ArticlePDF Available

Abstract and Figures

Manual control cybernetics aims to understand and describe how humans control vehicles and devices using mathematical models of human control dynamics. This “cybernetic approach” enables objective and quantitative comparisons of human behavior, and allows a systematic optimization of human control interfaces and training associated with manual control. Current cybernetics theory is primarily based on technology and analysis methods formalized in the 1960s and has shown to be limited in its capability to capture the full breadth of human cognition and control. This paper reviews the current state-of-the-art in our knowledge of human manual control, points out the main fundamental limitations in cybernetics, and proposes a possible roadmap to advance the theory and its applications. Central in this roadmap will be a shift from the current linear time-invariant modeling approach that is only truly valid for human behavior under tightly controlled and stationary conditions, to methods that facilitate the analysis of adaptive, and possibly time-varying, human behavior in realistic control tasks. Examples of key current developments in the field of cybernetics—human use of preview, predictable discrete maneuvering, skill acquisition and training, time-varying human modeling, and neuromuscular system modeling—that contribute to this shift are presented in this paper. The new foundations for cybernetics that will emerge from these efforts will impact all domains that involve humans in manual and semiautomatic control.
Content may be subject to copyright.
Manual Control Cybernetics:
State-of-the-Art and Current Trends
M. Mulder, D. M. Pool, Member, IEEE, D. A. Abbink, Senior Member, IEEE, E. R. Boer, Member, IEEE,
P. M. T. Zaal, Member, IEEE, F. M. Drop, Student Member, IEEE, K. van der El, Student Member, IEEE, and
M. M. van Paassen, Senior Member, IEEE
Abstract—Manual control cybernetics aims to understand and
describe how humans control vehicles and devices using math-
ematical models of human control dynamics. This ‘cybernetic
approach’ enables objective and quantitative comparisons of
human behavior, and allows a systematic optimization of human
control interfaces and training associated with manual control.
Current cybernetics theory is primarily based on technology
and analysis methods formalized in the 1960s and has shown
to be limited in its capability to capture the full breadth of
human cognition and control. This paper reviews the current
state-of-the-art in our knowledge of human manual control,
points out the main fundamental limitations in cybernetics, and
proposes a possible roadmap to advance the theory and its
applications. Central in this roadmap will be a shift from the
current linear time-invariant modeling approach that is only truly
valid for human behavior under tightly controlled and stationary
conditions, to methods that facilitate the analysis of adaptive, and
possibly time-varying, human behavior in realistic control tasks.
Examples of key current developments in the field of cybernetics
– human use of preview, predictable discrete maneuvering, skill
acquisition and training, time-varying human modeling, and
neuromuscular system modeling – that contribute to this shift
are presented in this paper. The new foundations for cybernetics
that will emerge from these efforts will impact all domains that
involve humans in manual and semi-automatic control.
Index Terms—Manual control, man-machine systems, cyber-
netics, dynamic behavior, modeling
CYBERNETICS is a system-theoretical, model-based ap-
proach to understand and mathematically model how
humans control vehicles and devices [1]–[6]. Most of current
cybernetics theory has been developed in the 1960s – for
1960s technology – and has been applied in aerospace [7]–
[29], automotive [30]–[46], other vehicles [47]–[52], robotics
[53]–[57] and medical applications [58]–[61]. The power of
cybernetics is evident from the seminal crossover model [2]–
[4], which captures the systematic adaptation of the Human
Controller (HC) to the dynamics of the controlled vehicle or
device, to achieve good feedback performance and robustness
which are largely invariant with the controlled system. By
M. Mulder, D. M. Pool, F. M. Drop, K. van der El, and M. M. van Paassen
are with the section Control and Simulation, Faculty of Aerospace Engi-
neering, Delft University of Technology, 2629 HS Delft, The Netherlands.
Corresponding author:
D. A. Abbink and E. R. Boer are with the department BioMechanical
Engineering, Faculty of Mechanical, Maritime and Materials Engineering
(3ME), Delft University of Technology, 2628 CD Delft, The Netherlands
P. M. T. Zaal is with the San Jose State University Research Foundation
performing collaborative research at the Human Systems Integration Division
of NASA Ames Research Center, Moffett Field, CA 94035, USA
revealing such key invariants and providing a means for
predicting manual control performance, classical cybernetics
theory has accelerated many innovations in human-machine
control system design, such as in aerospace [7]–[17].
Despite its many successes, cybernetics theory has also
often been shown to be limited in capturing the full breadth of
human cognition and control. Modern interface technologies,
such as three-dimensional visual displays [6], [62], [63] and
haptic (shared) control manipulators [43]–[46], [55], [64] are
rapidly expanding the way humans can interact with dynamic
systems. They also dramatically expand the factors that drive
human control adaptation. It is safe to say that, despite
haphazard attempts to update cybernetics theory, the progress
in technology has leapfrogged the classic cybernetics theory,
and our current tools and models fail to completely explain
and predict how humans interact with modern interfaces.
State-of-the-art cybernetics theory describes human con-
trollers as (quasi-)linear, time-invariant (LTI) feedback systems
[2]–[4], [65]–[77]. The most successfully applied models are
those that consider human behavior in the highly-constrained
compensatory tracking task [2], [69], without any preview of
future task constraints, allowing the operator only to react on
what happens (pure feedback). The time-invariance assump-
tion prevents us from modeling what is a defining attribute of
human controllers, namely their ability to adapt to changing
situations, which, in the age of increasing automation, is often
the only reason humans are kept in the control loop. The
same theoretical constraints that prevent us from studying and
understanding human learning, adaptation and the versatile set
of anticipatory feedforward control behaviors, also prevent us
to optimize current-day control interfaces in realistic tasks.
But this lack of understanding of realistic HC behavior is
not our only problem, as also our methodology and tools to
identify human manual control are limited to rather crude
experimental techniques. We can only identify the overall,
lumped response of a fully-trained human, based on pro-
longed measurements [69], [78]–[84]. This approach fuses all
cognitive and physiological adaptations and averages-out all
adaptation effects, preventing us from understanding design-
relevant aspects of human adaptation and learning.
In the past decade, we have come to the conclusion that the
intertwined theoretical and methodological limitations of the
state-of-the-art in cybernetics theory have become a limiting
factor in evaluating and improving our manual control inter-
faces. The inability to step-up from classical compensatory
control to more relevant real-life tasks means that we are
Published as:
Max Mulder, Daan M. Pool, David A. Abbink, Erwin R. Boer, Peter M. T. Zaal, Frank M. Drop, Kasper van der
El and Marinus M. van Paassen, Manual Control Cybernetics: State-of-the-Art and Current Trends (2017), in:
IEEE Transactions on Human-Machine Systems. doi:
currently able to model only the exception in manual control,
and not the rule. We see a striking similarity to the domain of
human visual perception, where in the 1950s the psychologist
James J. Gibson came to the conclusion that “the theory
of visual perception is all wrong” [85]. The theory and
experiments at that time studied visual perception performance
mainly through forcing human subjects to look at static scenes,
from a fixed position. Gibson was the first to conclude that
staring at static pictures is an exceptional case, as human
visual perception is all about dynamically perceiving (and
acting upon) the dynamics of environments, leading to the now
overarching ecological perspective on visual perception [86].
We strongly believe that cybernetics theory should step
up from studying merely the exception in manual control –
compensatory behavior – to the rule. Relevant control tasks
have preview of the future constraints and in many cases not
only allow, but actually require human adaptation. A targeted
research effort is needed to radically advance our theory, our
models, our tools. We must address some fundamental research
questions on human manual control. Examples are: 1) How
do humans use preview of future task constraints? 2) What
are the factors and mechanisms that drive adaptation, and
which invariants in adaptation exist? 3) To what extent are
measured human adaptations caused by physiological (e.g.,
neuromuscular) rather than cognitive adaptations? 4) What
are the temporal scales of human adaptation and learning in
changing situations? 5) What novel control theories and system
identification techniques exist that could allow us to study
time-varying and possibly non-linear manual control?
With this paper we aim to provide an overview of the field
of manual control cybernetics, elaborate on its fundamental
problems, provide a way forward, and show some of the
latest results in extending theory and applications. The paper
is structured as follows. In Section II we attempt to briefly
summarize the state-of-the-art; for some earlier summaries one
is referred to [2]–[4], [74], [76]. A roadmap to systematically
address the fundamental challenges is provided in Section III.
A number of key theoretical and methodological innovations
that follow from this roadmap will be discussed in Section IV.
Three novel applications of cybernetics theory, in haptic feed-
back design, multi-modal simulator fidelity evaluations, and
transfer-of-training studies are summarized in Section V. The
paper will end with a conclusions section.
The paper’s scope is intentionally kept limited, by mainly
focusing on classical control-theoretic frequency-domain ap-
proaches to modeling human control dynamics, and only
occasionally referring to other modeling perspectives that have
emerged in the past decades, such as those originating from
optimal, robust or satisfying control theory, and time-domain
analysis. In our experience, it is mostly this first class of
physical models that has prevailed, also because – the perhaps
in principle more generic and certainly intellectually appealing
optimal human control models [87]–[89] have shown to be
over-parameterized [90] meaning that they cannot be validated
experimentally. Further note that, in our discussion of inno-
vations and applications, we focus primarily on the ongoing
activities in our labs, as modernizing cybernetics theory is one
of our key objectives.
A. Successive Organization of Perception
In 1960, Krendel and McRuer [68] first introduced the Suc-
cessive Organization of Perception (SOP) hierarchy for human
manual control. The SOP postulates a framework describing
the development of skill-based manual control behavior, in
three stages: compensatory, pursuit and precognitive control,
see Fig. 1 for schematic representations. Depending on the
defining features of the control task, such as the display format
and the applied forcing functions, and training, human opera-
tors may apply compensatory, pursuit, or precognitive control
strategies, or could be switching between any combination of
these levels [68], [71]. The next subsections discuss the three
SOP levels in more detail.
Human controller
(a) Initial phase: single-loop compensatory behavior
Human controller
++u x
(b) Second phase: multi-loop pursuit behavior
Human controller
(c) Final phase: open-loop precognitive behavior
Fig. 1. Schematic representations of the three stages of control behavior
described in the Successive Organization of Perception (SOP), initially de-
scribed in [68], later adapted in [71]. These figures are reproduced, with
minor modifications, from [71].
B. Compensatory Tracking
In the compensatory stage, see Fig. 1(a), the human con-
troller (HC) acts solely on the error ebetween the reference
and the system output [67]. The HC responds only to the error,
either because it is the only perceivable signal, or because the
HC chooses to act on the error only. Compensatory control
has been studied extensively for control tasks were the HC
could only perceive the error, and all forcing functions were
unpredictable [2], [52], [65], [69], [79], [80]. Reasons for
retaining a compensatory organization in situations where
more signals can be perceived are: 1) a lack of experience, the
HC has not yet learned sufficiently to progress to the pursuit or
precognitive stages, 2) the HC is under stress, causing him/her
to ‘revert’ to a compensatory organization, or 3) a pursuit
Displays Controlled
Perceived Inputs,
Outputs and
Actions Outputs
Motion Feedbacks
In-Flight vs. Fixed-Base
Atmospheric Conditions
Experimental Design
Order of Presentation
Pilot Manipulator
Fig. 2. The variables that affect a closed-loop human controlled system, reproduced from [2].
or precognitive organization is not beneficial for improving
performance [68], [75].
Pioneering research into human tracking behavior by Tustin
[65] and Elkind [66], [91] led to a comprehensive framework
for the analysis and modeling of compensatory control be-
havior in the 1960s [2]–[4], [67]–[69]. Much of our current
knowledge stems from these investigations into human dy-
namics during single-loop compensatory tracking tasks [2],
[66], [69]. This research also showed the complexity of
studying the human controller, due to her or his capacity to
adapt to a myriad of task, environmental, operator-centered,
and procedural variables, as summarized in a comprehensive
overview compiled by McRuer and Jex [2], see Fig. 2.
For the most basic SOP level of compensatory control, the
well-known crossover model given by (1), in combination
with the verbal adjustment rules of [2], accurately describes
a crucial invariant of HC behavior in the systematic HC
adaptation to some of the key task variables: the controlled
system dynamics (Hc) and the bandwidth of the applied
forcing function spectra:
Hp()Hc() = ωc
ej ωτe(1)
To induce a compensatory control organization, and thus
force the HC into a mode where she or he cannot anticipate
on what comes next, the applied forcing functions must
be random-appearing [2], [66], [67], [69]. Typically, this is
achieved by using quasi-random multisine signals, sums of
a sufficient number of individual sinusoids that span the
frequency range of interest [69], [92], [93]. Not only do such
multisine forcing functions force compensatory control, they
also facilitate the straightforward identification of frequency-
domain describing functions of human dynamics in compen-
satory tracking tasks [69], [78], [80]. Using the quasi-linear
model assumption, the linear, time-invariant (LTI) part of the
HC can then be modeled. The remainder, called ‘remnant’, is
usually neglected, despite attempts to provide some rationale
for the remnant component as well [2], [94], [95].
Even though a number of different LTI models for compen-
satory HC dynamics have been proposed [15], [21], perhaps
the most-used is the precision model, which is given by (2)
in a form that, compared to its definition in [69], omits the
indifference threshold describing function.
Hp() =
z}| {
KpTL+ 1
TI+ 1
low-freq. lag-lead
z}| {
TK+ 1
K+ 1
z}| {
ejωτ ×
(TN+ 1) h
ωnm i2
+2ζnmj ω
ωnm + 1
|{z }
neuromuscular dynamics, Hnm
In this model, the main adaptation of the HC dynamics Hp
to the dynamics of the controlled system Hcis captured by
the equalization term of the model. Depending on what type
of equalization is required to satisfy (1) for a given Hc, the
lead-lag equalization form of (2) may reduce to a pure lead, a
pure lag, or a pure gain [69]. Furthermore, the precision model
includes an additional low-frequency lag-lead term, for cap-
turing low-frequency phase equalization found in describing
function data [2], [69]. Finally, the model includes terms that
account for characteristic HC limitations in a delay term ejωτ
and the neuromuscular actuation dynamics. In more recent
applications of the precision model, the low-frequency lag-lead
is often omitted and neuromuscular dynamics are simplified
to the second-order term only [12], [25], [27], [92], while
extended equalization was proposed for control of systems
with underdamped modes [27].
Theories and models for compensatory tracking have been
extended to multi- or multiple loop control tasks. Here, a
distinction is often made between control of 1) multiple
nested loops (e.g., aircraft pitch and altitude) [2], [32], [71],
[73], [96], 2) multiple (coupled) parallel loops (e.g., aircraft
pitch and roll) [2], [70], [73], [77], [79], [97]–[99], and
3) a single-loop task with a single controlled variable, but
with a multi-loop HC feedback organization (e.g., multimodal
visual/vestibular feedback) [21], [25], [100]. Multi-loop sce-
narios typically result in elaborate and often over-determined
HC models, requiring extended identification and modeling
methods to separate the different HC responses [79], [81]–
[83]. It is safe to say that current-day cybernetics theory and
methods, predominantly deal with single-loop compensatory
tracking. Only for this extremely simple task do we have
accepted, universal models, such as the crossover and precision
models, that allow us to predict how a (well-trained) HC adapts
to task variable settings.
C. Pursuit Tracking
In the pursuit stage, see Fig. 1(b), the HC utilizes a
combination of at least two of the following control strategies:
1) a feedforward response (Hpt) on the target ft[101], [102],
2) a compensatory feedback response (Hpe) on the error eas
in compensatory tracking [101], and 3) a feedback response
(Hpx) on the system output x[101]–[103]. The theoretically
optimal pure feedforward control law approximates the inverse
system dynamics, i.e., Hpt1/Hc[101], [104], while
feedback of the system output xis useful for mechanizing a
stabilizing “inner” control loop, mostly for tasks with sluggish
system dynamics [101]. Key is, however, that a “pursuit”
organization of HC behavior is not adopted in all tasks where
the feedbacks to support it are available [75], [101], [105]. The
opposite holds as well: a pursuit (or even precognitive, see Sec-
tion II-D) strategy may be developed even in a compensatory
tracking task – so, if no additional feedbacks are available –
for example when forcing functions are predictable.
Many studies report improved task performance when HCs
reach the pursuit stage [101], [105]–[109]. As proposed in
[101], the underlying change in HC control behavior can be
detected from the ‘effective open-loop describing function
– i.e., the describing function from the tracking error eto
the system output x– which shows strongly reduced low-
frequency phase lag in pursuit [75], [101]. While helpful for
detection, the effective open-loop lumps together all control
dynamics and thus does neither reveal the true adopted HC
control organization, nor the separate contribution of each
feedforward or feedback response (e.g., Hpt,Hpe, and Hpx).
Compared to the modeling of HC behavior in compensatory
tasks, pursuit tracking tasks have received meager attention
[75], [101], [102]. The main reason is that the multi-loop
control behavior in pursuit (see Fig. 1(b)), makes its modeling
significantly more complicated [75], [101]. In pursuit tasks,
HCs may choose to mechanize feedforward and/or feedback
control responses driven by the ft,e, and xsignals, see
Fig. 1(b). However, as e=ftx, only two of the three pos-
sible HC responses are independent, resulting in an inherently
overdetermined model structure. For modeling pursuit control,
model structures that include Hptand Hpe[2], [101], [102],
[110], Hpeand Hpx, or Hptand Hpx[111] have all been
proposed and applied. Furthermore, from an identification
perspective, the pursuit task requires two independent forcing
function signals (ftand fdin Fig. 1(b)) to separately estimate
the two independent describing functions [79], [81] and model
both using LTI model structures. Up until quite recently, this
has almost never been tried [104], [111].
What is stated for pursuit control, is even more true when
the HC has preview on the future task constraints, like the
future trajectory of the target signal [112]. With the direct
capacity for overcoming inherent HC control delays, preview
almost invariably results in improved task performance far
exceeding that of pursuit [80], [89], [113], [114]. In essence,
in tasks with preview, HCs adopt a pursuit control organization
with a strong feedforward Hptresponse driven by the future
target signal. From sampling and cueing theories, it is known
that HCs become almost optimal samplers with preview [89],
and that HCs’ internal representation [115] of task variables
greatly improves. The human response to preview is a con-
volved and very likely time-varying weighing of this future
information [112], which cannot be directly measured, as an
infinite number of different weighing mechanisms theoretically
yield the same control response. Even more than for pure
pursuit, the difficulty for preview control lies in the fact that,
when preview information becomes available, a multitude of
control strategies become possible [111], [112].
When considering realistic manual control tasks, it is diffi-
cult to think of tasks that better resemble pure compensatory
tracking than pursuit or preview tracking. Therefore, there is a
strong need for universal models for HC pursuit and preview
control, similar to those that are available for compensatory
control. Given the increased degrees-of-freedom in HC adapta-
tion, developing such universal models and sets of “adjustment
rules” for pursuit and preview control is extremely challenging.
Still, a firm grasp of how humans control in pursuit or preview
is one of the main crucial elements that is missing in the
current cybernetics state-of-the-art.
D. Precognitive Control
In the precognitive stage, see Fig. 1(c), the HC is assumed to
have complete knowledge of the target signal and to generate
a control input that results in perfect target tracking [72],
[108], [116]. In the precognitive phase, HCs may develop
purely open-loop control responses based on a fully-developed
internal representation of task demands, such as dominant
frequency components and the controlled system dynamics
[68], [71]. The HC does not actively rely on any feedback,
at least not for a particular time interval [71].
When the SOP was postulated, the hypothesized precogni-
tive level was not yet fully supported, mainly because direct
identification of human feedforward responses was lacking
[68]. Still, a broad collection of empirical observations and
recent data support the SOP’s precognitive phase. For ex-
ample, numerous studies report notably improved tracking
performance when following ‘predictable’ target target sig-
nals, in comparison with ‘unpredictable’ signals [72], [106],
[108], [117]–[120], even for signals with equivalent frequency
content and bandwidth. Further evidence for the development
of a precognitive control mode has been found with observed
response time delays and phase lags that are smaller than a
‘normal’ human reaction time [108], [121]. Finally, studies
involving temporal occlusion [116], where HCs tracked a sum-
of-two-sines target signal for a certain time, after which the
display was switched off, also report reasonably accurate con-
tinued tracking of predictable, repetitive signals only. Though
lacking a formal definition of subjective predictability and
empirical evidence for its limits, these observations provide
indirect evidence for a precognitive strategy.
To conclude, ample indirect and mostly qualitative evidence
of a precognitive level of manual control exists. Still, in
most cases the evidence is thin, with possible alternative
explanations that do not require the existence of a true
precognitive control strategy (e.g., strongly adapted feedback
control). Except for a rudimentary understanding and proposed
cybernetic models for the (partially) precognitive feedforward
control in ramp tracking tasks [104], [122], [123], we still lack
a structured, systematic understanding of the final level of the
SOP, as would be relevant to real world applications.
E. Neuromuscular Dynamics
Cybernetic theory emphasizes how control inputs to the
plant result from visual and vestibular cues. But McRuer [124]
already stated in 1966 that neuromuscular actuation properties
are “an essential element in the operators dynamic character-
istics”. He recognized that the neuromuscular system (NMS)
constitutes an inner loop that not only translates desired control
inputs to realized control inputs, but that can also provide very
fast reflexive feedback to forces on the control device, even
instantaneous responses from (co)contracted muscles and pas-
sive limb dynamics. Although subsequent work also took into
account this ability of the neuromuscular system to provide
force feedback [2], [87], [124], [125], this detail has been
neglected in later cybernetic studies. Often, the neuromuscular
system is viewed as a physical limitation, to account for the
fact that physical properties of our body coupled to the control
interface inherently limit the bandwidth of HC control inputs.
This limitation shows up in HC describing functions as a
distinctive peak around 2-4 Hz, with an ensuing decay.
Mathematical models of NMS dynamics were developed in
parallel to HC models [2], [87], [124], [125], see also (2). In
HC models, the combined manipulator and NMS dynamics
are typically accounted for with a single, lumped, low-order
model; generally an underdamped second- or third-order low-
pass transfer function [2], [124]. That no separate gain is
modeled, indicates the assumption that the NMS is fully
adapted to the control device dynamics [19], [126], [127], and
also avoids over-parameterization. The estimated parameters
of the cut-off filter have been shown to vary as a function of
manipulator characteristics [127], [128], the controlled system
dynamics [111], [123], and the presence of motion feedback
[25]–[27]. Simplification of the neuromuscular system as a
‘physical limitation’ described by a filter is valid for ap-
plications where the operator controls a system where the
control device receives no force feedback about plant states.
Such applications include fly-by-wire aircraft, rate-controlled
systems, or uni-directional telemanipulation.
For other control tasks, force feedback on the control
interface is essential for human operator performance. For
instance, during driving, forces and movements at the tires
are physically coupled to forces and movements at the steering
wheel, allowing the neuromuscular system to respond to force
perturbations from wind gusts or road properties – before these
perturbations change vehicle states enough to be observable by
visual or vestibular cues. The neuromuscular system then acts
as an inner-loop, responding to forces very quickly (through
reflexive feedback) or even instantaneously (through limb iner-
tia and visco-elasticity of co-contracted muscles). Frequency
response functions (FRFs) of the NMS can be estimated as
admittance”, a measure of the allowed limb displacement due
to an applied force [129]. HCs can adapt the admittance of
their NMS – i.e., how “stiff” or “compliant” their response to
forces is – which affects control performance in car driving
[40], [45], aircraft control [24], and the impact of biodynamic
effects in moving environments [130], [131].
Proposed models to describe NMS contributions to operator
control dynamics are based on theory about muscle and
arm dynamics [19]. Functional mathematical models typically
describe overall endpoint admittance by separating manipu-
lator dynamics from neuromuscular dynamics, which com-
prise passive limb dynamics (inertia, visco-elastic properties
of ligaments and (co)contracted muscles), reflex dynamics
(position and velocity feedback from muscle spindles and
force feedback from Golgi tendon organs) as well as their
interaction through cognitive processing [24], [132].
Clearly the NMS can increase or decrease admittance
through many mechanisms, whose interactions are complex
to determine. This means that NMS model structures are per
definition over-determined, making the parameters difficult to
extract from physical measurements. Regardless, the NMS
needs to be taken into account, to avoid attributing its con-
tribution to visual or vestibular control activity.
The cybernetics overview of the previous section clearly
showed that our knowledge and methods mostly cover highly-
constrained tasks – mainly compensatory tracking – that are
quite far from typical real-world manual control scenarios. In
this section we propose a five-step framework [84] to increase
our understanding of the learning and adaptive human con-
troller, see Fig. 3. It consists of five “steps”, each describing a
major extension of our knowledge of human control, that will
take the field of manual control cybernetics from its current
state-of-the-art (shown with the gray shaded area in Fig. 3) to
the level required for applications to real-world optimization
of human control interfaces and training.
Central in the framework is the concept of Internal Rep-
resentation (IR) [115] that, as shown with the purple blocks
in Fig. 3, is developed and refined during learning, when the
HC is exposed to the task constraints. For manual control,
primarily the task variables of Fig. 2 characterize the task,
especially key task variables such as the plant dynamics (P)
and the statistical properties of the target and disturbance
signals (T and D). Our premise is that it is the IR, the
quality of which increases with exposure and experience, that
is the critical driver behind human control adaptations. The IR
enables HCs to evolve through the different phases of the SOP
and thereby develop an optimal combination of feedforward
(FF) and feedback (FB) control to satisfy task constraints.
The following subsections describe the different fundamen-
tal steps of the proposed framework of Fig. 3 in more detail.
Fig. 3. A proposed framework for understanding the learning and adaptive human controller, reproduced from [84].
A. Steps 1 + 2: Understanding Pursuit and Preview
The first two steps to update our theory, see Fig. 3, focus on
developing validated and practical models and analysis meth-
ods for HC control at the pursuit level (Step 1), as well as for
human preview control (Step 2). While often seen as separate
levels of HC behavior, pursuit can be viewed as an extreme
(zero preview) case of preview control. Furthermore, both
pursuit and preview are characterized by a strong feedforward
component [75], [111], [112]. Hence, in our view, Steps 1 and
2 will be studied in unison. Similar to the crossover model for
compensatory tracking [2], [68], there is a need for a universal
model for pursuit and preview control, with an extensive set –
in fact a much more extensive set given the additional degrees-
of-freedom in HC adaptation – of adjustment rules for the key
HC control responses and parameters.
Developing this added understanding and modeling
“toolkit” will require a significant amount of new experimental
HC data, where human control is measured with a wide
variation in critical task variables, such as plant dynamics (e.g.,
linear vs. non-linear), target and disturbance signal properties
(e.g., spectrum, stochastic properties, predictability), and dis-
play and preview settings. Experiments can be preceded by
a theoretical analysis and computer simulations, e.g., through
assuming optimal control [87], [89], to explore the parameter
sensitivities and theoretically optimal information-weighing
strategies for human control in pursuit and preview tasks.
Steps 1 and 2 are required to ensure the applicability of
cybernetic models for the design of manual control interfaces
to support HCs in realistic, real-life control tasks, where our
current lack of understanding of how HCs actually control
leads to sub-optimal support systems. For example, this is
evident in the haptic shared control systems [44] that are
currently being developed to support car drivers, whose control
is strongly based on both visual preview of the road ahead and
a neuromuscular response to the guidance forces [43], [45].
B. Step 3: Isolating Neuromuscular Adaptations
The study of human control dynamics relies heavily on
their identification from measurements of HC “inputs” and
“outputs”, inherently resulting in a “lumped” insight into all
effects of HC adaptation to various task variables. Isolating
NMS contributions from the lumped adaptive HC data is
essential to lift the “blurring” effects of different parallel
modes of HC behavior. Also during learning, the HC dynamics
change not only due to “higher-level” cognitive adaptations, as
described in the SOP, but also due to “lower-level” underlying
physical adaptations in the neuromuscular system [133].
Motor control literature has shown the synergy between im-
proving internal models for limb movement and accompanying
reduction in co-contraction of relevant muscles [134], which
may also occur during driving: during repeated lane-changes
performance increases, while muscle co-contraction reduces
[39]. Hence, to understand the learning and adaptive nature of
HCs, we need to study the synergy between low-level NMS
adaptations and higher-level learning, see Fig. 3. That is: how
is the IR learned, and how does it drive adaptation of the HC’s
feedback, feedforward, and NMS dynamics?
This requires better understanding of the (time-varying)
nature of NMS adaptations in manual control and which NMS
parameters change the most, both captured in models of the
adapting HC dynamics – at “higher-level” and “lower-level”.
Essential here is to improve our measurement techniques, to
obtain more accurate and less intrusive estimates of the time-
varying NMS settings, for instance by taking additional non-
intrusive grip force measurements that are often related to
NMS admittance settings [135].
C. Step 4: Understanding Learning
Although closely related, we distinguish between learning
(Step 4) and adaptation (Step 5) as follows. Learning involves
how the novice human controller matures, for a fixed set of
task variables, to an expert controller, establishing the best
compromise between control effort and control performance.
Adaptation is seen as the process where a HC, proficient in
the manual control of the whole set of task variables under
investigation, switches from one control strategy to another
when one (or more) of the task variables change. Generally
speaking, training a learning HC to full proficiency is a
comparatively slow process when compared to the often rapid
HC adaptation response to a change in task variables.
An understanding of learning of the human controller can,
in our view, best be gained from investigating how the HC’s
internal representation (IR) of the task develops over time.
Fig. 3 illustrates that the IR evolves during learning, perhaps
even from scratch with novice controllers. The IR is used
by the brain to adapt the feedback and feedforward control
mechanisms and NMS dynamics (the purple parts in this
figure) to balance control effort and performance.
Where the majority of our current knowledge of cybernetics
is based on the control behavior of well-trained subjects
under steady-state task conditions, elucidating human control
learning requires a completely different approach: monitoring
the progress during the full learning curve, observing novice
HCs become expert controllers. This requires dedicated experi-
ments, which explicitly focus on training HCs, covering a wide
variety of constant tasks and task variables. This gives insight
into how IRs evolve in relation to specific task characteristics
and how HCs develop proficient control skills to deal with
combinations of different task variables.
Such data would facilitate “probing” the quality of the
evolving IR, to observe the extent to which novice controllers,
while gaining experience, develop an accurate IR of the
task constraints, to become experts. Of special interest for
understanding HCs’ learning are the possible limitations in
the evolving IR and especially the temporal scale of learning
for different key task variable combinations. The capability
to peek into the what is currently a “black box” of human
learning, and quantify the dynamics of experience, may have
great impact in all domains where humans are trained to
manually control dynamic systems.
D. Step 5: Understanding Adaptation
When task variables – which represent “situations” from a
control-theoretical perspective – change during manual con-
trol, proficient human controllers may detect these changes
because their expectation obtained from the IR (see Fig. 3)
does not match their observation. The plant will respond to
the control commands in a different way than expected, with
the expectation driven by the IR, resulting in an innovation
(the large iin Fig. 3). This mismatch then triggers cognitive
adaptations in the HC’s feedback and feedforward control
dynamics, as well as physiological changes in the NMS, as
indicated with the purple parts of Fig. 3.
When studying human control adaptation, intriguing ques-
tions include what external factors drive the IR adaptation,
to what extent do controllers detect these changes, and how
exactly the IR in turn drives the various adaptive parts of
human control behavior. Hence, we need experiments that
include systematic explicit time variations in task settings, to
gain full insight into whether, to what extent, and how fast,
HCs and their IRs adapt to such changes. Of special interest
would be ‘hysteresis’-effects that may occur when humans
adapt, back and forth, to varying task parameters.
Steps 4 and 5 both entail the development of a com-
pletely new theoretical framework for cybernetics, within
which human adaptive control capabilities can be interpreted
and predicted. A truly focused analysis of adaptive human
control not only requires focused experimentation, but also
significant methodological advances. Most notably, we need an
ability to explicitly capture the time-varying nature of human
controllers, perhaps even in real-time. The main thrust forward
towards understanding HC learning and adaptation would be to
move to intrinsically time-varying manual control identifica-
tion and modeling, for which novel excitation techniques and
test signals – with the lowest possible level of intrusiveness –
are definitely needed, to ensure the most reliable results.
Here we present three examples of current investigations
that contribute to the roadmap discussed above, which all
highlight the combination of theoretical and methodological
advances that is required. Examples include human preview
control, feedforward control with predictable target signals,
and time-varying behavior.
A. Manual Control with Preview
There is a need for a universal model for HC preview
control, together with a set of adjustment rules for HC adapta-
tion in preview tasks. Many HC preview control models have
been proposed (e.g., [20], [30], [31], [33]–[38], [41], [113]),
mostly based upon the pioneering work of Sheridan [112].
None of these preview models has been widely accepted,
mainly because the enormous variation in control organization
HCs can adopt in preview tasks is still poorly understood.
Even in constrained laboratory tracking tasks determining
these characteristics is difficult, as preview information allows
HCs to adopt separate responses to any part of the previewed
target trajectory ahead (Hpt), the controlled element output
10-1 100101
ω, rad/s
|Hpt|, -
10-1 100101
ω, rad/s
6Hpt, deg
10-1 100101
pursuit, non-par.
preview, non-par
ω, rad/s
|Hpx|, -
10-1 100101
pursuit, model
preview, model
ω, rad/s
6Hpx, deg
“near-viewpoint” reponse
equalization physical
feedback response
target response
human controller
Fig. 4. Illustration of the approach to derive the human controller model for preview tracking tasks in [111]. Based on non-parametric estimates (top left) of
the human’s target and feedback response dynamics (top right), the inputs, control organization, and control dynamics of the human controller model were
obtained (bottom).
(Hpx), and the error (Hpe), see Fig. 1(b). Therefore, single-
loop system identification techniques, which enabled the de-
velopment of models for compensatory tasks (see Section II),
no longer suffice. Moreover, it is impossible to independently
identify all three control responses, Hpt,Hpx, and Hpe,
due to the interdependence between the three input signals
(e=ftx) [101].
Recently the HC’s control dynamics in tracking tasks with
preview were estimated non-parametrically with multi-loop
system identification techniques [111]. Conditions included
both zero-preview pursuit tracking tasks, and tasks with 1 s
preview. Only the Hptand Hpxdynamics were estimated,
which are thus contaminated by the HC’s response to the
current error Hpe, if such a response is actually present [111].
Results from [111] are reproduced in Fig. 4. Based on the
non-parametric estimates of the HC dynamics (black and gray
markers in the Bode plots of Fig. 4), separate models for Hpt
and Hpxwere formulated, after which the model was restruc-
tured into the more intuitive form shown at the bottom of
Fig. 4. This model is the first that is based on objective multi-
loop measurements of HC’s input-output relation, without any
a priori assumptions on the HC dynamics.
The novel model provides a new view on preview tracking
behavior. Two distinctly different responses are initiated: a
near-viewpoint response with respect to a point on the target
τns ahead (typically 0.1-0.9 s), and a far-viewpoint response
with the target τfs ahead as input (typically 0.6-2 s). HCs
track low frequencies in the target signal (up to about 6-10
rad/s) predominantly with the far-viewpoint response, which
is a combined feedback/feedforward control mechanism on the
pursuit level of the SOP. The near-viewpoint response – an
open-loop control mechanism – is more effective at higher
frequencies. Note that the far-viewpoint response is the HC’s
main control mechanism in preview tasks, while the near-
viewpoint response is an optional additive response that can be
used to further improve high-frequency target-tracking [114].
The far-viewpoint “filter” provides a pre-shaped input to
an error feedback response, which is equivalent to the error
response in compensatory tracking tasks [2], see (2). However,
instead of responding to the current error e, the error e
in preview tasks is an internal (non-physical), time-advanced
error, based on the difference between the (possibly smoothed
and scaled) far-viewpoint and the controlled element output.
The far-viewpoint response includes a low-pass, or smoothing
filter 1/(1+Tl,f jω), with a bandwidth determined by the time
constant Tl,f (typically 0-1 s), to capture only the target’s
low frequencies. The far-viewpoint gain Kf(typically 0.5-
1.2) reflects how aggressive the HC tracks the target: Kf= 0
indicates that the HC completely ignores the target to focus
purely on stabilizing the controlled element. Note that, when
Kf= 1, and τf=Tl,f = 0 s, the internally calculated error
eequals the actual error e, and the far-viewpoint response
equals the precision model for compensatory tracking [69].
A large benefit of this novel model is that its parameters
have an intuitive physical interpretation, which can 1) provide
unique insights into possible invariants of HC behavior, and 2)
allow for predicting HC behavior. Working towards a universal
model for preview control tasks, current research focuses on
quantifying a set of adjustment rules for preview control,
including HC adaptation to controlled element dynamics [114],
preview time, and target signal characteristics (e.g., band-
width). Furthermore, our knowledge of human use of preview
information must be extended from tracking to more realistic
control tasks with preview, like car driving. Therefore, current
work also investigates how HC preview control is affected by
linear perspective [136] and inner feedback-loop closures due
to the presence of inertial motion and a visual flow field.
B. Feedforward on Predictable Target Signals
In situations where the HC does not have preview informa-
tion on the target, he/she might still have prior information
on the future course of the target through memory or predic-
tion. The HC might operate a feedforward response on the
target, in addition to a closed-loop feedback response, which
allows the HC to improve target-tracking performance without
sacrificing closed-loop stability; a key sign of effective HC
adaptation. In realistic control tasks, the desired trajectory
often has a simple waveform-shape, e.g., constant-velocity
ramp or constant-acceleration parabola segments, making the
target signal predictable and easy to memorize. Although
control responses involving a feedforward were frequently
hypothesized [2], [68], [71], [72] and empirical evidence was
presented [75], [101], [105], [108], [121], they were never
explicitly investigated with system identification and parameter
estimation methods until recently.
Established identification methods, such as the Fourier Co-
efficient method [79], [81], cannot be used with target signals
that have power at all frequencies, such as ramps. Studying
feedforward thus requires novel black-box HC identification
methods, e.g., based on LTI AutoRegressive with eXternal
input (ARX) models [137]. Fig. 5 shows identification results
obtained with the novel ARX-based method of [137] from a
human-in-the-loop tracking experiment featuring target signals
consisting of ramp segments [138]. Black-box identification
results as shown in Fig. 5, provide a means to objectively
detect the presence of feedforward HC control responses. Also,
they reveal the nature of the adopted feedforward control
dynamics, which enables the mathematical modeling of HC
feedforward behavior [104], [123].
Fig. 5(a) shows the estimated feedforward (Hpt) dynamics
for twelve participants who performed a ramp-tracking task,
compared to the theoretically ideal feedforward law, equal
to the inverse system dynamics 1/Hc. The range for which
the ARX identification results are valid, based on the lowest
and highest frequency component in the applied disturbance
signal fd, is indicated with two dashed vertical lines. At low
frequencies, the estimated feedforward dynamics evidently
approximate 1/Hc, except for a slight difference in gain.
For ω > 2rad/s the responses deviate from the theoretical
optimum, flattening as a low-pass filter, with considerable
spread between subjects. For most subjects, the phase response
rapidly becomes more negative, suggestive of a considerable
feedforward delay. For four participants, however, the phase
response is mostly flat or even becomes positive, indicating a
negative time delay and thus anticipation of the future course
of the target. From observations it can be deduced that the
feedforward path Hptof the HC model of Fig. 5 can be
ω, rad/s
Hpt, deg
ω, rad/s
Hpt|, -
(a) ˆ
ω, rad/s
Hpe, deg
ω, rad/s
Hpe|, -
(b) ˆ
HC dynamics
Fig. 5. HC model consisting of a feedforward path Hptand a feedback
path Hpewith estimates of the feedforward and feedback dynamics of twelve
subjects tracking a predictable target signal consisting of ramp segments. Data
are from [138], for Hc(s) = 1/s.
modeled with a gain, inverse system dynamics [101], a low-
pass filter [123], and a time delay:
Hpt(s) = Kpt
(TIts+ 1)2eτpts(3)
As is clear from (3), Hptapproximates the theoretically
optimal feedforward response, 1/Hc, for Kpt1,TIt0s,
and τpt0s. With clearly imperfect feedforward control (see
Fig. 5(a)), Fig. 5(b) shows that the feedback component Hpe
of the combined feedforward-feedback HC model is indeed
required. It can be modeled with a structure identical to well-
known models of compensatory HC behavior [69], [101],
[104], [123], such as the precision model of (2).
A key example of where feedforward HC models provide
increased understanding, is HCs’ sensitivity and adaptation to
predictable target signals [72], e.g., signals that consist of only
one or two sine waves [108], [121]. For instance, in [120]
HCs were asked to track three pairs of “harmonic” (H) and
“non-harmonic” (NH) multisine signals, consisting of 2, 3, or
4 sinusoids with a pursuit display. Analytical analysis with
a (linear) HC model as shown in Fig. 5 predicted identical
tracking performance for such H and NH signals, because
such a prediction is not sensitive to the predictability of the
target. Real HCs, however, performed distinctly better with
the harmonic signals. As shown in Fig. 6, this is explained
by an anticipatory feedforward response that is developed for
these more predictable signals: the feedforward gain Kptis
higher, and the feedforward delay τptis considerably smaller
and close to zero. Such data suggests that the predictability
Number of sines, -
Kpt, -
23 4
(a) Feedforward gain
Number of sines, -
τpt, s
2 3 4
(b) Feedforward delay
Fig. 6. Estimated feedforward HC model parameters from [120] for tracking
of harmonic (H) and non-harmonic (NH) reference signals.
profoundly affects HC behavior, and demonstrates the effec-
tiveness of feedforward HC models for quantifying the under-
lying HC adaptations. With only a severely limited database
of studies that explicitly focus on target signal predictability,
future studies should seek to understand how HC model
parameters change as a function of target signal properties.
In conclusion, the established feedforward HC model en-
ables unique insight into control strategies that involve feed-
forward, such as pursuit and precognitive control. Identifying
and modeling feedforward responses does, however, in itself
not reveal how the feedforward was established – e.g., pursuit,
preview, precognitive – or whether multiple parallel feedfor-
ward mechanisms coexist; we have thus not yet arrived at the
desired universal model of pursuit and preview. For developing
this universal model, it will be paramount to better understand
how the predictability properties of the target signal affect the
ability of the HC to utilize a feedforward. In realistic scenarios,
it is, however, likely that target signal predictability varies
considerably in time, possibly on a timescale of a few seconds,
calling for methods to identify time-varying HC adaptations.
C. Time-Varying Adaptations
Most of our knowledge on human control behavior is
restricted to stationary, time-invariant control tasks, where HCs
are considered as stationary, time-invariant controllers. In real-
ity, however, it is the adaptive nature of the HC, and how she
or he is able to respond to sudden changes in the environment,
that is of interest, yet still largely unknown [2], [110], [139].
Relevant real-world scenarios where HCs are forced to adapt
their control behavior are, for example, time-varying changes
in the controlled system dynamics (e.g., failure) [98], [99],
[140]–[144], instantaneously modified task constraints (e.g.,
decreased road width) [135], loss-of-control [145], automatic-
to-manual control transitions, and control with time-varying
information feedback (e.g., adaptive simulator motion feed-
back) [146]. Such time-varying HC adaptations are inherently
highly variable, nonlinear, short-duration, and strongly task-
dependent, making them immensely more complex to grasp
than LTI HC behavior. Both our current knowledge of HCs’
capabilities for temporally adapting control, as well as the ca-
pabilities of our methods for measuring adaptive HC dynamics,
are insufficient.
In studies on the adaptive HC, a distinction is often made
between “fast” adaptations in response to sudden changes in
the task or environment, and “slow” variations attributable to
factors such as fatigue, loss-of-attention, and learning [110].
While the latter can still be studied to some extent without
explicitly accounting for time-varying HC behavior, as shown
here in Section V-C, this does not hold for fast HC adaptations.
For certain fast adaptations, HC dynamics seem to remain
largely quasi-linear [98], [99], [147], but with time-varying
HC parameters, resulting in Linear Parameter Varying (LPV)
HC dynamics. However, in extreme scenarios such as loss-of-
control [145], HC dynamics are truly nonlinear, in addition
to time-varying. Understanding HCs’ capacity for adaptation
means grasping which of HCs’ control parameters are critical,
and what HCs’ limitations in the adaptation of these parame-
ters are. It is highly likely, but not yet proven, that some HC
parameters will change faster, while less critical parameters
may change more gradually.
Knowledge of the “life expectancies” of HC parameters,
and how this may vary for changes in (combinations of)
different critical task variables, is needed. This fact also
directly applies to certain control scenarios that are typically
studied with the assumption of time-invariant HC, such as
pursuit or preview tracking [111], [112], where in fact small,
local, time-varying adaptations in HC behavior are suspected
to occur. By assuming an LTI HC, temporal variations due to,
for example, the perceived difficulty of the applied test signals,
are averaged out, irrespective of how strongly they are present.
A thorough, explicitly time-varying, analysis of all HC data is
actually needed to prove that the “time-invariance” hypothesis
that is implicitly applied through the use of describing function
estimates and quasi-linear models is, in fact, valid.
To increase our knowledge of time-varying HC adaptations,
the traditional LTI framework for modeling and analyzing HC
behavior needs to be abandoned, as this requires methods and
model structures that inherently include additional degrees-
of-freedom to account for time-varying behavior. Given how
little we currently know about time-varying HC behavior, this
requires both methods for time-varying identification – i.e.,
detect and quantify time-varying changes in the HC with
preferably no a priori explicit assumptions on the nature of the
temporal variations – and time-varying parameter estimation
and model fitting, to extract high-accuracy time-varying HC
models from measured data. Also, we need to investigate
what excitation techniques and test signals will yield the most
reliable results, with the lowest possible level of intrusiveness.
Examples of time-varying identification methods are those
that rely on windowed LTI HC modeling [148], wavelets
[148], [149], recursive least-squares [147], [150] or Kalman
filtering [147], [151], [152]. Such methods are indispensable
for studying what actually varies in HCs and which “function
approximators” can best describe the adapting HC parameters.
Once known which time-variations in the HC need to be
modeled, promising approaches for the second step of fitting
intrinsically time-varying manual control models to measured
HC data are time-domain modeling [99], [146] or LPV model-
based methods [135], [153], [154]. The main challenge for
time-varying HC identification lies in developing methods that
are sufficiently sensitive and that can reliably pick out quick
and short-duration temporal variations in HC behavior from
inherently noisy data. Of great value to real-world applications
Theoretical (sigmoid)
Estimation (3 runs)
0 10 20 30 40 50 60 70 80
0 10 20 30 40 50 60 70 80
Time-varying HC dynamics, Hp(s, t)
Hnm(s)control, u
remnant, n
output, x
P(t) = P1+P2P1
1 + eG(tM)
Parameters vary according to sig-
moid functions with M=40 s,
G=0.5 s1:
error, e
target, ft
Time-varying system
dynamics, Hc(s, t)
Induced change in controlled sys-
tem dynamics:
s(s+ 6) =30
s(s+ 0.2)
Identified response in HC equalization parameter adaptation :
s= K
Fig. 7. Example of time-varying manual control identification for the scenario of [99] with a Kalman filter-based approach.
such as HC monitoring or adaptive support systems would be
methods that are suitable for recursive, real-time implementa-
Fig. 7 shows preliminary results from a current effort to
further develop time-varying HC modeling approaches based
on the Kalman Filter [147], [151], [152]. For three runs of
experimental HC data from a compensatory tracking task
with an induced change in the controlled system dynamics
Hc(s, t)(centered at t= 40 s) matching that of [99], Fig. 7
shows representative estimated HC equalization parameters:
the HC control gains on the tracking error (Ke) and error
rate (K˙e). Matching the expected “theoretical” time-variation
(sigmoid) to counter the induced change in the controlled
dynamics, Fig. 7 shows a distinct drop in Keafter the change
in Hc(s, t), and a notable increase in the error-rate gain
K˙e. However, Fig. 7 also clearly shows aspects of time-
varying HC behavior that are currently poorly understood:
1) a considerable variation over different runs of data, 2)
significant time-variations other than those in direct response
to the change in Hc(s, t), and 3) HC adaptations that clearly
lag behind the theoretically optimal responses.
In this section we will give three examples of novel appli-
cations of knowledge and models of human manual control
behavior. Presented are applications in haptic shared control,
simulator fidelity evaluations, and training.
A. Haptic Shared Control/Neuromuscular Adaptations
Understanding the contributions of the time-variant neuro-
muscular system to overall HC behavior is essential when
relevant forces are present on the control device, from external
perturbations (e.g., from wind gusts, potholes, turbulence),
biodynamic feedthrough (e.g., from undesired body move-
ments) [130] or support forces from haptic shared control
[44]. Current developments in NMS cybernetics focus on three
applications: 1) understanding fundamental motor control, by
enabling identification of both the NMS non-linearities and
time-variance, 2) enabling unobtrusive estimation of NMS
admittance during a flying or driving control task and 3) under-
standing co-adaptive systems in human-machine cooperation.
The first goal was already worked towards in early work
aimed at obtaining time-varying and non-linear identification
of NMS dynamics. Recent approaches used small-window
FRFs [42], wavelets [156], recursive least-squares algorithms
[150], and LPV methods [135], [157]. The second goal
requires perturbation techniques to estimate endpoint admit-
tance, which do not significantly influence manual control
behavior. This can be approached either by using small
rapid transient perturbations [158], or by using continuous
perturbations to estimate full-bandwidth admittance. The latter
technique has been used to estimate the arm NMS admittance
during aircraft control [159]. A particularly useful technique
to design continuous force perturbations is the Reduced-
Power Method [129], which allows full-bandwidth admittance
estimates while evoking unperturbed low-bandwidth control
behavior. It has been applied when comparing the NMS
admittance with and without haptic shared control, of the lower
limb during car-following [45] and of the arms while steering
a car [46]. Such analyses show that drivers can increase
their neuromuscular admittance to physically give way to the
guidance forces, thereby executing part of the control actions
suggested by the automation. An additional application for the
quantified NMS admittance is that it allows for a formal design
of the strength of the guidance forces of haptic shared control
[44], [64], as opposed to trial-and-error tuning. The third
goal is being pursued to understand physical co-adaptation of
two mutually adaptive controllers. Examples include human-
human physical interaction [56], the interaction between driver
and intelligent vehicle [44], and physical human-robot inter-
action [56]. Time-varying NMS identification techniques will
prove essential in all these efforts.
(a) Visual gain
Kv, V/deg
(0,0) (0.5,0.5) (1,0.5) (1,0) IF
(b) Visual lead constant
TL, s
subj. 1
subj. 2
subj. 3
subj. 4
subj. 5
subj. 6
subj. 7
(0,0) (0.5,0.5) (1,0.5) (1,0) IF
(c) Motion gain
(0,0) (0.5,0.5) (1,0.5) (1,0) IF
(d) Channel variance fraction
uv, %
(0,0) (0.5,0.5) (1,0.5) (1,0) IF
sensor dynamics equalization limitations
Visual response,Hpe
Motion response,Hpx
5.97(1 + 0.11s)
(1 + 5.9s)(1 + 0.005s)
|{z }
|{z }
s2+ 2ζnmωnms+ω2
| {z }
(a) (b)
Kv(1 + TLs)
Fig. 8. Multi-modal HC model, estimated HC model parameters, and visual and motion control contributions from the combined in-flight and simulator
experiment of [155]. Simulator conditions with varying motion filter (Hmf ) settings are indicated with (Kmf ,ωmf ), IF indicates in-flight data.
B. Flight Simulator Motion Cueing Fidelity
A key application of manual control cybernetics is evaluat-
ing the fidelity – realism – of virtual environments and vehicle
simulators. An example is the evaluation of flight simulator
motion cueing fidelity; something that simulator manufactur-
ers, engineers, and legislators still struggle with, even after
decades of experience in ground-based simulation [160]. One
of the biggest challenges in simulator motion cueing has
been finding the limits: when does the feedback supplied in
a simulator no longer induce “realistic”, representative, and
effective control behavior?
The known HC adaptation to critical task variables [2]
enables unique objective analysis of the effects of degraded
motion feedback quality. For example, an analytic control-
theoretic criterion based on pilot-vehicle system dynamics
that is sensitive to variations in motion cueing fidelity [22],
[23], [161] has been derived and successfully applied to
a range of different aircraft (both fixed-wing and moving-
wing) and flight maneuvers. In addition, multi-channel HC
modeling and identification techniques [79], [81], [83] can
be used to explicitly measure pilots responses to visual and
(simulator) motion cues during tracking tasks, to discover
under which motion washout filter settings pilots change their
control behavior [26], [29], [155]. This approach also enables
objective quantification of the behavioral discrepancies that
occur in flight simulators compared to real flight [18], [162]
and helps relate these discrepancies to the choices in motion
cueing [29], [155].
Fig. 8 shows multi-modal HC modeling results from [155],
where seven pilots performed an aircraft roll attitude tracking
task both in real flight and in a moving-base flight simulator
for a number of different settings of a first-order high-pass
motion filter, Hmf . In Fig. 8, the different simulator motion
conditions are indicated with “(Kmf ,ωmf )”, while “IF” is
the in-flight data. Kmf <1or ωmf >0rad/s results in
attenuated simulator roll cues. The multi-modal HC model
shown in Fig. 8 includes separate visual (Hpe) and motion
(Hpx) responses. Separating these contributions allows for
calculating metrics that provide unbiased insight into 1) how
pilots weigh visual and motion feedback for their control (Kv
and Km), 2) how much visual (lead) equalization they are
required to perform (TL), and 3) the overall contribution of
motion feedback to their control (σ2
The HC modeling results in Fig. 8 show that, in general,
pilot behavior is found to be strongly affected by degraded
simulator motion fidelity. With simulator roll motion cues that
are increasingly attenuated compared to the “perfect” (1,0)
case, pilots rely less on the presented motion information,
leading to a distinctly decreased contribution of the motion
feedback channel Hpx, see σ2
uvin Fig. 8(d). Consistent
for all pilots, this suboptimal control strategy is characterized
by reduced control gains (Kv) and increased visual lead
equalization (TL). Pilots are also not found to compensate for
lower magnitude motion feedback (Kmf <1) by a matching
increase of their motion response gain, Km. Finally, from the
in-flight (IF) data from [155], pilots were found to control
with a lower gain during in-flight tracking than for the 1-
to-1 simulator motion configuration, a result that might be
attributable to other factors than the quality of the supplied
motion feedback (e.g., environmental variables, see Fig. 2).
Overall, HC modeling results as shown in Fig. 8 are unique
in their ability to reveal the adaptation of pilot low-level
control behavior to reduced simulator cueing and have great
potential for the optimization of simulator motion cueing in
aircraft, but also automotive, simulation.
C. Control Skill Training
Another relevant application for manual control cybernet-
ics is evaluating the development of control skills during
training programs and verifying the overall effectiveness and
transferability of learned skills. Explicit quantification of HC
dynamics, as facilitated by cybernetic HC modeling tech-
niques, allows for opening-up the black box of human control
Curve fit M
Data NM Curve fit NM
Visual gain
Evaluation runs
Training runs
Data M
0 25 50 75 100 125 150 175
Visual lead time constant
Evaluation runsTraining runs
0 25 50 75 100 125 150 175
Training runs
Motion gain
Evaluation runs
0 25 50 75 100 125 150 175
++u x
(TLs+ 1)2
(TIs+ 1) ev
sensors equalization limitations
Human controller
Fig. 9. Estimated HC equalization parameters for visual and motion feedback responses from a training experiment with task-naive participants of [163].
Data from two training groups is presented: Group NM performed the first 100 runs in a non-moving (NM) simulator and was then transferred to a motion
setting (M), while Group M followed the opposite schedule.
adaptation, and observe the progression of HC feedback,
feedforward, and NMS dynamics through learning. This is
especially relevant for evaluating simulator-based training, as
for skill-based control HCs develop low-level automated re-
sponses to continuous feedback signals from the environment
[164]–[166]. This strong environmental dependency means
that a risk exists of teaching skills that do not fully transfer to
the real environment [164], [166].
In aircraft pilot training, the necessity for training simula-
tors that use a motion system to provide a physical motion
sensation as experienced during flight, continues to be a topic
of much debate [167]. The main reason for the continuing
controversy is the fact that collecting convincing and gen-
eralizable evidence regarding training effectiveness requires
reliable and quantitative data regarding trainees’ developing
skills. Most explicit transfer-of-training studies have relied on
ambiguous measures of task performance [167] or “lumped”
HC dynamics estimates [168], providing limited insight and
unconvincing conclusions. Only recently has explicit multi-
modal HC modeling been applied to verify the effectiveness
of simulator-based training of manual control skills in fixed-
base or limited-motion simulators [29], [163].
Fig. 9 shows the multi-modal HC model and data from the
experiment of [163], where 24 task-naive participants, divided
over two groups, were trained for a compensatory tracking task
with motion feedback, to investigate the need for motion in ab
initio skill training. The graphs in Fig. 9 present the estimated
values of key HC equalization parameters that quantify HCs’
use of motion feedback [15], [25], [29], [169]: the visual
response gain Kv, the visual lead time constant TL, and the
motion response gain Km. First, these data show that initial
control skill acquisition is a very slow process, with partici-
pants’ control parameter optimization – i.e., increasing control
gains (Kvand Km) and decreasing visual lead equalization
(TL) – continuing until after 75 runs of the tracking task.
Furthermore, Fig. 9 shows that for Group NM (no-motion
training) the HC equalization parameters indicate only minor
adaptation directly after transfer (run 101) and considerable
renewed adaptation during the 75 evaluation runs. Especially
the learning curves for Km, which are essentially identical
for both groups, provide clear evidence that training without
motion is not effective for training control skills to be applied
in an environment with motion feedback.
In conclusion, contrary to many earlier studies that relied
on performance metrics for training evaluation [167], a cy-
bernetic view on training in motion simulators, as shown in
Fig. 9, provides direct and objective evidence regarding the
effectiveness of such training. Applying the same methodology
to the training of other critical and realistic manual control
tasks (e.g., preview, feedforward) will greatly increase our
understanding of HC adaptation during training, and enable
fundamental training enhancements.
With this paper we attempted to give an overview of the
current state-of-the-art in manual control cybernetics research.
We identified several fundamental shortcomings and proposed
a new framework for bringing theory and methodology to
the level required for addressing current real-world issues.
In our view, this requires a special focus on the adaptive
characteristics of human control behavior in realistic control
scenarios. A crucial step forward would be to abandon the
linear time-invariant modeling framework altogether and move
to modeling structures and methods that inherently include
degrees of freedom to account for time-varying behavior. A
promising candidate could be the currently rapidly developing
linear parameter-varying (LPV) systems framework, to model
and identify intrinsically time-varying manual control models.
Developing an extended “toolkit” that will allow us to
identify, model and quantify adaptive human control, will lead
to (at least) three key innovations. First, the exploitation of
human controllers’ capability to adapt, and adaptation “invari-
ants”, is key to optimizing the multi-modal control interfaces
that our ever-advancing modern technologies permit. True
knowledge of the adaptive human controller will transform
the current trial-and-error tuning of such interfaces for the
“average” human to a systematic approach to create person-
alized support. Second, a model-based approach to quantify
progress in skill acquisition will be instrumental to improve
(simulator- or computer-based) training procedures and tech-
nologies, as it allows for a mathematical, and objectified,
optimization of training effectiveness. Third, understanding
and mathematically modeling human adaptive control will
enable designers of (semi-)automated systems to create high-
conformance human-like automation that is trusted and ac-
cepted in situations where control is either shared (e.g., haptic
shared control) or handed-over to a vehicle, robot, or computer.
And beyond the realm of supporting human-machine systems,
the insights gained in the unique human adaptation capabilities
could also serve as design inspiration for future generations
of fully autonomous, adaptive robots, ultimately equating the
control and communication in animal and machine [1].
[1] N. Wiener, Cybernetics: or Control and Communication in the Animal
and the Machine, 2nd ed. Cambridge (MA): MIT Press, 1961.
[2] D. T. McRuer and H. R. Jex, “A Review of Quasi-Linear Pilot Models,
IEEE Trans. on Human Factors in Electronics, vol. HFE-8, no. 3, pp.
231–249, 1967.
[3] D. T. McRuer and D. H. Weir, “Theory of Manual Vehicular Control,
IEEE Trans. on Man-Machine Systems, vol. 10, no. 4, pp. 257–291,
[4] D. T. McRuer and E. S. Krendel, “Mathematical Models of Human
Pilot Behavior,” AGARDograph AGARD-AG-188, 1974.
[5] R. J. Jagacinski and J. M. Flach, Control Theory for Humans: Quan-
titative Approaches to Modeling Performance. CRC Press, 2003.
[6] M. Mulder, M. M. van Paassen, and E. R. Boer, “Exploring the Roles of
Information in the Control of Vehicular Locomotion: From Kinematics
and Dynamics to Cybernetics,” Presence: Teleoperators and Virtual
Environments, vol. 13, no. 5, pp. 535–548, 2004.
[7] I. L. Ashkenas and D. T. McRuer, “A Theory of Handling Qualities
Derived from Pilot-Vehicle System Considerations,Aerospace Engi-
neering, vol. 21, no. 2, pp. 60–102, 1962.
[8] T. P. Neal and R. E. Smith, “A Flying Qualities Criterion for the Design
of Fighter Flight-Control Systems,” J. of Aircraft, vol. 8, no. 10, pp.
803–809, 1971.
[9] D. T. McRuer, D. H. Weir, and R. H. Klein, “A Pilot-Vehicle Systems
Approach to Longitudinal Flight Director Design,” J. of Aircraft, vol. 8,
no. 11, pp. 890–897, 1971.
[10] D. T. McRuer, “Development of Pilot-in-the-Loop Analysis,J. of
Aircraft, vol. 10, no. 9, pp. 515–524, 1973.
[11] R. A. Hess, “Application of a Model-Based Flight Director Design
Technique to a Longitudinal Hover Task,J. of Aircraft, vol. 14, no. 3,
pp. 265–271, 1977.
[12] ——, “Unified Theory for Aircraft Handling Qualities and Adverse
Aircraft-Pilot Coupling,” J. of Guidance, Control, and Dynamics,
vol. 20, no. 6, pp. 1141–1148, 1997.
[13] D. H. Klyde, D. T. McRuer, and T. T. Myers, “Pilot-Induced Oscillation
Analysis and Prediction with Actuator Rate Limiting,” J. of Guidance,
Control, and Dynamics, vol. 20, no. 1, pp. 81–89, 1997.
[14] R. A. Hess, “Theory for roll-ratchet phenomenon in high-performance
aircraft,” J. of Guidance, Control, and Dynamics, vol. 21, no. 1, pp.
101–108, 1998.
[15] R. J. A. W. Hosman and H. G. Stassen, “Pilot’s perception in the
control of aircraft motions,” Control Engineering Practice, vol. 7, pp.
1421–1428, 1999.
[16] R. A. Hess, “Analytical Assessment of Performance, Handling Quali-
ties, and Added Dynamics in Rotorcraft Flight Control,” IEEE Trans.
on Systems, Man, and Cybernetics Part A: Systems and Humans,
vol. 39, no. 1, pp. 262–271, 2009.
[17] H. J. Damveld, M. M. van Paassen, and M. Mulder, “Cybernetic
Approach to Assess Aircraft Handling Qualities,” J. of Guidance,
Control, and Dynamics, vol. 34, no. 6, pp. 1886–1898, 2011.
[18] H. J. Smith, “Human Describing Functions Measured in Flight and
on Simulators,” IEEE Trans. on Human Factors in Electronics, vol.
HFE-8, no. 4, pp. 264–268, 1967.
[19] R. A. Hess, “Model-Based Investigation of Manipulator Characteristics
and Pilot/Vehicle Performance,” J. of Guidance, Control, and Dynam-
ics, vol. 6, no. 5, pp. 348–354, 1983.
[20] R. A. Hess and K. K. Chan, “Preview Control Pilot Model for Near-
Earth Maneuvering Helicopter Flight,” J. of Guidance, Control, and
Dynamics, vol. 11, no. 2, pp. 146–152, 1988.
[21] R. A. Hess, “Model for the Human Use of Motion Cues in Vehicular
Control,” Journal of Guidance, vol. 13, no. 3, pp. 476–482, 1990.
[22] R. A. Hess and T. Malsbury, “Closed-loop Assessment of Flight
Simulator Fidelity,J. of Guidance, Control, and Dynamics, vol. 14,
no. 1, pp. 191–197, 1991.
[23] Y. Zeyada and R. A. Hess, “Computer-Aided Asessment of Flight
Simulator Fidelity,J. of Aircraft, vol. 40, no. 1, pp. 173–180, 2003.
[24] M. M. van Paassen, J. C. van der Vaart, and J. A. Mulder, “Model of
the neuromuscular dynamics of the human pilot’s arm,J. of Aircraft,
vol. 41, no. 6, pp. 1482–1490, 2004.
[25] P. M. T. Zaal, D. M. Pool, J. de Bruin, M. Mulder, and M. M. van
Paassen, “Use of Pitch and Heave Motion Cues in a Pitch Control
Task,” J. of Guidance, Control, and Dynamics, vol. 32, no. 2, pp. 366–
377, 2009.
[26] D. M. Pool, P. M. T. Zaal, M. M. van Paassen, and M. Mulder, “Effects
of Heave Washout Settings in Aircraft Pitch Disturbance Rejection,J.
of Guidance, Control, and Dynamics, vol. 33, no. 1, pp. 29–41, 2010.
[27] D. M. Pool, P. M. T. Zaal, H. J. Damveld, M. M. van Paassen, J. C.
van der Vaart, and M. Mulder, “Modeling Wide-Frequency-Range Pilot
Equalization for Control of Aircraft Pitch Dynamics,” J. of Guidance,
Control, and Dynamics, vol. 34, no. 5, pp. 1529–1542, 2011.
[28] G. Quaranta, P. Masarati, and J. Venrooij, “Impact of pilots biodynamic
feedthrough on rotorcraft by robust stability,J. of Sound and Vibration,
vol. 332, no. 20, p. 49484962, 2013.
[29] P. M. T. Zaal, D. M. Pool, M. M. van Paassen, and M. Mulder, “Com-
paring Multimodal Pilot Pitch Control Behavior Between Simulated
and Real Flight,” J. of Guidance, Control, and Dynamics, vol. 35, no. 5,
pp. 1456–1471, 2012.
[30] C. V. Kroll, “Preview-Predictor Model of Driver Behavior in Emer-
gency Situations,” Highway research record, no. 364, pp. 16–26, 1971.
[31] G. O. Burnham, J. Seo, and G. A. Bekey, “Identification of Human
Driver Models in Car Following,IEEE Trans. on Automatic Control,
vol. 19, no. 6, pp. 911–915, 1974.
[32] D. T. McRuer, D. H. Weir, H. R. Jex, R. E. Magdaleno, and R. W.
Allen, “Measurement of Driver-Vehicle Multiloop Response Properties
with a Single Disturbance Input,” IEEE Trans. on Systems, Man, and
Cybernetics, vol. 5, no. 5, pp. 490–497, 1975.
[33] D. T. McRuer, R. W. Allen, D. H. Weir, and R. H. Klein, “New Results
in Driver Steering Control Models,Human Factors, vol. 19, no. 4, pp.
381–397, 1977.
[34] E. Donges, “A Two-Level Model of Driver Steering Behavior,” Human
Factors, vol. 20, no. 6, pp. 691–707, 1978.
[35] C. C. MacAdam, “Application of an Optimal Preview Control for
Simulation of Closed-Loop Automobile Driving,” Systems, Man and
Cybernetics, IEEE Trans. on, vol. 11, no. 6, pp. 393–399, 1981.
[36] R. A. Hess and A. Modjtahedzadeh, “A Control-Theoretic Model of
Driver Steering Behavior,” IEEE Control Systems Magazine, vol. 10,
no. 5, pp. 3–8, 1990.
[37] G. Prokop, “Modeling Human Vehicle Driving by Model Predictive
Online Optimization,” Vehicle System Dynamics, vol. 35, no. 1, pp.
19–53, 2001.
[38] D. D. Salvucci and R. Gray, “A two-point visual control model of
steering,” Perception, vol. 33, no. 10, pp. 1233–1248, 2004.
[39] A. J. Pick and D. J. Cole, “Driver steering and muscle activity during
a lane-change manoeuvre,” Vehicle System Dynamics, vol. 45, no. 9,
pp. 781–805, 2007.
[40] ——, “A Mathematical Model of Driver Steering Control Including
Neuromuscular Dynamics,” J. of Dynamic Systems, Measurement, and
Control, vol. 130, no. 3, pp. 031 004–1 031004– 9, 2008.
[41] A. M. C. Odhams and D. J. Cole, “Application of Linear Preview
Control to Modelling Human Steering Control,” Proc. of the Institution
of Mechanical Engineers, Part D: Journal of Automobile Engineering,
vol. 223, no. 7, pp. 835–853, 2009.
[42] D. I. Katzourakis, D. A. Abbink, E. Velenis, E. Holweg, and R. Happee,
“Driver’s arms’ time-variant neuromuscular admittance during real car
test-track driving,IEEE Trans. on Instrumentation and Measurement,
vol. 63, no. 1, pp. 221–230, 2014.
[43] M. Mulder, D. A. Abbink, M. M. van Paassen, and M. Mulder, “Design
of a Haptic Gas Pedal for Active Car-Following Support,IEEE Trans.
on Int. Transp. Systems, vol. 12, no. 1, pp. 268–279, 2011.
[44] D. A. Abbink, M. Mulder, and E. R. Boer, “Haptic shared control -
smoothly shifting control authority?” Cognition, Work and Technology,
vol. 41, no. 5, pp. 1239–1249, 2011.
[45] D. A. Abbink, M. Mulder, F. C. T. van der Helm, M. Mulder, and
E. R. Boer, “Measuring neuromuscular control dynamics during car
following with continuous haptic feedback,” IEEE Trans. on Systems,
Man, & Cyb., Part B: Cyb., vol. 41, no. 5, pp. 1239–1249, 2011.
[46] D. W. van der Wiel, M. M. van Paassen, M. Mulder, M. Mulder, and
D. A. Abbink, “Driver adaptation to driving speed and road width:
Exploring parameters for designing adaptive haptic shared control,” in
Proc. of the IEEE Int. Conf. on Systems, Man, and Cybernetics, Hong
Kong, 2015, pp. 3060–3065.
[47] H. R. Jex, J. D. McDonnell, and A. V. Phatak, “A “Critical” Tracking
Task for Manual Control Research,” IEEE Trans. on Human Factors
in Electronics, vol. 7, no. 4, pp. 138–145, 1966.
[48] L. Bennett, “Powered wheelchair bucking,” J. of Rehabilitation Re-
search and Development, vol. 24, no. 2, p. 8186, 1987.
[49] M. R. Sirouspour and S. E. Salcudean, “Suppressing operator-induced
oscillations in manual control systems with movable bases,IEEE
Trans. on Control Syst. Tech., vol. 11, no. 4, pp. 448–459, 2003.
[50] S. S ¨
enyi and R. B. Gillespie, “Cancellation of Biodynamic
Feedthrough in Vehicle Control Tasks,” IEEE Trans. on Control
Systems Technology, vol. 15, no. 6, pp. 1018 – 1029, 2007.
[51] R. A. Hess, J. K. Moore, and M. Hubbard, “Modeling the Manually
Controlled Bicycle,” IEEE Trans. on Systems, Man and Cybernetics,
Part A: Systems and Humans, vol. 42, no. 3, pp. 545–557, 2012.
[52] J. J. Potter and W. Singhose, “Improving manual tracking of systems
with oscillatory dynamics,” IEEE Trans. on Human-Machine Systems,
vol. 43, no. 1, pp. 46–52, 2013.
[53] M. A. Goodrich and E. R. Boer, “Model-based Human-Centered Task
Automation: A Case Study in ACC design,IEEE Transactions on
Systems, Man, Cybernetics, Part A, vol. 33, no. 3, pp. 325–336, 2003.
[54] H. E. Tseng, J. Asgari, D. Hrovat, P. van der Jagt, A. Cherry, and
S. Neads, “Evasive manoeuvres with a steering robot,Vehicle System
Dynamics, vol. 43, no. 3, pp. 199–216, 2005.
[55] J. G. W. Wildenbeest, D. A. Abbink, C. J. M. Heemskerk, F. C. T.
van der Helm, and H. Boessenkool, “The impact of haptic feedback
quality on the performance of teleoperated assembly tasks,” IEEE
Trans. on Haptics, vol. 6, no. 2, pp. 242–252, 2013.
[56] G. Ganesh, A. Takagi, R. Osu, T. Yoshioka, M. Kawato, and E. Burdet,
“Two is better than one: Physical interactions improve motor perfor-
mance in humans,” Scientific reports, vol. 4, 2014.
[57] R. B. Warrier and S. Devasia, “Iterative learning from novice human
demonstrations for output tracking,” IEEE Trans. on Human-Machine
Systems, vol. 46, no. 4, pp. 510–521, 2016.
[58] R. D. Jones, I. M. Donaldson, and N. B. Sharman, “A Technique
for Removal of the Visuoperceptual Component from Tracking Per-
formance and Its Application to Parkinsons Disease,” IEEE Trans. on
Biomedical Engineering, vol. 43, no. 10, pp. 1001–1010, 1996.
[59] P. Soliveri, R. G. Brown, M. Jahanshahi, T. Caraceni, and C. D.
Marsden, “Learning manual pursuit tracking skills in patients with
Parkinson’s disease,Brain, vol. 120, no. 8, pp. 1325–1337, 1997.
[60] C. R. H. Innes, R. D. Jones, T. J. Anderson, S. G. Hollobon, and
J. C. Dalrymple-Alford, “Performance in normal subjects on a novel
battery of driving-related sensory-motor and cognitive tests,Behavior
Research Methods, vol. 41, no. 2, pp. 284–294, 2009.
[61] M. M. K. Oishi, P. TalebiFard, and M. J. McKeown, “Assessing
Manual Pursuit Tracking in Parkinsons Disease Via Linear Dynamical
Systems,” Annals of Biomedical Engineering, vol. 39, no. 8, 2011.
[62] A. J. Grunwald and S. J. Merhav, “Effectiveness of basic display
augmentation in vehicular control by visual field cues,” IEEE Trans.
on Systems, Man, & Cyb., vol. SMC-8, no. 9, pp. 679–690, 1978.
[63] M. Mulder and J. A. Mulder, “Cybernetic Analysis of Perspective
Flight-Path Display Dimensions,” J. of Guidance, Control, and Dy-
namics, vol. 28, no. 3, pp. 398–411, 2005.
[64] J. Smisek, E. Sunil, M. M. van Paassen, D. A. Abbink, and M. Mulder,
“Neuromuscular-system-based tuning of a haptic shared control inter-
face for uav teleoperation,IEEE Trans. on Hum.-Mach. Syst., 2016.
[65] A. Tustin, “The Nature of the Operator’s Response in Manual Control,
and its Implications for Controller Design,” J. of the Institution of
Electrical Engineers – Part IIA: Automatic Regulators and Servo
Mechanisms, vol. 94, no. 2, pp. 190–206, 1947.
[66] J. I. Elkind and C. D. Forgie, “Characteristics of the Human Operator
in Simple Manual Control Systems,” IRE Trans. on Automatic Control,
vol. 4, no. 1, pp. 44–55, 1959.
[67] D. T. McRuer and E. S. Krendel, “The Human Operator as a Servo
System Element,” J.o.t. Franklin Inst., vol. 267, no. 5, pp. 381–403,
[68] E. S. Krendel and D. T. McRuer, “A Servomechanics Approach to Skill
Development,J.o.t. Franklin Inst., vol. 269, no. 1, pp. 24–42, 1960.
[69] D. T. McRuer, D. Graham, E. S. Krendel, and W. J. Reisener, “Human
Pilot Dynamics in Compensatory Systems, Theory Models and Exper-
iments with Controlled Element and Forcing Function Variations,” Air
Force Fl. Dyn. Lab., Tech. Rep. TR-65-15, 1965.
[70] W. H. Levison and J. I. Elkind, “Two-Dimensional Manual Control
Systems with Separated Displays,” IEEE Trans. on Human Factors in
Electronics, vol. 8, no. 3, pp. 202–209, 1967.
[71] D. T. McRuer, L. G. Hofmann, H. R. Jex, G. P. Moore, A. V.
Phatak, D. H. Weir, and J. Wolkovitch, “New Approaches to Human-
Pilot/Vehicle Dynamic Analysis,” Air Force Fl. Dyn. Lab., Tech. Rep.
67-150, 1968.
[72] R. E. Magdaleno, H. R. Jex, and W. A. Johnson, “Tracking Quasi-
Predictable Displays Subjective Predictability Gradations, Pilot Models
for Periodic and Narrowband Inputs,” in Fifth Annual NASA-University
Conf. on Manual Control, 1969, pp. 391–428.
[73] W. H. Levison, J. I. Elkind, and J. L. Ward, “Studies of Multivariable
Manual Control Systems: A Model for Task Interference,” NASA CR
1746, 1971.
[74] D. T. McRuer, “Human Dynamics in Man-Machine Systems,Auto-
matica, vol. 16, no. 3, pp. 237–253, 1980.
[75] R. A. Hess, “Pursuit Tracking and Higher Levels of Skill Development
in the Human Pilot,” IEEE Trans. on Systems, Man, and Cybernetics,
vol. SMC-11, no. 4, pp. 262–273, 1981.
[76] D. T. McRuer, “Pilot Modeling,” in LS-157 Advances in Flying
Qualities. AGARD, 1988, pp. 2–1–2–30.
[77] D. T. McRuer and D. K. Schmidt, “Pilot-Vehicle Analysis of Multiaxis
Tasks,” J. of Guidance, Cont. & Dyn., vol. 13, no. 2, pp. 348–355, 1990.
[78] J. I. Elkind and D. L. Darley, “The Normality of Signals and Describing
Function Measurements of Simple Manual Control Systems,” IEEE
Trans. on Hum. Fact. in Electr., vol. 4, no. 1, pp. 52–55, 1963.
[79] R. L. Stapleford, D. T. McRuer, and R. E. Magdaleno, “Pilot Describing
Function Measurements in a Multiloop Task,” IEEE Trans. on Human
Factors in Electronics, vol. 8, no. 2, pp. 113–125, 1967.
[80] A. van Lunteren, “Identification of Human Operator Describing Func-
tion Models with One or Two Inputs in Closed Loop Systems,” Ph.D.
dissertation, TU Delft, Mechanical Engineering, 1979.
[81] M. M. van Paassen and M. Mulder, “Identification of Human Operator
Control Behaviour in Multiple-Loop Tracking Tasks,” in Proc. of
the Seventh IFAC/IFIP/IFORS/IEA Symp. on Analysis, Design and
Evaluation of Man-Machine Systems, Kyoto Japan, 1998, pp. 515–520.
[82] F. M. Nieuwenhuizen, P. M. T. Zaal, M. Mulder, M. M. van Paassen,
and J. A. Mulder, “Modeling Human Multichannel Perception and
Control Using Linear Time-Invariant Models,J. of Guidance, Control,
and Dynamics, vol. 31, no. 4, pp. 999–1013, 2008.
[83] P. M. T. Zaal, D. M. Pool, Q. P. Chu, M. M. van Paassen, M. Mulder,
and J. A. Mulder, “Modeling Human Multimodal Perception and Con-
trol Using Genetic Maximum Likelihood Estimation,” J. of Guidance,
Control, and Dynamics, vol. 32, no. 4, pp. 1089–1099, 2009.
[84] M. Mulder, D. M. Pool, D. A. Abbink, E. R. Boer, and M. M. van
Paassen, “Fundamental Issues in Manual Control Cybernetics,” in Proc.
of the 13th IFAC/IFIP/IFORS/IEA Symp. on Analysis, Design, and
Evaluation of Human-Machine Systems, Kyoto, Japan, 2016.
[85] J. J. Gibson, The Perception of the Visual World. Boston (MA):
Houghton Mifflin, 1950.
[86] ——, The Ecological Approach to Visual Perception. Hillsdale (NJ):
Lawrence Erlbaum Associates, 1986, originally published in 1979.
[87] D. L. Kleinman, S. Baron, and W. H. Levison, “An Optimal Control
Model of Human Response Part I: Theory and Validation,” Automatica,
vol. 6, no. 3, pp. 357–369, 1970.
[88] S. Baron and W. H. Levison, “An Optimal Control Methodology
for Analyzing the Effects of Display Parameters on Performance and
Workload in Manual Flight Control,IEEE Trans. on Systems, Man,
and Cybernetics, vol. SMC-5, no. 4, pp. 423–430, 1975.
[89] M. Tomizuka and D. E. Whitney, “The Human Operator in Manual
Preview Tracking (an Experiment and Its Modeling Via Optimal
Control),” J. of Dynamic Systems, Measurement, and Control, vol. 98,
no. 4, pp. 407–413, 1976.
[90] J. J. Kok and R. A. Van Wijk, “Evaluation of Models Describing Hu-
man Operator Control of Slowly Responding Complex Systems,” Ph.D.
Dissertation, Faculty of Mechanical Engineering, Delft University of
Technology, 1978.
[91] J. I. Elkind and L. T. Sprague, “Transmission of Information in
Simple Manual Control Systems,” IRE Trans. on Human Factors in
Electronics, vol. 2, no. 1, pp. 58–60, 1961.
[92] G. C. Beerens, H. J. Damveld, M. Mulder, M. M. van Paassen,
and J. C. van der Vaart, “Investigation into Crossover Regression in
Compensatory Manual Tracking Tasks,J. of Guidance, Control, and
Dynamics, vol. 32, no. 5, pp. 1429–1445, 2009.
[93] H. J. Damveld, G. C. Beerens, M. M. van Paassen, and M. Mulder,
“Design of Forcing Functions for the Identification of Human Control
Behavior,J. of Guidance, Control, and Dynamics, vol. 33, no. 4, pp.
1064–1081, 2010.
[94] W. H. Levison, S. Baron, and D. L. Kleinman, “A Model for Human
Controller Remnant,” IEEE Trans. on Man-Machine Systems, vol. 10,
no. 4, pp. 101–108, 1969.
[95] H. R. Jex and R. E. Magdaleno, “Corroborative Data on Normalization
of Human Operator Remnant,” IEEE Trans. on Man-Machine Systems,
vol. 10, no. 4, pp. 137–140, 1969.
[96] R. L. Stapleford, S. J. Craig, and J. A. Tennant, “Measurement of Pilot
Describing Functions in Single-Controller Multiloop Tasks,” Tech.
Rep. NASA CR-1238, 1969.
[97] R. Chernikoff, J. W. Duey, and F. V. Taylor, “Two-Dimensional Track-
ing with Identical and Different Control Dynamics in Each Coordinate,”
J. of Experimental Psychology, vol. 60, no. 5, pp. 318–322, 1960.
[98] R. A. Hess, “Modeling Human Pilot Adaptation to Flight Control
Anomalies and Changing Task Demands,” J. of Guidance, Control,
and Dynamics, vol. 39, no. 3, pp. 655–666, 2016.
[99] P. M. T. Zaal, “Manual Control Adaptation to Changing Vehicle
Dynamics in Roll Pitch Control Tasks,J. of Guidance, Control, and
Dynamics, vol. 39, no. 5, pp. 1046–1058, 2016.
[100] D. M. Pool, M. Mulder, M. M. van Paassen, and J. C. van der Vaart,
“Effects of Peripheral Visual and Physical Motion Cues in Roll-Axis
Tracking Tasks,” J. of Guidance, Control, and Dynamics, vol. 31, no. 6,
pp. 1608–1622, 2008.
[101] R. J. Wasicko, D. T. McRuer, and R. E. Magdaleno, “Human Pilot
Dynamic Response in Single-loop Systems with Compensatory and
Pursuit Displays,” Air Force Fl. Dyn. Lab., Tech. Rep. 66-137, 1966.
[102] R. W. Allen and D. T. McRuer, “The Man/Machine Control Interface–
Pursuit Control,” Automatica, vol. 15, no. 6, pp. 683–686, 1979.
[103] H. P. Bergeron, “Investigation of Motion Requirements in Compen-
satory Control Tasks,” IEEE Trans. on Man-Machine Systems, vol.
MMS-11, no. 2, pp. 123–125, 1970.
[104] F. M. Drop, D. M. Pool, H. J. Damveld, M. M. van Paassen, and
M. Mulder, “Identification of the Feedforward Component in Manual
Control With Predictable Target Signals,” IEEE Trans. on Cybernetics,
vol. 43, no. 6, pp. 1936–1949, 2013.
[105] A. Abdel-Malek and V. Z. Marmarelis, “Modeling of Task-Dependent
Characteristics of Human Operator Dynamics Pursuit Manual Track-
ing,” IEEE Trans. on Systems, Man, Cybernetics, vol. 18, no. 1, pp.
163–172, 1988.
[106] E. C. Poulton, “Perceptual Anticipation in Tracking with Two-Pointer
and One-Pointer Displays,” British Journal of Psychology, vol. 43,
no. 3, pp. 222–229, 1952.
[107] R. Chernikoff, H. P. Birmingham, and F. V. Taylor, “A Comparison
of Pursuit and Compensatory Under Conditions of Aiding and No
Aiding,” J. of Exp. Psych., vol. 49, no. 1, pp. 55–59, 1955.
[108] R. W. Pew, J. C. Duffendack, and L. K. Fensch, “Sine-Wave Tracking
Revisited,IEEE Trans. on Human Factors in Electronics, vol. HFE-8,
no. 2, pp. 130–134, 1967.
[109] P. D. Neilson, N. J. O’Dwyer, and M. D. Neilson, “Stochastic Pre-
diction in Pursuit Tracking: An Experimental Test of Adaptive Model
Theory,Biological Cybernetics, vol. 58, no. 2, 1988.
[110] L. R. Young, “On Adaptive Manual Control,” IEEE Trans. on Man-
Machine Systems, vol. 10, no. 4, pp. 292–331, 1969.
[111] K. van der El, D. M. Pool, H. J. Damveld, M. M. van Paassen,
and M. Mulder, “An Empirical Human Controller Model for Preview
Tracking Tasks,IEEE Trans. on Cybernetics, vol. 46, no. 11, pp.
2609–2621, 2016.
[112] T. B. Sheridan, “Three Models of Preview Control,IEEE Trans. on
Human Factors in Electronics, vol. 7, no. 2, pp. 91–102, 1966.
[113] K. Ito and M. Ito, “Tracking Behavior of Human Operators in Preview
Control Systems,” Electr. Eng. in Japan, vol. 95, no. 1, pp. 120–127,
[114] K. van der El, D. M. Pool, M. M. van Paassen, and M. Mulder, “Effects
of Preview on Human Control Behavior in Tracking Tasks with Various
Controlled Elements,” IEEE Trans. on Cybernetics, no. 99, online.
[115] H. G. Stassen, G. Johannsen, and N. Moray, “Internal Representation,
Internal Model, Human Performance Model and Mental Workload,”
Automatica, vol. 26, no. 4, pp. 811–820, 1990.
[116] G. Vossius, “Der kybernetische aspekt der willkurbewegung,Progress
in Cybernetics, 1965.
[117] E. C. Poulton, “Learning the Statistical Properties of the Input in
Pursuit Tracking,J. of Experimental Psychology, vol. 54, no. 1, 1957.
[118] M. Noble, D. Trumbo, L. Ulrich, and K. Cross, “Task Predictability
and the Development of Tracking Skill under Extended Practice,J. of
Experimental Psychology, vol. 72, no. 1, pp. 85–94, 1966.
[119] D. Trumbo, M. Noble, K. Cross, and L. Ulrich, “Task Predictability
in the Organization, Acquisition, and Retention of Tracking Skill,” J.
of Experimental Psychology, vol. 70, no. 3, pp. 252–263, 1965.
[120] F. M. Drop, R. J. De Vries, M. Mulder, and H. H. B¨
ulthoff, “The
Predictability of a Target Signal Affects Manual Feedforward Control,
in Proc. of the 13th IFAC/IFIP/IFORS/IEA Symp. on Analysis, Design,
and Evaluation of Human-Machine Systems, Kyoto, Japan, 2016.
[121] T. Yamashita, “Effects of Sine Wave Combinations on the Development
of Precognitive Mode in Pursuit Tracking,The Quarterly Journal of
Experimental Psychology, vol. 42A, no. 4, pp. 791–810, 1990.
[122] D. M. Pool, P. M. T. Zaal, M. M. van Paassen, and M. Mulder, “Identi-
fication of Multimodal Pilot Models Using Ramp Target and Multisine
Disturbance Signals,” J. of Guidance, Control, and Dynamics, vol. 34,
no. 1, pp. 86–97, 2011.
[123] V. A. Laurense, D. M. Pool, H. J. Damveld, M. M. van Paassen,
and M. Mulder, “Effects of Controlled Element Dynamics on Human
Feedforward Behavior in Ramp-Tracking Tasks,IEEE Trans. on
Cybernetics, vol. 45, no. 2, pp. 253–265, 2015.
[124] D. T. McRuer, “Remarks on Some Neuromuscular Subsystem Dynam-
ics,” IEEE Trans. on Human Factors in Electronics, vol. HFE-7, no. 3,
pp. 129–130, 1966.
[125] D. T. McRuer, R. E. Magdaleno, and G. P. Moore, “A Neuromuscular
Actuation System Model,” IEEE Trans. on Man-Machine Systems,
vol. 9, no. 3, pp. 61–71, 1968.
[126] D. T. McRuer and R. E. Magdaleno, “Human Pilot Dynamics with
Various Manipulators,” Tech. Rep. AFFDL TR-66-138, 1966.
[127] A. Morris, Jr. and D. W. Repperger, “Discriminant Analysis of Changes
in Human Muscle Function When Interacting with an Assistive Aid,
IEEE Trans. on Biom. Eng., vol. 35, no. 5, pp. 316–322, 1988.
[128] R. A. Hess, “Analyzing Manipulator and Feel System Effects in Air-
craft Flight Control,” IEEE Trans. on Systems, Man, and Cybernetics,
vol. 20, no. 4, pp. 923–931, 1990.
[129] W. Mugge, D. A. Abbink, A. C. Schouten, J. P. A. DeWald, and F. C. T.
van der Helm, “A rigorous model of reflex function indicates that
position and force feedback are flexibly tuned to position and force
tasks,” Experimental Brain Research, vol. 200, pp. 325–340, 2010.
[130] J. Venrooij, D. A. Abbink, M. Mulder, M. M. van Paassen, and M. Mul-
der, “A Method to Measure the Relationship Between Biodynamic
Feedthrough and Neuromuscular Admittance,” IEEE Trans. on Systems,
Man, and Cybernetics, Part B: Cybernetics, vol. 41, no. 4, pp. 1158–
1169, 2011.
[131] J. Venrooij, D. A. Abbink, M. Mulder, M. M. van Paassen, M. Mulder,
F. C. T. van der Helm, and H. H. B ¨
ulthoff, “A Biodynamic Feedthrough
Model Based on Neuromuscular Principles,” IEEE Trans. on Cybernet-
ics, vol. 44, no. 7, pp. 1141–1154, 2013.
[132] J. M. Winters and L. Stark, “Analysis of Fundamental Human Move-
ment Patterns Through the Use of In-Depth Antagonistic Muscle
Models,” IEEE Trans. on Biom. Eng., vol. 32, no. 10, pp. 826–839,
[133] D. W. Franklin, R. Osu, E. Burdet, M. Kawato, and T. E. Milner,
“Adaptation to stable and unstable dynamics achieved by combined
impedance control and inverse dynamics model.J. of neurophysiology,
vol. 90, no. 5, pp. 3270–82, 2003.
[134] R. Osu, D. W. Franklin, H. Kato, H. Gomi, K. Domen, T. Yoshioka,
and M. Kawato, “Short- and long-term changes in joint co-contraction
associated with motor learning as revealed from surface emg.J. of
neurophysiology, vol. 88, no. 2, pp. 991–1004, 2002.
[135] A. J. Pronker, D. A. Abbink, M. M. van Paassen, and M. Mulder,
“Driver Neuromuscular Admittance can be Estimated through Grip
Force,” in 20th IFAC World Congress, Toulouse, France, 2017.
[136] K. van der El, D. M. Pool, M. M. van Paassen, and M. Mulder, “Effects
of Linear Perspective on Human Use of Preview in Manual Control,
IEEE Trans. on Human-Machine Systems, Under review.
[137] F. M. Drop, D. M. Pool, M. M. Van Paassen, M. Mulder, and H. H.
ulthoff, “Objective Model Selection for Identifying the Human Feed-
forward Response in Manual Control,” IEEE Trans. on Cybernetics,
Online preprint available.
[138] F. M. Drop, D. M. Pool, M. M. van Paassen, M. Mulder, and H. H.
ulthoff, “The Effect of Target Signal Shape and System Dynamics on
Feedforward in Manual Control,” IEEE Trans. on Cybernetics, Under
[139] A. Karniel and G. F. Inbar, “Human Motor Control: Learning to Control
a Time-Varying, Nonlinear, Many-to-One System,IEEE Trans. on
Systems, Man & Cyb., Part C: Appl.& Rev., vol. 30, no. 1, 2000.
[140] J. D. McDonnell, “A Preliminary Study of Human Operator Behavior
Following a Step Change in the Controlled Element,” IEEE Trans. on
Human Factors in Electronics, vol. 7, no. 3, pp. 125–128, 1966.
[141] D. C. Miller and J. I. Elkind, “The Adaptive Response of the Human
Controller to Sudden Changes in Controlled Process Dynamics,” IEEE
Trans. on Hum. Fact. in Electronics, vol. 8, no. 3, pp. 218–223, 1967.
[142] A. V. Phatak and G. A. Bekey, “Model of the Adaptive Behavior of
the Human Operator in Response to a Sudden Change in the Control
Situation,” IEEE Trans. on Man-Machine Systems, vol. MMS-10, no. 3,
pp. 72–80, 1969.
[143] R. A. Hess, “Modeling Pilot Control Behavior with Sudden Changes
in Vehicle Dynamics,” J. of Aircraft, vol. 46, no. 5, pp. 1584–1592,
[144] ——, “Modeling Pilot Detection of Time-Varying Aircraft Dynamics,”
J. of Aircraft, vol. 49, no. 6, pp. 2100–2104, 2012.
[145] ——, “A model for pilot control behavior in analyzing potential loss-
of-control events,Proc. of the Inst. of Mech. Eng., Part G: J. of Aer.
Eng., vol. 228, no. 10, pp. 1845–1856, 2014.
[146] P. M. T. Zaal and D. M. Pool, “Multimodal Pilot Behavior in Multi-
Axis Tracking Tasks with Time-Varying Motion Cueing Gains,” in
AIAA Mod. & Sim. Tech. Conf., National Harbor (MD), no. AIAA-
2014-0810, 2014.
[147] E. R. Boer and R. V. Kenyon, “Estimation of Time-Varying Delay
Time in Nonstationary Linear Systems: An Approach to Monitor
Human Operator Adaptation in Manual Tracking Tasks,IEEE Trans.
on Systems, Man, and Cybernetics – Part A: Systems and Humans,
vol. 28, no. 1, pp. 89–99, 1998.
[148] P. M. T. Zaal and B. T. Sweet, “Estimation of Time-Varying Pilot
Model Parameters,” in Proc. of the AIAA Modeling and Simulation
Technologies Conference, Portland (OR), no. AIAA-2011-6474, 2011.
[149] P. M. Thompson, D. H. Klyde, and M. J. Brenner, “Wavelet-Based
Time-Varying Human Operator Models,” in Proc. of the AIAA At-
mospheric Flight Mechanics Conf. and Exhibit, Montreal (CA), no.
AIAA-2001-4009, 2001.
[150] M. Olivari, F. M. Nieuwenhuizen, H. H. B¨
ulthoff, and L. Pollini,
“Identifying Time-Varying Neuromuscular System with a Recursive
Least-Squares Algorithm: a Monte-Carlo Simulation Study,” in Proc.
of the 2014 IEEE Int. Conf. on Systems, Man, and Cybernetics, San
Diego (CA), 2014, pp. 3573–3578.
[151] J. R. Schiess and V. R. Roland, “Kalman Filter Estimation of Human
Pilot-Model Parameters,” NASA Langley Research Center, Hampton
(VA), Technical Report NASA-TN-D-8024, Nov. 1975.
[152] A. Popovici, P. M. T. Zaal, and D. M. Pool, “Dual Extended Kalman
Filter for the Identification of Time-Varying Human Manual Control
Behavior,” in Proceedings of the AIAA Modeling and Simulation
Technologies Conference, Denver (CO), no. AIAA-2017-3666, 2017.
[153] J.-W. van Wingerden and M. Verhaegen, “Subspace identification of
Bilinear and LPV systems for open- and closed-loop data,” Automatica,
vol. 45, no. 2, pp. 372–381, 2009.
[154] R. T ´
oth, V. Laurain, M. Gilson, and H. Garnier, “Instrumental variable
scheme for closed-loop LPV model identification,” Automatica, vol. 48,
pp. 2314–2320, 2012.
[155] D. M. Pool, P. M. T. Zaal, H. J. Damveld, M. M. van Paassen, and
M. Mulder, “Evaluating Simulator Motion Fidelity using In-Flight and
Simulator Measurements of Roll Tracking Behavior,” in Proc. of the
AIAA Modeling and Simulation Technologies Conf. 2012, Minneapolis
(MN), no. AIAA-2012-4635, 2012.
[156] M. Mulder, T. Verspecht, D. A. Abbink, M. M. van Paassen, D. C.
Balderas, A. C. Schouten, E. de Vlugt, and M. Mulder, “Identification
of Time Variant Neuromuscular Admittance using Wavelets,” in Proc.
of the IEEE Int. Conf. on Systems, Man and Cybernetics, Anchorage
(AK), 2011, pp. 1474–1480.
[157] S. Van Eesbeek, F. C. T. Van Der Helm, M. Verhaegen, and E. De
Vlugt, “LPV subspace identification of time-variant joint impedance,
Int. IEEE/EMBS Conf. on Neural Engineering, pp. 343–346, 2013.
[158] D. W. Franklin, G. Liaw, T. E. Milner, R. Osu, E. Burdet, and
M. Kawato, “Endpoint stiffness of the arm is directionally tuned to
instability in the environment,The Journal of Neuroscience, vol. 27,
no. 29, pp. 7705–7716, 2007.
[159] M. Olivari, F. M. Nieuwenhuizen, J. Venrooij, H. H. B¨
ulthoff, and
L. Pollini, “Methods for multiloop identification of visual and neuro-
muscular pilot responses,” IEEE Trans. on Cybernetics, vol. 45, no. 12,
pp. 2780–2791, 2015.
[160] R. J. A. W. Hosman and S. K. Advani, “Design and evaluation of the
objective motion cueing test and criterion,The Aeronautical Journal,
vol. 120, no. 1227, pp. 873–891, 2016.
[161] R. A. Hess and F. Marchesi, “Analytical Assessment of Flight Simulator
Fidelity Using Pilot Models,” J. of Guidance, Control, and Dynamics,
vol. 32, no. 3, pp. 760–770, 2009.
[162] M. F. C. van Gool and H. A. Mooij, “A Comparison of In-Flight and
Ground-Based Pitch Attitude Tracking Experiments,” in Proc. of the
Twelfth Annual Conf. on Manual Control, 1976, pp. 443–454.
[163] D. M. Pool, G. A. Harder, and M. M. van Paassen, “Effects of Simulator
Motion Feedback on Training of Skill-Based Control Behavior,J. of
Guidance, Control, and Dynamics, vol. 39, no. 4, pp. 889–902, 2016.
[164] J. R. Anderson, “Acquisition of Cognitive Skill,” Psychological Review,
vol. 89, no. 4, pp. 369–406, 1982.
[165] J. Rasmussen, “Skills, Rules, and Knowledge; Signals, Signs, and
Symbols, and Other Distinctions in Human Performance Models,”
IEEE Trans. on Systems, Man, and Cybernetics, vol. SMC-13, no. 3,
pp. 257–266, 1983.
[166] K. R. Lohse and A. F. Healy, “Exploring the contributions of declarative
and procedural information to training: A test of the procedural rein-
statement principle,” J. of Applied Research in Memory and Cognition,
vol. 1, no. 2, pp. 65–72, 2012.
[167] J. C. F. de Winter, D. Dodou, and M. Mulder, “Training Effectiveness
of Whole Body Flight Simulator Motion: A Comprehensive Meta-
Analysis,” The Int. J. of Aviation Psychology, vol. 22, no. 2, pp. 164–
183, 2012.
[168] E. A. Martin, “The Influence of Tactual Seat-motion Cues on Training
and Performance in a Roll-axis Compensatory Tracking Task Setting,”
Interim Report AFRL RH-WP-SR-2009-0002, 2008.
[169] R. S. Shirley and L. R. Young, “Motion Cues in Man-Vehicle Control
– Effects of Roll-Motion Cues on Human Operator’s Behavior in
Compensatory Systems with Disturbance Inputs,” IEEE Trans. on Man-
Machine Systems, vol. 9, no. 4, pp. 121–128, 1968.
Max Mulder (M’14) received the M.Sc. degree and
Ph.D. degree (cum laude) in aerospace engineer-
ing from TU Delft, The Netherlands, in 1992 and
1999, respectively, for his work on the cybernetics
of tunnel-in-the-sky displays. He is currently Full
Professor and Head of the section Control and
Simulation, Aerospace Engineering, TU Delft. His
research interests include cybernetics and its use in
modeling human perception and performance, and
cognitive systems engineering and its application in
the design of “ecological” interfaces.
Prof. Mulder is an Associate Editor of the IE EE TR ANS ACT ION S ON
Daan M. Pool (M’09) received the M.Sc. and Ph.D.
degrees (cum laude) from TU Delft, The Nether-
lands, in 2007 and 2012, respectively. He is currently
an Assistant Professor with the section Control and
Simulation, Aerospace Engineering, TU Delft. His
research interests include cybernetics, manual ve-
hicle control, simulator-based training, and math-
ematical modeling, identification, and optimization
David A. Abbink (M’12, SM’14) received the M.Sc.
and Ph.D. degrees in mechanical engineering from
TU Delft, Delft, The Netherlands, in 2002 and 2006,
respectively. He is currently an Associate Professor
with the Delft Haptics Laboratory, TU Delft. His
current research interests include neuromuscular be-
havior, driver support systems, and haptics.
Dr. Abbink is an Associate Editor of the IEE E
Erwin R. Boer received the M.Sc. degree in elec-
trical engineering from Twente University of Tech-
nology, Enschede, The Netherlands, in 1990, and
the Ph.D. degree in electrical engineering from the
University of Illinois, Chicago, in 1995.
From 1995 to 2001, he was a Research Scientist
with Nissan Cambridge Basic Research, Cambridge
(MA), where he conducted basic research on driver
behavior. In 2001, he founded his own research
consulting company – Entropy Control, Inc. – in San
Diego (CA). His research interests include manual
control, driver modeling, human operator performance assessment, driver
support system design, and haptic shared control.
Peter M.T. Zaal received his M.Sc. and Ph.D.
degrees in aerospace engineering from TU Delft,
The Netherlands, in 2005 and 2011, respectively. His
Ph.D. research focused on pilot control behavior dis-
crepancies between real and simulated flight. Peter
is currently a senior research engineer with the San
Jose State University Research Foundation, perform-
ing collaborative research at the Human Systems
Integration Division of NASA’s Ames Research Cen-
ter. His research focuses on the modeling of human
visual and motion perception, and cue integration in
manual control tasks, as well as the development of objective motion cueing
criteria for flight simulators. His research interests include human factors,
manual control, visual and motion perception, and flight simulation.
Frank M. Drop was born in 1986. He received his
M.Sc. (cum laude) and Ph.D. degrees in aerospace
engineering from TU Delft, The Netherlands, in
2011 and 2016, respectively. His M.Sc. and Ph.D.
research focused on human feedforward manual
control, in a collaborative project between TU Delft
and the Department of Human Perception, Cognition
and Action led by Prof. Heinrich H. B¨
ulthoff at
the Max Planck Institute for Biological Cybernetics,
ubingen, Germany. His research interests include
cybernetics and modeling of human control.
Kasper van der El received the M.Sc. degree in
aerospace engineering (cum laude) from TU Delft,
The Netherlands, in 2013, for his research on manual
control behavior in preview tracking tasks. He is
currently pursuing the Ph.D. degree with the section
Control and Simulation, Aerospace Engineering, TU
Delft. His Ph.D. research focuses on measuring and
modeling human manual control behavior in general
control tasks with preview. His current research in-
terests include cybernetics, mathematical modeling,
and system identification and parameter estimation.
Marinus (Ren´
e) M. van Paassen (M’08, SM’15)
received the M.Sc. and Ph.D. degrees from TU Delft,
The Netherlands, in 1988 and 1994, respectively, for
his studies on the role of the neuromuscular system
of the pilot’s arm in manual control.
He is currently an Associate Professor at the
section Control and Simulation, Aerospace Engi-
neering, TU Delft, working on human-machine inter-
action and aircraft simulation. His work on human-
machine interaction ranges from studies of per-
ceptual processes and manual control to complex
cognitive systems. In the latter field, he applies cognitive systems engineering
analysis (abstraction hierarchy and multilevel flow modeling) and ecological
interface design to the work domain of vehicle control.
Dr. van Paassen is an Associate Editor of the IE EE TR ANS ACT ION S ON
... As controllers, humans have unique adaptive behavior capabilities, which allow them to lessen the effects of any incidents [1][2][3][4][5][6][7][8][9][10]. Understanding how pilots adapt their control strategy to maintain stability has several advantages. ...
... First, the 'Human Operator' block includes the human operator's outer-(position) and inner-loop (rate) control gains, and , respectively. In the model, controls the proportional control performed by the human operator, while defines the magnitude of the rate feedback, i.e., human operator 'lead' [8,17]. Additionally, human operator limitations are included in the form of an added time delay , i.e., = − , and the human operator's neuromuscular dynamics , representing the dynamics of the operator's limb-manipulator interaction, modeled as a second-order mass-spring-damper system: ...
... The main changes we implemented are listed below, with our motivation for why they were implemented: 1) Added human operator delay: As already shown in Fig. 1, a key addition we made to the model is a pure time delay on the modeled human operator response. As it is well-known that human operators in a pursuit tracking task will generally show a latency in their control response equivalent to that seen in compensatory tasks [8,17], this was found to be a required addition to match the data from the human participants in [10]. 2) Removed saturation on CE input: As also shown in Fig. 1, the original model implementation included a saturation block on the CE input (human operator control input), to account for potential manipulator constraints. ...
... While tracking tasks have indeed proven to reveal differences in motor performance between healthy controls and PD patients, their true effectiveness for monitoring motor symptom progression in individual patients has, to the best of our knowledge, never been explored. A key challenge for using tracking task data for functional diagnostics is that motor performance is generally characterized by a number of different metrics (e.g., performance metrics, human control model parameters (Mulder et al., 2018)), meaning that the detection of degraded motor performance is a multivariate problem. Furthermore, performance in tracking tasks inherently shows significant, and strongly participant-dependent, day-today variation, which needs to be adequately accounted for to reliably detect adverse trends in disease progression. ...
... To approximate a clinical longitudinal data set, an experiment with 25 healthy participants aged between 55-75, who each performed 50 trials of a pursuit tracking task on a large touchscreen with both their dominant and non-dominant hands, was used to collect representative 'motor symptom free' reference data. Participants' control dynamics in the tracking task were quantified with a fitted human control model (Mulder et al., 2018). To simulate a reduction in motor performance, the experiment data was augmented with bootstrapped PD data, where trends in human control model parameters as representative for PD, based on earlier research (Pool et al., 2022), were implemented. ...
... (1) was used, which is a well-known model for describing control dynamics in compensatory, as well as pursuit, tasks with single integrator control dynamics (Mulder et al., 2018). Using this approach, we obtained estimates of a participant's control gain K p , time delay τ , and neuromuscular damping ratio and natural frequency (ζ nms and ω nms , respectively) for all measurement trials. ...
Full-text available
Parkinson's disease (PD) is a progressive nervous system disorder that affects movement. PD has a severely negative impact on the quality of life of patients and their caregivers. The timing of treatment depends, amongst others, on the quantification of patients’ motor performance. To date, the resolution used in scaling motor performance is too low to detect subtle behavioral changes over time. This paper investigates if ‘longitudinal’ data-sets of motor performance data obtained from tracking tasks can detect behavioural changes in motor performance data representative for PD symptoms. Such longitudinal data were approximated using a combined data-set based on 50 trials of collected experiment data from 25 healthy participants (age range 55-75 years), augmented with 25 bootstrapped samples scaled to represent ‘Mild’ or ‘Severe’ motor performance degradation. An approach based on general linear regression models was tested for its capacity to detect the adverse trends in typical tracking task metrics (Kp, τ, ζnms, ωnms, RMSe, and RMSu). Overall, it was found that with this approach in at least 50% of all participants, a simulated change in motor behaviour was successfully detected, a number that may increase to 97% for the most sensitive metric (ζnms) and consistent participant data. This indicates that the developed approach is promising towards the development of more objective and detailed monitoring of disease progression and treatments in PD patients.
... In this era of increasing automation, in which for many applications human controllers are increasingly removed from the control loop, human controllers' ability for 'fast' adaptations to changes in task dynamics remains an essential source of inspiration for adaptive automation design (Phatak, 1969;Mulder et al., 2018). Humans are still unsurpassed when it comes the effectiveness of their control adaptations, which are generally fully realized in 2-5 seconds depending on the type of transition (Young et al., 1964). ...
... Humans are still unsurpassed when it comes the effectiveness of their control adaptations, which are generally fully realized in 2-5 seconds depending on the type of transition (Young et al., 1964). Of special recent interest are human control adaptations to sudden changes in the dynamics of the system that is controlled, i.e., the Controlled Element (CE) (Hess, 2009;Zaal, 2016;Mulder et al., 2018;Plaetinck et al., 2019). Young et al. (1964) identified three main phases of the human control adaptation process: 1) detection, where the human operator detects a change in CE dynamics; 2) identification, where the human operator correctly identifies the new CE dynamics; 3) modification, where operators adapt their own control dynamics to the new CE. ...
... Young et al. (1964) identified three main phases of the human control adaptation process: 1) detection, where the human operator detects a change in CE dynamics; 2) identification, where the human operator correctly identifies the new CE dynamics; 3) modification, where operators adapt their own control dynamics to the new CE. The key driver of this adaptation process is believed to be an 'internal model ' of their task that human operators build up and that enables them to notice when, for example, the characteristics of occurring control errors (e) and error rates (ė) change (Miller and Elkind, 1967;Phatak and Bekey, 1969b;Niemela and Krendel, 1975;Hess, 2009;Weir and Phatak, 1967;Mulder et al., 2018). This hypothesis for modeling the human adaptive process is formalized in the 'supervisory control algorithm' (Phatak and Bekey, 1969b,a;Phatak, 1969), which defines explicit limits on tracking error and error rate that trigger human operators to detect, identify, and adapt to a change in CE dynamics. ...
Conference Paper
This paper presents the results of an experiment that was performed to verify the ‘supervisory control algorithm’, a well-known model of human operator adaptation to changes in controlled element dynamics. This model proposes that human adaptive behavior is triggered once the magnitudes of the tracking error or error rate exceed certain decision region limits. In the experiment, a compensatory tracking task with a sudden transition in the controlled element dynamics, as also tested in other recent experiments, was performed by six skilled participants. In addition to performing the control task, participants had to indicate with a button press when they detected a controlled element transition. The results indicate that the published detection limits for the ‘supervisory control algorithm’ are too conservative for our experiment data, as measured detections could be related to error or error rate occurrences that exceeded 2-6 times their respective pre-transition standard deviations. The effectiveness of new detection limits proportional to these pre-transition standard deviations was tested. The best match to our experiment data was obtained with limits at 3.9σ, for which in only 9.38% and 11.5% of cases a (false positive) too early detection or a (false negative) missed detection occurred, respectively. Overall, these results demonstrate that human operator adaptation can indeed be effectively predicted from statistical variations in tracking error and error rate.
... The model is based on McRuer and Jex (1967)'s 'precision model ', can be applied to a range of different tasks (Van der El et al., 2016 and uses intuitive and identifiable parameters, such as a human look-ahead time. Essential for the development of such valuable HO models are accurate and reliable identification techniques that enable explicit separation of human feedback and feedforward control contributions (Mulder et al., 2018). ...
... Since in manual control the closed-loop magnitude decreases at higher frequencies, effects of manipulating the weightings of the closed-loop fitting cost values at these frequencies must be further analyzed. (Mulder et al., 2018). ...
... While a thorough understanding of Human Operator (HO) feedback control behavior exists, as well as validated models for predicting this behavior (McRuer and Jex, 1967;Mulder et al., 2018), the complex combination of feedback and feedforward behavior that characterizes HOs' behavior in pursuit and preview tracking tasks is less well understood. Recently, Van der El et al. (2016) proposed a novel linear time-invariant model that can accurately explain and predict behavior in pursuit/preview tracking tasks. ...
Conference Paper
Better understanding of manual control requires more research on human anticipatory feedforward behaviour. Recent advances include a human control model for preview tracking, and a subsystem identification (SSID) technique that uses a candidate pool approach to identify the human feedforward and feedback responses. This paper discusses the performance of the SSID method when estimating the preview control model parameters. Through simulations of a preview task with two controlled element dynamics, the SSID performance with different remnant noise levels and candidate pool densities is quantified. We demonstrate its successful application to the preview model and show that its performance deteriorates for higher noise levels. While the feedforward parameters are estimated accurately, the high-frequency compensatory feedback dynamics cannot be reliably determined. Future work focuses on alternative formulations for using SSID to estimate preview model parameters. Since in manual control the closed-loop magnitude decreases at higher frequencies, effects of manipulating the weightings of the closed-loop fitting cost values at these frequencies must be further analyzed.
... Quantifying human manual control skill level is difficult. Cybernetics has shown to be a useful tool to assess skill level (Pool and Zaal, 2016;, but most methods suffer from the fundamental limiting assumption that the human controller (HC) is constant over a substantial time interval, ignoring any 'fast' changes in control behavior (Mulder et al., 2018). Although efforts have been made to model time-varying HC behavior (Zaal and Sweet, 2011;Duarte et al., 2017;Rojer et al., 2019), capturing short-duration behavioral variations from inherently noisy data remains challenging (Mulder et al., 2018). ...
... Namely, a time series X is fed to a classifier that predicts the probability P 1. INTRODUCTION Quantifying human manual control skill level is difficult. Cybernetics has shown to be a useful tool to assess skill level (Pool and Zaal, 2016;, but most methods suffer from the fundamental limiting assumption that the human controller (HC) is constant over a substantial time interval, ignoring any 'fast' changes in control behavior (Mulder et al., 2018). Although efforts have been made to model time-varying HC behavior (Zaal and Sweet, 2011;Duarte et al., 2017;Rojer et al., 2019), capturing short-duration behavioral variations from inherently noisy data remains challenging (Mulder et al., 2018). ...
... Cybernetics has shown to be a useful tool to assess skill level (Pool and Zaal, 2016;, but most methods suffer from the fundamental limiting assumption that the human controller (HC) is constant over a substantial time interval, ignoring any 'fast' changes in control behavior (Mulder et al., 2018). Although efforts have been made to model time-varying HC behavior (Zaal and Sweet, 2011;Duarte et al., 2017;Rojer et al., 2019), capturing short-duration behavioral variations from inherently noisy data remains challenging (Mulder et al., 2018). ...
Conference Paper
Mathematical human controller (HC) models are widely used in tuning manual control systems and for understanding human performance. Typically, quasi-linear HC models are used, which can accurately capture the linear portion of HCs’ behavior, averaged over a long measurement window. This paper presents a deep learning HC skill-level evaluation method that works on short windows of raw HC time signals, and accounts for both the linear and non-linear portions of HC behavior. This deep learning approach is applied to data from a previous skill training experiment performed in the SIMONA Research Simulator at TU Delft. Additional human control data is generated using cybernetic HC model simulations. The results indicate that the deep learning evaluation method is successful in predicting HC skill level with 85-90% validation accuracy, but that training the classifier solely on simulated HC data reduces this accuracy by 15-25%. Inspection of the results especially shows a strong sensitivity of the classifier to the presence of remnant in the simulated training data. In conclusion, these results reveal that current quasi-linear HC model simulations, and in particular the remnant portion, do not adequately capture real time-domain HC behavior to allow effective training-data augmentation.
... In [20], measured HO control behavior was quantified using "cybernetic" HO models as proposed in [21,22]. In compensatory tracking tasks, see also [20]. ...
... control response with the tracking error as the input [22,23]: ...
... In a recent overview, Mulder et al. (2018) argue that cybernetics theory has predominantly focused on the lowest SOP level of human control, compensatory tracking, culminating in the widely-used and universally-accepted crossover model (McRuer and Jex, 1967). This model, and most of those that followed, allows to describe the steady-state feedback response of operators to unpredictable quasi-random target signals. ...
... Until Drop (2016) and , universal HC models for tracking with pursuit and preview displays, which allow the HC to exert a strong feedforward (and with predictive target signals even precognitive) response barely existed. Mulder et al. (2018) concluded that much more efforts should be put into understanding human control behavior in these operationally relevant tasks, and strive for describing the learning, adaptive human capabilities as these are the main reason humans are still included in the loop. Modeling adaptive behavior inevitably means moving ahead, from the current predominant linear time-invariant (LTI) models (Zaal et al., 2009) to models that allow including time-varying or even non-linear behavior. ...
Conference Paper
Mathematical human control models are widely used in tuning manual control systems and understanding human performance. Human behavior is commonly described using linear time-invariant models, averaging-out all non-linear and time-varying effects, which are gathered into the remnant. These models are limited in their capability to capture particular tracking strategies that an experienced subject may learn to use. In this paper, we consider manual control from a different perspective, namely through investigating the probability densities of the tracking error for different regions of the target signal amplitude. Results show that distinct strategies become apparent for compensatory, pursuit and preview tracking tasks. Effects of these strategies are often averaged-out by current models and can only be captured in situation-dependent models. Modeling this systematic human adaptation not captured in linear models could potentially lead to better model fits and explain/reduce part of the remnant.
... Modeling human behaviour in manual control has been studied for over 70 years, see review papers by McRuer and Jex (1967) and Mulder et al. (2018). Krendel and McRuer (1960) distinguish three stages of human control: compensatory, pursuit and precognitive. ...
... This model has served for decades to tune manual control devices and to understand human bio-dynamics and other phenomena. Mulder et al. (2018) argue, however, that research has focused too much on modeling feedback-only (FB) compensatory tracking, the exception in human control. More efforts are needed to study the rule of human control, that is, to model the versatile, adapting, biological system, using more relevant pursuit and preview displays (Drop, 2016;Van der El, 2018). ...
Conference Paper
Cyberneticists develop mathematical human control models which are used to tune manual control systems and understand human performance limits. Neuroscientists explore the physiology and circuitry of the central nervous system to understand how the brain works. Both research human visuomotor control tasks, such as the pursuit tracking task. In this paper we discuss some commonalities and differences in both approaches to better understand the adapting human controller. Special attention is given to Adaptive Model Theory, which studied adaptive human control using several linear and nonlinear control engineering techniques. The insights gained yield schemes and concepts which pave the way for key future work on this topic.
This paper investigates the potential of using a manual pursuit tracking task for quantifying loss of motor skills due to Parkinson's disease (PD), by applying human controller (HC) modeling techniques. With this approach, it is possible to obtain detailed quantitative data on motor performance in terms of control gain, response delay, stiffness and damping. Pursuit tracking data was collected from seven early-stage PD patients and a matched control group at the Erasmus MC in Rotterdam. Tracking performance was significantly worse in the PD group compared to the controls. Furthermore, the PD patients showed significantly lower control gains and degraded neuromuscular damping and bandwidth, which indicates that early-stage PD is associated with loss of quick and fast arm movements. While the PD patients showed less consistent and linear control behavior in the task, their data could still be modelled at high accuracy. Using HC models to quantify PD patients’ fine motor skill abilities may contribute to improved (early) detection of motor skill loss in PD, as well as detailed monitoring of symptom development and intervention effectiveness.
Full-text available
In the field of motor control, two hypotheses have been controversial: whether the brain acquires internal models that generate accurate motor commands, or whether the brain avoids this by using the viscoelasticity of musculoskeletal system. Recent observations on relatively low stiffness during trained movements support the existence of internal models. However, no study has revealed the decrease in viscoelasticity associated with learning that would imply improvement of internal models as well as synergy between the two hypothetical mechanisms. Previously observed decreases in electromyogram (EMG) might have other explanations, such as trajectory modifications that reduce joint torques. To circumvent such complications, we required strict trajectory control and examined only successful trials having identical trajectory and torque profiles. Subjects were asked to perform a hand movement in unison with a target moving along a specified and unusual trajectory, with shoulder and elbow in the horizontal plane at the shoulder level. To evaluate joint viscoelasticity during the learning of this movement, we proposed an index of muscle co-contraction around the joint (IMCJ). The IMCJ was defined as the summation of the absolute values of antagonistic muscle torques around the joint and computed from the linear relation between surface EMG and joint torque. The IMCJ during isometric contraction, as well as during movements, was confirmed to correlate well with joint stiffness estimated using the conventional method, i.e., applying mechanical perturbations. Accordingly, the IMCJ during the learning of the movement was computed for each joint of each trial using estimated EMG-torque relationship. At the same time, the performance error for each trial was specified as the root mean square of the distance between the target and hand at each time step over the entire trajectory. The time-series data of IMCJ and performance error were decomposed into long-term components that showed decreases in IMCJ in accordance with learning with little change in the trajectory and short-term interactions between the IMCJ and performance error. A cross-correlation analysis and impulse responses both suggested that higher IMCJs follow poor performances, and lower IMCJs follow good performances within a few successive trials. Our results support the hypothesis that viscoelasticity contributes more when internal models are inaccurate, while internal models contribute more after the completion of learning. It is demonstrated that the CNS regulates viscoelasticity on a short- and long-term basis depending on performance error and finally acquires smooth and accurate movements while maintaining stability during the entire learning process.
Conference Paper
Full-text available
Humans can rapidly change their low-frequency arm dynamics (i.e., stiffness) to resist forces or give way to them. Quantifying driver’s time-varying arm dynamics is important for the development of steer-by-wire systems and haptic driver support systems. Conventional LTI identification, and even time-varying techniques such as wavelets, fail to capture rapidly-varying low-frequency dynamics. In this study, we propose to estimate driver admittance in real-time, using grip force measurement of the hands on the steering wheel and linear parameter-varying (LPV) modeling techniques. We hypothesized that grip force is strongly correlated to neuromuscular admittance, and can serve as an appropriate scheduling variable for an LPV model. We performed an experiment in which 18 subjects performed a boundary tracking task, and applied torque perturbations to the steering wheel to perform a baseline LTI identification. Six different boundary widths were used to evoke changes in admittance, while their grip force was measured with pressure gloves. A global LPV model is identified by linear interpolation between the local LTI models identified for each boundary width. The estimated stiffness and damping parameters varied proportionally with the grip force. Although small between-subject variations in grip force levels are found, we conclude that grip force can indeed serve as an appropriate scheduling variable for a global LPV model, which is capable of tracking fast-changing admittance changes. Future work focuses on using the LPV model in realistic driving tasks, permitting admittance estimates to be obtained without the need to apply external disturbance torques on the steering wheel.
Full-text available
Due to linear perspective, the visual stimulus provided by a previewed reference trajectory reduces with increasing distance ahead. This paper investigates the effects of linear perspective on human use of preview in manual control tasks. Results of a human-in-the-loop tracking experiment are presented, where the linear perspective's horizontal and vertical