Human Processor Models to Outline the Pilot Assistance Required for Single Pilot Operations

Table of contents

Abstract

Nomenclature

1 Introduction

2 State of research
2.1 Human-factors engineering
2.2 Flight phases
2.3 Cockpit systems
2.4 Operating procedures

3 Model Human Processor
3.1 Perceptual System
3.1.1 Perceptual Memories
3.1.2 Perceptual Processor
3.2 Motor System
3.2.1 Motor Processor
3.3 Cognitive System
3.3.1 Cognitive Memories
3.3.2 Working Memory
3.3.3 Long-Term Memory
3.3.4 Cognitive Processor
3.4 Movement - Fitt’s Law

4 Model
4.1 Structure
4.1.1 Definitions
4.2 Model design
4.2.1 Assumptions
4.2.2 Identifying procedures, tasks and actions
4.2.3 Representation in spreadsheet
4.3 Simulation
4.3.1 Structure
4.3.2 Import
4.3.3 Sums of imported values
4.3.4 Distributions
4.3.5 Movement - Fitt’s Law
4.3.6 Procedure times
4.4 Results
4.4.1 Total action times
4.4.2 Procedure times for exemplary flight sectors
4.4.3 Workload of Pilot Flying
4.4.4 Summary
4.5 Validation

5 Pilot assistance
5.1 Interfaces
5.1.1 Control panels
5.1.2 Microphone
5.1.3 Touch screen
5.1.4 Loudspeaker
5.1.5 Head up display
5.1.6 Eye tracker
5.2 System outline
5.2.1 Cross checks
5.2.2 Check lists
5.3 Information design
5.3.1 Perceptual System
5.3.2 Cognitive System
5.3.3 Conclusion
5.4 Simultaneousness
5.4.1 Input
5.4.2 Output
5.5 Limitations

6 Conclusion

References

List of figures

List of tables

7 Appendix
7.1 Model Human Processor
7.2 Procedure representation
7.3 Matlab Code
7.4 Distributions
7.5 Results

Abstract

The object of this thesis is to outline prospective assistance systems enabling a pilot to fly an airliner single-handedly. A cognitive modelling technique called Model Human Processor is introduced. Procedures and tasks involved in the operation of an aircraft are identified. Assumptions with respect to the single pilot design alternative are made. A simulation is implemented in Matlab in order to assess the pilots’ workload. Results allow for a procedure time and workload comparison of the two flight crew alternatives. The outcome of this analysis facilitates the design of potential additional pilot support systems that can reduce workload and improve situational awareness.

Keywords: Model Human Processor, Single Pilot Operations, workload, avionics

Nomenclature

Abbildung in dieser Leseprobe nicht enthalten

Chapter 1 Introduction

Historically, as aircraft became larger requiring more engines and complex systems to operate, the workload on its two pilots became excessive during certain critical parts of the flight. Therefore, the flight engineer made up the crew as the person who moni- tored and operated the complex aircraft systems.¹ Nowadays the opposite tendency is observed. Avionics have reduced the pilot’s workload by taking over numerous duties and continue to be improved regarding their performance. As a consequence avionics have not only made the flight engineer superfluous but also allowed manufacturers to produce unmanned aerial vehicles.² Today pilots on civil airliners are progressively becoming flight managers mainly programming and monitoring the tasks performed by the avionics. This trend suggests that in the future even passenger jet aircraft will be able to fly autonomously remotely monitored from ground control stations. How- ever, with the concept being radical from a contemporary point of view the next step of transition will most likely be an aircraft operated by a reduced crew consisting of just one pilot. Research on the so called Single Pilot Operations (SPO) is carried out in order to evaluate the feasibility of the concept. The pilot still plays a central role in SPO and is a subject of research since human failure has been the cause of many accidents. In order to mitigate the risk the pilot must not be overburdened by having to take over too many duties originally assigned to the co-pilot. Therefore, this study focusses on the assessment of the pilot’s workload as well as assistance systems that may optimize it.

Owing to the fact that Single Pilot Operations on airliners is still a future concept empirical data are not available. Therefore, and because of flight simulators not being available for the study, the workload is simulated using a human engineering approach called the Model Human Processor (MHP) and is described in detail in chapter three.

Chapter four starts with the development of the model used in this study. Definitions and assumptions are made in order to simplify the complex nature of operating an aircraft. The identification of the tasks involved forms the basis of the Matlab simula- tion that is explained subsequently. Results regarding procedure times as well as the workload of the pilots are presented, assessed and validated at the end of the fourth chapter. The following chapter begins with the introduction and evaluation of useable input and output interfaces. Moreover, the outcome of the simulation and the knowl- edge gained from the MHP are integrated. This facilitates outlining assistance systems that can help lower the workload of the pilot in SPO while also providing flight safety. The appendix consists of additional information about the Model Human Processor, the spreadsheet containing the data used for the simulation, the Matlab Code, figures explaining the implementation of the MHP in Matlab as well as more detailed results. However, first of all, the following chapter will give an overview of the research that has already been carried out and of the technology currently available.

Chapter 2 State of research

The evaluation of pilot workload has represented a complex measurement problem since the earliest days of manned flight. Traditional approaches have included a series of techniques, many of which have not proved successful. The most popular method has employed post-flight questionnaires or interviews.¹ In order not to rely on these subjective parameters evaluated by the pilots themselves, scientists have tried to find additional techniques to quantify workload. Some of these are the measurement of pupil diameter, speech², heart rate, respiratory rate³ and brain activity.⁴

However, assessing the workload of a pilot single-handedly operating an aircraft requires a new approach. Since the necessary flight simulator and qualified pilots are not available for this study, a way of modelling workload has to be found:

2.1 Human-factors engineering

Human engineering is an applied science that coordinates the design of systems and physical working conditions with the capacities of the worker. It involves cognitive modelling methods which allow a system designer to predict the time it takes a person to complete a task without performing experiments.⁵

Model Human Processor

One basic and often used model is the Model Human Processor, which draws an analogy between the processing and storage areas of a computer with the perceptual, motor, cognitive and memory areas of the human and will be explained thoroughly in the following chapter 3. It has been successfully utilized for the optimization of driver

workload in different driving situations as well as for comparing the performance of young and old adults⁶ and the determination of pilot’s workload. Wu, Tsimhoni and Liu integrated two complementary approaches to cognitive modelling: the queueing network approach and the MHP. Thereby the three discrete serial stages of the Model Human Processor expands into three continuous-transmission subnetworks. Multitasking performance emerges. A driver performance model was created and interfaced with a driving simulator to perform a vehicle steering and a map reading task concurrently and in real time. The performance data of the model were similar to human subjects performing the same tasks.⁷

In order to evaluate the feasibility of a single pilot cockpit design alternative, Graham, Hopkins, Loeber and Trivedi developed a human performance model based on the Model Human Processor. Using the Flight Crew Operating Manual of an Avro RJ100 aircraft as their main source the researchers from George Mason University concluded that a single pilot operating the aircraft would experience a workload increase while the total processing time would decrease.⁸

2.2 Flight phases

Statistics reveal that accident-rich phases of flight are takeoff and climb, with 31% of all accidents and, even more so, approach and landing, with 43% of the fatal accidents occurring within 16% of the average flight time.⁹ Studies focussing on workload, sit- uation awareness and fatigue have shown that workload is especially high for ascent (/takeoff) and descent (/landing).¹⁰ Both of these findings suggest that focussing on these phases makes sense since flight safety is the most important concern in aviation and high or even critical workload increases the risk of human error.¹¹

2.3 Cockpit systems

Airliners are equipped with avionic systems which include communications, naviga- tion, the display and management of multiple systems, and the hundreds of systems that are fitted to aircraft to perform individual functions. The cockpits of most jets in use today are fitted with pushbuttons, switches, levers and sticks. Nowadays, con- ventional manual fight controls are replaced by an electronic interface. Input given by the pilot via levers, switches etc. is converted to electronic signals and analyzed by a computer how to move the actuators correspondingly. This system is called fly-by-wire (FBW). Wide-ranging pilot support systems including automatic stability and autopi- lot systems assist the pilot in controlling the aircraft. State-of-the-art jet airliners are equipped with so called glass cockpits featuring multi-function displays that can show any needed flight information. Recently, head up displays have been introduced for latest generation commercial aircraft like the Boeing 787 or Airbus A350 XWB and are now available for other airliner families as retrofit equipment,¹² whereas integrated touch screens have only been used in combat and business aircraft cockpits.¹³

2.4 Operating procedures

Every flight is different and every pilot acts differently. But since an accurate determination of the procedures performed by the pilots is essential for the assessment of their workload, these procedures have to be standardized. This is accomplished by choosing the Flight Crew Operating Manual (FCOM) as the basis of the developed model. It comprises information about the tasks involved in the operation of an aircraft, including normal as well as emergency procedures.¹⁴ The vehicle considered in this study is the Airbus A320, part of the world’s best-selling single-aisle aircraft family. It features a glass cockpit, but is not fitted with a head up display.¹⁵

Chapter 3 Model Human Processor

The Model Human Processor is a cognitive modelling method used to compute the expected time a human being would need to perform a certain task. It is a human engineering model which was developed by Card, Moran and Newell in 1983 as a way to summarize decades of psychology research.¹ It can be seen as a high-level view at the cognitive abilities of a human being.

The MHP consists of three parts: the perceptual, cognitive and the motor subsys- tem each of which - similar to a computer - has memories and processors. Input by the eyes and ears is taken by the perceptual processor and dropped into the visual image store and the auditory image store, respectively. As a computer hardware analogy, these memories are like frame buffers, storing a single frame of perception. The motor processor takes instructions from the working memory and runs these on the muscles. The cognitive processor operates on data from all the memories, including long-term memory, and puts its results back in the working memory. Figure 3.1 shows a scheme illustrating this structure. In the case of tasks like pressing a key in response to a light, the human model has to behave as a serial processor with each of the subsystems working successively, whereas for tasks like reading or typing integrated, parallel oper- ation of the subsystems is more suitable. The memories and processors are described by parameters, the most important ones are:

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.1: Model Human Processor scheme

3.1 Perceptual System

Sensations detected by the body’s sensory systems are carried into internal represen- tations of the mind by using integrated sensory systems. A good example of such an integration can be found in the visual system: To be able to continuously follow a visual scene the perceiver has to combine central vision, peripheral vision, eye movements and head movements. The eye is in continuous movement in a sequence of saccades, each taking about 30 ms³ to jump to the new point of regard and dwelling there 60 700 ms for a total duration of⁴ Eye − movement (travel + fixation) = 230 [70 − 700] ms (The number 230 ms represents a typical value and the numbers in brackets indicate that values may range from 70 ms to 700 ms depending on conditions of measurement, task variables or subject variables.)⁵ Whenever the target is more than 30 degrees away from the central vision, head movements occur in order to reduce the angular distance.

3.1.1 Perceptual Memories

After visual or auditory stimuli, a representation of these appear in the Visual Image Store or Auditory Image Store, respectively. These memories then contain information coded physically, i.e., as an analogue to the external stimulus. Physical properties such as intensity affect the code. Without going into further detail the code types can be determined as:⁶⁷

Abbildung in dieser Leseprobe nicht enthalten

For example, the Visual Image Store representation of the number 2 contains features of curvature and length.

The half-life is used to get a parameter of memory decay time and defined as the time after which the probability of retrieval is less than 50”%”. Even though exponential decay is not necessarily implied by the use of the half-life, studies haven shown that in many cases it is a good approximation to the observed values. The Visual Image Store has a half-life of about⁸

Abbildung in dieser Leseprobe nicht enthalten

but the Auditory Image Store decays more slowly⁹:

Abbildung in dieser Leseprobe nicht enthalten

This is consistent with the fact that auditory information must be interpreted over time. The capacity of the Visual Image Store is hard to fix, but rough working purposes may be taken to be about

Abbildung in dieser Leseprobe nicht enthalten

The capacity of the Auditory Image Store is even more difficult to fix, but estimated to be around

Abbildung in dieser Leseprobe nicht enthalten

3.1.2 Perceptual Processor

The cycle time of the Perceptual Processor is identified by the time response of the visual system to a pulse of light, the so called unit impulse response, and is in the order of¹⁰

Abbildung in dieser Leseprobe nicht enthalten

I.e., if the retina is stimulated at a certain time t = 0 and the image is available in the Visual Image Store 140 ms later, the cycle time of the Perceptual Processor for this case is

Abbildung in dieser Leseprobe nicht enthalten

Table 3.1 summarizes processor and memory parameters of the Perceptual System.

Abbildung in dieser Leseprobe nicht enthalten

Table 3.1: Parameters of the Perceptual System

3.2 Motor System

The motor system finally translates thought into action by activating muscles. The most important sets of effectors are the arm-hand-finger system and the head-eye sys- tem.

3.2.1 Motor Processor

Since movement consists of a series of discrete micromovements, it is not continuous. Each micromovement requires about¹¹

Abbildung in dieser Leseprobe nicht enthalten

This time is identified as the cycle time of the Motor Processor. With the feedback loop from action to perception being 200-500 ms long, rapid behavioral acts such as typing must be executed in bursts of preprogrammed motor instructions. In the case of the user making a mistake this feedback loop has to be considered. The total time is higher due to the perceptual, cognitive and motor system requiring

Abbildung in dieser Leseprobe nicht enthalten

Table 3.2: Parameters of the Motor System

3.3 Cognitive System

The cognitive system merely serves to connect the perceptual to the motor system for the simplest tasks. But with most tasks being complex and involving learning, retrieval of facts or the solution of problems, memories and processor for the cognitive system are more complicated than those of the other systems.

3.3.1 Cognitive Memories

The cognitive system consists of two important memories: the Working Memory for holding the information under current consideration, and the Long-Term Memory for storing knowledge for future use. Structurally, Working Memory consists of a subset of the elements in Long-Term Memory that have become activated. What with most tasks frequently performed by people involving auditory or visual sensations for the Model Human Processor the predominant code types are considered to be

Abbildung in dieser Leseprobe nicht enthalten

The activated elements of Long-Term Memory define Working Memory and consist of symbols, called chunks, which may themselves be organized into larger units.¹² These can be seen as nested abstract expressions:

Abbildung in dieser Leseprobe nicht enthalten

What constitutes a chunk depends on the user as well as on the task, because the contents of user’s Long-Term Memory play a major role in chunking. For example, most people are not able to repeat the following sequence of nine letters back:

Abbildung in dieser Leseprobe nicht enthalten

However, consider the list below:

Abbildung in dieser Leseprobe nicht enthalten

Especially when spoken, this sequence will be chunked into CBS IBM RCA (by the average American college student) and thus be easily remembered because of consisting of only three chunks. This is how random lists of symbols can be mapped into prepared chunks if the user can perform the necessary recoding fast enough. However, the user will not be able to recall the chunks for an infinite amount of time. Sooner or later the Working Memory will be empty.

3.3.2 Working Memory

The half-life is taken as a working value. Since the general decay half-life of the working memory¹³

Abbildung in dieser Leseprobe nicht enthalten

has a wide range and is particularly sensitive to the number of chunks in the recalled item, the decay rates of different item sizes are useful as well:

Abbildung in dieser Leseprobe nicht enthalten

When recalling information a few seconds after hearing it, people use both Working Memory and Long-Term Memory. These two systems have been separated in experiments showing with the results showing a pure capacity of Working Memory¹⁴

Abbildung in dieser Leseprobe nicht enthalten

and an effective capacity of Working Memory¹⁵

Abbildung in dieser Leseprobe nicht enthalten

which is the pure capacity augmented by the use of Long-Term Memory.

3.3.3 Long-Term Memory

Long-Term Memory holds the user’s available knowledge which consists of a network of related chunks, accessed from the contents of the Working Memory. These contents are not only facts, but procedures and history as well. Apparently, there is no erasure from Long-Term Memory, so

Abbildung in dieser Leseprobe nicht enthalten

However, successful retrieval of a chunk depends on whether associations to it can be found. The importance of the necessary links between chunks in Long-Term Memory is the reason why the predominant code type is

Abbildung in dieser Leseprobe nicht enthalten

3.3.4 Cognitive Processor

On each cycle, the contents of Working Memory initiate linked actions in the Long- Term Memory (recognize), which in turn modify the contents of the Working Memory (act), starting the next cycle. Organized behaviour, such as plans or procedures, are built out of a set of recognize-act cycles by the Working Memory and the Long Term Memory. Still, the Cognitive Processor has a cycle time similar to those of the other processors:¹⁶

Abbildung in dieser Leseprobe nicht enthalten

Table 3.3: Parameters of the Cognition System

3.4 Movement - Fitt’s Law

Moving the hand towards a target can be understood, and an expression for move- ment time can be derived, using the Model Human Processor plus some assumptions. Suppose a person wishes to move his hand to a target of width S at distance D. The movement of the hand is not continuous, but consists of a series of microcorrections, each with a certain accuracy. Making a correction takes at minimum one cycle of the Perceptual Processor to observe the hand, one cycle of the Cognitive Processor to de- cide on the correction, and one cycle of the Motor Processor to perform the correction:

Abbildung in dieser Leseprobe nicht enthalten

The time to move the hand to the target is then the time needed to perform n of the these corrections:

Abbildung in dieser Leseprobe nicht enthalten

Let Xi be the distance remaining to the target after the i th corrective move and

Abbildung in dieser Leseprobe nicht enthalten

be the starting point. Assume that the relative accuracy of movement is constant, i.e., that

Abbildung in dieser Leseprobe nicht enthalten

is the constant error. On the first cycle the hand moves to

Abbildung in dieser Leseprobe nicht enthalten

On the second cycle, the hand moves to

Abbildung in dieser Leseprobe nicht enthalten

The hand stops moving when it is within the target area, i.e. when 1

Abbildung in dieser Leseprobe nicht enthalten

Hence the total movement time Tpos is given by

Abbildung in dieser Leseprobe nicht enthalten

Equation 3.7 is called Fitt’s Law. It states that the time needed to move a hand to a target depends only on the relative precision required, i.e., the ratio between the target’s distance and its size.¹⁷

The constant ε has been found to be about 0.07, so IM can be evaluated:

Abbildung in dieser Leseprobe nicht enthalten

Several methods have been used to measure the correction time. Based on several experiments the value is set to¹⁸¹⁹

Abbildung in dieser Leseprobe nicht enthalten

The problem of the points off the line for low values of log 2 /D S andtheslightcurvature can be resolved adopting a variant of Fitt’s Law:

Abbildung in dieser Leseprobe nicht enthalten

Chapter 4 Model

4.1 Structure

The MHP itself only provides the times a human being needs to process certain information. It can not be used to calculate the time needed to perform a whole procedure directly. Thus, in order to be able to use the MHP for the calculation of workload, a tree structure represented in figure 4.1 is used. This is beneficial for a better lucidity as well as for the later programming.

Abbildung in dieser Leseprobe nicht enthalten

Figure 4.1: Model structure tree

The respective procedure represents the root node (first level) of the tree. Pilot Flying and Pilot Not Flying are the nodes forming the second level. Tasks 1-3 are the nodes building the third level while the actions are the leaves of the tree.¹

4.1.1 Definitions

In order to elaborate an unambiguous and understandable procedure model, the terms used to describe it need to be clear. This is particularly important since all the words are known and some of them are used almost synonymously in everyday life. Therefore, in the following the most important terms of the developed model will be defined and explained.

Action

The term action or action time is used as a collective term for the parts of the MHP as described in chapter 3 (namely eye movement, perception, cognition, motor and including movement according to Fitt’s Law, section 3.4) plus communication as a factor that is not considered in the original Model Human Processor, but does play a major role for the operation of an aircraft.

In figures 4.1 and 4.2 actions are represented as yellow circles on the bottom level of the tree structure since they embody the basis of the model.

Task

Tasks comprise several actions as can be seen by the arrows connecting the action with the task level in figure 4.1. Tasks may but need not include all six actions.

Entity

The model implements two entities: Pilot Flying and Pilot Not Flying, both defined on the following page. This nested structure is necessary since part of this work’s goal is to compare the pilot’s workloads for different team compositions. This way the model yields results for tasks performed by the Pilot Flying and the Pilot Not Flying, respectively.

Procedure

Procedures integrate one or several tasks performed by one or both entities. In figure 4.1 this is shown by the fact that all arrows add up to one procedure after connecting on the task and entity levels.

Example Figure 4.2 shows the exemplary procedure ”FLAPS 1”. The goal is to move the respective selector to the required position. Since the procedure is part of a check list routine, the Pilot Not Flying, who is reading the list, gives the order. This task called ”Ordering FLAPS 1” involves cognition, motor and communication.

Abbildung in dieser Leseprobe nicht enthalten

Figure 4.2: Exemplary procedure tree

The Pilot Flying receives the order, moves his eyes to focus the selector, perceives and processes the image and moves his hand to select ”FLAPS 1” position. Thus, this task called ”Selecting FLAPS 1” involves eye movement, perception, cognition, motor and movement actions. Finally, the Pilot Flying confirms the completion of the task. This again requires cognition, motor and communication. The figure shows how the ordering is associated with the Pilot Flying as the performing entity while the selection and confirmation is done by the Pilot Not Flying. On the top of the scheme everything adds up to one procedure.

Flight Phase

A flight can be separated into several segments consisting of different procedures per- formed by the pilot. These segments are called flight phases. Because of the limited time frame of this work it was not possible to account for every procedure performed from cockpit preparation to securing the aircraft. Instead, exemplary especially crit- ical flight phases such as Before Takeoff, Takeoff, instrument landing system (ILS) Approach, Engine Failure and Engine relight have been chosen for this study.

Workload

The quantitative workload as measured by the amount of working time assigned to a pilot. To assess the qualitative workload parameters like pupil diameter, heart rate -r brain activity are needed.² Since these parameters can not be calculated using the chosen model and because the difficulty of work highly depends on the individual’s perception, the qualitative workload is not taken into consideration in this study.

Pilot Flying

Commercial aircraft like those of the narrow-body, short- to medium range Airbus A320 family considered for the model are operated by a flight crew consisting of the Captain, who is the pilot designated as the Pilot-In-Command (PIC) as well as the highest ranking member of the flight crew, and the First Officer (FO, also called copilot). Normally the FO is seated to the right of the captain. Before commencement of each flight sector, the Captain decides which pilot will take direct responsibility for flying the aircraft for the complete flight or for particular parts of it such as the approach or landing. This pilot is called the Pilot Flying (PF).⁴

Pilot Not Flying

The other pilot is then the designated for that part of the flight as the Pilot Not Flying (PNF) who is responsible for monitoring the flight management and aircraft control actions of the PF and carrying out support duties such as communications and check- list reading.

4.2 Model design

4.2.1 Assumptions

Due to the enormous range of different flight scenarios and conditions, the complexity of the operation of an aircraft as well as the limited applicability of the Model Human Processor, the following simplifying assumptions have to be made.

Normal flight conditions

The flight takes place under perfect conditions, i.e. it is not influenced by bad weather or similar circumstances.

Expert skills

Both pilots have excellent aircraft operating skills. Human failure is ruled out.

No additional communication

The pilots communicate only to co-ordinate their actions, to perform a task or for navigation and air traffic control (ATC) contact.

FCOM sufficiency

The procedures included in the FCOM represent the flight completely.

Autopilot used

Pilots use the autopilot whenever possible instead of flying the aircraft manually using the sidesticks.

MHP approximation

The model based on the MHP can approximate the pilots’ actions in cockpit. The operation of the aircraft is a case of human-computer interaction.

Discrete procedures and tasks

Continuous procedures such as monitoring the environment are not taken into consid- eration because these last as long as the flight itself. Thus they can not be represented using the MHP. Also continuous procedures need not necessarily be regarded for the comparison of the workload because the respective times are constant irrespective of the cockpit occupation.

No simultaneousness

Procedures and / or tasks are not performed simultaneously. Otherwise the model would require a much more sophisticated approach as well as many more assumptions when it comes to deciding which procedures or tasks can be (and actually are) performed simultaneously.

Cockpit occupation

The workload is simulated and compared for the case of one and two pilots operating the aircraft, respectively. The results are used to outline possible assistance systems. For each of the three cases assumptions have to be made:

Two pilots: The aircraft is operated by two equally skilled pilots. The designation of the PF and PNF does not change during the respective exemplary flight sectors.

[...]

---

¹ Stringman 1984

² ICAO 2011

¹ Federal Aviation Administration 2012

² Wickens 1992

³ S. Miller 2001

⁴ Song et al. 2011

⁵ Card, Moran, and Newell 1983

⁶ Jastrzembski and Charness 2007

⁷ Liu, Feyen, and Tsimhoni 2004

⁸ Graham et al. 2014

⁹ Ebermann and Scheiderer 2013

¹⁰ Diken 2011

¹¹ Dhillon 2007

¹² Albert 2014

¹³ Collinson 2011

¹⁴ Airbus 1998b

¹⁵ Airbus 2015

¹ Jastrzembski and Charness 2007, page 226

² Card, Moran, and Newell 1983, pages 24-25

³ Russo 1978, pages 89-109

⁴ Busswell 1922, page 31

⁵ Card, Moran, and Newell 1983, page 28

⁶ Averbach and Coriell 1961, pages 309-328

⁷ Darwin, Turvey, and Crowder 1972, pages 255-267

⁸ Sperling 1963, page 22

⁹ Darwin, Turvey, and Crowder 1972, pages 259-262

¹⁰ Harter 1967, pages 47-58

¹¹ Fox and Stansfield 1964, pages 317-320

¹² Simon 1974, pages 482-488

¹³ Murdock 1961, page 280

¹⁴ Tulving and Colatla 1970, pages 86-98

¹⁵ G. A. Miller 1956, pages 81-97

¹⁶ Requin 1978, pages 89-111

¹⁷ Card, Moran, and Newell 1983, pages 51-55

¹⁸ Pierce and Karlin 1957, pages 497-516

¹⁹ Fitts 1954, pages 381-391

¹ Knuth 1998

² Song et al. 2011

³ Wickens 1992

⁴ SKYbrary 2015