patterns of pilots in terms of frequency of fixations seem to be related to the
instruments’ importance. The length of fixations, however, is related to the difficulty in
obtaining/interpreting information from instruments [15];
• An increase in workload is accompanied by increased fixation times [16] [17];
• The decrease of the duration and the number of eye blinks are strongly correlated to visual
demand [18] [9];
• Low frequency “pupillary oscillations” are linked to fatigue [19];
• The pupillary response is related to mental workload [20] [21] [22];
• In many cognitive processes such as language processing, perception, memory, complex
reasoning and attention, the pupil diameter grows with the difficulty of the task [23] [24];
• Pupillary responses also provide clues on the emotional state [25] [26] and pupil size may vary
on a continuum according to emotional valence [27] or reflect the emotional activation or
arousal [28].
Our long-term goal is to develop an onboard system able to predict the pilot’s performance
through the analysis of the aircraft state vector, the pilot-aircraft interactions state vector and the
pilot’s physiological state vector. However, the integration of this latter state vector implies to assess
the feasibility and the acceptability of a non intrusive physiological device. Thus, this preliminary study
proposes to assess the usability of an on-board eye tracker in real flights and aims at assessing the
benefits of this tool for human factor concerns.
II. METHODS
A. Participants
A permit to fly was given by the European Aviation Safety Agency (number 856/2007 – EASA
PTF.A07.0232) to conduct the experimentation with the restriction that ISAE flight instructors only
were authorised to fly the airplane with the on-board eye tracker. Six ISAE flight instructors, all males,
could participate to the experiment. Their mean age was 43 years (range, 35-58). Their mean flying
experience was 5896 hours (range, 1480-13000). The six participants were qualified to fly the Aquila
AT01 aircraft (two-seated light airplane, 100 horsepower).
B. Scenario
The flight scenario starts at nightfall at Lasbordes airfield and ends before the beginning of the
aeronautical night3. It is divided into two consecutive sequences (cf. fig 1):
• The first sequence consists in a classical visual traffic pattern : take off (1) – cross wind leg (2) –
down wind leg (3) - base leg (4) – last turn (5) – final leg (6) - touch and go (7) . This sequence is
the nominal condition;
• The second sequence starts after the previous touch and go (7) and consists in flying back
toward Lasbordes airfield at an altitude of 2500 feet. Once over the airfield, the pilot had to
chop the throttle so as to perform an engine failure exercise (8) and then to plane down
toward the airfield (9-12). This sequence is the degraded condition.
This scenario is presented to the pilot one hour before the beginning of the experimentation.
During the briefing, it is clearly exposed to each pilot that he decides the moment of the engine failure
exercise and that he may use the throttle at any moment if flight safety is altered, see Figure 28.
Figure 28: The two flight sequences: the nominal and the degraded one. The nominal sequence ends after the landing (6), the degraded sequence starts just after, with the take off (7)
C. Oculometry
A non intrusive eye tracker Tobii x50 was used for the purpose of the experimentation. This device
has 0.5 degree of accuracy and a 50 HZ sampling rates. It also provides instant re-acquisition after
3 The aeronautical night begins in France thirty minutes after sunset.
extreme head motion. It had to be adapted to be easily set in the Aquila aircraft without any
modification of the airplane and without provoking any disturbance for the pilot (e.g. no visual
scanning disturbance).
The eye tracker was placed below the left part of the instrument panel in front of the pilot’s seat
(cf. Figure 29). A scene camera was centrally placed under the fix part of the canopy. Data
synchronization and processing was done via an analog/numerical video converter, a Tobii external
card and a Sony Vaio laptop. These three light devices were situated in the luggage compartment.
Figure 29: The eye tracker (ET) was fixed under the instrument panel and the data processing system was placed in the luggage compartment. This latter processed and surimposed in real time both the eye tracker data and the video data coming from the scene camera.
The technical characteristics of the x50 eye tracker and its particular location allowed to track the
pilot’s eye gaze on the left part of the instrument panel where are the primary flight beacons (i.e.
airspeed, altimeter, horizon…) As shown in the Figure 30 and in the Figure 31, it is not possible to
determine the pilot’s eye gaze out of this area (i.e. the eye gaze out of the cockpit, the eye gaze on the
beacons situated on the right part).
Figure 30: The blue dot represents the gaze fixation. Note that the pilot is focusing on the airspeed instrument to check the rotation speed (Vr), just before taking-off
1) Area of interests A dedicated analysis software provides in real time data such as the timestamps, the (x,y)
coordinates of the pilot’s eye gaze on the left instrument panel and the pupil diameter. Moreover it is
possible to determine the number and the duration of fixations in specific “areas of interests”. In this
sense and in order to study the pilot’s ocular behaviour, fourteen areas of interest corresponding to
the fourteen beacons of the left instrument panels have been respectively defined: (1) outside
temperature, (2) compass, (3) manifold pressure, (4) alarm, (5) airspeed, (6) horizon, (7) altimeter,
(8) tachymeter, (9) bank and turn indicator, (10) directional gyroscope, (11) vertical speed, (12) VOR,
(13) switches (14) flaps (cf. fig 4).
Figure 31: The fourteen rectangles define the different area of interests.
2) Pupil diameter variations As the eye gaze, the pupil size is recorded continuously. In practice, establishing mean physiological
values for a group of subjects for an entire task is meaningless because of inter-individual variability.
We use delta values (differences between the mean pupil diameter during the concerned flight
sequence and the one calculated on the whole experiment) for measuring the pupil variations.
Luminosity measurements were performed. Indeed the pupil regulates the amount of light that
enters the eye, and thus, its size variations are highly sensitive to the luminosity changes. Thanks to a
lux-meter the ambient luminosity was continuously recorded in order to identify the flight sequences
where the light remains reasonably constant. In accordance with Gupta’s work [30] on pupil size
variations in function of the ambient luminosity, we have limited the pupil variations analysis to
sequences where the visible light was inferior to 25 lux.
III. RESULTS
As the sample size was small and the data did not follow normal distributions, nonparametric
statistical methods for dependent sample were used. Overall analyses were performed with the
Friedman Anova. Wilcoxon signed rank test was used for paired-samples tests. The analyses were
performed with Statistica 7.1 (© StatSoft).
A. Behavioural results
The different fixation times, expressed in percentage of the total time spent on the 14 defined
instruments during the nominal landing sequence vs. the landing sequence with the simulated failure
(the degraded landing sequence), are presented in Figure 32.
Figure 32: Mean fixation durations in percentages on the areas of interest of the six pilots during the nominal and the degraded landing, from base leg until the flare.
The results showed a reduction of the number of instruments gazed during the degraded sequence
in comparison to the nominal one. During the nominal sequence, all the instruments were looked
(except for the alarm panel) whereas only 10 instruments were looked during the degraded sequence.
More precisely, compass, switches and tachymeter weren’t gazed. Moreover, there was an increase of
the relative fixation time on the airspeed during the degraded sequence regarding to the nominal
(77.49% vs. 58.12%).
Finally, the time spent on instruments appeared to be lower during the degraded sequence (Figure
33), showing that pilots focused more on external information.
Figure 33: Mean fixation time on the airspeed instrument and all other instruments during the nominal landing and the degraded landing (time in sec)
Below is presented the official “cruise checklist” of the Aquila AT01 (TABLE VIII. ) and the fixation
times in percentages (Figure 34) obtained during the cruise check list (generally performed at an
altitude of 2000 feet).
TABLE VIII. TABLE 1: AQUILA ATO1 OFFICIAL “CRUISE CHECKLIST” [31]
Trim setChronometer Top and estimatedAltimeter setDirectional CheckedGPS Use & Stby stateEngine instruments CheckedManifold pressure 25 inchesTachymeter 2000 RPM
Figure 34: Mean fixation times in percentages on the 14 zones of interest during the “cruise checklist”. Note that “airspeed” stands for “airspeed indicator”, “directional” stands for directional gyroscope, “horizon” stands for gyro horizon and “bank and turn” stands for “bank and turn indicator” (mean total duration = 7.27 sec)
The analysis shows that the tachymeter is the most looked instrument (46.71%), and then comes
the manifold pressure (29.78%) and the airspeed (14.12%).
B. Pupillary response
In spite of the fact that all the experiments were conducted at nightfall, the luminosity variation
did not allow to analyse pupil diameter variations during all the sequences (TABLE IX. )
Considering this pitfall, only the degraded sequences of four pilots (pilot ID: 1 to 4) were included
in the pupil diameter variation analysis. Indeed, mean luminosity variations during the considered
sequences were only of 5.75 lux for the four pilots.
TABLE IX. LUMINOSITY MEASUREMENTS FOR EACH PILOT AT THE BEGINNING AND THE END OF THE TWO FLIGHT SEQUENCES (E.G. PILOT ID 1 STARTED THE FIRST FLIGHT SEQUENCE WITH 127 LUX AND ENDED IT WITH 40 LUX ; HE THEN STARTED THE SECOND SEQUENCE WITH 20
LUX AND ENDED IT WITH 7 LUX).
The Friedman’s ANOVA showed a strong significant difference (p=0.001) concerning the delta
pupillary diameter among the four flight phases (Figure 35). However, Wilcoxon post hoc paired-
samples analysis didn’t show any difference.
Figure 35: Pupillary diameter changes (in mm) regarding the four flight phases during the second flight sequences, respectively one minute before the failure (1), the failure and the crosswind (2), the base leg (3), from the last turn to the final touch (4)
IV. DISCUSSION
The introduction of a new onboard device for human factors purposes must fulfil three
requirements: 1) the device does not disturb the pilot’s activity, 2) the device is able to work correctly
Luminosity variations (in lux)
Pilot ID first sequence second sequence
Subject 1 127-40 20-7
Subject 2 82-36 8-4
Subject 3 90-35 10-6
Subject 4 93-33 10-12
Subject 5 473-230 220-91
Subject 6 610-600 520-380
in real flight conditions, 3) the device improves significantly the human-machine interface, and
therefore the flight safety. The preliminary neuroergonomics work that is presented here gives some
clues about the capabilities of an eye tracker as a device onboard a small aircraft.
Observation of the six pilots showed that no perturbation was generated by the eye tracker that
remained totally unnoticed after the preliminary set up phase. However, this must be confirmed on
non expert pilots and on other types of aircrafts.
Concerning the usability of the eye tracker in real flight conditions, the results are more equivocal.
Because of light issues, we weren’t able to compare pupil dilation during the nominal sequence vs. the
degraded one in all pilots. Moreover, AOIs can be only defined at places that are constrained by the
eye tracker’s location in the plane. This can be overcome by helmet eye trackers, however with an
intrusiveness of the device in the pilots’ activity. Technological progress must thus be accomplished to
allow the generalisation of onboard eye tracking experiments.
In spite of these drawbacks, the preliminary results highlight the possibility of deriving interesting
measures of the pilot’s activity from eye tracker data.
Analysis of the areas of interest captured by the eye tracker is a way to assess its accuracy in real
conditions. In particular, analysis of critical events such as the in-flight checklist sequences is a key to
evaluate the reliability of this tool:
• The visual scanning is codified by an official procedure in the flight manual than can be used as a
model of reference;
• These sequences are very short (less than ten seconds) and flight parameters to be checked by
the pilot are vital.
Such constraints lead the pilots to relevant eye fixations during these periods and allow assuming
that the areas of interest observed are also a priori the result of a real voluntary attentional activity.
In this perspective, the analysis of the areas of interest during the “cruise checklist” (cf. Fig 5)
shows that the visual scanning of the six pilots is limited to eight flight instruments. More precisely the
pilots have focused on the tachymeter, the manifold pressure, the airspeed indicator, the altimeter,
the directional gyroscope, the compass, the bank and turn indicator and the gyro horizon. These areas
of interest are consistent with the ones defined in the official Aquila “cruise checklist”: the tachymeter
and the manifold pressure have to be set to particular values and the results of these adjustments
have to be implicitly checked on the airspeed indicator, the altimeter has to be checked as the “cruise
checklist” starts at an altitude of 2000 feet, and the value of the directional gyroscope has to be
checked, which is done thanks to a quick comparison with the value of the compass.
Though it is not explicitly expressed in this checklist, it is totally consistent that flight indicators
such as gyro horizon and turn and bank indicator are supervised by the pilots in order to be stabilised
perfectly on the three axes (roll, pitch and yaw) to perform an optimal checklist. One may notice that
some actions of the checklist are not detected by the eye tracker but:
• The chronometer and GPS settings were not performed as the pilots were not engaged in a
complex navigation task but stayed close to the airfield;
• The trim or the engine instruments (e.g. oil pressure indicator) are located on the right part of
the instruments panel where no eye tracking could be established due to the limitation in angle
of the Tobii x50.
The analysis of the fixation times on the areas of interest during this checklist showed that the
pilots focused particularly on the tachymeter (46.71 % fixation time on this instrument during that
checklist), the manifold pressure (29.78 %) and the airspeed indicator (14.12 %). Interviews with the
six pilots have confirmed that the engine management during this checklist requires a certain amount
of attentional demand: very accurate and careful adjustments were needed on the tachymeter,
manifold pressure and the speed. The pilots spent less time on checking instruments as the altimeter
(4.48%), the directional gyroscope (1.82 %) or the compass (1.32 %). Indeed, discussion with the pilots
revealed that the altimeter was rapidly looked to check that a 2000 feet altitude was reached (i.e. to
start the cruise checklist). They also just glanced at the directional gyro and the compass: as the pilots
were not about to perform a navigation task, the cross-checking of these two indicators were of less
importance. In this sense, these findings are consistent with research conducted on the correlation
between attentional demand and time fixations [15] [16] [17]: important information implies longer
time fixations.
The analysis and the comparison of the pilots’ areas of interest during the landing in nominal and
degraded conditions (cf. fig 6 and 7) have revealed different ocular patterns. First of all, the total
duration of fixations in nominal conditions was more than two time higher to the total duration of
fixations in degraded conditions (13.59 seconds vs. 5.53 sec). This suggests that in degraded
conditions, the pilots spent more time looking outside the cockpit to assess and adapt their trajectory
in reference with the airfield. Another major difference between the two landings relied on the fact
that the pilot’s areas of interests were less distributed in the degraded conditions than in the
nominal one with a particular focus on the airspeed indicator (77.49% of total fixation times in
degraded landing vs. 58.12% in nominal landing). A first consideration is to take into account the fact
that in the degraded condition, the pilots had no more interest to supervise the tachymeter and the
manifold pressure due to the engine failure. Another consideration is given by the pilots who all
agreed that in the degraded condition they had faced troubles to manage their speed as they were
surprised by the high lift-to-drag ratio of the Aquila. In this sense, this led them to focus especially on
the airspeed indicator and particularly to take a key decision: maintaining the landing or going
around.
The analysis of pupil diameter variations shows some evolutions according to the different flight
sequences. More precisely, mean delta pupil diameter was of -0.25 mm before the simulated failure,
+0.47 mm during the failure and the cross wind leg, -0.10 during the base leg and +0.12 from the last
turn to the final touch. These results are consistent with the pilots’ interviews that report a high
anxiety and cognitive demand due to the management of the aircraft during the few early minutes of
the simulated failure, a lower anxiety during the base leg because of the successful stabilization of the
aircraft, and finally another increase of anxiety and cognitive demand during the landing sequence
because of the required precision and the potential go-around in case of unsafe approach.
Furthermore, the literature classically reports a high cognitive demand and a high ANS arousal during
the landing [32], which is consistent with the increase of pupil diameter observed during the last turn
and the final touch.
During the flying activity, pilots are confronted with numerous stressors that can deplete their
performance, such as time pressure, increased anxiety, and unexpected failure. Whereas a growing
literature [33] [34] sheds light on the effect of complex flight scenarios or anxiety on pilots’
performance and physiological parameters, real flight experiments remain extremely rare. Therefore,
on-board eye tracking offers promising perspectives in term of real condition monitoring of both
pilot’s actions and physiological states, although this ecological approach shows some technical
limitations. Firstly, the analysis of the areas of interest shows the reliability of the tool and its
capability to predict behaviours. Indeed, during a takeoff, it has been possible to link the absence of
visual scanning on the flaps with an omission of a required action on them later. Moreover, the AOI
analysis allows bringing to light differential visual behaviours according to the landing condition
(nominal or degraded). Secondly, the measurement of the pupil dilation gives clues on the pilot’s
emotional state and/or cognitive workload. The pupil response seems to evolve differently during
the four flight phases. The pupil diameter appears to be higher just after the simulated engine
failure. This observation is coherent with the increase of mental demand and/or anxiety during this
particularly critical flight phases. In addition, the occurrence of differential patterns during the failure
phases vs. the nominal ones seems to emphasize pupil diameter results. Further work should be
conducted by night to totally get rid of the luminosity variations and to attempt to produce more
results, in particular concerning the comparison of degraded and nominal conditions.
ACKNOWLEDGMENT
The authors wish to thank all the pilots from ISAE and Fabrice Bazelot, Frank Yvars and Thierry
Louvet for the time they spent on setting up the eye tracker onboard the Aquila, Eric Absil for his great
job on the eye tracker, and Christian Colongo for his support. The eye tracker Tobii x50 has been
kindly lent to ISAE by the firm “Pilot Vision” (Alexander Seger) during the experimentation.
REFERENCES [1] M. Heymann, A. Degani, On formal abstraction and verification of human-machine interfaces: the discrete event case. NASA
Technical Memorandum, 2001. [2] T. Callantine, The crew activity tracking system: Leveraging flight data for aiding, training, and analysis. Proceedings of the 20th
Digital Avionics Systems Conference, Daytona Beach, FL, October, 5.C.3-1--5.C.3-12 (CD- ROM), 2001. [3] C. Lesire, C. Tessier, Particle Petri nets for aircraft procedure monitoring under uncertainty. 26th International Conference on
Application and Theory of Petri Nets and Other Models of Concurrency (ATPN), Miami, Florida, USA, 2005. [4] F. Dehais, C. Lesire, A. Goudou, C. Tessier, Toward an anticipating agent to help pilots AAAI Fall Symposium "From reactive to
anticipatory cognitive embodied systems", Arlington, VA, USA, 2005. [5] F. Dehais, C. Tessier, L. Chaudron, GHOST: experimenting conflicts countermeasures in the pilot's activity. IJCAI-03, Proccedings of
the Eighteenth International Joint Conference on Artificial Intelligence, Acapulco, Mexico, 2003. [6] HD. Critchley, Electrodermal responses: what happens in the brain? Neuroscientist, 8(2):132-42, 2002. [7] E. Granholm, SR. Steinhauer, Pupillometric measures of cognitive and emotional processes. International Journal of Psychophysiology,
52(1):1-6, 2004. [8] M. J. Skinner, P. A. Simpson, “Workload issues in military tactical aircraft,” Int. J. Aviat. Psychol., vol. 12, no. 1, pp. 79–93, 2002. [9] J.A. Veltman, A.W.K. Gaillard, (1996) Physiological indices of workload in a simulated flight task, Biological Psychology ,42, 323-42,
Elsevier, 2002. [10] D. De Waard, The measurement of drivers' mental workload. PhD thesis, University of Groningen. Haren, The Netherlands: University
of Groningen, Traffic Research Centre, 1996. [11] R.J. Mumaw, N.B. Sarter, C.D. Wickens, Analysis of pilots’ monitoring and performance on an automated flight deck. Proceedings of
the 11th biennial meeting of the International Symposium on Aviation Psychology, Columbus, OH, 2001. [12] P. Kasarskis, J. Stehwien, J. Hichox, A. Aretz, C. Wickens, "Comparison of expert and novice scan behaviors during VFR flight," 11th
International Symposium on Aviation Psychology, 2001. [13] J.A. Veltman, A Comparative Study of Psychophysiological Reactions During Simulator and Real Flight. International Journal of
Aviation Psychology,Volume 12, 2002. [14] P.K. Hughes, B.L. Cole, The effect of attentional demand on eye movement behaviour when driving. In A.G. Gale, M.H. Freeman,
C.M. Haslegrave, P. Smith & S.P. Taylor (Eds.), Vision in vehicles-II (pp. 221-230). Amsterdam: North-Holland, 1988. [15] G.F. Wilson, F.T. Eggemeier, Psychophysiological assessment of workload in multi-task environments. In D.L. Damos (Ed.), Multiple-
task performance (pp. 329-360). London: Taylor & Francis, 1991. [16] R.D. O'Donnell, F.T. Eggemeier, Workload assessment methodology. In K.R. Boff, L. Kaufman & J.P. Thomas (Eds.), Handbook of
perception and human performance. Volume II, cognitive processes and performance. (pp 42/1-42/49). New York: Wiley, 1986. [17] R.W Backs, L.C. Walrath, Eye movement and pupillary response indices of mental workload during visual search of symbolic displays.
Applied Ergonomics, 23, 243-254, 1992. [18] L.O. Bauer, R. Goldstein, J.A Stern, Effects of information-processing demands on physiological response patterns. Human Factors,
29, 213-234, 1987. [19] B. Wilhelm, H. Wilhelm, Das Pupilverhalten verrät Übermüdung. Zeitschrift für Verkehrssicherheit, 41, 116-118, 1995. [20] T.I. Shamsi, X. Sam Zheng, B.P. Bailey, Task-evoked pupillary response to mental workload in human-computer interaction. CHI
2004, Vienna, Australia, 2004. [21] L.T.M. Hoeks, The pupillary response as a measure of mental processing load: with application to picture naming. PhD Thesis,
Nijmegen, The Netherlands: University of Nijmegen, 1995. [22] J. Hyönä, J. Tommola, A.-M Alaja, Pupil dilation as a measure of processing load in simultaneous interpretation and other language
tasks. The Quarterly Journal of Experimental Psychology, 48A, 598-612, 1995. [23] J.L. Andreassi, Pupillary response and behaviour. In: Mahwah, N.L.E.A.(Ed.), Psychophysiology. Human behav. Physiol.response,Vol.
10. Oxford University Press, New York, pp. 218–233, 2000. [24] J.R. Jennings, M.W. Van der Molen, Preparation for speeded action as a psychophysiological concept. Psychol. Bull. 131 (3), 434–459,
2005. [25] J. Hytintk, J. Tommola, A-M. Alaja, “Pupil Dilation as a Measure of Processing Load in Simultaneous Interpretation and Other
Language Tasks”. The Quarterly Journal of Experimental Psychology, 48A (3): 598-612, 1995. [26] M. Causse, B. Pavard, J-M. Sénard, J-F. Démonet, J. Pastor, Emotion Induction through Virtual Avatars and its Impact on Reasoning:
Evidence from Autonomous Nervous System Measurements and Cognitive Assessment. Actes de la Conférence VRIC’06, Laval, France, 18-20 Avril, 2007.
[27] E. H. Hess, “Pupillometrics”. In N. S. Greenfield and R. A. Stembach (Eds.) Handbook of Psychophysiology, pages 491- 531. Holt, Richard & Winston, New York, NY, 1972.
[28] MP. Janisse, “Pupil Size, Affect and Exposure Frequency”. 128 Social Behavior and Personality, 2 (2): 125-146, 1974. [29] R. Parasuraman, M. Rizzo, Neuroergonomics: The Brain At Work, Oxford University Press, 2007. [30] A. Gupta, J. Schwiegerling, J. Straub, Design and use of an infrared Pupilometer for real-time pupil mapping in response to incremental
illumination level. Optical society of America, 2000. [31] A. Schimederer, Aquila AT01 Flight Manual, Technical report AFM-AT01-1010-100-E, AQUILA GmbH, 2005. [32] Y-H, Lee, B-S. Liu, Inflight workload assessment: comparison of subjective and physiological measurements. Aviat Space Environ
Med; 74:1078–84, 2003. [33] C. Dussault, J-C. Jouanin, M. Philippe, C-Y. Guezennec, EEG and ECG changes during simulator operation reflect mental workload
and vigilance. Aviat Space Environ Med; 76:344 –51, 2005. [34] T. Macuda, T. Poolman, P. Keller, Workload Assessment in Flight using Dense Array Eeg Schnell. 25th Digital Avionics Systems
Conference, IEEE/AIAA Volume , Issue , Oct. 2006 Page(s):1 – 11, 2006.