A model-driven approach to the a priori estimation of operator...
Transcript of A model-driven approach to the a priori estimation of operator...
![Page 1: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/1.jpg)
A model-driven approach to the a priori estimation of operatorworkload using Mental Representation Theory
Olivier Grisvard1,2 & Dhouha Kbaier Ben Ismail3
1Thales Airborne Systems, Radar & Electronic Warfare Competence Centre2Institut Mines-Télécom/Télécom Bretagne, Logics of Use, Social & Information
Sciences Department3University of Bedfordshire, Department of Computer Science & Technology
1. Introduction
The concept of cognitive workload has several possible meanings. Despite the
interest in the topic for the past half-century, it seems that no common definition may
be found. Hart and Staveland [1] argue that the “workload is not an inherent
property, but rather it emerges from the interaction between the requirements of a
task, the circumstances under which it is performed, and the skills, behaviours, and
perceptions of the operator”. The definition proposed by Wickens [2] gives a good
idea of the concept since he considered the workload as a relationship between the
cognitive demand and the cognitive resources of an operator executing a task.
According to Cain [3], workload aspects can be separated in three categories: the
amount of work, the time and the subjective psychological experience of the human
operator. A simplistic, but realistic, way to look at workload measurement is that “if
the person feels loaded and effortful, he is loaded and effortful, whatever the
behavioural and performance measure show” (Johannsen et al., 1979) [4]. Therefore,
the most frequently used technique for estimating the cognitive workload involves
asking the person in charge to complete the task directly.
![Page 2: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/2.jpg)
2 Olivier Grisvard and Dhouha Kbaier Ben Ismail
Workload measurement has been applied to a number of military and industrial
problems. There are three major types of workload measures: performance-based,
physiological and subjective. Performance-based measures can be subdivided into
primary-task and secondary-task measures. Primary-task measures provide a direct
indication of performance on the task under consideration. However, performance on
the primary task may be insensitive to workload changes if the operators compensate
by an increased effort. The secondary task is an additional measure to the primary
task. “The basic idea of a secondary task is that it measures the difference between
the ‘mental capacity’ consumed by the main task, and the total available capacity”
(Mulder, 1979) [5]. The major problem that may occur when secondary tasks are used
to measure workload is that they may disrupt primary task performance (Colle &
Reid, 1999; Sirevaag et al., 1993) [6, 7]. For example, a verbal secondary task may not
interfere with a spatial primary task, even if the primary task is very demanding.
Physiological measures are based on the assumption that workload will induce
physical changes. These changes are measured in cardiac activity, brain activity,
breath activity, speech measures, and eye activity. An overloaded operator may
experience changes such as an increase in heartbeat rate and skin conductance. Often,
a large volume of data is collected, requiring unfortunately sophisticated analysis.
Finally, subjective measures are used to reflect the amount of information used in
working memory (Yeh & Wickens, 1988) [8].
Although physiological measures of workload may be more precise, subjective
measures are more practical. Furthermore, subjective tests are flexible for different
people with different capabilities. “Because subjective ratings take into account
individual differences in ability, state, and attitude – differences that may be
obscured in objective measures of performance until breakdown makes them obvious
– they are valuable because of, not despite, their subjectivity” (Muckler & Seven,
1992) [9]. Even though subjective and objective measures of workload are very
![Page 3: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/3.jpg)
Papers dedicated to Anne Reboul 3
different, it has been shown that subjective measures correlate with physiological
measures of workload such as heartbeat rate variability (Tattersall & Foord, 1996)
[10]. Adding to that, an increasing number of studies have found operator ratings to
be a more direct indicator of workload than physical measures. Subjective measures
are considered to be the least intrusive, most flexible, most convenient, least time
consuming, and least expensive form of evaluating workload (Yeh & Wickens, 1988)
[8]. As such, subjective measurement is often based on the use of scales to measure
the amount of workload a person is feeling.
A variety of subjective measures have been developed and applied in many
studies, particularly those of the flight deck. Several rating scales have been
specifically developed for the purpose of measuring subjective workload. The NASA
Task Load Index (Hart & Staveland, 1988) [11] is one of the most widely used scales.
The NASA-TLX has been implemented in many aviation studies since it is considered
to be a good multidimensional scale. It uses six dimensions to assess workload:
mental demand, physical demand, temporal demand, performance, effort and
frustration. The six subscales can be divided into three groups, according to three
factors assumed to generate workload. First, the mental, physical and temporal
demand sub-scales are properties assumed to be associated with the task. Then, the
performance and effort sub-scales are assumed to be characteristic of behaviour and
skill. Finally, the frustration is assumed to be characteristic of the individual.
In this paper, we present a new method to estimate workload that calls back these
three important factors: characteristics of the task, behavioural characteristics and
individual characteristics. In fact, we argue that workload can be inferred from
analysis of the tasks required of human operators. However, individual differences
must be taken into account. For example, a novice and expert will obviously
experience different levels of workload when performing the same task. For this
reason, we explore the following important parameters: task complexity, time load,
![Page 4: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/4.jpg)
4 Olivier Grisvard and Dhouha Kbaier Ben Ismail
experience, knowledge and abilities compared to task requirements. Since
‘frustration’ and ‘physical demand’ have only shown a small relevance for workload
(Pfendler and Widdel, 1988; Sepehr, 1988; Veltman and Gaillard, 1993) [12-15], these
subscales of the NASA-TLX have not been included in our analysis.
The remainder of this paper is organized as follows. In section 2, we go through
the operational analysis of a maritime surveillance (SurMar) operation and focus
particularly on the role of the TACtical COordinator (TACCO). We also introduce the
Medusa project and how the human behaviour is integrated in nowadays SurMar
systems. Section 3 is an insight into Mental Representation Theory. Then, in section 4,
we describe our approach that proposes to use Mental Representations of tasks,
human actors, human roles, knowledge and abilities. The new workload estimation
method is presented in section . The approach is illustrated with examples, through a
scenario, a workload chart and the results of experiments. Finally, Section 6 draws
some conclusions.
2. Operational analysis
In system engineering, the Human View (HV) is required to explicitly represent
the human and to document the unique implications humans bring to the system
design. It provides a way to integrate the human system into the mainstream
acquisition and system engineering process by promoting early and frequent
consideration of human roles. The purpose of a HV is to capture the human
requirements and to inform on how humans interact with systems. The NATO
Architecture Framework (AF), which builds on the United States Department of
Defence AF (DoDAF) and the United Kingdom Ministry of Defence AF (MoDAF), is
the most extensive and complete HV [16, 17]. The Operational Analysis (OA) is the
entry point to the analysis of human activities and constraints, called human
integration process.
![Page 5: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/5.jpg)
Papers dedicated to Anne Reboul 5
2.1. Human activity in an operational analysis
The OA consists in identifying operational requirements from the operational
concepts. Scenarios with representative data are strongly recommended to support
the identification of the dynamic aspects of the human interaction with the system.
They are mandatory to be able to apply most of the metrics. The human activities are
identified in the Operational View (OV) and the HV is produced as follows. First of
all, activities are refined to produce a task model. The tasks are enriched with data
and Knowledge, Skills and Abilities (KSA) requirements. Then, the human roles are
identified in the OV and the KSA requirements for the roles can be inferred from the
tasks. In a multi-user case, the role structure is also determined. After allocating the
human roles to human entities, the required KSA for these entities are deduced. Once
the HV is available, the application of metrics allows verifying the capacity of the
entities to perform the tasks they have been allocated, updating the HV and iterating
the process. When a User Interface (UI) prototype is ready, an implementation of the
simulation of scenarios through the UI prototype is necessary in order to perform
experiments with users. Depending on the results of the experiments, a better HV
can be produced and some modifications can be applied to the UI prototype before
repeating experiments. For further analysis in the context of situation awareness, see
Blash et al. [18-19]. We now describe the operational analysis that has been carried
out in the context of the Medusa project.
![Page 6: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/6.jpg)
6 Olivier Grisvard and Dhouha Kbaier Ben Ismail
2.2. Operation activity excerpt from the Medusa case study
At a time when we are witnessing an explosion in publicly available information
technology, the development of human-centred rather than machine-centred
applications is becoming a priority. The workload of the crews of operational
surveillance aircrafts for example is considerably increased by the range of on-board
sensors: radar, electro-optical and Forward Looking InfraRed cameras (FLIR), radar
detectors, ultraviolet scanners to detect deliberate pollution, AIS for ship
identification and new means of communication for getting rapidly in touch with
decision-makers and relevant public authorities. The aim of the Medusa project, an
applied research project conducted between 2010 and 2014 and funded by the French
FUI1, was to introduce behavioural aspects of user-system interaction upstream in the
design of new systems that deal with maritime emergencies. An iterative
methodology has been defined in Medusa to reconcile the need to make a system
user-friendly, easy-to-learn and efficient with its complexity and the multiplicity of
interactions involved. Intended for use by the French government maritime initiative
(Action de l’État en Mer), and for managing maritime shipping, Medusa enhances
the operators’ responsiveness in stressful situations and facilitates decision-making.
We go through the operational analysis of a SurMar operation and focus
particularly on the role of the TACCO. We suppose that a unique operator holds this
role and thus manages the tactical situation. As shown in Figure 1, the role of TACCO
consists mainly of four tasks: produce FLIR video, produce Radar video, manage track list
and classify tracks. The operator can switch over manage track list or track classification
and iterate them as many times as necessary.
1Fonds Unique Interministériel (French inter-ministry industrial research funding program).
![Page 7: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/7.jpg)
Papers dedicated to Anne Reboul 7
Figure 1: Timeline chart for the role of TACCO.
Produce FLIR video and produce Radar video are sub-automatic tasks. The only
contribution of the operator consists in switching on/off the FLIR camera or the
Radar respectively. The two other tasks are part of the tactical situation management
activity, as shown in Figure 2. We hereafter explain in details these two tasks. To
manage the track list, the operator builds a Track Wide Scan (TWS) zone (zone of
automatic radar tracking) and updates the list. He has also to configure the Radar:
configuration of the emitter, the receiver, the antenna’s behaviour, the wavelength,
the scanning strategies, etc. The operator can create and delete tracks according to the
positions and classes of all vessels of interest within the defined area of surveillance.
He can perform manual tracking or request automatic tracking on Radar echoes,
either for living tracks or tracks in dead reckoning status (not updated by the Radar
anymore), through an estimation of the direction and distance travelled by a ship on
the basis of the ship's and the previous path and speed.
![Page 8: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/8.jpg)
8 Olivier Grisvard and Dhouha Kbaier Ben Ismail
Figure 2: Track management in the tactical situation management diagram.
Once he has created a set of tracks within the defined area of surveillance, the
operator selects some tracks for classification. Both Radar and FLIR classifications can
be performed as many times as necessary for each selected track. In order to compute
the ship length, the human actor requests a Dynamic Range Profile (DRP imaging
Radar mode) image or an Inverse Synthetic Aperture Radar (ISAR imaging Radar
mode) image depending on the position of the targeted ship relatively to the aircraft,
and proposes a Radar classification. If close enough, in order to get a live image, the
operator can request the FLIR image of the target. Figure 3 is a flow diagram of FLIR
classification: the FLIR camera is first adjusted and pointed on the track. Then, when
the FLIR image is available, the operator decorates the ship image and proposes a
FLIR classification.
![Page 9: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/9.jpg)
Papers dedicated to Anne Reboul 9
Figure 3: FLIR classification flow diagram.
3. Mental Representation Theory
Although the classic models of knowledge representation are not recent, they
remain widely used. First, Belief Desire Intention (BDI) architectures [20-21] describe
the internal state of an agent through its mental attitudes of beliefs, goals and
intentions. BDI theories provide a conceptual model of the knowledge, the goals and
the commitments of an agent.
In the late nineties, Ross Quillian introduced semantic networks as a method of
modelling the structure and storage of human knowledge in the shape of a graph
[22]. Quillian wanted his system to explore the meaning of English words by the
relationships between them. In particular, Quillian's system sought to compare words
and express the results of those comparisons. Thus, a semantic network is a structure
of directed graph, without any circuit, which encodes the taxonomic knowledge by
objects as well as their properties by a double labelling. The nodes represent concepts
or words that the system “knows” about. Each arc between two nodes represents a
semantic relation between two concepts such as the “is-a” relationship, a
modification (adjective or adverb), a conjunction (and), a disjunction (or), similarity,
![Page 10: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/10.jpg)
10 Olivier Grisvard and Dhouha Kbaier Ben Ismail
consequence, etc. In this way, the representation of the knowledge (of common sense)
is simpler and more natural than with the predicate logic. Adding to that, the ease of
search for information, necessary for certain reasoning and inferences, explains also
the popularity of semantic networks. Nevertheless, their semantics remains vague:
quantization problems; the transcription of a sentence in semantic network is a
delicate problem, since there is no unique (universal) solution. Another defect in this
model is that it is not planned to represent correctly formal semantics such as the
inference. Even though relations of inference are used for the inheritance, semantic
networks do not admit a correct transitivity.
However, there are two derived models with less vague and more formal
semantics. These are the conceptual graphs (Sowa, 1976) [23] and the description
logic [24]. Finally, neural networks started in 1943 by the presentation of McCulloch
and Pitts [25] about the formal neuron which is an abstraction of the physiological
neuron. Neural networks yet require too much processing to be functional. As
exposed above, the formalization of knowledge can rapidly become very complex. In
our analysis, we propose to use mental representations.
3.1. Mental representations: definition and ontology
Mental Representation Theory (MRT), see Reboul & Moeschler 1998 [26], aims at
implementing Relevance Theory (Sperber & Wilson 1986) [27] on the issue of the
enrichment of logical form when reference assignment is concerned. As such, its
main goal consists in proposing tools for reference resolution, in the context of
natural language processing. MRT depends on some cognitive hypotheses, the most
important one being the cognitive functioning. It postulates that reference
assignment is never entirely done at the linguistic level and makes the strong
hypothesis that reference assignment goes through Mental Representations (MRs).
MRs are structured representations which gather heterogeneous information, visual,
![Page 11: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/11.jpg)
Papers dedicated to Anne Reboul 11
spatial, linguistic and encyclopaedic. As such, reference resolution generally consists
in creating, modifying and manipulating MRs, i.e., representations of external reality,
on which a finished number of simple operations can be applied. Although MRs
were designed to solve specific computational linguistic issues (such as plurals,
associative and evolutional anaphora, non truth-conditional expressions, generics,
etc.), they have proven to be a powerful tool for conceptual modelling. We can
distinguish between two sorts of MRs: generic and specific. A generic MR
corresponds to what we classically call a concept and allow determining a category
satisfying the information contained in the generic representation. A specific MR
identifies an individual.
There are different types of MRs organized hierarchically as illustrated in Figure
4. Any MR inherits from one of the two basic MRs-parents: object and eventuality.
The starting point of Eventuality-MRs is the ontology of eventualities proposed by
Vendler (1957) [28] who distinguishes between two major types of eventualities:
states and events. In fact, there are two main reasons to represent events in MRT.
First, events can be designated by referring expressions (the classification of the track,
the configuration of the Radar, etc.) and, given that a basic principle in MRT is that
referring expressions are resolved on MRs, events must have MRs corresponding to
them. Furthermore, objects can be designated through present or past states (the
deleted track, the track that was classified) and these states are the consequences of
various events.
![Page 12: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/12.jpg)
12 Olivier Grisvard and Dhouha Kbaier Ben Ismail
Figure 4: Hierarchy of mental representations and ontology of eventualities.
The category of events is subdivided into three sub-categories: accomplishments,
achievements and activities. According to Vendler, activities and accomplishments
are distinguished from achievements in that the former allow the use of continuous
and progressive aspects. Activities and accomplishments are distinguished from each
other by boundedness. Activities are events that take time but have no inherent
temporal endpoint built into them point (a point before which the activity cannot be
said to have taken place, and after which the activity cannot continue – for example
“The TACCO drew a TWS zone”); these events could go on indefinitely – at least, if
real world limitations or conventions were not a consideration. Accomplishments,
corresponding to result verbs, also describe events that take time, but, in contrast
with activities, these events have an inherent temporal endpoint: the time when the
result state comes about. Of achievements and accomplishments, achievements are
instantaneous and take place immediately (such as in recognize or find) whereas
accomplishments approach an endpoint incrementally (as in classify track or adjust
FLIR parameters). Vendler claims that all achievements encode the inception or
termination of an act and “occur at a single moment”. Thus, achievements are
punctual events. Here are a few examples, (1) to (4) are respectively a state, an
activity, an accomplishment and an achievement:
(1) This operator is well trained and experienced;
![Page 13: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/13.jpg)
Papers dedicated to Anne Reboul 13
(2) He works as a TACCO on Falcon 50 surveillance aircraft;
(3) During this mission, he has built many tracks within the TWS zone;
(4) He has successfully classified all the selected tracks.
We now present the internal structure of a MR and the operations that can be
performed with Mrs.
3.2. Mental representations: structure and operations
The MRs have the following constituent elements:
An address representing at the same time the name of the MR and a pointer to
access the information it contains;
A logical entry containing the logical relations this MR maintains with others
MRs, that is part-whole and whole-part relations (composition);
An encyclopaedic entry offering an access to the concept (generic MR) of
which the object (or eventuality) comes out and its category-specific
information. The encyclopaedic entry also includes a notation field containing
characteristic information of the object that makes possible to isolate the
objects of the corresponding category from objects of other categories;
A visual entry including information relative to the present and past
appearance of the object (or eventuality);
A spatial entry indicating the intrinsic orientation of the object (or
eventuality), as well as its position compared with the other objects in a
common space. This spatial entry keeps a record of the movements of the
objects;
Finally, a lexical entry containing the expressions that have been used or could
be used to designate the object (or eventuality). Thus, the lexical entry
![Page 14: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/14.jpg)
14 Olivier Grisvard and Dhouha Kbaier Ben Ismail
indicates the linguistic expressions used to refer to the object and their
possible morphological derivations.
The visual and spatial entries are additions to the composition of concepts given
in Relevance Theory, since operations on MRs can be triggered by perception as well
as by discourse. The operations on MRs are the following: creation, modification,
fusion, duplication, grouping and extraction.
4. Proposed mental representation-based model
We present here our model of human entities, human roles, tasks, knowledge and
abilities. We illustrate this model with examples using graphical representations of
MRs, taken from a SurMar operation, the target application in Medusa.
4.1. Model of human actors and roles
In domains like SurMar, roles are generally distinct from human actors. On the
one hand, an operator can perform several roles. On the other hand, a given role can
be distributed across several operators. For example, let us suppose that the SurMar
crew is composed of five operators in total representing the cockpit crew and the
cabin crew. Piloting the aircraft is the unique role assigned to the pilot whereas
piloting is only a part of the co-pilot’s responsibilities who also commands the
aircraft. The role of observing is shared between two different operators. Finally,
mission command and sensor management are allocated to the TACCO operator who
elaborates the tactical situation.
We propose to model a human entity using a concrete, animate and human object
concept. As such, we propose to model a human actor by a concrete object MR.
Figure 5 shows the corresponding MR’s internal structure. In the conceptual entry,
operator[1] refers to the category and the cardinal. The notation field contains four
![Page 15: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/15.jpg)
Papers dedicated to Anne Reboul 15
properties. In fact, we characterize an operator by his first name, his last name, the
knowledge he has acquired and the abilities he has developed.
Figure 5: Internal structure of a human actor as an object MR.
In Figure 6, an operator has been instantiated. The encyclopaedic entry includes a
notation field containing characteristic information of the object. Here, the operator is
characterized by four properties: first name (toto), last name (operator1), knowledge
and abilities. The operator has acquired the knowledge @knowlege<275>, a group of
six MRs: knowledge of identification systems, radar systems, FLIR systems, the
maritime domain, meteorology and communication systems. He also has developed
some abilities @abilities<276>, a group of four MRs: ability to switch on/off the FLIR
camera, draw a TWS zone and configure the Radar.
![Page 16: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/16.jpg)
16 Olivier Grisvard and Dhouha Kbaier Ben Ismail
Figure 6: Human operator example.
A role is modelled by an abstract object MR. Since a role consists in the execution
of several tasks over time, we have defined a partition of all the possible tasks in the
logical entry of the object MR. As for an actor, some knowledge and abilities are
required to perform a role. A role has a complexity computed in terms of number of
MRs, as well as a temporal structure, similarly to tasks (see below). The temporal
structure of a role is like a detailed graph grouping all the possible executions of the
allocated tasks. In the example of Figure 6, the operator’s role consists in the
execution four possible tasks, already presented in section (see Figure 2). The
corresponding partition is available in the logical entry of the object MR. The task
model is presented in details in the next subsection.
4.2. Task model
We propose to model tasks using events MRs. Thus, a task can be an achievement,
an activity or an accomplishment. For example, switch FLIR camera on is an
achievement, adjust FLIR camera an activity and perform FLIR classification an
accomplishment. We represent tasks by conceptual events and we decompose each
![Page 17: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/17.jpg)
Papers dedicated to Anne Reboul 17
task into subtasks. All the parts should be described and the corresponding partition
is available in the logical entry of the MR. Furthermore, each task has a temporal
structure. This structure is a complex propositional content corresponding to the
application of a loop (iteration) or binary operators (such as conjunction, disjunction
and conditional clause) to one or several propositions. As an example, for task
perform_FLIR_classification, the temporal structure is the conjunction of three
propositions corresponding to the previously mentioned subtasks:
{get_boat_FLIR_image & decorate_FLIR_image & propose_FLIR_classification} (see Figure
3 in section 2.2). As for all MRs, we refer to the category and the cardinal in the
conceptual entry. In the notation, we implement the same properties already defined
for roles: required knowledge, required abilities, temporal structure and complexity.
Adding to that, an event has a temporal location and a begin time that can be found
in the Time field of the conceptual entry.
Figure 7 is an application of the task model in the case of camera_on, one of the
produce_flir_video subtasks. In order to switch on the FLIR camera, a human actor acts
on the camera concept, and this object participant is defined in the conceptual entry
of camera_on (argD). The required knowledge and abilities are also compulsory to
evaluate the complexity of a task. Below, we address these issues in details and we
introduce the mental model we have adopted for knowledge and abilities.
![Page 18: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/18.jpg)
18 Olivier Grisvard and Dhouha Kbaier Ben Ismail
Figure 7: Internal structure of a task as an event MR.
As an example, we apply the model to the task manage track list. The task is
divided into nine subtasks: create_TWS_zone, exploit_Radar_video, configure_Radar,
build_track, delete_track, automatic_tracking, manual_tracking, dead_reckoning and
end_track_list_update. This partition is represented in the logical entry of the parent
MR, where the parts can be differentiated according to their generic concept and the
begin time of the corresponding event. The temporal structure of this task can be
described as follows: {P1 & {LP ({P2 ||| {P3 V P4 V P5 V P6 V P7 V P8 V P9}})}}
where:
P1 = @create_TWS_zone<25>;
P2 = @exploit_Radar_video<62>;
P3 = @configure_Radar <33>;
P4 = @build_track<20>;
P5 = @delete_track<28>;
P6 = @automatic_tracking<40>;
![Page 19: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/19.jpg)
Papers dedicated to Anne Reboul 19
P7 = @manual_tracking<41>;
P8 = @dead_reckoning<63>;
P9= @end_track_list_update<64>.
LP represents an iterative process, V is a disjunction, & is a conjunction and |||
shows that subtask exploit_Radar_video is simultaneously executed with all the
remaining subtasks (parallelism). Let us suppose that, during a SurMar operation,
the TACCO operator first created a TWS zone. Then, he configured the Radar and
created two tracks. He performed automatic tracking for the former and manual
tracking for the latter. Finally, he ended his tracks list update. The temporal structure
is a complex proposition representing the conjunction of seven propositions as
follows: create_zone<256> & {@exploit_Radar_video<261> ||| {radar_menu<257> &
build_track<258> & build_track<259> & automatic_tracking<260> &
manual_tracking<261> & end_track_list_update<260>}}.
In Figure 8, we apply the mental model to the task classify_tracks. The logical entry
of the MR contains a partition of several events corresponding to the decomposition
of this task into seven subtasks. Then, the temporal structure can be expressed in the
following way: {P1 & {LP ({P2 & {LP ({P3 & {P4 V P5 V P6}})}})} & P7} where:
a. P1 = @start_track_classification<80>;
b. P2 = @select_next_track_to_classify<81>;
c. P3 = @need_refinement<82>;
d. P4 = @get_FLIR_classification<89>;
e. P5 = @get_RADAR_classification<88>;
f. P6 = @perform_track_classification<83>;
g. P7 = @enhance_tactical_situation<84>.
![Page 20: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/20.jpg)
20 Olivier Grisvard and Dhouha Kbaier Ben Ismail
We assume that this task has been carried out by the TACCO. The execution in
Figure 8 shows that the operator used both the Radar and the FLIR to perform one
track classification. The corresponding temporal structure provides more information
about the sequence of events: {P1 & P2 & P3 & {P4 ||| {P5 & P6 & P7 & P8 & P9}} &
P10 & {P11 & P12 & P13 & P14} & P15 & P16 & P17 & P18} where:
P1 = @start_track_classification<284>;
P2 = @select_next_track_to_classify<285>;
P3 = @need_refinement<286>;
…;
P18 = @enhance_tactical_situation<307>.
In fact, the TACCO had been exploiting the FLIR video; and meanwhile he
adjusted three times the camera and pointed it once on the track. Furthermore, he
requested DRP image and proposed a Radar classification. At the end of the process,
he performed the classification.
![Page 21: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/21.jpg)
Papers dedicated to Anne Reboul 21
Figure 8: Task example.
What about the required knowledge and abilities? The next subsection is
dedicated to answer that question.
4.3. Model of knowledge and abilities
Knowledge designates the familiarity with information, facts and descriptions.
Abilities designate aptitudes and intelligence. They are competences to perform an
activity. We do not include skills in the model since we consider them as reflex
behaviours. As discussed in section 3, extensive literature on the subject shows that
the formalization of knowledge and abilities can rapidly become very complex.
However, we need a simple formalism that enables matching between roles and
![Page 22: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/22.jpg)
22 Olivier Grisvard and Dhouha Kbaier Ben Ismail
human actors. For these reasons, we consider knowledge and abilities as mental
states and we model them by state MRs, as shown in Figure 9.
The knowledge model has two participants and one property: agent, know and
level, respectively. The agent participant refers to the operator executing the current
task. The human actor has acquired some knowledge. The required level for this
knowledge can be A (basic), B (intermediate) or C (proficient) depending on the
difficulty of the task. For example, in order to be able to configure the Radar, the
operator should have acquired some knowledge of Radar systems (required
knowledge) and should know how to perform a configuration (required ability). The
required level of knowledge differs from one task to the other. For example, A-level
(basic) Radar systems knowledge is enough to produce Radar video. However,
manage track list requires C-level (proficient) Radar systems knowledge as it requires
to configure the Radar, this configuration including the emitter, the receiver, the
antenna, the wavelength, the scanning strategies, etc., and exploit efficiently the
Radar video (strategic selection of the tracks, etc.).
Figure 9: Generic internal structures of knowledge and abilities as state MRs.
We propose a similar structure to model abilities. The state MR has two
participants (agent and know how to) and one property (required expertise level). The
![Page 23: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/23.jpg)
Papers dedicated to Anne Reboul 23
expertise level property takes the values of novice, advanced beginner, competent,
proficient or expert, inspired from the “novice to expert” model from Dreyfus [29].
For example, an operator can perform a Radar classification if he has developed the
corresponding ability with the highest expertise level (i.e. expert). As a more
complete example, Figure 10 provides the internal structure of threat_analysis_ability.
This structure can be interpreted as follows: to enhance the tactical situation, the
operator must be an expert in threat analysis. If an operator has not developed this
ability, the role of TACCO cannot be assigned to him.
Figure 10: Ability example.
To make sure a given role runs smoothly, it is necessary to check that the human
actor has acquired some concepts and can put his abilities to the best use. The role
allocated to the operator will hopefully go off to a good end. Thus, roles and entities
matching is a two-folded matching:
Matching the required knowledge and the operator acquired knowledge;
Matching the minimum required and the operator acquired expertise level for
all the abilities.
In fact, to carry out the role of TACCO, it is necessary to have acquired the
following knowledge: identification systems (AIS, IFF), radar systems (radar, ISAR,
![Page 24: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/24.jpg)
24 Olivier Grisvard and Dhouha Kbaier Ben Ismail
TWS, DRP, SAR and ISAR library), FLIR systems (FLIR, FLIR library), maritime
knowledge (maritime environment), meteorology, communications knowledge
(navigation systems, GPS), etc. The level of knowledge differs from one task to the
other. For example, a B level (intermediate) maritime knowledge is enough for
manage_track_list. However, performing track classification requires a C level
(proficient) maritime knowledge.
With regard to abilities, a minimum expertise level is required for each
competence. For example, if an operator is an expert in switching on/off the FLIR
camera but only proficient in FLIR classification; then the matching of abilities fails.
In fact, he must do some training in order to become expert in FLIR classification.
Only in the case where the matching goes off smoothly, we can allocate the role to the
concerned operator. We can then observe the operator in situation, estimate the
mission completion time, estimate and analyse the operator’s workload. Applied
metrics allow to verify the capacity of the human actor to perform the tasks he has
been allocated. Sometimes, it seems necessary to redesign the system in case model
performance does not meet requirements by applying a different distribution of tasks
between operators and system or increasing the number of operators. In Medusa,
several workshops and interviews were held with TACCO operators in order to
accurately estimate subjective factors such as the level of experience and the required
expertise level. Figure 11 summarizes the use of the various categories of MRs in our
model and the ontology designed for the SurMar use-case. In the following section,
we now focus on workload calculation issues, on the basis of that model.
![Page 25: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/25.jpg)
Papers dedicated to Anne Reboul 25
Figure 11: Ontology of the SurMar use-case and position of the model.
![Page 26: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/26.jpg)
26 Olivier Grisvard and Dhouha Kbaier Ben Ismail
5. Estimation of the operator’s workload
We propose to use the mental representations of human actors, roles, tasks,
knowledge and abilities as a basis for the estimation of workload. In other words, the
operator’s workload is expressed in terms of the number of MRs used to perform the
role. The workload W of an operator is calculated by means of the formula
)(11 1
MRsCWt in
i
TL
jjj
, where:
tn is the number of tasks considered within a given role;
Factor j is linked to the operator's experience. In fact, the more experienced
an operator is, the less complex the task is for that operator and the less load
over time it requires;
Task complexity jC depends on both the level of required knowledge for the
task (A, B or C) and the required expertise levels of the different abilities
(novice, advanced beginner, competent, proficient, expert). The task
complexity remains unchanged if the expertise level is novice or advanced
beginner. However, it is doubled for a competent required level, tripled for
proficient and quadrupled for expert. According to the knowledge level, the
complexity is doubled for A-level (basic), tripled for B-level (intermediate) and
multiplied by four for C-level (proficient).
Factor iTL represents the load over time; and is computed according to the
temporal structure of the task.
![Page 27: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/27.jpg)
Papers dedicated to Anne Reboul 27
It is important to differentiate TL from C. Although subtasks build fifty tracks
during two hours and build fifty tracks in five minutes have the same complexity, they
have different loads over time.
Let us consider the produce_Radar_video task for example. The corresponding
event MR contains a partition of four subtasks in the logical entry: radar_on,
provide_drp, provide_isar and radar_of. A differentiation criterion, based on the
category and the begin time of the subtasks, is used to isolate each part from the
others. The temporal structure is the conjunction of four propositions corresponding
to the subtasks, and it is expressed as follows: {@radar_on<58> & @provide_drp<60>
& @provide_isar<61> & @radar_off<59>}. Thus, the time load TL of the
produce_Radar_video task is computed by means of the complex proposition above,
that is 4TL . As far as task complexity is concerned, we analyse the required
knowledge and abilities. For radar_on and radar_of subtasks, the operator should
know the concept of Radar. As only A-level (basic level) knowledge of this concept is
required, the complexity is then doubled. The corresponding object MR (argD
participant) has a property on/off which is also updated. Furthermore, an advanced
beginner is the minimum threshold identified as required expertise level to switch
on/off the Radar. With this property value, the complexity of the subtask remains
unchanged. Then for radar_on and radar_of the complexity is 51121 C MRs,
corresponding to the MRs of the event itself, A-level, advanced beginner and argD
participant values. Similar reasoning is applied for subtasks provide_isar and
provide_drp. The operator should have acquired A-level knowledge of the concepts
isar and drp respectively. The argD participants are also updated and there is no
ability for these automatic subtasks. In this case, 4121 C MRs, which
corresponds to the MRs of the event itself, A-level and argD participant values. As
![Page 28: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/28.jpg)
28 Olivier Grisvard and Dhouha Kbaier Ben Ismail
produce_Radar_video is a sub-automatic task, it does not need an expert to be
accomplished. For this reason, we consider factor 0 , since the experience and
training of the operator slightly influence the workload in the example. Finally, the
estimated workload for produce_Radar_video is )(184252 MRsW .
We have developed a graphical interface in order to put our approach into
practice. First of all, we can instantiate an operator and match his knowledge and
abilities with those required to carry out the assigned role. If the matching goes off
without any incident, we can test the operator in situation. The graphical interface
gives access to the tasks associated with the role and makes possible to simulate a
scenario. The main contribution of the interface is to trace and analyse the workload
graphs, especially the causes of workload peaks.
5. 1. Role execution: a scenario example
Let us assume that we have observed an experienced and well-trained operator
SurMar operator on Dassault’s Falcon 50 aircraft. We allocate the role of TACCO to
this operator and suppose that the matching between the operator and the role goes
off smoothly. We make use of the graphical interface to simulate the execution for the
role of TACCO (see Figure 12).
Figure 12: Instantiation of an operator for the role of TACCO.
![Page 29: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/29.jpg)
Papers dedicated to Anne Reboul 29
Figure 13 represents the graphical interface used to simulate the execution for the
role of TACCO. Below is a selection among the tasks he has accomplished during the
mission. First of all, the operator switched on the FLIR and the Radar. Then, he
configured the Radar. He exploited the displayed video and drew a TWS zone. Then,
he built five tracks. He decided to delete one of them. After that, he classified one
track based on a FLIR classification. A Radar and FLIR classifications were necessary
to classify the second track. The operator classified the third track after two Radar
image-based classifications (ISAR and DRP). Finally, he updated the tracks list by
setting a living track as dead reckoning.
Figure 13: Role execution via the graphical interface: a SurMar scenario example.
The workload graph of Figure 14 shows a number of peaks. In fact, even for a well
trained and experienced operator, track classification remains the most demanding
task. The three first peaks reflect the amount of time spent by the operator to propose
both Radar and FLIR classifications of the first three tracks. These were the most
complex tasks throughout the scenario. Next, the operator recognized easily two
tracks and we actually observe a consequent workload reduction for the Radar
classifications from 152 MRs to 66 MRs. Finally, as the last tracks required both FLIR
and Radar classifications, the workload necessary to carry out the corresponding
tasks is higher compared to the previous occurrence and varies from 71 to 112 MRs.
Note that, even though building tracks does not represent a complex task, its
![Page 30: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/30.jpg)
30 Olivier Grisvard and Dhouha Kbaier Ben Ismail
repetition about ten times entails a significant time load, nearly 20% of the total
mission duration.
Figure 14: Sample workload graph from the execution of the SurMar scenario.
In the next subsection, we present how the model has been validated through a
series of experiments.
5.2. Results of the experiments
A series of experiments were conducted at Thales Airborne Systems (Brest,
France) with a SurMar crew in December 2013 and February 2014. The SurMar crew
was composed of a pilot, a co-pilot, two observers and a radar operator. Both the
radar operator and the observer on the left side of the aircraft shared the role of
TACCO. The Radar operator was equipped with several sensors – such as a
contactless eye tracker, an electrocardiogram (ECG)-enabled armband, a wireless
heart rate monitor, etc. – to measure the psycho-physiological signals. The ECG has
been widely accepted in the literature for the assessment of mental workload [30].
This approach is based on the evidence that varying task difficulty influences the
psycho-physiological signals. Figure 15 shows one of the raw electrocardiograms
recorded during the experiments.
Figure 16 shows a twenty-minute portion of the raw ECG recorded during the
experiments. We have associated the corresponding tasks executed by the Radar
operator. We notice that the high peaks correspond to demanding tasks such as
![Page 31: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/31.jpg)
Papers dedicated to Anne Reboul 31
tracks classification or tactical situation enhancement. Consequently, a reduction of
the workload corresponds to simple tasks such as displaying the cartography,
zooming in/out, creating or deleting tracks, etc. Therefore, the ECG is coherent with
the workload graph generated by the model.
Lastly, the table in Figure 17 below presents two interesting peaks (extracted from
the ECG) and associates the corresponding tasks executed by the radar operator. In
these considered samples of the experiment, it was observed that the classification of
the tracks is one of the most demanding tasks. These lead to high peaks in the ECG
record, whereas simple tasks such as creating tracks engender a consequent
reduction of the workload; which confirms the proposed model of workload based
on mental representations
Figure 15: A sample of the recorded ECG raw signal of the radar operator.
![Page 32: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/32.jpg)
32 Olivier Grisvard and Dhouha Kbaier Ben Ismail
Figure 16: Sample of the ECG recorded during experiments on the SurMar scenario.
![Page 33: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/33.jpg)
Papers dedicated to Anne Reboul 33
Peaks Corresponding tasks
Low peak Focus on
Zoom in (6 times)
Create track
High Peak FLIR queuing
AIS track menu
FLIR queuing
Zoom in
Classify
Classify
FLIR queuing
Tactical
Tactical
Classify
Classify
Figure 17: Two peaks from the ECG and the corresponding tasks executed by the TACCO.
6. Conclusion
In this paper, we have proposed a new method to measure the workload. Our
approach calls for three important parameters: the experience and training of the
human actor, the complexity of the task and the time load. Our model is inferred
from tasks analysis. In order to measure the complexity of the task, we have analysed
the required knowledge and abilities for the allocated role. We have proposed mental
representations of human entities, human roles, tasks, knowledge and abilities. The
required knowledge and abilities for each task affect the corresponding complexity.
In fact, the higher the required expertise level is the more complex the task is. We
![Page 34: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/34.jpg)
34 Olivier Grisvard and Dhouha Kbaier Ben Ismail
have investigated the mental representations as well as workload issues to model a
SurMar operation and illustrated our approach for the role of TACCO. Finally,
experiments with SurMar operators were carried out to give a concrete expression to
the predictive workload estimation and validate the proposed analysis. These
experiments on sophisticated SurMar simulators allowed validating the results,
enabling real time processing and a more conveying workload analysis.
The proposed approach is generic and is currently being applied to another use-
case (airborne SIGnal INTelligence – SIGINT) for further validation. In addition, the
model is being refined based on some interviews we conducted with TACCO
operators. The main perspective consists in enriching the theory, adding the influence
of the physical environment and the physiological constraints such as heat, noise,
vibration, stress, etc.
Note: the approach described here has been previously presented at the IEEE
CogSIMA 2015 conference and published as two separate papers in the proceedings
of this conference [31, 32].
References
[1] Hart, S.G., Staveland, L.E.: Development of a multi-dimensional workload rating scale, Human mental workload, 139-183 (1988).
[2] Wickens, C.D.: Multiple resources and performance prediction, Theoretical issues in ergonomics science, 159-177 (2002).
[3] Cain, B.: A review of the mental workload literature, NATO RTO-TR-HFM-121-PartII, 4-1-4-34 (2007).
[4] Johannsen, G., Moray, N., Pew, R., Rasmussen, J., Sanders, A., Wickens, C.: Final report of experimental psychology group. Mental workload: Its theory and measurement, New York: Plenum, 101-114 (1979).
[5] Mulder, G.: Mental load, mental effort and attention. Mental Workload. Its Theory and Measurement, 299-326 (1979).
![Page 35: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/35.jpg)
Papers dedicated to Anne Reboul 35
[6] Colle, H., Reid, G.: Double trade-off curves with different cognitive processing combinations: Testing the cancellation axiom of mental workload measurement theory. Human Factors: The Journal of the Human Factors and Ergonomics Society, SAGE Publications, 41, 35-50 (1999).
[7] Sirevaag, E., Kramer, A., Wickens, C., Reiswerber, W., Stayer, D., Grenell, J.: Assessment of pilot performance and mental workload in rotary wing aircraft. Ergonomics, Taylor & Francis, 36, 1121-1140 (1993).
[8] Yeh, Y., Wickens, C.: Dissociation of performance and subjective measures of workload. Human Factors: The Journal of the Human Factors and Ergonomics Society, SAGE Publications, 30, 111-120 (1988).
[9] Muckler, F., Seven, S.: Selecting performance measures:" Objective" versus" subjective" measurement. Human Factors: The Journal of the Human Factors and Ergonomics Society, SAGE Publications, 34, 441-455 (1992).
[10] Tattersall, A., Foord, P.: An experimental evaluation of instantaneous self-assessment as a measure of workload. Ergonomics, Taylor & Francis, 39, 740-748 (1996).
[11] Hart, S.G. Staveland, L.E.: Development of NASA-TLX (Task Load Index): Results of empirical and theoretical research, Human mental workload, Amsterdam, Holland, 1, 139-183 (1988).
[12] Pfendler, C., Widdel, H.: Gedächtnisleistung und Beanspruchung beim Wiedererkennen von farbigen und schwarzen Reizmustern auf elektronischen Anzeigen. FAT Bericht (1988).
[13] Sepehr, M.: Assessment of subjective mental workload using NASA-Task load index. Proceedings of the 7th European Annual Conference on Human Decision Making andManual Control, 69-75 (1988).
[14] Veltman, J., Gaillard, A.: Indices of mental workload in a complex task environment. Neuropsychobiology, Karger Publishers, 28, 72-75 (1993).
[15] Veltman, J., Gaillard, A.: Measurement of pilot workload with subjective and physiological techniques. In Proceedings of the Workload Assessment and Aviation Safety Conference (1993).
[16] NATO Human View Handbook.
[17] NATO Human View Quick Start Guide.
[18] Blasch, E., Breton R., and Valin P.: Information Fusion Measures of Effectiveness (MOE) for Decision Support, Proc. SPIE 8050 (2011).
![Page 36: A model-driven approach to the a priori estimation of operator …reboul.isc.cnrs.fr/GrisvardKbaier.pdf · 2016-05-11 · A model-driven approach to the a priori estimation of operator](https://reader030.fdocuments.in/reader030/viewer/2022040109/5e6e0f0d4965d5587c49d98e/html5/thumbnails/36.jpg)
36 Olivier Grisvard and Dhouha Kbaier Ben Ismail
[19] Blasch E. P., Bosse E., and Lambert D. A.: High-Level Information Fusion Management and Systems Design, Artech House, Norwood, MA (2012).
[20] Bratman, M., Israel, D., Pollack, M.: Toward an architecture for resource-bounded agents. SRI and Stanford Univ., Stanford, CA, Centre for the Study of Language Information, CSLI, 87-104 (1987).
[21] Rao, A., Georgeff, M.: Modeling Agents Within a BDI-Architecture. In R. Fikesand E. Sandewall, editors, Proc. of the 2rd International Conference on Principles of Knowledge Representation and Reasoning (KR'91), Cambridge, Mass, 473-484 (1991).
[22] Quillian, M. R.: Semantic memory. In M. Minsky, (Ed.), Semantic information processing. Cambridge, MA: MIT Press, 216-270 (1968).
[23] Sowa, J.F.: Conceptual graphs for a database interface. IBM Journal of Research and Development, 20:4, 336-357 (1976).
[24] Brachman, R., Schmolze, J.: An overview of the KL-ONE knowledge representation system. Cognitive science, Elsevier, 9, 171-216 (1985).
[25] McCulloch, W., Pitts, W.: A logical calculus of the ideas immanent in nervous activity. Bulletin of mathematical biology, Springer, 5, 115-133 (1943).
[26] Reboul, A., Moeschler, J.: Pragmatique du discours. De l'interprétation de l'énoncé à l'interprétation du discours (1998).
[27] Sperber, D., Wilson, D.: Relevance: Communication and cognition, Blackwell, Oxford (1986).
[28] Vendler, Z.: Verbs and times. The philosophical review, JSTOR, 143-160 (1957).
[29] Dreyfus, S.E., Dreyfus, H.L.: A five-stage model of the mental activities involved in directed skill acquisition, DTIC Document (1980).
[30] Kilseop, R., Rohae, M.: Evaluation of mental workload with a combined measure based on physiological indices during a dual task of tracking and mental arithmetic, International Journal of Industrial Ergonomics 35(11): 991-1009 (2005).
[31] Kbaier Ben Ismail, D., Grisvard, O.: A model-driven approach to the a priori estimation of operator workload. IEEE International Multi-Disciplinary Conference on Cognitive Methods in Situation Awareness and Decision Support (CogSIMA), March 9-12, Orlando, USA (2015).
[32] Kbaier Ben Ismail, D., Grisvard, O.: Model-driven estimation of operator’s workload for usage centred design of interactive systems, IEEE International Multi-Disciplinary Conference on Cognitive Methods in Situation Awareness and Decision Support(CogSIMA), March 9-12, Orlando, USA (2015).