AN AIR TRAFFIC CONTROLLER PSYCHOLOGICAL MODEL WITH AUTOMATION

JOSÉ JUAN CAÑAS, PEDRO FERREIRA

Mind, Brain and Behaviour Research Centre, University of Granada,Granada, Spain

delagado@ugr.es, ferreira.pnp@gmail.com

EVA PUNTERO, PATRICIA LÓPEZ, ELENA LÓPEZ

CRIDA A.I.E ATM R&D + Innovation Reference Centre,Madrid, Spain

{epuntero,pmldefrutos,elopezb}@e-crida.enaire.es

FERNANDO GÓMEZ

Aerospace Systems, Air Transport and Airports Department, Polytechnic University of Madrid,Madrid, Spain

fernando.gcomendador@upm.es

Abstract

This paper presents research conducted on a psychological model within the AUTOPACE project (Grant 699238) funded by the SESAR Joint Undertaking as part of SESARa 2020 Exploratory Research Programme within the framework of the EU’s Horizon 2020 research and innovation programme. The model aims to predict the effects of automation on the Air Traffic Controller (ATCo) Mental Workload (MWL) in the future scenarios where automated systems will perform tasks that are currently under the responsibility of an ATCo. Following the new attentional theories, the model predicts that automation will not only have consequences on the mental resources demanded by the task but it will also increase or decrease the ATCo mental activation and engagement with the task. To analytically validate the hypotheses of the model, a computational prototype is being researched. The paper presents some results obtained with an intermediate version of this computational prototype based on the modelling of the mental demanded resources. AUTOPACE ATCo psychological model will be used for investigation on the required new competences and the training strategies to ensure that the ATCo MWL is appropriate to ensure a safe operation.

Keywords ATCo mental workload; automation; computational modelling; trust on automation.

1. Introduction: Automation in the future air-traffic scenarios

The foreseen increase in the air traffic density (EUROCONTROL, 2013a, 2013b; AIRBUS, 2011) will have as one of its major consequences a greater complexity of the Air Traffic Controller (ATCo) tasks. This greater complexity will mean an increase in the workload that will necessarily affect the performance of the ATCo task. To counteract this increase in workload and its consequences on safety, automatic systems are being introduced to help the controllers in performing their tasks or even to perform some functions that have been up to now assigned to them. However, the introduction of these automatic systems presents serious performance drawbacks due to the risk of the “out of the loop” and “fear of automation” effects whose consequences are especially severe in case of automation failure occurring when the ATCo controller is disconnected from the task or has lost certain skills needed to face up with the failure situation. For this reason, a number of research projects are analysing these possible negative consequences and searching for ways to address them.

a As the technological pillar of the Single European Sky initiative, SESAR aims to modernise and harmonise air traffic management in

Europe. The SESAR Joint Undertaking (SESAR JU) was established in 2007 as a public-private partnership to support this endeavour. It

does so by pooling the knowledge and resources of the entire ATM community in order to define, research, develop and validate innovative

technological and operational solutions. The SESAR JU is also responsible for the execution of the European ATM Master Plan which

defines the EU priorities for R&D and implementation. Founded by the European Union and Eurocontrol, the SESAR JU has 19 members,

who together with their partners and affiliate associations represent over 100 companies working in Europe and beyond. The SESAR JU also

works closely with staff associations, regulators, airport operators and the scientific community. www.sesarju.eu.

Amongst them, AUTOPACE is a project researching for a psychological model based on established theories of attentional resources to predict the effects of automation on ATCo Mental WorkLoad (MWL). This model will be used as the base for investigation on the required new competences and training strategies to reach the appropriate ATCo MWL to ensure a safe operation when the future automatic systems are introduced. A safe operation implies that the controllers in their new roles of supervision and monitoring functions can always keep themselves “in-the-loop” to initiate an efficient decision making process, especially to deal with possible unforeseen operational conditions and malfunctions of automation. AUTOPACE is a project within the SESAR 2020 Exploratory Research Programme focused on fundamental science and outreach, thus the aim of the project is to propose the hypotheses that would be validated in subsequent empirical research.

2. AUTOPACE scenarios: 2050 Time Horizon

Due to the level of uncertainty on how automatic systems will be in 2050 (AUTOPACE time frame horizon), two different levels of automation are described in AUTOPACE Future Automation Scenarios (De Crescenzio. et all, 2016): High Automation and Medium Automation. The references used for the definition of those scenarios have been SESAR and other R&D references (European Commission 2011; ACARE, 2012; EREA 2012 and HALA!, 2014) identifying future trends in ATM operational and technological environment. For the purpose of AUTOPACE research, the definition of AUTOPACE Future Automation Scenarios is focused on the responsibilities that are expected to be allocated to ATC actors that are the ATC system and the ATCo. ATCo responsibilities have been described by using three actions: Apply, Approve, and Monitor:

Monitor. It is used when the ATC system is assuming the major tactical actions and the ATCo has to monitor its behaviour to prevent system deviations. Monitoring or vigilance is the activity that an operator performs to acquire Situational Awareness (SA) (Endsley, 1995). The ATCo monitors ATC system actions in both High and Medium Automation Scenarios. Additionally, the ATCo will always need some monitoring of the situation prior to approving or applying any action.

Approve . Once the ATC system has proposed a solution for an ATC intervention, the ATCo must approve it in order to be implemented. Approving requires monitoring as it is also an evaluation of the correctness of the decision of the system. Approving does not imply the implementation of the action, but it does require that the ATCo considers the consequences of the action carried out by the system.

Apply . The ATCo analyses the situation, decides and implements the most suitable solution from those proposed by the ATC system according to the information provided by the ATC tools. Applying also requires monitoring but, in contrast to approving, it is the ATCo who needs to elaborate the solution to the problem and then identify and implement the necessary actions to be carried out.

In general terms, in nominal situations, the ATCo is expected to have the responsibility of monitoring or monitoring and approving in the provision of the majority of the ATC services in AUTOPACE High Automation Scenario. Nevertheless, in AUTOPACE Medium Automation Scenario the ATCo will be responsible not only for monitoring and approving but also for applying many of the ATC services after analysing the proposals made by the ATC system (therefore monitoring, approving and sometimes applying). Table I shows the responsibilities allocation to the ATCo depending on the automation scenario.

Table I. ATCo Responsibilities for the High and Medium Automation Scenarios.

ResponsibilitiesHigh

AutomationMedium

AutomationIdentify conflict risks between aircraft Monitor MonitorProvide flight information to all known flights Monitor MonitorProvide information on observed but unknown flights that may constitute traffic for known aircraft Monitor MonitorRelay to pilots SIGMETS that may affect the route of a flight Monitor MonitorProvide Alerting Service (ALRS) to all known flights according to the following three different phases (INCERFA, ALERFA, DETRESFA)

Monitor Monitor

Check flight-plans/RBT/RMTs for possible conflicts and complexity issues within its area of responsibility Monitor MonitorPlan conflict-free flight path through its area of responsibility Monitor Monitor

Table I (Continued)

ResponsibilitiesHigh

AutomationMedium

AutomationProvide early conflict detection and resolution if the early resolution brings operational benefit (either on the ground side or the airborne side)

Monitor Approve

Assign specified headings, speeds and levels Monitor ApproveRe-route flights to avoid non-nominal or hazardous weather areas Monitor ApproveProvide sequencing between controlled flights Monitor ApproveResolve boundary problems by re-coordination Monitor ApproveImplement solution strategies by communicating trajectory changes to the aircraft through the concerned ATC Controller/System via Data Link

Monitor Approve

Provide separation between controlled flights Monitor ApplyApply appropriate separation to all controlled flights departing his area of jurisdiction Monitor ApplyInput data into the flight data processing system regarding tactical route modification, modification of flight level, etc.

Monitor Apply

Transfer control of aircraft to the appropriate Controller/System when clear of traffic within his area of jurisdiction

Monitor Apply

Co-ordinate with adjacent Controllers/Systems (exit and entry conditions) Monitor ApplyInput tactical trajectory changes into the Flight Data Processing System Monitor ApplyCommunicate with pilots by data link Monitor ApplyMonitor flights regarding adherence to flight plan/RBT/RMT Monitor ApplyMonitor the air situation picture Monitor ApplyMonitor the weather conditions Monitor ApplyMonitor information on airspace status, e.g. activation of segregated airspace Communicate with pilots by data link

Monitor Apply

Monitor aircraft equipment status as provided by the system Monitor ApplyCo-ordinate with adjacent control areas/sectors for the delegation of airspace or aircraft Monitor ApplyIn coordination with the ATC Supervisory or Local Traffic Management roles determine the need for Complexity Solution Measures in the case of overload situations forecast

Approve Apply

Issue holding instructions Approve Apply

3. AUTOPACE ATCo Task Model

After defining the automation scenarios and the ATCo responsibilities in High and Medium Automation, the second step in the project was to model the relevant ATCo tasks for achieving the corresponding responsibilities. In a psychologically valid model of a task such as air traffic control, the responses of a person are organised into structures as part of reactive and, above all, proactive strategies to cope with airspace control events (Sperandio, 1971, 1978). Following this approach, AUTOPACE uses three conceptual units for constructing ATCo task model:

The Control Event: The situation of the environment in which the ATCo is at a given time and that constitutes the input that will trigger an ATC task or the situation that results from carrying out a task. The event is the psychological stimulus to which the ATCo responds. Therefore, the situation of the environment refers to two moments in the dynamics of a situation: (1) the configuration of the situation that will require the completion of the task in response to the Control Event; (2) and the situation that results from carrying out the task. Two examples of Control Events are: “assume a flight” and “transfer a flight”.

Typical Control Situation (TCS): A Typical Control Situation is a task as defined by Vicente (1999): "actions that can or must be performed by one or more actors to achieve a particular goal". Therefore, a task is described in AUTOPACE as a set of actions that are related to each other so that all together serve to achieve a goal. It is possible to make explicit that the actions carried out within a task have a structure that governs these relations: if some actions must follow others or if some actions are optional or should be done in parallel to others, etc. For example, the Control Events “Tactical Conflict resolution by applying re-routing, a level change, a speed adjustment or a direct to a point” will trigger the actions of the Typical Control Situation called “Evaluate and decide a solution”. and Figure 2 show this Typical Control Situation in the High Automation and Medium Automation Scenarios. The representation of the TCS uses several constructors to indicate the relation between actions: sequence (SEC), Alternative (ALT), Parallel (PAR),

Simultaneous (SIM). Shadowed boxes are used for actions that the ATCo will execute in both scenarios while white boxes refer to actions that are executed differently.

Fig

ure

1:

Typical Control Situation "Evaluate and decide a solution" in High Automation

Figure 2:

Typical Control Situation "Evaluate and decide a solution" in Medium Automation

Actions: They are observable behaviours that can be defined as the "behaviour of an actor directed to an objective" (Vicente, 1999). The actions of the ATCo are carried out with the implication of cognitive processes that consume mental resources and are responsible for the mental load supported by the ATCo. The actions that the ATCo will need to implement within each TCS depend on the distribution of responsibilities between the ATCo and the ATC system (see and Figure 2). The cognitive processes associated to these actions are described in the following section.

4. AUTOPACE ATCo psychological model

Once the ATCo task is fully identified and characterised for the two levels of automation (High and Medium), the third step in the project was to propose an ATCo psychological model to predict the effects of automation on the ATCo MWL in the future scenarios. For that purpose, two sub-models are considered: (a) the functional structure of the cognitive system and (b) the mental resources needed to ensure its functioning. It is very important to differentiate between these two aspects (functional structure and functioning of the cognitive system) to understand AUTOPACE hypotheses about the effects of automation on ATCo performance.

4.1. Functional structure of the ATCo cognitive system-Cognitive Processes

The cognitive system has some structural components whose functions are the processing of information from outside, the storing of the results of that processing, and the responding to the environment. This model has been represented in many different ways and AUTOPACE takes as reference the model proposed by Histon y Hansman (2008) (see Figure 3). This model incorporates several interesting aspects of the recent theoretical developments in Human Factors to better represent the ATCo task. In particular, the incorporation in the model of the levels of processing constitutes what is called Situation Awareness (SA) (Endsley, 1995): perception, comprehension and projection. After the three components of SA, the model assumes that decision making processes are finally put in place leading to the action that is executed. The information that the ATCo receives and that comes from the traffic situation is processed by combining it with the one that the ATCo has stored in its memory after its learning and experience in the task of air traffic control. That process allows the controller to understand the actual situation. Then, the ATCo should project into the future to predict how the traffic situation will be and, finally, make decisions about what to do to correctly perform the tasks (execution).

Summarising, the cognitive processes associated to the ATCo tasks that underpin the ATCo Cognitive Functional Structure can be classified in five dimensions: Perception, Comprehension, Projection (SA

Figure 3: Functional Structure for the ATCo Cognitive System, Histon y Hansman (2008)

Processes), Decision Making (Decide Process), Execution (Execution Processes). In case of an ATCo, the two main channels for perception are visual and auditory, and the two main channels for execution are manual and verbal.

In order to link the cognitive processes required to perform ATCo Tasks and Actions (ATCo Task Model), Behavioural Primitives are used to express actions in basic psychological terms or elementary actions. In some models, such as the MIDAS model (Tyler, S. W., et all, 1998), the "Behavioural Primitives" are the ones defined in Table II

Table II. Behavioural primitives grouped by cognitive channels

Cognitive Dimensions Cognitive Channels Behavioural Primitives

Perception

Visual

Fixate Object

Track Object

Search with pattern

Scan with pattern

AuditoryListen

Monitor Audio Signal

ComprehensionRecall

Recognise

Projection

Select

Compare

Compute

Decision Making Decide

ExecutionManual

Reach Object

Press with foot

Move with pattern

Continuous adjustment

Grasp

Mark/Point to an Object

Touch (screen)

Press and release (mouse)

Adjust by rotation

Write

Type

Verbal Say a message

4.2. Functioning of the ATCo cognitive system- Mental Workload Concept

Human behaviour and mental activity require energy. In the same way that a car needs petrol, human beings need energy to perform mental and motor activities. In a sense, research has followed a mechanistic paradigm according to which human machinery function depends on supplied energy (Rabinbach, 1992). It is assumed that the performance of a task will be improved or deteriorated depending, among other things, on the quantity and quality of the energy (resources) supplied (Kahneman, 1973). In the tradition of Attentional Theories and Human Factors and Ergonomics the energy is called “mental resources”. Therefore the cognitive processes associated to the performance of ATCo actions and their related Behavioural Primitives will require mental resources. The amount of resources required by every cognitive channel will depend on the ATCo responsibilities (monitor,

apply, approve) and on the degree of automation of the processing by that channel (Wickens (2002). As an example, and only for illustration, the following Table III shows the assignment of behavioural primitives to actions and their corresponding mental resources (scale 1 to 10).

Table III Example of Actions, Behavioural Primitives and Resources

ActionsBehavioural Primitives Resources

weight

Evaluate the solutions to early conflict resolution proposed by the system (planning phase)

Fixate Object 3Recall 4

Recognise 5Select 7

Compare 7Compute 8

Decide the early conflict resolution from the proposed ones (by the system) (select a solution)

Decide 8

Psychological models have explained the functioning of the human cognitive system by using two concepts, demanded and available resources. While demanded resources are those required by the task and essentially dependent on the task complexity, available resources are the resources that the ATCo has that could be used to perform the task. The fundamental premise of all the models is that the functional structures, such as those described above, will work with an efficacy that will depend on the relationship between the demanded and the available resources. This relation is called the Mental WorkLoad (MWL) and is expressed according to the following formula.

MWL=(Demanded Resources)/( Available Resources)……….(1)

At a certain point three possible situations can occur. If the amount of available resources equals the amount of resources required for performing the task can

be performed optimally. If the amount of available resources is less than the amount of demanded resources, the operator

experiments Mental Overload and the task cannot be performed optimally and the effort to perform the task will affect their physical and mental health.

If the available resources are greater than the demanded resources, the operator may want to perform the task in an optimal way and still has spare resources available to devote to another simultaneous task. If the discrepancy were very large it would produce boredom and finally distraction or drowsiness. This situation is called Mental Underload.

4.2.1 Demanded Resources

The quantification of the cognitive demand resources is calculated using Wickens’s Theory (2002) that was refined in Wickens’s and McCarley’s Theory (2008). When tasks overlap in time, the demanded resources depends on two factors:

The resources demanded for processing for each cognitive channel: perception, comprehension, projection, decision-making, manual or verbal actions;

When two tasks are performed in parallel and use the same pool of resources there would be interferences that increase the demanded resources. This increase is the second component of the formula. The interference could be reduced by the prioritisation of tasks.

These two factors could be expressed as in the Formula 2 after Wickens’ (2002) that it is now used by most researchers for calculating demanded resources.

𝐷𝑒𝑚𝑎𝑛𝑑𝑒𝑑 𝑟𝑒𝑠𝑜𝑢𝑟𝑐𝑒𝑠= 𝑤𝑐𝑢

𝑐=1 + 𝑖(𝑐,𝑑)𝑁𝑑=𝑐+1

𝑛𝑐=1

(2)

where

wc = resources demanded by channel C(c,d) = interference between channels c and d

This formula reflects the assumption that the demand for resources depends on the sum of the weights associated with the demand of the different cognitive channels involved in a task and the sum of the values of interference between channels. Therefore, the second component of this formula represents the cost that the interference between channels has in the calculation of the demanded resources.

4.2.2 Available Resources

The concept of available resources traditionally used in Human Factors has been reviewed in the recent years and is a key aspect of AUTOPACE research. The available mental resources are considered as a pool of resources that a person has at their disposal to perform a task. In the traditional view in Human Factors research it has always been considered that the person confronting the task uses the whole pool of available resources. However, this pool could be of different dimensions containing more or less available resources depending on a number of factors such as stress, fatigue, emotions, etc., all being factors that affect the level of activation or arousal. Therefore, according to these researchers, the whole pool of available resources enters in the denominator of the mental workload equation meaning that the person allocates all the mental resources he has to the task depending on the level of activation.

However, for many reasons, a person might decide to use only a set of the whole pool of available resources. This new understanding of the available resources is behind the recent interests of researchers for concepts such effort or engagement (Wickens, 2014 and Endsley, 2017). Engagement affects the amount of resources that the person will make available to performing a task: the more engaged in the task the more available resources would be allocated to it. We can describe engagement as a continuum. At on one end of the engagement scale, there is a “passive cognitive engagement” that leads to allocating a small amount of resources. On the other end, there is an “active cognitive engagement” that increases the amount of available resources allocated to the task. Therefore, engagement might affect the size of the pool of available resources, but more important than that, it would determine how much of those available resources are dedicated to the task.

5. Effects of automation on AUTOPACE ATCo Psychological Model

Finally, once set the ATCo Psychological Model, AUTOPACE research has been focused to analyse how automation impact on the above-mentioned model based on two different approaches:

By using an existing computational prototype to estimate the demanded mental resources in the two automation scenarios (High, Medium). This analysis provides interesting findings on how the ATCo Congitive processes (funtional strucuture) are affected by automation.

By applying classical and new attentional theories to elaborate a set of hypotheses on how automation affects the demanded resources (task complexity) and available resources (level of activation and engagement), that is the funtioning of the ATCo cognitive system. This analysis provides some contradictory effects to be investigated in future research.

5.1. Automation effect on Functional structure of the ATCo cognitive system

COMETA (COgnitive ModEl for aTco workload Assessment) is a computational prototype developed by CRIDA (Suarez N. et all, 2014) that currently estimates the demanded resources required to perform the controller activity. The Demanded Resources are calculated based on Wickens and McCarley Algorithm (Applied Attention Theory, 2008). Typically COMETA inputs are the Control Events generated in real or simulation environments along with the ATCo Task Model expected in the scenario under study. For AUTOPACE, the Control Events have been generated by a Fast Time Simulation Tool (RAMSb) where AUTOPACE Scenarios Environment (airspace and procedures) have been modelled. The ATCo Task Model

b RAMS: Reorganized ATC Mathematical Simulator. Is a FTS developed by ISA Software (http://ramsplus.com – taken on 06-03-2017)

(Tasks associated to Events, Actions, Behavioural Primitives and Mental Resources) has been adapted to the control activity expected in AUTOPACE Scenarios (High and Medium automation) in nominal and non-nominal situations. COMETA presents the results not only as a global figure for the demanded resources but also the apportionment to every dedicated cognitive processes and dimensions.

Figure 4 shows some results obtained with COMETA related to the functional structure evolution and the expected cognitive process in the future automation scenarios, all of them compared with current ATC paradigm. As observed, the distribution of the functional structure of the cognitive system changes drastically with automation. While current ATCo uses the cognitive dimensions (visual, comprehension, projection, decision making and verbal resources) in a balanced way, the future ATCo shall focus his/her cognitive effort in mainly comprehension and projection. The ATCo needs to project what is going to happen to understand the system performance without missing Situational Awareness. In Medium Automation Scenario where main actions are not only monitor and approve but also apply, Projection is more relevant than Comprehension as ATCo needs to invest more resources on project future pictures to correctly select the options given by the system (approve) and his/her own instructions (apply). In the High Automation Scenario, the contrary occurs as what is important is to have a more robust mental picture of what is occurring to monitor system performance (monitor) and to approve system proposals (approving), namely a better Comprehension than Projection.

Figure 4: Cognitive Processes evolution in the Current, Medium Automation, and High Automation scenarios (Cañas, J. et all, 2017).

5.2. Automation effect on Functioning of the ATCo cognitive system (Mental Workload)

The prediction of the effect of automation on the ATCo MWL depends on the atteentional theories supported by researchers, apparently with contrary effects if the attentional theories are referred to the Level of Activation/Arousal or to the Engagement with the task as it will be described in the current section.

5.2.1 Classical Attentional Theory

On the one hand, the psychological effects of automation are explained by the classical attentional resource theories (Kahneman, 1973), by assuming that automatic systems only reduce the demanded resources by the task and, therefore, reduce the mental load and avoid the overload. According to these classical theories, automation does not affect the available resources.

Figure 5: Effects of automation on demanded and available resources from the point of view of classical theories of attentional resources

5.2.2 MART Attentional Theory-Level of Activation

On the contrary, there are alternative theories, such as the Malleable Attentional Resources Theory (MART) (Young and Stanton, 2002) that assume that the task complexity would affect not only the demanded resources but also the available ones depending on operator’s expectations. For example, when the operator expects that the task will be easy in the near future, she/he will reduce available resources.

Figure 6: Effects of automation on demanded and available resources from the point of view of new theories of attentional resources

This reduction will occur at the same time as a reduction on demanded resources. Then, there would be more risk of overload or no change of MWL at all if the reduction is equal in the demanded and the available resources. However, if the ATCo feels fear of automation failures, the stress would increase and then there will be an increase in the amount of available resources due to an extra activation. Therefore, the available resources would be greater than the demanded resources and underload could be observed. In this case the extra resources would be dedicated to the task when they are not needed for performing it. Therefore, when the controller feels fear of failure -or untrust the automatic system- underload, disorientation, overacting or erratic behaviour would be observed.

Trust on the system is the key factor for the ATCo level of activation. If the ATCo trusts the systems, he/she will reduce the amount of resources allocated to the main task and therefore, potentially reduce available resources. Interestingly, this reduction of available resources increments the risk of lack of SA and OOTL. The ATCo tust in the sytem can be either the system is not too complex/opaque (situational trust) or there is a positive experience with a very reliable system (learned trust) (Hoff and Bashir, 2015). Based on situational trust and learned trust, the more is the trust, the less is the level of activation. From a Situational Trust point of view, the level of activation for the High and the Medium Scenarios will be different. As the High Automation Scenario is more complex and opaque, the trust of the operator might be lower in a high automation scenario than in a medium automation scenario. As the operator will invest more available resources (denominator of the formula) in high automation scenarios (higher level of activation) than in

medium automation, for the same level of demanded resources in both scenarios, the ATCo will experiment less MWL in a High Automation Scenario than in a Medium Automation Scenario.

Meanwhile, Learned Trust acts as a shaper of the Situational Trust. Independently of the scenario (High or Medium), if the ATCo fully trust the system as no failure has experienced before, the level of activation can be so low that might create an OOTL effect. If the task complexity is also low no safety problem would occur (same level of demanded and available resources) but if a failure, occurs the task complexity would be higher creating Mental Overload. If the ATCo does not trust the system, the level of activation can be so high that panic, desorientation or erratic behaviour can be observed. From a safety point of view this situation is not desirable. Considering the MWL model, the available resources in the dennominator would be high and therefore the MWL would be very low creating a situation of Underload. Discarding the psychological negative effects, if a failure occur (increase of demanded resources) the operator would be in a better position to cope with the task complexity increment as he/she had more avaialble resources.

5.2.3 Cognitive Engagement

Different new researchers (new different attentional theories) claim that no all the available resources that a person has are allocated to the task. Independently on how many available resources the ATCo has (due to activation or arousal), he/she might allocate only one part of the total amount depending on the effort dedicated to the task and how much he is engaged in the task (Endsley, 2017).

Figure 7: Effects of automation on available resources from the point of view of new theories of attentional resources

It is assumed that engagement with the task and SA would vary with responsibility (Metzger and Parasuraman. 2001). As can be seen in Figure 8, the responsibility for “apply” actions would require more engagement than just “approve” automated decisions and actions. The responsibility that requires less engagement is “monitor”. However, each responsibility relates to different amplitude of variability on engagement. For example, when applying an action, the engagement does not suffer a significant variability, i.e. the ATCo has to be engaged with the task. However, “monitor” could vary significantly. The ATCo could decrease his/her level of vigilance during monitoring in the case that nothing occurs during a period of time.

That means that in a High Automation Scenario, where the ATCo mainly performs “monitor” and “apply” actions, he/she would be less engaged with the task than in Medium Scenario. As a consequence he/she would have poorer SA and lower decision making and the performance in detecting a system failure and shifting to the correct course of action would be poorer (more difficult, take longer time). In other words, the ATCo will allocate less available resources in the High Scenario than in the Medium and for the same level of demanded resources, the ATCo MWL in High Automation Scenarios would be higher than in the Medium

Automation Scenario. This effect is apparently in contradiction with the predictions coming from the attentional theories predicting the effects of automation on the level of activation.

Figure 8: Engagement and situation awareness as a function of the three responsibilities in High Automation versus Medium Automation

6. Conclusions and future research

As a final conclusion, AUTOPACE sets the base to develop a new psychological model to predict the effects of automation on the ATCo performance in the Future Automation Scenarios. The model is composed of two main sub-models: (a) the cognitive functional structure model reflecting how the ATCo acquires the Situational Awareness (perception, comprehension, projection), makes decisions and executes the tasks manually and verbally and (b) the functioning of the cognitive system modelled through the mental resources concept to analyse the ATCo MWL associated to those cognitive processes when performing the task. The MWL concept is understood as the relationship between the demanded resources by a task (task complexity) and the available resources allocated to perform a task.

AUTOPACE preliminary findings on how the automation impacts on the functional structure of the cognitive system show radical changes with respect to the current situation. While current ATCo uses visual, comprehension, projection, decision making and verbal resources in a balanced way, in highly automated environments the ATCo shall focus his/her cognitive effort on mainly comprehension and projection. The ATCo needs to understand the situation and to project what is going to happen to understand the system performance without missing Situational Awareness. These changes in the use of cognitive process with automation might support the design of the systems to facilitate the SA acquisition.

In turn, AUTOPACE contributes to better understand how automation impact on the demanded and available resources by incorporating new attentional theories that also help comprehend negative psychological effects of automation such as OOTL or panic to failure. As already exposed, classical attentional theories only predicts the

automation effects on demanded resources saying that the task complexity is reduced with automation and therefore the MWL decreases. On the contrary, the new attentional theories consider that automation also affects the available resources not only on the pool of available resources to perform a task (level of activation) but also on the amount of the available resources that the ATCo wants to allocate to these tasks (engagement). When incorporating these new attentional theories to the model some of them apparently predict opposite effects with automation on ATCo MWL.

The level of activation (arousal) increments the pool of available resources and decreases with the operator trust on the system. The more is the trust, the less is the level of activation (the less is the pool of available resources). As High Automation implies complex systems and opacity (situational trust), the ATCo tends to not trust the system and as a consequence the level of activation is higher. But independently of the level of automation (high or medium), if the ATCo fully trust the system (learned trust), the reduction of the level of activation creates an OOTL effect being in an unsafe situation in case of system failure. When the ATCo does not trust the system, the level of activation can be so high that panic, disorientation or erratic behaviour can be observed.

The engagement with the task makes ATCo to allocate more available resources to the task. The more active participation on the task performance, the more engagement. As High Automation implies more “passive” activities such as monitor, the ATCo is less engaged with the task and therefore allocate less available resources to them.

As a summary, if the demanded resources remains the same in High and Medium Scenarios, the following TableIV reflects the effect of automation on the ATCo MWL according to the new Attentional Theories.

Table IV Automation Effect on the ATCo MWL according to new Attentional Theories.

High Automation Medium Automation

Available Resources MWL Available Resources MWL

Situational Trust ↑ ↓ ↓ ↑Learned Trust – Full Trust (OOTL) ↓↓ ↑↑ ↓↓ ↑↑Learned Trust – No Trust (Panic) ↑↑ ↓↓ ↑↑ ↓↓Engagement ↓ ↑ ↑ ↓

As the total effect of automation on available resources is a combination of the level of activation (situational and learned trust) and the ATCo engagement with the task, further research is needed to validate how the combination of these factors will be. Once this is done, a development of a computerised model to be used for prediction of automation on ATCo performance will be closer to reality.

Finally, the ultimate goal of AUTOPACE is to indicate requirements for training competences resulting from the analysis carried out about the effect of automation on the ATCo psychologial model. In the future automation scenarios some new training strategies and competences will be needed to cope with the effects of “out-of the loop”, stress, disorientation, panic, etc. to ensure that the ATCo performance is optimum. Therefore, the research done in AUTOPACE based on the ATCo Psychological Model will support future research on system design to balance the use of the different cognitive processes and new training strategies to cope with the negative effects of automation.

References

ACARE (Advisory Council for Aviation Research and Innovation in Europe. (2012). SRIA – strategic research and innovation agenda – executive summary No. 44)

Cañas, J.J., Ferreira, P., Puntero, E. Rodriguez, S. Gómez, F. López E. Dominguez, D., & López, P. AUTOPACE, D3.1 ATCo psychological model with automation

De Crescenzio, F., and, & Lucci, F. (2016). AUTOPACE, D2.1 future automation scenariosEndsley, M. R. (1995). Toward a theory of situation awareness in dynamic systems. Human Factors: The Journal

of the Human Factors and Ergonomics Society, 37(1), 32-64.Endsley, M. R. (2017). From here to autonomy: lessons learned from human–automation research. Human

factors, 59(1), 5-27.EREA (Association of European Research Establishments in Aeronautics. (2012). From air transport system

2050 vision to planning for research and innovation No. 3)Elsevier B.V.EUROPEAN COMMISSION. (2011). Flightpath 2050, europe’s vision for aviation - report of the high level

group on aviation researchHALA! SESAR Research Network. #7. HALA! position paper 2014 version rev 1 dic finalHiston, J. M., & Hansman, R. J. (2008). Mitigating complexity in air traffic control: The role of structure-based

abstractions. report no. ICAT-2008-05. Cambridge, MA: Department of Aeronautics and Astronautics, Massachusetts Institute of Technology.

Hoff, K. A., & Bashir, M. (2015). Trust in automation integrating empirical evidence on factors that influence trust. Human Factors: The Journal of the Human Factors and Ergonomics Society, 57(3), 407-434.

ISA. Innovation for Sustainable Aviation. RAMS plus 6.09.Kahneman, D. Attention and effort. 1973. Englewood Cliffs, N~ Prentice-Hall,Metzger, U., & Parasuraman, R. (2001). The role of the air traffic controller in future air traffic management: An

empirical study of active control versus passive monitoring. Human Factors, 43(4), 519-528.Parasuraman, R., & Riley, V. (1997). Humans and automation: Use, misuse, disuse, abuse. Human Factors: The

Journal of the Human Factors and Ergonomics Society, 39(2), 230-253.Rabinbach, A. (1992). The human motor: Energy, fatigue, and the origins of modernityUniv of California Press.SESAR Joint Undertaking. (a). SESAR concept of operations step 2 edition 2014SESAR Joint Undertaking. (b). WP4 detailed operational description (DOD) step 2Sperandio, J. C. (1971). Variation of operator's strategies and regulating effects on workload. Ergonomics, 14(5),

571-577.Sperandio, J., C. (1978). The regulation of working methods as a function of work-load among air traffic

controllers. Ergonomics, 21(3), 195-202.Suárez, N., López, P., Puntero, E., & Rodriguez, S. (2014). Quantifying air traffic controller mental workload.

Fourth SESAR Innovation Days,Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science (New York,

N.Y.), 185(4157), 1124-1131.Tyler, S. W., Neukom, C., Logan, M., & Shively, J. (1998). The MIDAS human performance model.

Proceedings of the Human Factors and Ergonomics Society Annual Meeting, , 42. (3) pp. 320-324.Vicente, K. J. (1999). Cognitive work analysis: Toward safe, productive, and healthy computer-based work.

Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers, Mahwah, NJ.Wickens, C. D. (2002). Multiple resources and performance prediction. Theoretical Issues in Ergonomics

Science, 3(2), 159-177.Wickens, C. D. (2014). Effort in human factors performance and decision making.Human Factors: The Journal

of the Human Factors and Ergonomics Society, 56(8), 1329-1336.Wickens, C. D., & McCarley, J. S. (2007). Applied attention theory CRC press.Young, M. S., & Stanton, N. A. (2002). Malleable attentional resources theory: A new explanation for the effects

of mental underload on performance. Human Factors, 44(3), 365.