Cognitive Control Modes and Mental Workload: An Experimental

Approach

Philippe Rauffet

, Farida Said

, Amine Laouar

1,3

, Christine Chauvin

and Marie-Christine Bressolle

Lab-STICC, Université Bretagne Sud, 17 bd Flandres-Dunkerque, Lorient, France

LMBA, Université Bretagne Sud, 17 bd Flandres-Dunkerque, Lorient, France

IYDN Department, Airbus, Toulouse, France

Keywords: Cognitive Control Modes, Mental Workload, MATB, fNIRS.

Abstract: This study aims to examine the relationships between cognitive control modes and mental workload. It uses

the Multi-Attribute Task Battery (MATB-II) microworld, which reproduced tasks carried out in an aircraft.

Twenty participants performed a main task in different conditions defined by the level of complexity and the

absence or presence of a secondary task. Two types of physiological data were considered as indicative of

mental strain: cardiac activity and oxygenation and deoxygenation of the prefrontal cortex. Besides a classic

relationship between mental stress and mental strain, this study draws attention to a relationship between the

level of complexity and the control modes, which is highly significant for the tactical mode. Furthermore, this

mode is associated with a significantly lower oxygen concentration than that found in the other modes,

indicating lower mental strain. Hence, in this study, the tactical mode is found to be the most efficient one,

since it is associated with a satisfying performance and with low mental strain. It is also the most impacted

by task complexity. This finding should prompt an investigation of possible ways of supporting this mode in

naturalistic situations.

1 INTRODUCTION

Monitoring and process control activities carried out

in dynamic situations are characterized by the

management of uncertainty and risk, the multiplicity

of tasks, and the complexity of the controlled systems

(Hoc, 1996). From a cognitive perspective, these

activities call for diagnostic/prognostic and decision-

making processes that use both internal data

processing (i.e. mental models relating to the

controlled systems or to environmental dynamics)

and external data processing (i.e. information that is

available in the environment or interfaces). Hence, in

dynamic environments, the pilot of a mobile vehicle

(e.g. aircraft, car, ship) is more or less proactive

(when he/she relies on mental models to act) or more

or less reactive (when his/her actions are mainly

driven by external data). This behavioral flexibility is

closely linked to the notion of cognitive control,

which is at the center of two important models in the

https://orcid.org/0000-0002-6179-4348

https://orcid.org/0000-0002-8670-9584

https://orcid.org/0000-0003-3721-8831

field of cognitive ergonomics: Rasmussen’s Skills-

Rules-Knowledge (SRK) model (Rasmussen, 1983)

and Hollnagel’s Contextual Control Model

(COCOM) (Hollnagel, 1993). Those models are

well-known. However, the different modes of control

they identify have rarely been “quantified” or

evaluated from a neurophysiological point of view

(Borghini et al., 2017). Several questions arise

concerning the relationships between the control

mode that operators are likely to adopt and their

performance but also concerning the control mode

adopted and the operators’ workload. This article

investigates these issues.

2 COGNITIVE CONTROL AND

MENTAL WORKLOAD

Cognitive control is one of the key concepts in

contemporary cognitive neuroscience. It refers to

Rauffet, P., Said, F., Laouar, A., Chauvin, C. and Bressolle, M.

Cognitive Control Modes and Mental Workload: An Experimental Approach.

DOI: 10.5220/0010011600170026

In Proceedings of the 4th International Conference on Computer-Human Interaction Research and Applications (CHIRA 2020), pages 17-26

ISBN: 978-989-758-480-0

processes that allow information processing and

behavior to vary adaptively from moment to moment,

depending on current goals, rather than remaining

rigid and inflexible. Research conducted in cognitive

neuroscience has identified two control modes

(Braver, 2012). The proactive control mode is

characterized by the maintenance of goal-relevant

information in working memory, which optimizes

attention, perception, and response preparation. It

relies on a sustained activity of the dorsolateral

prefrontal Cortex (dlPFC). In the reactive control

mode, attention is mobilized as part of a late

correction mechanism, and decision making is

guided by stimuli (Mäki-Marttunen, Hagen, &

Espeseth, 2019a). This mode of control is linked to a

more transient activation of the dlPFC (Ryman, El

Shaikh, Shaff, Hanlon, Dodd, Wertz, C. & Abrams,

2019). As summarized by Braver (2012, p. 106),

“proactive control relies upon the anticipation and

prevention of interference before it occurs, whereas

reactive control relies upon the detection and

resolution of interference after its onset”. In the field

of cognitive neuroscience, different experimental

studies have shown a link between cognitive

workload and cognitive control, as a heavy workload

leads to the adoption of a reactive control mode

(Mäki-Marttunen, Hagen, & Espeseth, 2019b).

In the field of cognitive ergonomics, two models

of cognitive control have been proposed to account

for the behavior of operators in dynamic situations:

the SRK taxonomy of Rasmussen (1983) and the

COCOM model proposed by Hollnagel (1993). As

Hoc and Amalberti (2007) point out, these two

models focus on two different aspects of cognitive

control. The SRK taxonomy considers the level of

abstraction of the data processed during supervision

activities (sub-symbolic vs. symbolic data), whereas

the COCOM model accounts for the more or less

reactive or proactive nature of the observed

behaviors.

The taxonomy defined by Rasmussen

distinguishes three different levels of control. The

skill-based level results in the implementation,

without conscious attention, of cognitive

automatisms and automated and strongly integrated

patterns of actions. At the rule-based level, behavior

is guided by known rules or procedures. The

knowledge-based level is brought into play to solve

new problems requiring the definition of new rules,

innovation, and creativity. Cegarra et al. (2017) have

shown that the skill-based level is associated with a

lower mental load than the rule-based level.

The COCOM model puts the emphasis on

temporality. It distinguishes among four main control

modes (Hollnagel, 1993, 2002): strategic, tactical,

opportunistic, and scrambled.

The strategic mode is used only when there is

considerable time available. It involves managing

several goals simultaneously and using predefined or

generated plans in order to address a situation. Hence,

it requires considerable attentional resources. The

tactical mode is based upon using known rules and is

used to process a limited number of goals. When the

available time is only just sufficient, operators are

likely to use an opportunistic mode that focuses on

managing one goal only. Hence, the resulting choice

of action is determined by the most salient

information. Finally, the scrambled control mode is

used when the time available is extremely limited. In

that case, planning is impossible and the choice of

action is random; consequently, the operators no

longer control the situation.

A number of studies have already used the notion

of control modes to account for operators’

performance in dynamic situations (Stanton,

Ashleigh, Roberts, & Xu, 2001; Eriksson & Stanton,

2017; Chauvin, Said, & Langlois, 2019), but, to our

knowledge, the relationship existing between the four

modes of the COCOM model and the mental

workload of operators has not been investigated yet.

The present study deals with this issue. It uses the

Multi-Attribute Task Battery (MATB-II, Santiago et

al., 2011) microworld to meet two goals:

distinguishing different control modes and

examining the relationship between control mode and

mental workload.

Following the ergonomics principles of standard

DIN ISO 10075-1:2017 (2018), mental workload is

viewed from both aspects of mental stress (i.e. the

constraints imposed upon operators) and mental

strain (i.e. the cognitive cost of the task for the

operators).

3 METHOD

The experiment used the MATB-II microworld,

which has already been used to examine the relations

between cognitive control modes and mental

workload (Cegarra, Baracat, Calmettes, Matton &

Capa, 2017). In this previous study, the notion of

cognitive control was viewed from the perspective of

Rasmussen's (1983) SRK taxonomy. Hoc and

Amalberti (2007) explain that the SRK taxonomy

deals with an aspect of cognitive control that involves

the level of abstraction of the data processed as part

of monitoring activities (i.e. sub-symbolic or

symbolic data). In the present study, another aspect

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications

of cognitive control is focused on, namely the source

of the data being processed: proactive control

involves internal data (i.e. mental models), whereas

reactive control involves external data.

The experiment entailed asking participants to

execute a main task for which optimum performance

would require adopting a strategic mode; the task

involved managing the content of fuel tanks. The task

was repeated three times, and its complexity (i.e.

mental stress) increased each time.

3.1 Participants

Twenty participants in the 18-21 age group (M =

18.55; SD = 0.83), all male, were recruited from

among the student population of Université Bretagne

Sud. All showed normal hearing and normal vision

(or corrected to normal vision). The participants were

informed of their rights and provided written consent

for their participation, in line with the Helsinki

Declaration.

3.2 Experimental Set-up

Participants were asked to perform tasks in the

MATB-II environment shown in Figure 1. MATB-II

is a microworld that enables people to execute four

tasks that are characteristic of flying an aircraft. In

this experiment, one of these tasks was used as a

"main task".

The communication task was excluded from the

protocol because it involves listening to an audio

message in English, which could bring about a bias

effect due to the heterogeneity of the linguistic level

of our participants.

The main task, called "Resource management",

simulated process monitoring. It involved the fuel

management of a civil aircraft, using a set of six fuel

tanks and the pumps that connect them. The

instructions were to keep both upper tanks at a stable

level of 2,500 units (symbolized by blue marks on the

tanks), keeping in mind that the level was

automatically reduced to simulate the fuel

consumption of the engines. To do so, both tanks

needed to be continuously supplied from the other

tanks through the pumps that, however, could break

down.

The secondary tasks were as follows: a tracking

task that involved keeping a target at the center of a

marker, and a system monitoring task that involved

spotting anomalies in the position of markers (see the

dial in the upper left corner of Figure 1). For this

monitoring task, six parameters needed to be

Figure 1: Screen capture of the MATB-II window.

monitored; they related to the colors of the boxes and

the position of the marks on the scale. The top two

boxes should normally be green for the left one, and

white for the right one. The bottom four marks should

be approximately in the middle of each scale.

Deviations could be observed: the top two boxes

could change color (red for the left one, green for the

right one), and the bottom marks could move and

touch the extremities of each scale. These deviations

represent abnormal situations that must be remedied

by pressing the key corresponding to the incriminated

element: F5 and F6 for the top boxes, F1 to F4 for the

bottom scales. Participants needed to press these keys

within a time budget of 30 seconds.

3.3 Experimental Protocol

A preliminary phase was used to explain the tasks to

be executed during a 15-minute training session

followed by a test session aimed at ensuring that

participants had fully understood the instructions.

The experimental phase was broken down into

three 7-minute sequences (see Figure 2).

The first sequence involved the main task only

(referred to as “resman”), the second one involved

the main task and the secondary system monitoring

task (named “with track”), and the third one required

participants to execute both the main task and the

secondary tracking task (named “with sysmon”).

Cognitive Control Modes and Mental Workload: An Experimental Approach

Figure 2: Three experimental sequences, each with two

levels of complexity of the main task.

It should be noted that the main task (“resman”)

was a continuous task, since the participants had to

manage the levels of the two reservoirs that changed

every second, and whose dynamics (filling or

emptying) could change when failures occurred.

Furthermore, as explained by Philips et al. (2007) and

Gutzwiler and Wickens (2015), we can also

distinguish the two different secondary tasks of our

scenario: tracking is a continuous task, which

requires permanent control of the trajectory, whereas

the monitoring system is a discrete task, consisting of

acknowledging alarms when they appear on the

screen. Thus, the succession of sequences in our

scenario resulted in an increase in difficulty: first

there was a continuous task alone, then a continuous

task with a discrete task (generating "discrete"

stimuli occasionally disturbing the participants in the

main task), and finally two continuous tasks (which

required the control of two processes whose dynamic

evolution must be managed).

Furthermore, Figure 2 shows that each sequence

itself was broken down into two periods: one 3-

minute period during which executing the main task

was less complex and a second 4-minute period

during which the long breakdown of one pump made

the task more complex.

At the end of each 7-minute sequence, a NASA-

TLX questionnaire was given to the participants

through the MATB interface.

3.4 Measures and Coding

The performance of the main task was coded to

identify the modes of cognitive control likely to be

adopted by the operator. To do this, we examined

whether the participants complied with the

instructions for the task (i.e. keeping the level of each

of the two reservoirs between 2000L and 3000L), and

how they managed their safety margin in relation to

the low threshold of 2000L (the low threshold was

more difficult to comply with than the high threshold,

since pump failures accelerated the emptying of the

reservoirs). Table 1 shows the characteristics of the

operations used to operationalize each control mode

likely to be used for the main task of fuel tank

management. In accordance with Hollnagel’s model,

the strategic and the tactical modes are associated

with a satisfying performance, whereas the

opportunistic mode is associated with some errors

and the scrambled mode with a poor performance.

Table 1: Characteristics of the control modes.

Control mode Performance

Strategic Complying with instructions and high

margins for at least one of the two tanks

(maximum upper value between 2,750 and

3,000)

Tactical Complying with instructions and lower

margins for both tanks (values oscillate

between 2,000 and 2,750 around the target

value of 2,500).

Opportunistic Errors for at least one tank; the participant

takes action when the minimal value

(between 2,000 and 1,950) is exceeded

Scrambled Serious errors for at least one tank; the

minimum value (inferior to 1,950) or the

maximum value (superior to 3,050) is

exceeded by a large margin when the

participant intervenes.

Two types of physiological data were collected

and analyzed as indicative of mental strain: cardiac

activity with Bioharness 3 belt (Zephyr, Medtronic,

Ireland), and oxygenation and deoxygenation of the

prefrontal cortex with the 8-channel functional near-

infrared spectroscopy (fNIRS) system (Octamon,

Artinis Medical, Netherlands). These sensors were

Sequence 1 (S1) - 7 minutes

Main task only (Resource Management)

Sequence 2 (S2) - 7 minutes

Main task (Resource Management) and secondary

task (System monitoring)

Sequence 3 ( S3) - 7 mi nutes

Main task (Resource Management) and secondary

task (Tracking task)

S1.1 (3 min)

Low complexity of

the main task

S1.2 (4 min)

High complexity of the

main task

S2.1 (3 min)

Low complexity of

the main task

S2.2 (4 min)

High complexity of the

main task

S3.1 (3 min)

Low complexity of

the main task

S3.2 (4 min)

High complexity of the

main task

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications

especially chosen for their known and robust

relationships with mental strain (see Table 2), as well

as their ease of implementation in real world

application (without too many interferences with

ambient factors, such as light variations. The

processing of fNIRS data was performed using a

bandpass filter (0.01hz-0.09hz). To select the cutoff

frequencies, we followed the approach of Pinti et al.

(2019), which advocates a low frequency of 0.01hz

and a high frequency lower than the Mayer wave

frequency (0.01hz).

Table 2: Relationships between neurophysiological

indicators and mental strain.

Indicators

Cardiac

activity

Heart rate variability (HRV), computed within

time-domain parameters with the standard

deviations over 100 successive RR intervals

Relationship with mental strain

Decreases with an increased mental workload

(Malik, 1996, Durantin et al., 2014)

Prefrontal

Cortex

activity

Concentrations of oxygenated hemoglobin

(HBO2) and deoxygenated hemoglobin (HBB)

on the 8 optodes of the Fnirs. T1 to T4 capture

relative changes in cerebral activation of the

right hemisphere of the prefrontal cortex (PFC),

and T5 to T8 capture changes in the left

hemisphere. Specifically, T1-T7 capture changes

in the dorsolateral PFC, T2-T8 capture changes

in the ventrolateral PFC, T3-T5 capture changes

in the ventromedial PFC, and T4-T6 capture

changes in the orbitofrontal PFC.

Relationship with mental strain

Neuronal activity is associated with an increase

in concentration of oxygenated hemoglobin and

a decrease in deoxygenated hemoglobin

(Fairclough et al., 2018, Causse et al., 2019)

3.5 Data Analysis Method

The analyses conducted are part of an exploratory

study. We used a two-step methodology to analyze

the data. Bhapkar and McNemar analyses were

conducted to investigate possible links between

mental stress (viewed as an independent variable)

and control mode (viewed as a dependent variable).

Additionally, we used R (R Core Team, 2012), and

especially the lme4 package (Bates, Maechler &

Bolker, 2012), to perform linear mixed effects

analyses of the relationship between sequence and

complexity (viewed as independent variables) and

the neurophysiological indicators presented in Table

2 (viewed as dependent variables). Visual inspection

of residual plots did not reveal any obvious

deviations from homoscedasticity or normality. As

fixed effects, we entered sequence (single or double

task), complexity (low or high, corresponding to

whether the main task had few or many incidents on

the pumps) and cognitive control mode (with

interaction terms) into the full model. As random

effect, we had intercept for participants. Regarding

fixed effects, a stepwise model selection by AIC

(stepAIC) was conducted. During each step, a new

model was fitted, in which one of the model terms

was eliminated and tested against the former model.

4 RESULTS

4.1 Effect of Mental Stress upon the

Control Modes

First, we conducted a multinomial logistic regression

between control modes and the two factors related to

mental stress (sequence and complexity). No effects

of interaction were found between these two factors.

Next, we examined the effect of the complexity of the

main task by comparing the control modes adopted

when complexity is low (first period) and when it is

more important (second period) as shown in Table 3.

Table 3: Contingency table crossing complexity levels and

cognitive modes of control.

Scrambled

mode

Opport.

mode

Tactical

mode

Strategic

mode

Low

Complexity 4 3 26 23

High

Complexity 15 10 3 28

We observed that tactical and strategic modes are

largely adopted when the task complexity is low. The

strategic mode is still used when the complexity

increases but the tactical mode disappears.

A Bhapkar test revealed that the level of

complexity has a significant effect on the control

mode regardless of the secondary task: resman (χ2(3,

19) = 30.38, p < 0.001), resman with sysmon (χ2(3,

19) = 11.08, p = 0.01), and resman with track (χ2(3,

18) = 20.79, p < 0.001). McNemar post-hoc tests with

Bonferroni adjustment revealed that, for the main

task alone (resman), the scrambled mode is

significantly more frequent when the level of

complexity is high (p = .03). Besides, the tactical

mode is significantly more frequent in tasks with low

complexity level than in tasks with high complexity

Cognitive Control Modes and Mental Workload: An Experimental Approach

level: resman (p <.001), resman with sysmon

(p=.043) and resman with track (p = .019).

The probability of moving from an X mode when

the complexity is low to a Y mode when the

complexity is higher was calculated from a transition

matrix (see Table 4).

Table 4: Matrix of transition between the mode adopted

when complexity is low and the mode adopted when

complexity is high.

Low /

High 

Sc O T St Total

Sc 4 4

O 3 3

T 5 9 3 9 26

St 3 1 19 23

(Sc = Scrambled, O = Opportunistic, T = Tactical, St =

Strategic)

Examining the transitions between the two

periods (hence between the two complexity levels)

shows (see Figure 3) the stability of the strategic

mode (among the 23 participants who adopted the

strategic mode in the first period, 19 maintained it in

the second one) and the instability of the tactical

mode (among the 26 participants who adopted the

tactical mode in the first period, only 3 maintained it

in the second one).

Figure 3: Transitions between modes between the periods

of low complexity of the main task (S1.1, S2.1, S3.1) and

the periods of greater complexity (S1.2, S2.2, S3.2).

In contrast, the comparisons conducted for each

complexity level between sequence 1 (main task

alone) and sequences 2 and 3 (main task and

secondary tasks of system monitoring and tracking)

do not show any negative effect of the secondary task

upon the control modes. As a matter of fact, the

majority of participants kept the control mode they

had adopted for sequence 1 (main task alone), or else

they adopted a more effective control mode, which

shows the effect of learning.

4.2 Relations between Mental Stress

and Mental Strain

We conducted different linear mixed-effect analyses

to test the effects of task, complexity and cognitive

control modes (CCM) on physiological responses.

The significant effects (figures in bold in Table 5) are

given relative to the reference condition. For

example, we can observe the effect of the sequence

on HRV, which decreases by 13.72 ms between the

single task condition “resman” only” and dual task

condition “resman with sysmon”, and by 12.46 ms

between the single task condition and “resman with

tracking” (see Table 5 and the left part of Figure 4).

These analyses show that HRV can be explained

by mental stress, i.e. by sequence, complexity and

their interactions (see Table 5 and Figure 4).

Figure 4: Interactions of sequence and complexity on HRV.

We found a significant main effect of complexity,

with HRV more likely to decrease in the high

complexity than in the low complexity condition (β=-

24.03, SE=3.60, t(63)=-6.67, p<0.001). Moreover,

there is also a significant effect of sequence. We

found lower HRV when the main task was carried out

with the secondary tracking task (β=-12.46, SE=3.69,

t(63)=-3.38, p<0.01) or with system monitoring task

(β=-13.72, SE=3.60, t(63)=-3.81, p<0.001), than

when it was conducted as a single task.

This effect of sequence upon operator strain is

also observed, on the mental demand dimension of

the NASA-TLX. A one-way between subjects

ANOVA shows that the single task condition

involved a significantly lower mental demand than

the double task conditions (F(2,48)=4.32, p<0.05).

The interaction between complexity and sequence is

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications

Table 5: Estimates of fixed effects from linear mixed-effect model for HB02 on the 8 fNIRS optodes and for HRV.

HBO2 T1

~ CCM

+ Sequence

HBO2 T2

~ CCM

HBO2 T3

~ CCM

HBO2 T4

~ CCM

+ Sequence

HBO2 T5

~ CCM

HBO2 T6

~ CCM

HBO2 T7

~ CCM

HRV

~ Complexity

* Sequence

(Intercept) 0.13 (1.66) 3.08 (0.95)

0.26 (2.56) 1.39 (1.16) 0.36 (2.53) 2.66 (0.95)

1.35 (0.94) 7.25 (4.66)

CCM (reference = Tactical)

Opportunistic 0.94 (0.21)

***

1.11 (0.32)

***

0.69 (0.30)

0.99 (0.34)

0.48 (0.29) 0.97 (0.32)

0.59 (0.28)

Scrambled 1.08 (0.24)

***

0.71 (0.36)

1.26 (0.33)

***

0.84 (0.39)

0.62 (0.33)

0.85 (0.35)

0.71 (0.32)

Strategic 0.63 (0.18)

***

0.78 (0.27)

0.76 (0.25)

0.53 (0.29)

0.71 (0.25)

0.74 (0.27)

0.79 (0.24)

Sequence (reference = single task resman)

with sysmon 0.23 (0.15) 0.30 (0.24) -13.72 (3.60)

***

with track 0.48 (0.15)

0.69 (0.24)

-12.46 (3.69)

***

Complexity (reference = low complexity)

high complexity -24.03 (3.60)

***

Sequence:Complexity

with sysmon :

high complexity

13.07 (5.09)

with track:

high complexity

13.24 (5.19)

Num. obs. 100 100 100 100 100 100 100 82

Num. groups:

Participant_

17 17 17 17 17 17 17 14

Var:

Participant_

(Intercept)

46.61 14.82 110.76 21.72 108.17 14.62 14.52 212.55

Var: Residual 0.36 0.82 0.70 0.93 0.69 0.80 0.64 90.84

Note. The fixed factors resulting from the stepwise model selection by AIC are indicated below the response variables; the

reference condition is indicated in brackets for each explanatory factor;

***

p < 0.001,

p < 0.01,

p < 0.05.

also found to be significant, with a higher contrast

between single task and double task conditions when

the complexity is low. Moreover, there is no

significant correlation between neurophysiological

indicators and NASA-TLX scores.

4.3 Relations between Control Modes

and Mental Strain

The linear mixed-effect analyses also showed a

significant effect of the control modes (CCM) upon

the concentration in oxy-hemoglobin (HBO2).

According to the stepwise model selection by AIC,

HBO2 can be explained by control modes only, for

optodes T2, T3, T5, T6 and T7, whereas HBO2 for

optodes T1 and T4 can be explained by two main

fixed effects, CCM and sequence. We followed the

same procedure for concentration in deoxy-

hemoglobin (HBB), but no significant results were

found.

It should be noted that, for all the optodes from

T1 to T7, the tactical control mode (set as reference

condition in the linear mixed model) always produces

a significant lower HBO2 concentration, in

comparison with the less effective modes (scrambled

and opportunistic control) or the more anticipatory

one (strategic control).

5 DISCUSSION

The study findings are both theoretical and

methodological in nature.

First, regarding the stress-strain relationship, we

observed a significant effect of mental stress on

HRV, which is unsurprising. There is a main effect

of the complexity of the reservoir management task

on cardiac activity. The other constraint factor of

sequence (i.e. the addition of a secondary task) also

Cognitive Control Modes and Mental Workload: An Experimental Approach

plays a significant but lesser role when the

complexity of the main task is low. We should also

note that the neurological indicators (fNIRS) are not

or not very sensitive to the constraint.

In addition, we investigated, in a more original

way, the relation between the cognitive control

modes and mental workload, from the perspective of

both mental stress and mental strain. Our analyses

reveal two main theoretical contributions.

On the one hand, there is a significant effect of

task complexity on the adoption and the variation of

control modes. In particular, we found an instability

of the tactical mode, showing attraction between this

mode and low complexity, and repulsion between

this mode and higher complexity. This instability of

the tactical mode was also analyzed with the finer-

grained analysis of transitions between the

consecutive periods of low and higher complexity.

We observed that an increase in complexity mainly

leads to transitions from the tactical mode to a less

effective mode (54% of the transitions). In contrast,

the strategic and scrambled modes were mostly stable

(respectively 83% and 100% of participants in one of

this mode remained in the same mode, between low

and high complexity periods within a given

sequence). Furthermore, and congruent with the

study of Stanton et al. (2001), we observed that a

major part of the transition is between two

“close” modes (70% of transitions from tactical to

opportunistic or strategic modes, and 100% of

transitions from opportunistic to scrambled modes).

This result suggests that people move between

control modes in a linear manner.

On the other hand, we found links between the

modes of control and operator strain, as it was shown

by Cegarra et al. (2017). The present study indicates

that the tactical mode is associated with lower mental

strain, when considering the HBO2

concentrationndicator of mental workload. As stated

by Leon-Carrion et al. (2008), “the hemodynamics of

inter-individual differences in this region may reflect

different cognitive strategies used in task resolution”.

Our study shows that the tactical mode is the most

efficient one, since it is associated with a satisfying

performance and with the lowest mental strain off all

control modes.

This result calls attention to the advantage of

studying brain activity to detect changes in control

mode. If, in our study, the cerebral activity seems

little correlated with mental stress variations, we

nevertheless observe, on almost all the zones of the

prefrontal cortex, a significant difference in HBO2

concentration between the tactical mode and other

modes. Hence, an increase in cortical activation

could help reveal the shift away from the tactical

mode towards less effective and more reactive

control (the opportunistic or scrambled mode, where

control of the situation is no longer guaranteed) or on

the contrary towards more proactive control (the

strategic mode, requiring more anticipation). This

potential detection ability opens new perspectives to

design and trigger assistance aimed at keeping

operators in the tactical mode, since it appears to be

the most efficient one. Such perspectives are worth

considering in all areas where operators have to

control dynamic situations and where they have

therefore to make, in real time, compromises between

speed and efficiency, between performance and risk,

or between understanding and action (Hoc, 2000).

Such circumstances cover the field of transport but

also the supervision of industrial processes, the

medical field (anesthesia in particular), or the field of

crisis management.

Finally, it should be noted that this research work

has some limitations. The experiment was run with

novice participants only, who may be more

heterogeneous in terms of cognitive control than an

expert population. One may thus wonder whether

some individual factors might not explain the

propensity of some participants to adopt a particular

control mode. Therefore, it would be necessary to

verify whether the same findings would apply to

experts (e.g. a population of aircraft pilots).

In addition, our study, which involved only male

participants, may hide gender effects on the adoption

of control methods. Moreover, we coded the four

control modes of the COCOM according to

operators’ performance on the main task and not the

overall performance in the case of double task

situations. When participants had to carry out

multiple tasks, there could have been phenomena of

focusing on or prioritizing the main task. This focus

may have led to the maintenance of an effective

cognitive control on the management of the

reservoirs, to the detriment of the control of the

secondary tasks. Hence, in future research studies, it

would be worth considering cognitive control by

adopting an approach modeling operators’ multi-task

management on MATB-II, as proposed by

Gutzwiller et al. (2014).

Finally, we did not control the chronobiological

aspects in this study (i.e. food or caffeine intake

before the test, or sleep duration the night before the

experiment). We also did not measure the level of

fatigue during the different phases of the experiment.

It would be worth considering these elements in the

future to analyze the relationship between fatigue and

control modes and to explain the performance

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications

variations observed over time during the

experimental scenario.

6 CONCLUSION

As part of an experiment on the MATB-II platform,

this study investigates the relationship between

cognitive control modes and mental workload. As

shown in Figure 5, the control modes of the COCOM

could enable regulating operators’ mental workload,

which would moderate the stress-strain relationship

(Hockey, 1997; Kostenko et al., 2016; Cegarra,

2017). Higher mental stress may induce operators to

leave the tactical mode, which can be detected

through the fNIRS system. Leaving the tactical mode

towards more degraded and reactive control (i.e. the

scrambled or opportunistic modes), or on the contrary

more elaborate and more proactive control (i.e. the

strategic mode), could also explain the increase in

operators’ strain that is observed at the physiological

level with the HRV indicator. Finally, the potential

ability of fNIRS system to detect the tactical mode

with HBO2 concentration could, in the future, help

trigger adaptive assistance in order to keep operators

in this efficient mode.

Figure 5: Towards a moderating effect of control mode on

the stress-strain relationship.

REFERENCES

Bates, D.M., Maechler, M., Bolker, B., 2012. lme4 : Linear

mixed-effects using S4 classes. R package version

0.999999-0

Borghini, G., Aricò, P., Di Flumeri, G., Cartocci, G.,

Colosimo, A., Bonelli, S., ... & Pozzi, S., 2017. EEG-

based cognitive control behaviour assessment: an

ecological study with professional air traffic

controllers. Scientific reports, 7(1), 1-16.

Braver, T. S., 2012. The variable nature of cognitive control:

a dual mechanisms framework. Trends in cognitive

sciences, 16(2), 106-113.

Causse, M., Chua, Z. K., & Rémy, F., 2019. Influences of

age, mental workload, and flight experience on

cognitive performance and prefrontal activity in private

pilots: a fNIRS study. Scientific reports, 9(1), 1-12.

Causse, M., Chua, Z., Peysakhovich, V., Del Campo, N., &

Matton, N., 2017. Mental workload and neural

efficiency quantified in the prefrontal cortex using

fNIRS. Scientific reports, 7(1), 1-15.

Cegarra, J., Baracat, B., Calmettes, C., Matton, N., Capa

R.L., 2017. A neuroergonomics perspective on mental

workload predictions in Jens Rasmussen’s SRK

framework, Le travail humain, 80, 7-22.

Chauvin, C., Said, F., Langlois, S., 2019. Does the Type of

Visualization Influence the Mode of Cognitive Control

in a Dynamic System?, in International Conference on

Intelligent Human Systems Integration, (pp. 751-757).

Springer, Cham.

DIN EN ISO 10075-1, 2018. Ergonomic principles related

to mental workload—Part 1: general issues and

concepts, terms and definitions (ISO 10075-1:2017).

Durantin, G., Gagnon, J. F., Tremblay, S., & Dehais, F. ,

2014. Using near infrared spectroscopy and heart rate

variability to detect mental overload. Behavioural brain

research, 259, 16-23.

Eriksson, A., Stanton, N.A., 2017. Driving performance

after self-regulated control transitions in highly

automated vehicles, Human Factors, 59, pp. 1233-

1248.

Fairclough, S. H., Burns, C., & Kreplin, U., 2018. FNIRS

activity in the prefrontal cortex and motivational

intensity: impact of working memory load, financial

reward, and correlation-based signal improvement.

Neurophotonics, 5(3), 035001.

Gutzwiller, R. S., Wickens, C. D., & Clegg, B. A., 2014.

Workload overload modeling: An experiment with

MATB II to inform a computational model of task

management. In Proceedings of the Human Factors and

Ergonomics Society Annual Meeting, Vol. 58, No. 1,

(pp. 849-853). Sage CA: Los Angeles.

Hoc, J.M., 1996. Supervision et contrôle de processus : la

cognition en situation dynamique. Presses

universitaires de Grenoble. Grenoble.

Hoc, J. M., 2000. La relation homme-machine en situation

dynamique. Revue d’intelligence artificielle, 14(1), 55-

Hoc J.M., Amalberti, R., 2007. Cognitive control dynamics

for reaching a satisfying performance in complex

dynamic situations, Journal of Cognitive Engineering

and Decision Making, 1, 22–55.

Hockey, G. R. J., 1997. Compensatory control in the

regulation of human performance under stress and high

workload: A cognitive-energetical framework.

Biological psychology, 45(1-3), 73-93.

Hollnagel, E., 1993. Human Reliability Analysis – Context

and Control. Academic Press. London.

Hollnagel, E., 2002. “Time and time again”. Theoretical

Issues in Ergonomics Science, 3, 143-158.

Cognitive Control Modes and Mental Workload: An Experimental Approach

Kostenko, A., Rauffet, P., Chauvin, C., & Coppin, G., 2016.

A dynamic closed-looped and multidimensional model

for Mental Workload evaluation. IFAC-PapersOnLine,

49(19), 549-554.

Kostenko, A., Rauffet, P., Moga, S., & Coppin, G., 2019,

November. Operator Functional State: Measure It with

Attention Intensity and Selectivity, Explain It with

Cognitive Control. In International Symposium on

Human Mental Workload: Models and Applications (pp.

156-169). Springer, Cham.

León-Carrion, J., Damas-López, J., Martín-Rodríguez, J. F.,

Domínguez-Roldán, J. M., Murillo-Cabezas, F., y

Martin, J. M. B., & Domínguez-Morales, M. R., 2008.

The hemodynamics of cognitive control: the level of

concentration of oxygenated hemoglobin in the superior

prefrontal cortex varies as a function of performance in

a modified Stroop task. Behavioural brain research,

193(2), 248-256.

Mäki-Marttunen, V., Hagen, T., & Espeseth, T., 2019a.

Proactive and reactive modes of cognitive control can

operate independently and simultaneously. Acta

psychologica, 199, 102891.

Mäki-Marttunen, V., Hagen, T., & Espeseth, T., 2019. Task

context load induces reactive cognitive control: An

fMRI study on cortical and brain stem

activity. Cognitive, Affective, & Behavioral

Neuroscience, 19(4), 945-965.

Malik, M., 1996. Heart rate variability: Standards of

measurement, physiological interpretation, and clinical

use: Task force of the European Society of Cardiology

and the North American Society for Pacing and

Electrophysiology. Annals of Noninvasive

Electrocardiology, 1(2), 151-181.

Pinti, P., Scholkmann, F., Hamilton, A., Burgess, P., &

Tachtsidis, I. (2019). Current status and issues

regarding pre-processing of fNIRS neuroimaging data:

an investigation of diverse signal filtering methods

within a general linear model framework. Frontiers in

human neuroscience, 12, 505.

Phillips, C. A., Repperger, D. W., Kinsler, R., Bharwani,

G., & Kender, D. (2007). A quantitative model of the

human–machine interaction and multi-task

performance: A strategy function and the unity model

paradigm. Computers in biology and medicine, 37(9),

1259-1271.

R Core Team, 2012. R : A language and environment for

statistical computing. Foundation for statistical

Computing. Vienna, Austria.

Rasmussen, J., 1983. Skills, rules, and knowledge - signals,

signs, and symbols, and other distinctions in human

performance models. In IEEE Transactions on Systems,

Man, and Cybernetics, vol. 13, pp. 257-266.

Ryman, S. G., El Shaikh, A. A., Shaff, N. A., Hanlon, F.

M., Dodd, A. B., Wertz, C. J., ... & Abrams, S., 2019.

Proactive and reactive cognitive control rely on flexible

use of the ventrolateral prefrontal cortex. Human brain

mapping, 40(3), 955-966.

Santiago-Espada, Y., Myer, R. R., Latorella, K. A., &

Comstock Jr, J. R., 2011. The multi-attribute task battery

ii (matb-ii) software for human performance and

workload research: A user's guide

Stanton, N.A, Ashleigh, M.J., Roberts, A.D., & Xu F.,

2001. Testing Hollnagel’s contextual control model:

Assessing team behaviour in a human supervisory

control task, Journal of Cognitive Ergonomics, 5, 21-

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications