The relationship between pilots' manual flying skills ...

The relationship between pilots’ manual flying skills and their visual behavior: a flight simulator study using eye tracking Andreas Haslbeck1, Ekkehart Schubert2, Patrick Gontar1, Klaus Bengler1 1

Institute of Ergonomics, Technische Universität München Munich, Germany [email protected]

2

Section Flight Guidance and Air Transportation, Technische Universität Berlin Berlin, Germany

ABSTRACT This paper presents an experimental evaluation of pilots’ ability to support their manual flying skills through visual behavior. To this end, two groups of pilots with different levels of practice and training are compared in a full flight simulator. Different visual information acquisition strategies are used during the flight phases. In flight, pilots must direct their attention towards monitoring, while in a manual flying phase (approach and landing), a more frequent and accurate panel scan is imperative. The gaze data collected during this high-taskload flight period makes it possible to detect the differences between these two groups. Keywords: manual flying, eye tracking

1

INTRODUCTION With highly automated ‘fly-by-wire’ aircraft and the introduction of glass cock-

3131

pits (e.g. types A320 and A340 from Airbus, which were chosen for this experiment), pilots’ management duties in the cockpit have begun to squeeze out manual flying tasks from the cockpits. In general, this process should not be considered negative. Long-haul flights, for example, are very monotonous and require a very low level of action from the pilots. So automation can support pilots in their monitoring tasks during such uneventful flying phases. For critical phases of a flight, however, an active pilot is necessary. Manual flying does not just mean “flying by hand.” When defined as the opposite of automated flight, manual flying covers cognitive processes from information acquisition to cognitive processing, ending with response execution (information implementation). Information is acquired via sensory perception and includes processes such as visual, auditory, and vestibular perception. Verbal communication defines a meta-level of some auditory information. In the next step, all of the information is processed. A broad variety of models exists for this core element of cognition. One frequently used example of these models was created by Rasmussen (1983). The information is implemented via eye-hand coordinative activities, such as trajectory control via side sticks or yokes, and the speed is set by the thrust levers. Another output is also communication. There is some evidence that today’s pilots have lost a significant degree of their manual flying skills. Haslbeck et al. (2012) have discussed these degradation effects and introduced an experimental design to measure such effects. They presented core requirements to find variance in pilots’ manual flying abilities within a realistic but difficult landing scenario. This paper introduces the results of the aforementioned work with a focus on the relationship between pilots’ manual flying skills and their visual behavior. In this experiment, pilots had to perform a landing scenario with eye tracking. The measurements were taken during different approach phases under different automation conditions and taskload levels.

2

PAST AND PRESENT RESEARCH

There are only a small number of studies analyzing visual behavior in modern glass cockpits. Anders (2001) analyzed eye movements as well as head movements of pilots during flight trials in a full flight simulator (Airbus A330). All in all, 16 pilots and thus eight crews took part in this experiment, flying up to four standard missedapproach scenarios. Anders worked out the average attention time allocated to different areas of interest. Thereby he found out, that the gaze allocations on the several indicators differ. The attitude indicator, for example, is fixated more often than speed and altitude indicator. In contrast, they could not find any outstanding differences between the four approaches on the miscellaneous airports or between the different pilots. Another experiment conducted by Diez et al. (2001), asked five Boeing B777 pilots to perform two flight scenarios on a Boeing B747-400 desktop simulator (single-pilot operation). One scenario involved the takeoff, climb, cruise, descent and

3132

approach phases, while a second scenario involved the descent and approach phases. The authors found that the proportion of the gaze allocation to the different areas of interest depended on the flight phase and on specific environmental events. Furthermore, they came to the conclusion that the pilots especially failed to remember the current FMA modes. Sarter, Mumaw, and Wickens (2007) asked ten volunteer Boeing B747-400 flight crews to fly their scenario. The study’s focus was on analyzing mode awareness and appropriate behavior when “wrong” modes were selected. Their results clearly show that a large number of pilots failed at this: only 47% of all required FMA checks were conducted. Furthermore, they found “that such failures were attributed, at least in part, to inappropriate or incomplete knowledge.” Frische, Osterloh, and Lüdtke (2011) conducted a study in an experimental flight simulator cockpit according to the Airbus A350 XWB layout. The tasks’ focus for 15 German airline pilots was on planning und making decisions with a newly built flight management system. The analysis of gaze distributions showed a significant change in the importance of the primary flight display (PFD) among all areas of interest (AOIs). The newly presented flight management system became the most important AOI for several flight tasks. Only during final approach did the PDF remain the most frequently observed AOI. None of these studies referred to manual flight in detail. So there remains a need to link visual behavior on the cognitive processes’ input side with manual flying activities on the output side.

3

EXPERIMENTAL DESIGN AND METHOD

The experiment presented in this paper was motivated by the question of whether the level of practice and training influences manual flying skills. The independent variable, the level of practice and training, was varied by using two different groups of German commercial airline pilots. With a l ow occurrence of this variable, 27 long-haul captains (CPTs) participated. Representing a high level of practice and training, 30 short-haul first officers (FOs) took part in the experiment. All of the subject pilots were analyzed during their regular duties, so participation was not voluntary. The participants were the pilots responsible for flying (pilot flying). For pilot monitoring, the same FO/CPT was always on duty and played a rather passive role – but did not provoke any errors. Manual flight performance was measured as a dependent variable. A full flight simulator objectively measured the actions according to information implementation. Here, deviations from the ideal localizer, glide slope during approach and from the ideal touchdown point, vertical speed, and gforces while touching down were recorded. A DIKABLIS eye tracking system recorded visual behavior (Lange, Spies, Bubb, and Bengler 2010). In combination with audio records, this allows important elements of the pilots’ information acquisition to be analyzed. The participants had been asked for some explanatory variables such as perceived tiredness and workload. Haslbeck et al. (2012) gave a comprehensive overview of the experimental design. The flight scenario started about half an hour before the scheduled touchdown.

3133

After an uneventful approach, the participants were forced to go around because of a strong emerging tailwind after passing an altitude of 1.000ft. After performing the missed approach, an automation defect made a manually flown landing necessary.

4

RESULTS

4.1

GAZE ALLOCATION

For the four different flight phases listed in table 1, relevant AOIs were defined and gaze allocation was analyzed afterwards. The primary flight display shown on figure 1 was separated into five different AOIs: attitude indicator (ATT) with flight director (FD) when engaged, airspeed indicator (SPD), heading (HDG), flight mode annunciator (FMA), and altitude (ALT).

Figure 1 Airbus A340 primary flight display with markers for the DIKABLIS eye tracking system Table 1 Overview of different flight phases Phase I

II III

IV

Content

Start / End

Automated descent on runway 08L EDDM: autopilot (AP) and flight director (FD) engaged Continuous manual flown approach with FD engaged Go-around due to strong tail wind component while ATC changes landing direction Manual landing (raw data approach) without AP and without FD engaged (RWY 26R)

Start: ATC instruction HDG 110 (vectored ILS approach on RWY 08L) End: PF disconnects the AP Start: PF disconnects the AP End: call-out “go-around” Start: call-out “go-around” End: ATC instruction HDG 080 (vectored ILS approach on RWY 26R) Start: malfunction (disconnect of AP and FD) End: touch-down

It was not possible to distinguish between the five different areas of the FMA,

3134

nor between altitude and vertical speed, due to the resolution capabilities of the eye tracker. Table 2 Gaze allocations for CPTs and FOs , flight phase I - IV flight pilots phase

ATT

SPD

ALT

FMA HDG chi-square test

CPTs

.177

.104

.084

.052

.009

FOs

.178

.124

.069

.048

.006

CPTs

.312

.027

.003

.011

.000

FOs

.495

.078

.027

.016

.001

CPTs

.261

.121

.040

.034

.011

FOs

.233

.160

.056

.044

.002

CPTs

.425

.089

.045

.028

.070

FOs

.367

.123

.139

.024

.108

I II III IV

Χ²(4)=49.45; p