Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

Journal of Motor Behavior, Vol. 53, No.

6, 2021
# Taylor & Francis Group, LLC

REVIEW ARTICLE
Collective Variables and Task Constraints in Movement
Coordination, Control and Skill
Karl M. Newell1, Yeou-Teh Liu2
1
Department of Kinesiology, University of Georgia, Athens, Georgia, USA. 2Department of Athletic Performance, National
Taiwan Normal University, Taipei 116, Taiwan.

ABSTRACT. In this paper we review studies that have identi- challenges that the DF problem held for the prevailing
fied collective variables (order parameters) in movement coord- representational accounts of motor control and skill
ination, control and skill with emphasis on whole-body acquisition (engram, motor program, plan, schema). In
multiple joint degree of freedom (DF) tasks. Collective varia-
bles of a dynamical system have been proposed formally and our view, Bernstein’s problem remains fundamental and
informally from a diverse set of perceptual-motor tasks, from open today even if one considers as Latash (2012) has
which we emphasize: bimanual coordination, locomotion (ped- that there are positive even blissful outcomes from a sys-
alo, walking, running, bicycle riding), roller ball task, static tem with many DF, that include the facilitation of com-
(quiet standing) and dynamic (moving on a ski-simulator) bal-
ance, grasping, and juggling. Several types of candidate collect-
pensatory behavior (see also Lipsitz, 2002; Newell &
ive variables have been identified, including: relative phase, Vaillancourt, 2001; Huys et al., 2014 on the adaptive use
frequency ratio, number of hands active in grasping, synchrony, of multiple DF).
learning rate and relative timing. There is a strong influence of The coordination and control of DF became a depart-
the task goal in determining the collective variable that can be ure point for the introduction of a self-organization the-
body or environment relative. The emergence of the task rele-
vant collective variable is typically in the early stage of skill ory of movement coordination (Kugler et al., 1980;
learning where subjects through practice adapt movement Kugler & Turvey, 1987) “with minimal recourse to an
organization to realize a never previously produced movement intelligent regulator”. The umbrella construct of self-
coordination pattern. Throughout, the paper elaborates on open organization provided a natural basis for implementation
theoretical, experimental and analysis issues for collective vari-
ables in the context of task constraints and Bernstein’s (1967) of the principles and tools of nonlinear dynamics to the
view of skill acquisition as learning to master redundant DF. analysis of movement coordination and control. The
foundation of this approach, now known as coordination
KEYWORDS: Coordination, degrees of freedom, degeneracy,
whole body actions, movement, posture, learning, early stage
dynamics (Kelso, 2009), was the experimental findings
of learning, motor tasks, PCA on phase transitions in bimanual finger (Kelso, 1981)
and hand (Kelso, 1984) coordination, and the subsequent
creation of the HKB (Haken et al., 1985) model based
on Haken’s (1983) self-organization theory of synergetics
in physical systems.
Introduction
Indeed, the role of relative phase as the collective vari-
he introduction of Bernstein’s (1967, 1996) writings able in HKB and the evolving principles of coordination
T into the western world provided a new biophysical
framework of theorizing about motor control and skill
dynamics formed originally on the bases of within-per-
son movement of fingers, hands, arms and legs have
acquisition that contrasted sharply with the then develop- been shown to generalize to between-person and between
ing cognitive/representation accounts of movement and stimuli (auditory, visual, haptic) under a variety of rhyth-
skill (Adams, 1971; Fitts, 1964; Keele, 1968; Schmidt, mic movement settings (Fuchs & Kelso, 2018). At the
1975). A central feature of Bernstein’s agenda was the same time, there has been a broadening of the coordin-
challenge of how the many DF of the hierarchical system ation dynamics framework to the learning and perform-
at different levels of analysis (tonus, synergies, space ance of tasks other than the bimanual 2 DF protocol.
and action) are harnessed to produce coordinated and The mapping of brain dynamics to the movement coord-
skilled (dexterous) movement in action. Indeed, for ination dynamics has also become a research emphasis.
Bernstein (1967, p. 127), the problem of motor skill Our review is centered to the theoretical, experimental
acquisition is the mastery of redundant DF. and analytical developments of studies of collective
Turvey et al. (1978) elaborated on Bernstein’s DF variables (order parameters) in movement coordination,
problem by outlining estimates of the number of DF of control and skill and the learning and performance of
the perceptual-motor system: a number that rapidly whole-body multiple joint DF tasks. It will be evident
increases as one moves consideration from the macro
Correspondence address: Yeou-Teh Liu Department of
level of behavior and joint motions to the more micro
Athletic Performance, National Taiwan Normal University, 88,
levels, respectively, of muscles, motor units and neurons. Ting-Zhou Road Section 4, Taipei 116, Taiwan. Email:
Their theorizing pointed up the logical and material yeouteh@ntnu.edu.tw

770
Collective Variables

that there have been several research emphases to the analysis and quantifying the qualitative macroscopic
investigations of collective variables in motor learning movement forms of the system at hand (Haken, 1996).
and control: including formal dynamical systems model The construct of a collective variable has also been
construction, tests of macro level variables in coordin- used in the motor control domain as a synonymous term
ation dynamics, together with formal and informal identi- to that of an order parameter (Sch€oner & Kelso, 1988a,
fication of the collective variable for a given task. 1988b, 1988c). Like an order parameter, the construct of
collective variable originated in physics but has taken on
Collective Variables in Movement Theory subtle variations of interpretation according to the discip-
and Experiment line of inquiry. Mitra et al. (1998) defined a collective
variable in the movement domain as one that captures, in
Essential Variables, Order Parameters and unitary fashion, the pattern of coordination and its spe-
Collective Variables cific spatial and temporal details (see also Turvey, 2004).
It has been conjectured that the collective variable is
Bernstein (1967, 1996) considered coordination to
unlikely to be a basic variable of biomechanics but rather
reflect an activity that guarantees that a movement has
an abstract relational (informational) variable specific to
homogeneity, integration and structural unity. The appli-
the coordination for the action in question (Kelso, 1994;
cation of systems’ approaches to human movement has
Mitra et al., 1998). Kelso (2009) through the develop-
facilitated the integrated analysis of the qualitative and
ment of coordination dynamics interpreted collective var-
quantitative dynamics and the decomposition of the role
iables generally as relational quantities that are created
of system variables in motor control. There is a small set
by the cooperation of the individual parts of the system
of related strands of theoretical influence to the deter-
(see also Kelso & Fuchs, 2016).
mination of the variables that organize the macroscopic
Collective variables have been hypothesized by Mitra
dynamics of human movement.
et al. (1998) to reflect the topological structure formed in
Kugler et al. (1980) originally drew on the complex
the early coordination stage of skill acquisition that has
system formulations of Gel’fand and Tsetlin (1962) to
decompose the system variables into those that are essen- long been loosely interpreted, without experimental
tial and those that are nonessential. In this view, the examination, as a cognitive stage focused on ‘getting the
essential variables determine the qualitative properties of idea’ of the task (Fitts, 1964; Gentile, 1972). From a
the coordination pattern (as reflected in the relational dynamical viewpoint, however, the participant in the
variables of coordination – Turvey, 1990), whereas the early stage of learning is searching directly or indirectly
nonessential variables produce changes in the scalar val- through practice to realize the task goal and relevant
ues of the function. In essence, Kugler et al. (1980) pro- movement pattern of the coordination dynamics as cap-
vided an embedded account of the emergent properties tured in the emergence of the collective variable at the
of coordination, control and skill (Newell, 1985). macroscopic level (Liu et al., 2019).
Coordination is the function that constrains the poten- There may be a need for more than one collective
tially free variables into a behavioral unit. Control is the variable to adequately capture the global structure of a
process by which values are assigned to the variables in system (Haken, 1983). In the perceptual-motor skill
the function (i.e., parameterizing the function). Skill domain this proposition seems particularly relevant for
reflects the approximation of optimal values be assigned so-called whole-body multiple DF tasks. To date, how-
to the controlled variables. ever, the respective claims from papers are all one as to
The HKB (1985) model applied the construct of an the number of collective variables, except for the 2-order
order parameter to movement coordination and control parameter model of Frank et al. (2009) for one and two
that was expressed in the stability and instability of hand grasping and the PCA analysis of walking and run-
bimanual relative phase (see Haken, 1996 for more back- ning (Lamoth et al., 2009) that are discussed later in the
ground and detail on the order parameter construct). paper. Thus, in certain perceptual-motor skills we may
Relative phase is a coordination variable in that it cap- be searching for collective variables, though still a low
tures the space-time coupling between the motion of the dimensional solution (emphasis 1 to 3 collect-
individual component DF (e.g., index fingers) and ive variables).
reflects the macroscopic organization of the bimanual Bernstein (1967) considered skill as the mastery of
task. The notion of an order parameter as a macroscopic redundant DF and focused on the potential planar and
variable has its origins in physics and the transitions range of motions of the component joint DF. In contrast,
between solids, liquids, and gases where it has been in coordination dynamics the emphasis has been on the
expressed as a measure of the degree of order across the dynamical or functional DF of the attractor dynamics
boundaries in a far from equilibrium system. The effi- rather than the configuration of joint space. The number
cacy of the order parameter construct depends on formal- of dynamical DF will typically be smaller than the num-
izing the relevant microscopic and macroscopic levels of ber of joint DF engaged in the action (Kay, 1988; Kay

2021, Vol. 53, No. 6 771


K. M. Newell & Y.-T. Liu

et al., 1991). This distinction also allows the DF of the the individual movement repertoire prior to learning the
component joint space (elements) to retain their integer new task (Zanone & Kelso, 1992, 1997).
whole number status while the functional DF of the Warren (2006) within the ecological theory of percep-
attractor dynamics can take on partial numbers reflecting tion and action has proposed task categories that are
its fractal dimension (Newell & Vaillancourt, 2001). based on the stability and in-stability of the movement
Presently, the terms essential variable, order parameter dynamics. The task categorization framework hinges on
and collective variable are used interchangeably in the the relative contribution of physical constraints and infor-
movement domain. We adhere to using the term collect- mational constraints to the control of the task. Three
ive variable in the remainder of this paper. classes of perceptual-motor tasks were identified. One is
stabilizing a passively stable system such as in bouncing
Task Influence on the Embedding of Coordination, a ball on a racket (Sternad et al., 2001). In this task type,
Control and Skill the physics of the context dominates control of the action
and the coordination solution is found through percep-
The constructs of coordination, control and skill offer
tual-motor search strategies. A second categorization is
a framework for the categorization and decomposition of
active stabilization of an unstable system, such as in bal-
a particular class of motor tasks. There have been mul-
ancing a pole (Foo et al., 2000). In this task category
tiple theoretical and practical efforts to categorize motor
there is strong contribution of both physical and informa-
skills ostensibly for the purposes of understanding the
tional constraints for a successful task outcome. Third,
generalization of motor learning and control across tasks
and finally, there is the active stabilization of a neutrally
(e.g., Fleishman et al., 1984; Gentile, 1987; Poulton,
stable system, such as in braking while driving a car.
1957; Schmidt & Lee, 2012; Warren, 2006 – see Magill
Motor learning and control in this task category is domi-
& Anderson, 2014 Chapter 1 for an overview).
nated by task relevant information and, in our view, has
Nevertheless, task categorization frameworks have been
generally (though unwittingly) been the task category of
largely descriptive and not directly relevant for a dynam-
choice in the study of perceptual-motor skills.
ical coordination perspective.
The learning of representative motor tasks from the 3
In motor learning, we can view coordination tasks as
model categories of Warren (2006) is likely to induce
those that require directly or indirectly the learning of a
different functions of change in the dynamics of learning
new coordination pattern as defined by a relational prop-
(Newell et al., 2001). The relation between task catego-
erty of movement dynamics not previously realized by
ries, attractor dynamics and behavioral change still has
the participant (Kugler et al., 1980; Newell, 1985;
many open issues, however, including the hypothesized
Saltzman & Kelso, 1987; Turvey, 1990). Scaling tasks
reciprocal (or circular) causality of the collective variable
are those that require a different parameterization to a
with the motions of the individual DF (Kelso, 1995).
learned coordination function that the individual can
Thus, in spite of the general recognition of the role of
already produce – as, for example, in the related but dif-
task constraints in motor learning and control the formal
ferent assumptions of schema theory on generalization
link to the dynamics of behavior beyond the 2 DF of
within a movement class (Schmidt, 1975). The majority
bimanual control, is still in its early stages. Nevertheless,
of motor learning and control research has focused on
the dynamical systems approach to task decomposition
the scaling of an already formed movement coordination
provides a framework to model task dynamics within the
pattern and a collective variable has not typically been
coordination, control and skill perspective (Beek & van
hypothesized or sought.
Santvoord, 1992; Newell, 1986; Newell et al., 2001;
The task types reflect the embedded processes of
Saltzman & Kelso, 1987).
coordination, control, and skill that are constrained by
the task goal in conjunction with the individual and
Ecological Theory of Perception and Action
environmental constraints (Newell, 1986) that collect-
ively may specify directly or indirectly the respective The perception/action links between Gibson’s (1979)
variables for a solution of the coordination function. This theory of direct perception and Bernstein’s (1967) bio-
background is relevant because recent studies have physical account of movement provided the beginnings
shown theoretically and experimentally that the function of the ecological theory of perception and action (Turvey
of learning is different for the fixed-point attractor & Kugler, 1984; Turvey, 1992; Turvey et al., 1981).
dynamic tasks of movement scaling (Newell et al., 2001, From this background, there has been rich theoretical
2006; Newell, Liu, et al., 2009; Newell, Mayer-Kress, and experimental development to many aspects of move-
et al., 2009) than for those tasks that induce transitions ment (Haken et al., 1985; Kelso, 1995; Sch€oner &
(Liu et al., 2006, 2010; Liu et al., 2012). In a related Kelso, 1988c; Turvey, 1990) and perception (Lee, 2009;
approach, Kostrubiec, Zanone, Fuchs, and Kelso (2012) Profeta & Turvey, 2018; Turvey, 1992, 2004; Warren,
have shown that the rate of change (drift or shift) in 2006) in action. Indeed, in the theory of ecological
learning a new pattern of coordination is dependent on psychology, perception and action are complementary

772 Journal of Motor Behavior


Collective Variables

constructs to the extent that it is difficult to investigate theory of movement and action by interpreting coordin-
them independently in the perception-action cycle ation and control in the 5 task categories from a multiple
(Gibson, 1979; Profeta & Turvey, 2018; Kugler & levels of construction perspective of brain-behavior
Turvey, 1987; Warren, 2006). Historically, nevertheless, organization.
theories of perception and action have been largely inde- Indeed, in this view certain kinds of tasks are con-
pendent and many factors have led to what Turvey trolled at the level of space, for example, in catching a
(2004) called ‘the great divide’ of perception and action. fly baseball (Fink et al., 2009; Zaal & Michaels, 2003),
Sch€
oner and Kelso (1988a, 1988b, 1988c) in early the neuromuscular synergies of leg motions will be con-
papers from the evolving domain of coordination dynam- strained to a background role rather than a leading role
ics proposed a dynamic that was continuous in its formu- in the hierarchy of control. That the Bernstein (1996,
lation of perception and action as in a M€obius band 1947/in press) levels of space and action can take on a
(Turvey, 2004). A significant feature for the theme of more leading role according to the environmental and
this paper is that it brings the constructs of perception task demands (Profeta & Turvey, 2018) is a proposition
and action together under the constraint of the collective consistent with the perceptual theory of Gibson (1979),
variable (f). The proposed dynamic from Sch€ oner and but has received little experimental investigation. This
Kelso (1988c) is: line of theorizing also supports the position that task
 constraints (Newell, 1985, 1986) play an important role
f_ ¼ L f, ci (1) in determining the relevant collective variable. In prin-
where the temporal change of the collective variable is a ciple, however, task constraints are not synonymous with
function of the variable itself subject to both specific and collective variables, though as we amplify later in the
nonspecific parameter influences, ci. The collective vari- paper they could be in the special circumstances of
able f reflects the abstract but coherent relations among some tasks.
the parts and processes of the system and is context The ecological theory of perception and action
dependent. Kelso and Fuchs (2016) have proposed a together with the related domain of coordination dynam-
coordination dynamics model of baby – environment ics has provided a foundation for a dynamical systems
coupling in the learning of the Rovee-Collier mobile account of motor learning (Beek, 1989; Huys et al.,
contingent reinforcement task. 2004a, 2004b; Fowler & Turvey, 1978; Sch€oner &
The emphasis in the ecological framework of princi- Kelso, 1988a, 1988b; Sch€oner et al., 1992; Mitra et al.,
ples from Gibson’s (1979) theory of direct perception 1998; Newell, 1985; Newell et al., 2001; Sch€oner, 1989;
leads to the position that assessing collective variables in Zanone & Kelso, 1992) and development (Adolph &
actions generally will need to consider not only move- Hoch, 2019; Kugler, 1986; Kugler et al., 1982; Newell,
ment relative to the body, as in the HKB model, but also 1986; Thelen et al., 1987; Thelen & Smith, 1994; Thelen
to the environment (Turvey, 1992, 2004; Warren, 2006). & Ulrich, 1991). Central to this effort is a coherent
In other words, the collective variable will need to cap- framework in the form of time scales to investigate the
ture movement coordination of the participant to proper- various processes of change over time associated with
ties of the environment in the context of the task motor learning and motor development (Newell et al.,
demands (Newell, 1986). Moreover, the collective vari- 2001). The more recent research in these areas has
able for some perceptual-motor tasks will not necessarily extended the common dynamical ground of change proc-
be based directly on the neuromuscular or component esses in motor learning and development to a lifespan
joint motion DF, that have long been viewed in biomech- view of the dynamics of movement development (Newell
anics and movement physiology as the foundational & Liu, 2014; Newell & Morrison, 2016).
dimensions of analysis for motor control.
In this context, it is relevant that Bernstein’s (1996) The HKB (1985) Model of Bimanual Coordination
anatomically based hierarchy of control in action had 4 The HKB (1985) model was developed based on the
inter-related levels of organization: namely, tonus, syn- novel experimental findings of Kelso in the bimanual
ergy, space and action (see Profeta & Turvey, 2018 for coordination of rhythmic horizontal index finger (Kelso,
an ecological psychology contemporary view on 1981) and hand (Kelso, 1984) movements. The essence
Bernstein’s levels of control). This framework has been of the experiments was similar with a metronome pacing
revised in Bernstein (1947/in press) to 5 levels adding at the two fingers or two hands to oscillate initially at an
the top of the hierarchy - symbolic coordinations. in-phase or anti-phase pattern with the frequency being
Different task emphases in movement coordination and increased every few seconds from 1.25 Hz up to 3.5 Hz
control, however, can change the relative influence of in small steps for the finger protocol and from 1-5 Hz in
levels in the hierarchy and, in particular the level of con- the hand experiment. The instructions to subjects empha-
trol that is leading momentarily the system organization. sized maintaining one full cycle of motion for each beat
In effect, Bernstein brought task constraints into his of the metronome. And, that should they feel that the

2021, Vol. 53, No. 6 773


K. M. Newell & Y.-T. Liu

finger or hand movement pattern may change, subjects


should not prevent it from doing so but rather adopt the
most comfortable pattern in preserving the beat.
The experiments revealed four key phenomena that
were the basis for determining the equation structure of
the HKB model (Haken et al., 1985). First, there are
only two stable bimanual coordination patterns between
the fingers and also between the hands – anti-phase and
in-phase. Second, there is an abrupt transition from anti-
phase to in-phase at a critical frequency. Third, beyond
the transition frequency only the in-phase condition is
stable. Fourth, when the oscillation frequency is reduced
the initially prepared anti-phase condition does not return
to an anti-phase pattern.
A central factor for designating relative phase as the
collective variable was that participants stably repro- FIGURE 1. Landscape view of the attractor dynamics
duced the in-phase and anti-phase conditions under low arising from the bimanual HKB model (Haken
et al., 1985).
frequency initial conditions but when the frequency was
scaled up to a critical point the anti-phase condition tran-
sitioned to in-phase. In contrast, when the initial condi-
tion was in-phase it was preserved over the full preceding the transition, in addition to evidence of hys-
frequency range for that coordination pattern. The quali- teresis, both indices of a nonlinear dynamical system.
tative asymmetrical transition in the candidate collective Equation (2) reflects the potential landscape of the stabil-
variable from anti-phase to in-phase (and not vice-versa) ity/instability of the collective variable shown in Figure 1:
as a function of control parameter frequency provides V ð/Þ ¼ acosð/Þ  bcosð2/Þ (2)
strong evidence, to supplement the theoretical rationale,
that relative phase is the collective variable for this where V is the potential function, / is the relative phase
bimanual system. between the fingers and the ratio of b/a is related to the
The experimental transition presaged the creation of a frequency of finger oscillation.
formal experimental strategy for determining the collect- The bimanual task with two joint DF provides a con-
ive variable for a given perceptual-motor skill. As theor- trolled but restricted context to determine the collective
etical rationale, HKB were influenced by the finding in variable but this operational benefit in turn limits the
many physical systems that phase is an accurate reflec- generality of the dynamical analysis in the context of
tion of the cooperativity among the system components. Bernstein’s (1967) proposition of skill acquisition being
This observation also relates to a principle from the the- the mastery of redundant DF. In the experiments of
ory of synergetics (Haken, 1996) that the configuration Kelso (1981, Kelso, 1984) the task is a body relative
of the subsystems specifies their phase, and conversely, skill with the collective variable of relative phase of the
that the phase variable specifies the spatial-temporal bimanual effectors that also is the task goal and the
ordering of the components (a relation labeled reciprocal bimanual synergy for the task (Liu et al., 2019). The
causality – Kelso, 1994, 1995). Another influential linking of relative phase to the goal of the task is, how-
observation in arriving at the position of relative phase ever, an inference because relative phase is not men-
as the collective variable for bimanual tasks was that tioned directly in the instructions to subjects. Indeed,
phase remains invariant across transformations of many what the subjects perceive as the goal in this task and
motor activities. This phenomenon is consistent with the motor skill tasks more generally is an understudied prob-
principle that the collective variable tends to change lem in spite of the prevalence of oral instructions in
more slowly than the individual components to the influ- laboratory experiments and, moreover, in clinical practice
ence of control parameters. Moreover, non-linear coupled by change agents in the contexts of teaching, coaching
oscillator models fit well with relative phase as the col- and therapeutics.
lective variable. These experimental constraints give emphasis to the
The HKB model provides for potentially two stable role of the intrinsic dynamics in the coordination task
relative phase regimens (in-phase and anti-phase) (Sch€oner & Kelso, 1988a, 1988b). However, they restrict
depending on frequency that forms the landscape of the in a number of ways the possible mappings between the
intrinsic dynamics of the collective variable. variable(s) reflecting the task goal, the bimanual neuro-
Furthermore, the stochastic version of the model muscular coordination synergy and the collective variable
(Sch€oner et al., 1986) predicts enhanced fluctuations that affords degenerate coordination solutions evident in

774 Journal of Motor Behavior


Collective Variables

many perceptual-motor skills. Indeed, task decomposition critical value separating regimens of integration and sep-
has revealed the potential of several layers of structure in aration between groups.
the movement dynamics and outcome each with their The modeling of the data contrasted the extended
own time scale (Beek & van Santvoord, 1992; Huys HKB model with the longstanding large-scale model of
et al., 2004a; Newell et al., 2001) to the extent that the Kuramoto (1984). The outcome was that the resultant
distinction of task goals/constraints and the candidate integrative model that captures all key experimental
collective variable can be difficult to determine. observations happens to also connect theories of small-
The HKB (1985) model of bimanual coordination and large-scale biological coordination in a single math-
introduced the relative phase between the oscillatory ematical formulation. This study shows not only the
horizontal movements of the two index fingers as the generalization of HKB to social communication but the
order parameter that reflected the macroscopic organiza- core place of the extended version of HKB and relative
tion of the coordination solution in terms of its stability/ phase in biological coordination.
instability (see also Kelso, 1990). The oscillation fre- Coordination dynamics has elaborated the synergetic
quency of the fingers was used as the control parameter pattern formation framework from motor control into the
that moved the system through stable/unstable regions of broader but complementary consideration of the self-
state-space including at a critical frequency inducing in organization of brain and behavior (Haken, 1996; Kelso,
the anti-phase preparation condition a transition to in- 1995, Kelso, 2012; Kelso & Engstrom, 2006). The cen-
phase. Since then there have been many experimental tral thesis of the brain-movement-behavior program is
tests of features of the HKB model (e.g., Fuchs & Jirsa, that “the brain is fundamentally a pattern forming self-
2008; Jirsa & Kelso, 2004; Kelso, 1995, 2009; Molenaar organized system governed by potentially discoverable
& Newell, 2003) and experimental extensions (e.g., nonlinear dynamical laws” (Kelso, 1995, p. 257). The
theoretical and experimental approach, as in motor con-
visuo-motor tracking – Wimmers et al., 1992; isometric
trol, is mesoscopic or macroscopic depending on the
force tracking, Lafe et al., 2016).
brain recording apparatus with the emphasis on pattern
The HKB model has also led to several theoretical
formation of brain cell assemblies, transitions between
elaborations (e.g., stochastic - Sch€oner et al., 1986; sym-
states, and how macro states of the brain emerge out of
metry breaking – Kelso & Jeka, 1992; oscillators with
cooperative processes.
different eigenfrequencies – Fuchs et al., 1996; multifre-
A series of brain-behavior experiments on phase tran-
quency tapping – Haken et al., 1996). And, by extending
sitions has shown that the experimental search strategy
the tenets of coordination dynamics, Huys et al. (2014)
for the emergent collective variable(s) of a brain cell
developed a framework for the analysis of sequential
assembly and its transition in principle follows the same
motor behavior (e.g., handwriting) based on the con-
pathway of change as discussed previously seeking the
structs of structured flows on manifolds and units of
dynamical principles of behavioral motor control (see
behavior. These theoretical and experimental contribu- Appendix A). A strength of the experiments is that they
tions, including the role of relative phase as a collective build from previous and substantial experimental work in
variable, have contributed to solidifying and extending the behavioral manipulations thus inspiring confidence in
the scope of HKB and coordination dynamics. the expected behavioral outcomes and associated brain
A more recent focus of coordination dynamics has activity. The major question in initial work was whether
been social coordination between people (Zhang et al., the relative phase profile of brain activity would match
2018; Zhang et al., 2019). The coordination dynamics of that of the behavioral data in a single finger syncopation
HKB has typically drawn on coupled nonlinear oscillator task (Kelso, 1997). The initial stimulus tone started at
models where the number of oscillators is N ¼ 2 given 1 Hz increasing every 10 tones in small steps.
the bimanual task. At the other extreme are models of Using Squid (superconducting quantum interference
populations of individual oscillators where the number device) Technology for measurement of brain activity it
of oscillators can be quite large, for example, in swarms was found that there was a spontaneous transition in
of fireflies and fish. In a novel protocol dubbed the coordination that occurred at the same critical frequency
‘human firefly’ experiment Zhang et al. (2018, 2019) in both motor behavior and brain signals (Jirsa et al.,
had subjects engage in a midscale social coordination Kelso, 1994; Kelso, 1995). There is a critical frequency
protocol in an ensemble (N ¼ 8) group. The subjects where subjects cannot sustain the syncopation that leads
could not see each other but they could see the flash to a transition to synchronize with the stimulus.
from their own tap and that of other group members. Moreover, there was a strong relation in a low dimen-
The tapping task was done under regimens of varying sional frame of reference between the macroscopic ana-
frequencies and continuation tapping segments without lysis of the brain activity and that of the movement
the stimulus. Quantitative changes (diversity) in fre- profile. The role of both anticipation processes and sen-
quency led to qualitative changes in coordination with a sorimotor coordination was evident in the simple

2021, Vol. 53, No. 6 775


K. M. Newell & Y.-T. Liu

experimental protocol. These findings support the inter- strategy that has been implemented and operationalized
pretation that the coherence of both brain and behavior most fully in Kelso (1984) may be found in
signals can be captured by the same collective variable, Appendix A.
namely, relative phase and, that there is an isomorphism In spite of this explicit experimental strategy, there
between brain and behavior events that cuts across the has been modest generalization of the construct of a col-
fact that different things are being coordinated (Kelso, lective variable from HKB to other motor tasks. One rea-
1995, 1997). Fuchs and Kelso (2018) provide an over- son for this we conjecture is that the identification of the
view of relative phase as the collective variable in a collective variable for tasks with multiple redundant joint
number of experimental task protocols, including an indi- space DF is not such a straightforward enterprise as the
vidual ballet dance sequence. determination of relative phase of 2 DF in the restricted
2 planar finger motions of the bimanual task (Haken,
Collective Variables Beyond the HKB Model 1996). Even contemplating an ‘ansatz’ in some tasks is a
As noted previously, there could be more than one, if challenge. Moreover, actions that engage multiple joint
not several macroscopic variables, that are candidate col- space DF are more likely to have a larger pool of candi-
lective variables for a given perceptual-motor skill. The date collective variables. Laboratory tasks still tend to be
challenge is to isolate a candidate collective variable or informationally impoverished and motorically restricted
variables, evaluate validity for the respective function leaving them poorly placed or in the extreme case unable
and distinguish it (as appropriate) from the component to approach conditions to investigate Bernstein’s (1967)
and synergy variables in realizing a given task outcome. position of the centrality of the redundant DF problem to
This can be achieved formally through the theoretical motor skill acquisition.
and experimental principles of coordination dynamics
(Kelso, 1995, Kelso, 2009). Or it can be pursued infor- Relative Phase
mally by the experimenter via an ‘ansatz’ as to the col- Although the HKB model continues to hold its rele-
lective variable for a given task. It is probable that these vance and robustness there is no a priori reason to
informal and formal approaches can and do work assume that a particular relative phase will necessarily be
together where the informal strategy provides a first pass the collective variable of the perceptual-motor system of
hypothesis (conjectured relation between an independent interest. On the other hand, as Kelso (1995, p. 52) indi-
variable and a dependent variable) that can be put subse- cated, relative phase has the virtue of being a single vari-
quently to more formal experimental test. Theory, of able that captures not only the observed movement
course, also provides the operational motivation for pattern but also transitions between them. And, in this
hypothesis testing. view only the phase relation appeared to fulfill these
requirements for the collective variable as outlined in
Informal Haken et al. (1985). This observation may turn out to be
The informal strategy has the experimenter make an task specific in the sense that to date it fits well with
‘ansatz’ of a macroscopic property that may reflect the nonlinear coupled oscillator models of rhymical move-
collective variable for the task. It follows that one also ment but it may be less relevant in other task categories,
interprets the data through this working assumption with- such as discrete arm movements in prehensile tasks (see
out necessarily any validity checks that the experimental forthcoming section on grasping).
tests support a particular macroscopic property as the However, even if one pursues a relative phase as the
collective variable. This can be limiting or even prob- collective variable in the multivariate DF action context
lematic because identifying a macroscopic variable for one is still left with the question of which relative phase
the system dynamics is a necessary but not sufficient – that is, how does one know the right one? This ques-
means to determining a collective variable in a dynam- tion is relevant because in whole body actions such as
ical systems framework. In other words, the identified quiet standing with many joint space DF engaged there
macroscopic variable can by definition be macroscopic are likely to be several possible relative phases of move-
but it may not be the actual collective variable for the ment kinematics that could be considered as a macro-
task at hand. scopic variable (Bardy et al., 1999; Kilby et al., 2015).
The relative phase (collective variable) of the two fin-
Formal ger motions in HKB (Haken et al., 1985) is that of a
The formal identification of the collective variable(s) body relative coordination property. Indeed, the classic 2
for a task requires a substantial experimental effort DF bimanual protocol of the HKB model is a closed
guided by the theoretical modeling of dynamical sys- skill (Poulton, 1957) except for the change in frequency
tems. Sch€oner et al. (1992) and Kelso (1995) have pro- conditions, and the initial trial conditions of the move-
vided a set of procedures for the identification process of ment pattern (anti-phase or in-phase) are in essence the
the collective variable. A precis of the experimental goals of the task. The limited component DF of two

776 Journal of Motor Behavior


Collective Variables

bimanual finger motions reduces the options on a candi- As a consequence, it is useful to heed that the statistic-
date collective variable but it also leads to the circum- ally derived principal components are not necessarily the
stance that the task goal can be redundant with the same as the collective variables that are identified from
synergy relation and collective variable between the the dynamics of physical systems, such as relative phase
bimanual effectors. in the HKB model. In PCA the experimenter is, in effect,
Many perceptual-motor skills are, however, open skills letting the statistics on the variance and covariance of
organized with respect to their relation to the environ- the dependent variables run their course to determine
ment (Poulton, 1957), that has varying degrees of unpre- post hoc the number and nature of the independent prin-
dictability. The ecological theory of perception and cipal components. In the dynamical systems framework
action holds that there is an individual-environment rela- by comparison, the candidate collective variable is con-
tive coordination property that captures the collective jectured a priori through theoretically based task decom-
variable of the task (Turvey, 1992, 2004; Warren, 2006). position and determined by subsequent dynamical tests.
And, in parallel to the ecological framework, for this set The uncontrolled manifold (UCM – Scholz & Sch€oner,
of constraints to action, the spatial-environmental 1999) follows this dynamical approach but, in addition,
demands of the task can become the leading level of links a priori a relevant restricted task space of stability
control in Bernstein’s (1996) hierarchy. Moreover, in this to analyze the contributions of the joint motions.
situation the task goal and the neuromuscular synergy Haas et al. (1995) used PCA to determine the number
motions will typically be distinct from that of the collect- of components as a function of skill level in the pedalo
ive variable, because the mapping between these action locomotion task (see also Haken, 1996). It was shown
frames of reference is not necessarily direct. that, under the single step cycle approach to analysis
rather than the more typical gait time series over many
Principal Component Analysis (PCA) step cycles, that only one component was required with
The linear multivariate statistic of principal component a ‘relatively’ skilled performer to characterize the per-
analysis (PCA) has been used as a formal strategy to formance. This implies, in effect, that all the torso and
compress the number of variables in a movement ana- limb motions were highly synchronous in performing
lysis (Daffertshofer et al., 2004; Forner-Cordero et al., the task. The data on the pedalo task also showed that
2005). Of additional relevance here is that PCA can also the number of components reduced with practice and the
enhancement of skill.
act in a complementary way to revealing features of
In contrast, Chen et al. (2005) found that 3–5 compo-
coordination dynamics that inform about the macroscopic
nents were required by individuals to capture 90% of the
variable(s) of a movement system including those with
performance variance in the pedalo task. There were
many DF (Daffertshofer et al., 2004; Haken, 1996; Witte
probably many differences in the experimental set-ups of
et al., 2009). Principal component analysis, also known
Haas et al. (1995) and Chen et al. (2005) including skill
as the Karhunen-Loeve (KL) expansion, or singular
level of the participants. Moreover, there are a consider-
value decomposition, has been used extensively to deter-
able number of analytic decisions involved in executing
mine low dimension spatial patterns to brain and move- and interpreting a PCA (Daffertshofer et al., 2004;
ment time series (e.g., Jirsa et al., 1998). One can also Molenaar et al., 2013), including the possibility that the
run PCA directly in the frequency domain as opposed to divergent findings arise from one study using the correl-
the time domain (Molenaar et al., 2013). ation matrix and the other study the co-variance matrix.1
With the PCA technique, the many variables of input The net outcome is that the determination of the col-
to the analysis (e.g., joint motions) can be compressed lective variable(s) is or can look rather different in the
into lower dimensional independent components that relative phase and PCA approaches to a given task. The
reflect the variance of a linear combination of the input relative phase approach whether formal or informal has
variables with maximal variance accounted for. In this to date provided a single variable as the collective vari-
approach, the number of components and their variance able. With PCA one derives a number of independent
accounted for provide information as to the nature of the components that are a linear combination of the variance
system organization (dimension and weighting of each accounted for by the input variables. In most cases, the
variable to the component collective). It also provides a single variable relative phase approach will tend to be
basis for constructing the projection of the components easier to interpret than the PCA component strategy
into a dynamical frame of reference and potentially (Forner-Cordero et al., 2005) but it may be a simplifica-
evaluate in terms of relative phase or some other dynam- tion that over compresses the contribution of relevant
ical variable (Daffertshofer et al., 2004; Lamoth et al., variables. On the other hand, the intuitive hypothesis of
2009). The strengths and limitations of this approach to the collective variable may be the relevant simplification
deriving relative phase from the PCA remain, however, for determining the macroscopic structure of the move-
to be evaluated. ment output.

2021, Vol. 53, No. 6 777


K. M. Newell & Y.-T. Liu

The nature of what preliminary information the com- tool to not only compress a data set with many variables
ponents provide in terms of a collective variable, how- (its original basic function) but also to provide prelimin-
ever, becomes more challenging the greater the number ary information that is relevant to understanding the
of variables in the component linear combination and the coordination dynamics, including the macroscopic prop-
more irregular the variable time series. This is in contrast erties of the task. This is particularly the case in terms of
to the more regular motions of in-phase and anti-phase interpreting the relations of the projections of the princi-
that tend to map well with the outcome of a PCA. In the pal components in dynamical terms (Daffertshofer et al.,
former PCA situation, the clarity provided by the 2004; Lamoth et al., 2009).
‘ansatz’ of the relative phase of two variables, is masked.
Expressed another way, with the PCA approach, the Collective Variable of Learning Rate
number of higher order components required to capture Frank et al. (2008) have formalized a quantitative
the variance can be determined but the interpretation for dynamical systems approach to the development of a
motor learning and control of what these components self-organization principle and collective variable equa-
represent through the variable weightings and projections tions of a training regimen for motor skill learning called
of the components can remain a challenge. differential learning (Sch€ollhorn et al., 2009).
Differential learning is assumed by Frank et al. (2008) to
Haken, Friedrich, Uhl Order Parameter Analysis be a self-organized process that results in the emergence
Building on the PCA approach, Haken (1996) outlined of subject- and context-dependent attractors. This theor-
a dynamical strategy (the Haken, Friedrich, Uhl order etical framework it is argued contrasts with that of trad-
parameter analysis) to the determination of a proposed itional learning that is taken to be driven by external
order parameter(s) for the whole-body motion (multiple factors and the formation of environmentally specified
joint space DF) pedalo task. The Haas et al. (1995) PCA attractors (Frank et al., 2008).
analysis to the pedalo decomposition had provided pre- The modeling is based on an experimental investiga-
liminary evidence that a few if not just one component tion of traditional and differential learning regimens in
captured the data and that the number of significant com- subjects performing the shot putt task (Beckmann, 2003;
ponents was reduced with practice. From this preliminary Beckmann & Sch€ollhorn, 2008), where the differential
PCA background analysis of the movement data regimen showed a greater rate of change in performance
Equation 3 was created for examination in the context of over training time. Performance improvement during
the pedalo data: post-training periods is explained by a hysteresis effect.
The derived collective variable equation is a fourth order
nu ¼ Aeixt þ h:o: (3)
polynomial potential. The collective variable with this
where nu is a single complex order parameter, x is the approach is in effect the learning rate.
fundamental frequency ¼ 2p/T, T is the duration of one The modeling strategy here to the determination of the
period and h.o. is the higher harmonics that are assumed collective variable differs from the physical systems rela-
to be negligible. The dynamical analysis led to the con- tive phase agenda introduced earlier (Kelso, 1995), and
clusion that one collective variable was enough on many as we will see in upcoming sections, also other
trials to represent the data that were single step cycles approaches to collective variables in motor learning and
rather than the more usual continuous time series con- control. Indeed, learning rate is traditionally a derived
taining a series of cycles of locomotor activity. variable from the change in performance over time.
Witte et al. (2003, 2009) used the Haken – Friedrich – Learning rate is not a macroscopic variable of coordin-
Uhl method in providing evidence that human walking ation that reflects the structural integrity of the action
was captured by a single order parameter. Again, the sin- (Bernstein, 1967; Turvey, 1990) though it may be a
gle step cycle approach was used in the data analysis. In product of it. Moreover, there are multiple processes in
contrast, Lamoth et al. (2009) reported that a PCA of learning and as a consequence multiple time scales of
timeseries of human walking and running found both change (Newell et al., 2001).
movement forms were dominated by the variance of
essentially the same first two principal components,
Experimental Examples of Collective Variables in
whereas a small level of variance in principal compo-
Perceptual-Motor Skills
nents three and four distinguished the gait patterns.
Clearly, there remain a number of methodological There have been two primary experimental protocols
issues in the use of PCA in the study of coordinated used in the determination of the collective variable in
movement that need to be addressed to adequately exam- perceptual-motor skills. One protocol is a learning para-
ine the theoretical issues of collective variables and their digm and the investigation over practice time of the
proposed role in reciprocal causality of multiple DF emergence of the collective variable that characterizes
movement coordination. The PCA is a useful analytic the macroscopic integrity of the coordination mode. The

778 Journal of Motor Behavior


Collective Variables

FIGURE 2a. Example of a successful roller ball trial that increases the ball velocity of the initial conditions.

2021, Vol. 53, No. 6 779


K. M. Newell & Y.-T. Liu

FIGURE 2b. Example of an unsuccessful roller ball trial – the ball velocity did not get above the velocity of the initial
ball conditions.

780 Journal of Motor Behavior


Collective Variables

HKB model. As anticipated the participants through


practice reduced error around the 90-degree task goal
either in an incremental change rate labeled ‘drift’ or a
more abrupt change rate labeled ‘shift’.
To assess change in the overall landscape of the intrin-
sic dynamics as a function of practice on the 90 degrees
task goal a scanning technique was devised to determine
the change in the variability of relative phase over the
landscape of zero to 180-degree phase relation. This
technique was used to assess the stability/instability pro-
file of relative phase over a wide range of phase rela-
tions and determine how the landscape changes as a
function of practice and learning at the 90-degree task
(Kelso, 1995; Sch€oner et al., 1992; Zanone & Kelso,
1992). Evidence was provided that the whole HKB land-
scape changed with practice and not just the local
dynamics at 90-degree relative phase.
More recent experiments in this paradigm investigated
the learning of phase relations other than 90 degree and
the impact of transfer conditions on relative phase as a
FIGURE 3. Kinematic time series of a single subject
function of the intrinsic dynamics of the system (Zanone
single trial of joint and body motions to the influence of & Kelso, 1997). From these studies an additional gener-
an oscillating platform that was increasing in frequency alization of the HKB model for these bimanual learning
to the same amplitude (Adapted from Ko et al., 2014). and transfer relative phase phenomena has been pre-
sented (Kostrubiec et al., 2012). Moreover, and import-
antly for the focus here, the learning protocols of
coordination dynamics provided further evidence that
second protocol is a performance paradigm where the
relative phase is the collective variable for the bimanual
participants can perform the perceptual-motor skill com-
coordination tasks.
petently and the investigator seeks to determine the col-
lective variable (s) for this task and participant skill
Roller Ball Task (Liu et al., 2019)
level, together with the relevant control param- The roller ball (Gulick & O’Reilly, 2000; Higbie,
eter influences. 1980) provides practice and learning conditions in which
the participant has typically not previously produced the
Motor Learning and the Emergence of
collective variable for the task. In this situation, the par-
Collective Variables
ticipants are not successful in the early practice trials and
Here we highlight example studies of learning to form producing the task goal of increasing the initial speed of
a novel coordination pattern in perceptual-motor tasks. the ball. Indeed, in contrast to many laboratory experi-
We give emphasis to three papers that have followed to ments, considerable practice is generally required to get
varying degrees the set of procedures from coordination the ‘idea of the task’ as reflected in searching to main-
dynamics for formally identifying the collective variable. tain or enhance the ball velocity from its initial level
All of these experimental examples concluded with a sin- (Liu et al., 2006; 2010), and substantial further practice
gle relative motion measure as the collective variable for to become skilled or dexterous in Bernstein’s
the respective motor task. (1996) terms.
We were attracted to the roller ball task as a research
Learning 90 Degree and Other Relative Phase tool because our preliminary testing indicated that there
Relations in Bimanual Coordination (Kelso, 1995; was a transition associated with learning the task that
Sch€ oner et al., 1992; Zanone & Kelso, 1992) took participants from the stable state of not being able
This bimanual learning paradigm was a natural out- to perform the task through the unstable state of search-
growth of the experiments of Kelso (1981, Kelso, 1984) ing for the task relevant coordination pattern that cap-
and the resulting stability/instability landscape of the tures the structural integrity for successfully performing
intrinsic dynamics of bimanual coordination. The new the task, and finally to the new learned stable state of
task to be learned was a particular relative phase (e.g., task success. The task requires participants to hold with
90 degree) that was unstable (greater fluctuations in rela- one hand the outer shell of a ball and to turn that
tive phase) as revealed in the intrinsic dynamics of the through motion of complex arm-hand kinematic chain,

2021, Vol. 53, No. 6 781


K. M. Newell & Y.-T. Liu

FIGURE 4. In 4a and 4b shows platform motion on trial 1 day 1 and a late practice single trial 20 of day 7 of a novice
subject. Figure 4c shows the cophase of the individual novice subjects on the first 3 practice trials of day 1 (adapted from
Dutt-Mazumder & Newell, 2018).

so as to try to increase the acceleration of the initial more direct test of whether relative phase is the collect-
inner ball speed that was the initial condition for the ive variable for the task rather than relying on a prob-
task. The early experiments used probability of task suc- ability of success score as we had done in the earlier
cess as the performance outcome from the control experiments. Figure 2a and b show typical individual tri-
parameters and independent variables of initial hand con- als of success and failure and their time series character-
figuration, ball speed, and amount of practice. istics, respectively.
The results were consistent in showing that partici- The findings provide strong support for the synchrony
pants went through a practice induced nonequilibrium (relative frequency) of the inner ball and outer shell as
phase transition at the task outcome level (Liu et al., the collective variable in this task. The initial segment of
2010). This phase transition led to the task outcome per- a trial is focused on searching for the way to control ball
formance function of a sigmoid curve and what is known speed as reflected in the random-like phase wrapping
traditionally as S-shaped learning (Liu & Newell, 2015) (see Figure 2a and b). The more skilled the subject the
– a learning function that is atypical given the prevalence shorter this exploratory search duration at the beginning
of an exponential or power-law function for the learning of the trial (Newell et al., 1989) and the time to the onset
of laboratory scaling tasks (Newell, Liu, & Mayer- of ball-shell synchrony. The kinematic analysis showed
Kress, 2006). that the spatial property in relative phase of the coupling
In more recent experiments we have implemented sen- of the inner ball and outer shell was not critical to the
sors to record in real time the velocity of the inner ball task leaving synchrony (timing) as the collective variable
and outer shell motions (Liu et al., 2019). This allows a capturing the relation between motion of the inner ball

782 Journal of Motor Behavior


Collective Variables

(environment) and outer shell (mapped to body by fixed better than the novice group – a performance difference
position of the shell to the hand). Synchrony is an estab- that was reduced considerably over practice. Indeed, the
lished collective variable in the physical sciences interpretation was that with further practice the obtained
(Kuramoto, 1984; Strogatz, 2003; Strogatz & group effect of experience on performance would be
Stewart, 1993). eliminated altogether.
The experiment of Liu et al. (2019) also revealed Of primary interest in the study, however, was the
degeneracy in movement organization whereby the par- effect of experience and practice on the changing coord-
ticipants used one of three neuromuscular synergies (par- inative dynamics and in particular the relative phase of
ticular wrist, elbow, shoulder organizations) to execute platform and CoM. On the initial trials of day 1 the
performance. This shows that the collective variable can experienced group all showed anti-phase motion of CoM
be distinct from the neuromuscular synergy and the task and platform. In contrast three participants of the novice
goal in perceptual-motor skills that require the coordin- group showed initially on trials 1-3 of day 1 an in-phase
ation of several DF, including functional coupling with relation of CoM and platform. With an in-phase relation
the environment (Warren, 2006). We used synergy in the of CoM and platform the participant was acting in an
Bernstein (1967) tradition of muscular-articular configu- inverted pendulum mode that restricts whole body lateral
rations while recognizing that this construct has since motion but magnifies it relatively in the torso and head.
taken on more general interpretations (Kelso, 2009; However, these three participants by trial 4 had all
Latash, 2008; Profeta & Turvey, 2018; Turvey, 2007). switched to anti-phase motion of CoM and platform
allowing the transition to a hanging pendulum organiza-
Moving on a Ski Simulator (Dutt-Mazumder & tion of the system whereby though the amplitude of plat-
Newell, 2017, 2018) form motion is large the motion of the torso and head is
The ski-simulator is a laboratory task that requires per- relatively small (see Figure 3). That the transition of the
formers to stand on a platform and to generate through collective variable took place after very few trials sug-
their own body side to side (medial-lateral) motion with gests that this stable movement pattern was available to
smooth, large amplitude and high frequency oscillations be assembled by the individual even on trial 1 (as in the
of the platform. The task has been used to examine a HKB model).
number of issues in motor learning and control including This study did not follow all of the experimental steps
Bernstein’s (1967) DF problem (Hong & Newell, 2006; of the Kelso (1995) framework for determining the col-
Teulier et al., 2006; Vereijken, 1991; Vereijken et al., lective variable. In effect, an ‘ansatz’ was tested from
1992). The dynamic balance ski-simulator task not only the anticipated generalization of findings from an exter-
allows but requires implicitly in a performance context nally pulsed platform (Dutt-Mazumder & Newell, 2018;
that the center of mass (CoM) for segments of the ski- Ko et al., 2014) and a participant pulsed platform of a
simulator cycle be outside of the stability boundary typ- learning paradigm (Hong & Newell, 2006; Vereijken,
ically associated with the task constraint of 1991; Vereijken et al., 1992). Furthermore, practice con-
quiet standing. sidered as a control parameter reduces the capacity of
It was with this background that Dutt-Mazumder and the experimenter to control the trial to trial dynamics and
Newell (2018) investigated the learning of a candidate performance of the participant. And, unlike other experi-
collective variable for the ski-simulator task. mental control parameters such as frequency in bimanual
Extrapolating from postural transition studies on a com- coordination, one cannot necessarily reverse with a short
puter controlled moving platform (Ko et al., 2014) the time scale the consequence of adaptation effects
hypothesis was tested that the relative phase of CoM to from practice.
platform in the medial-lateral plane was the collective
variable for this task. It was anticipated that the ski- Collective Variables of Performance Tasks
simulator task requires an anti-phase relation between Here we highlight experimental examples from the
CoM and platform for the production of a successful per- study of posture, locomotion and grasping that have
formance (Hong & Newell, 2006) because an in-phase reported evidence of an emergent collective variable for
relation limits the amplitude of platform motion that pre- the performance of a given perceptual-motor task.
serves a stable solution.
In Dutt-Mazumder and Newell (2018) experienced and Static and Dynamic Balance
novice down-hill skiers practiced the ski-simulator task The maintenance of balance in standing posture is one
over 7 days of practice. The standard task outcome of the fundamental motor skills that infants typically
measures of platform amplitude and velocity showed a learn through the sequence of prone progression in first
practice by experience interaction whereby both groups 18 months of life and continues to evolve over a range
improved their movement outcomes over practice but the of environmental conditions and task demands through
performance of the experienced group was initially much the life span. Winter (2009) and colleagues have outlined

2021, Vol. 53, No. 6 783


K. M. Newell & Y.-T. Liu

the basic control features of standing balance and the frequency with the individual standing aligned to anter-
laboratory protocol that is now known experimentally as ior-posterior (AP) motion (Ko et al., 2014). The focus
‘quiet standing’. was on the coupling between the center of mass (CoM)
The experimental work on posture has focused on the and center of pressure (CoP) as a candidate collective
motion of the CoM and that of the center of pressure variable that supports maintaining balance on a sinus-
(CoP) in quiet standing (static balance). This is because oidal oscillating platform in the AP plane and was con-
the stability of standing posture is viewed as dependent tinuously scaled up and then down across a frequency
on the preservation of the vertical projection of the CoM range from 0.2 Hz to 1.2 Hz. The CoM-CoP coordination
within the base of support. In this framework, the CoP changed from in-phase to anti-phase and anti-phase to
reflects the location on the surface of support of the col- in-phase at a critical frequency (0.4 Hz to 0.6 Hz,
lective neuromuscular input (reactive force) to control respectively) in the scaling up or down of the support
the motion of CoM. Thus, the received position is that surface frequency. The CoM-CoP relative phase also
the CoM is the variable that is controlled while CoP showed hysteresis as a function of the direction of fre-
reflects the controlling variable. Indeed, the low fre- quency change and critical fluctuations at the transition
quency (< 1 Hz) components of CoM and CoP within a region (see Figure 4). These findings of CoM-CoP
quiet standing trial tend to be highly correlated (Wang coupling have been replicated in a similar oscillatory
et al., 2014; Winter, 2009). platform motion experiment in the ML plane (Dutt-
Does it follow, therefore, that the CoM-CoP relative Mazumder & Newell, 2017). The approach outlined here
motion is the collective variable for standing posture? to postural analysis allows the consideration of static and
The CoM by definition is a macroscopic variable reflect- dynamic balance in the same frame of reference rather
ing the location of the collective of forces on the neuro- than the traditional strategy of considering them as dis-
muscular system in the action under consideration. tinct tasks.
Similarly, the CoP reflects the location at the base of
support of the collective reactive force to standing from Walking, Running
the inputs of all the joint space DFs. It is important to It is established that speed of walking relative to leg
note that unlike the variables discussed to date in regard length can act as a control variable for the transition to
to collective status, CoM and CoP are not joint motions the gait pattern of running (Alexander, 1992). Less theor-
or muscle activity DF, although both CoM and CoP etical and experimental attention has been given to deter-
result from the collective influence of the joint motions. mining the collective variable(s) for walking and running
This observation highlights the need to consider carefully perhaps because of the biomechanical emphasis on the
the frame of reference for the coordination problem at kinematic and kinetic details of the limb and torso
hand and the determination of the collective variable. motions, rather than their macroscopic organizational
In our view, the long-standing approach of using properties. Here, as an exemplar, we consider in more
standard deviation of CoM or CoP motion alone is not detail than in an earlier section, the PCA study of walk-
sufficient to characterize the macroscopic topological ing and running by Lamoth et al. (2009) for insights as
torso and joint motion relations of standing posture. It to the collective variable for walking and running. These
would not, for example, distinguish quiet standing from authors did not explicitly couch or interpret their study
quiet hanging from a bar. A further limitation masking a in terms of collective variables but as noted previously
unique determination is that quiet standing is typically in the Haken (1996) framework the component analysis
studied in young adults that can already stand under of PCA provides a window into possible collective varia-
relatively natural conditions and, indeed, have done so bles for the task.
over many years of practice. A manipulation in infants Lamoth et al. (2009) investigated the multi-segment
of practice as a control parameter through the sequential and stride characteristics of walking and running on a
stages of postural stability, instability, and stability of treadmill. Their main focus was to examine the coordin-
postural control has not been conducted. In quiet stand- ation patterns and associated variance as a function of
ing, anterior-posterior (AP) CoP motion is typically walking and running speed, particularly around the gait
in-phase to CoM while medial-lateral (ML) motion is transition of walking to running. There were 10 belt
anti-phase to CoM. speeds that were incrementally increased and then
Our studies in dynamic standing balance tasks have decreased with 60 s at each speed. 3D markers at 16
provided preliminary evidence of CoM-CoP phase rela- standard anatomical sights provided 48 time series for
tion as an integrative organizing (collective) variable for each trial. Participants, as in the original finger wiggling
standing posture (Dutt-Mazumder & Newell, 2017; Ko experiment (Kelso, 1981), were instructed to not resist
et al., 2014; Wang et al., 2014). Subjects stood on a changing gait patterns should they feel this occurring.
force platform located a top a computer-driven moving Standard stride characteristic data were analyzed, in add-
platform that oscillated a given amplitude and scaled in ition to, a PCA analysis of whole-body segment motions.

784 Journal of Motor Behavior


Collective Variables

PCAs were conducted for walking (PCAw), running joint motions and other action variables. But according
(PCAr) and walking to running (PCAwr) trials. The anal- to the research question this omission will be more or
yses showed that 4 components accounted for 65-70% of less material. There is, also no reason why both measure-
the variance for walking and running, respectively. ment approaches, relative phase and PCA, could not both
Moreover, the first 2 components were similar in both be implemented. Finally, one can get the step cycle ratio
walking and running while the first component repre- for walking and other gait patterns with a reduced ana-
sented the fundamental frequency of the respective gait lysis of the whole-body kinematics. Hoenkamp (1978) in
pattern. The results also showed a small but consistent a perceptual judgment experiment has shown that young
effect of speed on the 3rd and 4th principal components. adult subjects can perceive the gait pattern (walking, run-
In essence, the study found that the coordination patterns ning, skating) of stick figure simulations through the
of walking and running are largely similar with small relative motion of the hip/knee in the step cycle.
differences in the definition and composition
of components. Grasping
Lamoth and colleagues emphasized the advantage of a The hand-arm skeletal-muscular complex has many
formal multivariate algorithmic analysis to determine the structural-functional DFs that provide a robust but flex-
kinematic patterns in walking and running, and their ible system for coordination and control of a broad rep-
transition, rather than an informal ‘ansatz’ of the relevant ertoire of fine motor skills involving reaching and
collective variable. The focus here is their PCA deter- grasping (Connolly, 1998; Jones & Lederman, 2006).
mination of the independent components of walking and Most of the joint DFs of the hand-arm complex are in
running as this relates to the theme of collective varia- the digits of the hand but the wrist, elbow and shoulder
bles (see also Phinyomark et al., 2015; Witte et al., joints provide additional movement potential and flexibil-
2010). Lamoth and colleagues (2009) recovered the rele- ity for a broad and rich diversity of actions (Bernstein,
vant relative phases in the coordination pattern through 1967; Zatsiorsky, 1998), including musical skills, hand-
projections of the principal components, but, as said, the writing, typing, drawing, painting and neurosurgery. The
interpretation of the PCA findings to collective variable prehensile activities provide contexts and challenges to
determination was not explicitly addressed. address Bernstein’s (1967) DF problem that on the sur-
The relative phase of variables in the frames of refer- face may seem distinct from that for the whole-body
ence of work and energy may also be relevant to identi- activities but that have increasingly been interpreted and
fying the collective variable, beyond the established experimented with similar theoretical constructs
kinematic invariants of gait patterns. For example, walk- and methods.
ing is characterized by out-of-phase oscillations of kinet- The ecological approach to perception and action
ics and gravitational potential energy of the body center stimulated several lines of the experimental study of
of mass (CoM), whereas in running these mechanical hand function, including the grasping of objects (e.g.,
energy components fluctuate in-phase as in the inverted Cesari & Newell, 1999, 2000; Newell et al., 1989;
pendulum and mass-spring paradigms (e.g., Segers et al., Newell et al., 1989; van der Kamp et al., 1998;
2007; Segers et al., 2013). There has also been dynam- Wimmers et al., 1998). These experiments on body-scale
ical modeling of locomotion-respiration coupling information for grasping were motivated from the theory
(Daffertshofer et al., 2004). Mapping the dynamics of direct perception (Gibson, 1979) and the construct of
between the kinematic and work/energetic levels of the animal-environment niche - the fit of the individual
walking and running remains to be investigated, as it to the environment. The experimental finding of the tran-
does for most actions and the learning thereof. sition of the perception of the task-relevant mode of stair
A contrasting approach to the PCA decomposition of climbing, as a function of scaling riser height (Warren,
walking is the ‘ansatz’ route of proposing what the rele- 1984), also stimulated body scaling movement experi-
vant collective variable is for walking. Thelen and col- ments in action more broadly (e.g., aperture passing,
leagues (Thelen & Ulrich, 1991; Thelen & Smith, 1994) reaching, hurdle clearing).
pursued this strategy in a study of infants (7–9 months A central question in these studies of grasping was
old) learning to walk on a treadmill. They decomposed whether body-scaled information specifies a particular
the skill of walking on the basis that the task has step grip configuration for the affordance or goal of the
alternation and essentially symmetry across the left and action? And, given a positive position on this - what is
right steps of the body. Thus, a 0.5 phase relation in feet the nature of that body-scaled information? In the grasp-
motion reflects these two criteria for walking. Indeed, ing experiments, the participant looks at the to-be-picked
executing this ratio in the step cycle is consistent with up object before the hand begins to move toward it
the definition of walking. through to the completion of moving the object to a new
Of course, this pursuit of step cycle relative phase as location. Thus, it is information that is picked up visually
the collective variable gives up information details about that is specifying the task-relevant properties for reaching

2021, Vol. 53, No. 6 785


K. M. Newell & Y.-T. Liu

and grasping together with the transitions of grip problem of the coordination and control of multiple DF.
configurations. These studies include some formal but predominantly
The above experiments on grasping had the relative informal analyses of a candidate collective variable:
size (ratio) of the object to the hand formally or infor- social movement coordination (synchronization between
mally as the control parameter. This control parameter people of common body motion, Schmidt, Fitzpatrick,
manipulated over a sufficient range of relative object Caron, & Mergeche, 2011); table tennis (stroke action -
sizes can induce transitions so that multiple grip mode forehand or backhand, Sorensen, Ingvldsen, & Whiting,
configurations of hand(s) and digits can emerge particu- 2001); badminton - speed scalar product between
larly in the middle range of size parametrization. Thus, motions of pairs of players (Chow et al., 2014); tennis
in certain parameter ranges of the control parameter, the (relative phase of the motions of two players, Palut &
grip configuration pattern transitions in terms of the use Zanone, 2005); horse-back riding (Witte et al., 2009);
of one hand, two hands, and digit configuration (# and and the game of basketball (offensive rating, Garcia-
type) within and between hands. Complementary experi- Rubio et al., 2015).
ments on transitions in prehension from the framework
of nonequilibrium phase transitions and coordination Challenges in Determining and Interpreting the
dynamics have also been conducted (Kelso et al., 1994). Collective Variable Agenda
Frank et al. (2009) on the basis of the above experi-
mental findings created a dynamical systems model of In this section we address some open issues arising in
grip transitions when object size was scaled as a control regard to collective variables in movement coordination,
parameter. The study modeled the collective variable control and skill. The emphasis is toward the role of col-
dynamics of body scale hysteresis and grip mode transi- lective variables in the context of Bernstein’s (1967) pos-
tions of two experimental conditions: one to two hands tulation that skill acquisition is the mastery of redundant
and two hands to one hand. The model was derived from DF. Although, Bernstein (1967) and Bernstein, 1996) did
the Haken (1991) neural network model for pattern rec- not discuss the general notion of macroscopic variables
ognition that has been used extensively in experiments in motor learning and control or the particular instanti-
on facial recognition. ation of essential variables as outlined by Gel’fand and
The model for grasping has two collective variables: Tsetlin (1962).
one hand and two hands that are acting as nominal varia-
bles providing the structure to the range of grip configu- Emergence of Collective Variables in the Early Stage
rations over object size. This leads to development of the of Motor Learning
landscape for the probability of a one or two hands solu- Mitra et al. (1998) proposed that the formation of the
tion. The model provides a novel solution in the motor collective variable was the dynamical instantiation of
domain in that there are two collective variables and that getting the “idea of the task” (Fitts, 1964; Gentile,
they are not relational measures. Thus, the model is a 1972). This early stage of emerging coordination has
qualitative solution that has given up some details pro- been largely by-passed in motor learning research due to
vided from the prior experiments on the hand to object the task selection for experimental study that has typic-
ratio. In the model the deterministic aspect of grasping ally required merely the rescaling of a coordination
behavior is completely realized by the two types of mode that the participant has already produced. This
fixed-point dynamics that specify one hand and two hand strategy has restricted the scope of the study of percep-
grasps. The collective variable in this grasping protocol tual-motor skill learning, in effect, to the acquisition of
is formed on a different basis from that of relative phase non-optimal control (Newell, 1985) or intermediate
and the dynamics of HKB and bimanual control. motor learning (Mitra et al., 1998). Thus, the experimen-
Finally, it should be noted that the Frank et al. (2009) tal studies of motor skill acquisition have focused on a
modeling of grasping places the visually induced grasp- narrow segment of the dynamical landscape as reflected
ing transitions inspired from Gibson (1979) into the in the prevalence of monotonic performance curves with
same action category as the motor control non-equilib- practice and learning (Newell et al., 2006).
rium phase transition approach of HKB. This proposed In contrast, the field of motor development has given
integration is open to experimental test in a grasp- emphasis to the progressive development from concep-
ing protocol. tion to maturation of the fundamental movement skills
(Gesell, 1952; Thelen & Smith, 1994; Wickstrom, 1977).
Other Collective Variables This emphasis gives focus to the emergence of new
There have been a number of other perceptual-motor movement forms or macroscopic coordination patterns
tasks studied in the context of identifying collective vari- and more generally the qualitative change over time in
ables. Several of these are sport tasks where the whole- dynamics of children’s movement behavior. Young chil-
body movement demands invoke the Bernstein (1967) dren, therefore, are progressively learning and adapting

786 Journal of Motor Behavior


Collective Variables

the fundamental motor skills through, in effect, the chal- locomotion is only relevant in the context of water but
lenge of realizing new functional goals through action this may not be the case in space.
(e.g., Thelen et al., 1987). The study of infant motor skill There are likely additional dynamical aspects to this
acquisition through the emergence of the fundamental getting “the idea of the task” stage that are separate but
movement forms lends itself to a dynamical decompos- related to the emergence of the collective variable. These
ition (Newell et al., 2003; Newell, 2020; Thelen & would include perceiving the task goal and the relevant
Smith, 1994; Warren, 2006). constraints on performance and learning (Newell, 1986),
A question arising is whether the collective variable and in Gibson’s (1979) terms, perceiving what the situ-
for a given task changes through the life span of an indi- ation affords for action. Nevertheless, in our view, the
vidual’s movements in action and skill acquisition. emergence of the task relevant collective variable is the
Given the redundancy and degeneracy of the system foundation of the embedded constructs of movement
(Bernstein, 1967; Edelman & Gally, 2001) and the devel- coordination, control and skill. It provides the qualitative
opmental changes that occur particularly in children and dynamical structure within which the quantitative details
the elderly, it would be premature to conclude that there of the motions of the component DF and their local cou-
is only 1 or the same number of collective variable(s) for plings are embedded.
This dynamical interpretation of the early stage of
a participant in a given task that is preserved over a life
learning implies that the traditional constructs of learning
span of (Newell & Morrison, 2016). Thelen and Smith
through the use of instructional strategies as control
(1994) proposed that there would be a developmental tra-
parameters (Davids et al., 2008; Schmidt & Lee, 2012)
jectory to the 0.5 relative motion designation for the pro-
should reorient to the fundamental challenge of facilitat-
gression of infant locomotion. Nevertheless, in the
ing early practice and the emergence of the task-relevant
absence of significant changes in the constraints on
coordination pattern and collective variable. Learning a
action (Newell, 1986), we anticipate that most percep- task would be enhanced by the decomposition of the
tual-motor skills will have the same collective variable behavioral dynamics of movement coordination, control
for a given individual considering stage of skill level and and skill, the task relevant variables reflecting what is
age of the participant. And, moreover, that in many controlled, and the determination of the most potent con-
tasks, the collective variable will be common across trol parameters that effect functional continuous and dis-
healthy individuals from the same cohort group (e.g., continuous change in performance (Beek & van
walking by definition will hold at 0.5 relative motion of Santvoord, 1992; Newell, 1985; Saltzman & Kelso,
two legs in the step cycle). 1987). For example, the effectiveness of different kinds
Beyond the infant motor sequence and the fundamen- of augmented information to facilitate search behavior in
tal movement patterns in early childhood we intuit that early and subsequent learning will depend on the nature
individuals rarely are required to learn a new pattern of of the skill to be learned and the stage of performance
coordination to realize the outcome of an action. In competence of the learner (Newell, 1991, 1996).
effect, there comes a developmental stage of the life
span where individuals in day to day activities are pre- Collective Variables and Elite Performance
dominantly only scaling an already produced coordin-
There is an advanced level of the motor learning and
ation mode with the same collective variable. The
performance continuum that reflects highly skilled or elite
exceptions perhaps are in sports, such as gymnastics,
performance (Ericsson et al., 1993). There has been lim-
high board diving, figure skating, and the X-games that ited experimental study in any task of the elite levels of
have tasks where even the adult participant has to learn motor skill performance, though there is recent research
to form a here-to-fore not produced task relevant coord- activity built on the application of coordination dynamics
ination pattern that is also the goal of the task. Learning to the original learning through to elite performance of
to play a musical instrument may provide another area of dance and sport skills (cf. Davids et al., 2008; Fuchs &
exception reflecting novel qualitative change in move- Kelso, 2018; Renshaw et al., 2015; Vicinanza et al.,
ment organization. 2018). Indeed, the study of sport skills provides a rich
Modification to the environment and/or health and dis- context to investigate collective variables together with
ease issues in an adult could also change contexts suffi- movement coordination, control and skill development.
ciently so as to induce the search for a new task relevant As previously outlined, we anticipate that typically
coordination pattern and the emergence of a new collect- there will be a low dimensional solution to the number
ive variable. For example, switching to a gravity free of collective variables at this macroscopic level of the
environment such as the space shuttle changes the earth- task learned through the lifespan with the redundancy
bound movement forms of human locomotion and the preserved at other lower component and synergy system
mode of transport of the body from one location to levels. The persistence of the collective variable reflects
another. On earth, swimming as a functional form of the retention of the qualitative dynamics of movement

2021, Vol. 53, No. 6 787


K. M. Newell & Y.-T. Liu

coordination over long retention intervals. Indeed, in practice through the progressive harnessing of the many
practice, it is often recognized anecdotally that people do component DF and subsystems (Kelso, 1995; Mitra
not forget how to swim or ride a bicycle, even over et al., 1998). It has been proposed that the macroscopic
retention time periods as long as 20 years. But, there is collective variable is a relational (informational) property
only limited experimental evidence that the qualitative of the system dynamics rather than a standard biomech-
properties of the dynamics reflected in the collective anical kinematic or kinetic variable that are commonly
variable are well retained over substantial retention peri- used to describe movement and posture. Based on the
ods of years (Nourrit-Lucas et al., 2013). Nevertheless, initial and preliminary data this relational property seems
these observations are consistent with the proposition more tangible in the relative phase of the bimanual 2 DF
that there is a distinct and special role for topological or case than the more abstract outcome from the Haken –
more generally qualitative attractor dynamics in percep- Friedrich - Uhl method to determine the collective vari-
tual-motor learning. able in the multiple DF case of the pedalo.
The emergence of a low dimensional collective vari- A related foundational principle of coordination
able as a function of practice is the general experimental dynamics is that of reciprocal (circular) causality (Haken
finding in regard to Bernstein’s (1967) DF problem et al., 1985; Kelso, 1995, Kelso, 2009). This holds that
(Mitra et al., 1998). This reduction of functional DF in the collective variable provides macroscopic constraint to
motor skill acquisition, however, may occur only on the the motions and local couplings of the component DF
collective variable with the motion of individual DF and and that the motions of the latter in turn influence the
their couplings exhibiting a higher dimension of organ- stability of the collective variable. We are not aware of
ization than that of the collective variable. A further any experimental examinations in motor control and
challenge of understanding the effects of practice is that learning of this proposition from coordination dynamics
the dimension of the dynamical DF may not change uni- on reciprocal causality. It follows from Haken’s (1983)
formly in the single direction of a reduction for all tasks theory of synergetics that the separation of the time
as postulated by Mitra et al. (1998). scales of the individual DF with those of the collective
Newell and Vaillancourt (2001) proposed that with variable(s) provides an essential entry point to this
practice the dimension of a continuous isometric force understudied aspect of coordination dynamics2.
output could increase or decrease according to the The principle of reciprocal causality is a construct of
dynamics of the task demands. For example, practice the synergetics theory to the self-organization of nonliv-
increases the correlation dimension in a constant force ing systems (Haken, 1996). Our intuition, however, is
output task and decreases it in an oscillatory force sinew- that in the domain of biological motion the reciprocal
ave task (Newell et al., 2003). Similarly, the task goal in causality construct is in a theoretical sense potentially
dance or upper-limb motion tasks, for example, to maxi- more relevant in tasks with multiple and redundant joint
mize through practice the dimension of the joint space space DF to be organized (Bernstein, 1967). This is
dynamics (Deutsch & Newell, 2004; Newell et al., because unlike the 2 DF bimanual paradigm the collect-
2000), could lead to a direction of change from the ini- ive variable in this context can be distinct from the task
tial conditions opposite to the general assumption of a goal and the individual DF motions (Liu et al., 2019).
reduced dimension with practice.
These examples provide further indications of the Collective Variables and Task Constraints
strong influence of task constraints on the organization of
The nature of the task constraints holds a strong rela-
collective variables in movement coordination, control and
tion to the nature of collective variables in movement
skill. Several questions arise in the context of Bernstein’s
coordination. We are a long way, however, from identi-
(1967) proposition of skill as the mastery of the redundant
fying task categories with types of collective variables.
DF. Does the variability of the collective variable decrease
This is in part because the emergent behavioral dynamics
with practice and the advance of skill level? Or, does/can
of an action are channeled by more than the task con-
the variability also increase with practice and skill learn-
straints alone.
ing due to adaptation to the redundancy in the task rele-
Three sources of constraint to movement in action
vant pathways of change in the organization of the
have been proposed (Newell, 1986): namely, the individ-
individual component DF? How does practice change the
ual, environment and task. In this framework, task con-
relative variability of the collective variable, pair-wise
straints include: a) the goal of the task - which leads to
joint couplings and the individual joint motions?
movement outcome being a primary variable; b) rules or
instructions that specify or constrain how the task goal
The Nature of the Collective Variable and its
should or could be realized; and c) implements (tools)
Reciprocal Influence
and machines that are used in action. The relative contri-
A central proposition from coordination dynamics is bution of these three classes of task constraints deter-
that a task relevant collective variable emerges from mines the degree to which a particular movement pattern

788 Journal of Motor Behavior


Collective Variables

of the joint space DF is required to be successful in dynamics of bicycle riding, Mitra et al. (1998) following
the task. Stewart (1989) consider the problem as that of a rider-
There are perceptual-motor tasks where the movement bicycle 10 DF component system and the search for low
pattern is, in effect, the goal of the task. Certain closed dimensional control of this 10-dimension space.
skills of sport (Poulton, 1957), such as gymnastics, high Equation (4) could be interpreted from a dynamical
board diving, and ice skating, fit into this category. As view with a relation (to be determined) of a and r as the
noted earlier, we also interpret the HKB bimanual proto- candidate collective variable while considering, v the
cols (Haken et al., 1985) as having the initial task goal speed of the bicycle, the control variable. The equation
as synonymous with the collective variable: namely, rela- also reveals why the bicycle is more stable at higher
tive phase. In this subset of perceptual-motor skills the speeds through what we would postulate as a greater sta-
task goal can be interpreted as mapping directly to the bility region of the a and r relation. This inverse influ-
collective variable. ence of bicycle speed is what makes learning to ride a
Nevertheless, in the broader context of the many bicycle so difficult for the beginner who is typically fear-
classes of perceptual motor skills the task goal and con- ful of falling from faster initial speeds of the bicycle that
straints are distinct from the collective variable, although they assume they are less likely to be able to control.
related in some way. To point up examples of the chal- A complementary approach to determining the collect-
lenge of separating task constraints and collective varia- ive variable in bicycle riding is to run a PCA analysis as
bles we consider briefly two well-known motor skills has been done in young adults learning to ride a unicycle
where task constraints in the form of rules have been (Lee et al., 2017). The PCA showed that two compo-
proposed for given actions: namely, riding a bicycle nents accounted for the majority of the variance in uni-
and juggling. cycle riding and it was interpreted through the PCA
weightings and projections that the two components
Riding a Bicycle reflected the macroscopic properties of balance and pro-
Polanyi (1958) developed the case that in riding a pulsion. Thus, learning to ride a bicycle or unicycle may
bicycle the rider was unknowingly through tacit know- be about coupling of balance and propulsion with syner-
ledge following a particular rule. The rule observed by getic leg motions playing a background role. As an
the cyclist is this. When he starts falling to the right he aside, Lee et al. (2017) found that some young adult
turns the handle-bars to the right, so that the course of beginners failed to learn the unicycle even after 28 days
the bicycle is deflected along a curve toward the right. of practice revealing again that learning a new movement
This results in a centrifugal force pushing the cyclist to coordination pattern can be a different and more difficult
the left and offsets the gravitational force dragging him challenge than the typical laboratory scaling motor tasks
down to the right. This maneuver presently throws the (see also in juggling, Huys et al., 2004a, 2004b; roller
cyclist out of balance to the left, which he counteracts ball learning, Liu & Newell, 2015).
by turning the handlebars to the left; and so he continues
to keep himself in balance by winding along a series of Juggling
appropriate curvatures. Claude Shannon (1983)3 presented his now well-
The behavior of bicycling can, therefore, be expressed known formula for juggling. Equation (5) is
as:
ðF þ D Þ H ¼ ðV þ D Þ N (5)
r  v =a
2
(4)
where: F ¼ how long a ball stays in the air, D ¼ how
where for a given angle of imbalance (a) the curvature long a ball is held in a hand, H ¼ number of hands,
of each winding (r) is inversely proportional to the V ¼ how long a hand is empty or vacant, and
square of the speed (v) at which the cyclist N ¼ number of balls being juggled. The formula is gen-
is proceeding. eral and provides a distributional analysis of the juggling
This description of the bicycle kinematics emerges of a different number of balls with a different number of
from the confluence of constraints on action - individual, hands. The Shannon equation also holds for periodic jug-
environment and task (Newell, 1986) and departures gling patterns in which a hand never contains more than
from this rule will lead to a loss of stability and even a one ball (as in so-called multiplexing).
fall from the bicycle. This task rule from Polanyi is typ- This formula for juggling would seem to provide the
ically not explicitly apparent or stated in instructions for basis for determining the relevant collective variable(s).
riding a bicycle nor is it assumed to be known by most The equation is, however, not strictly a dynamical equa-
cyclists. The rule, however, does hold properties similar tion that specifies the time evolutionary properties of the
to those of a collective variable – a low dimensional hand-ball dynamics in juggling, though it does relate to
description that captures properties of the topological that goal. An approach through acceleration of the balls
organization of the skill. In a broader view of the task and hands might be able to provide a dynamical equation

2021, Vol. 53, No. 6 789


K. M. Newell & Y.-T. Liu

for juggling (see Beek, 1989), and the relevant collect- associated with the stability of juggling and thus holds
ive variable. parallels to a collective variable.
The Shannon Equation (5) reflects a task constraint k is a variable that captures the ratio of the time a
that needs to be satisfied if the performer is to retain the hand is loaded relative to the time of the complete hand
integrity of the act of juggling (Beek, 1989; Beek & van cycle. It is similar to the timing analysis of handwriting
Santvoord, 1992). Put another way, if this formula is not by Viviani and Terzuolo (1980) who showed from a
followed by the performer (when juggling standard pat- motor program perspective that the relative time to write
terns) he/she will not be juggling successfully. Learning a letter within a word is a constant ratio of the time to
to juggle and realizing the Shannon equation in perform- write the complete word. Relative timing is a key vari-
ance reflects one of the more challenging perceptual- able in the contrasting theoretical positions of general-
motor skills for both the learner and experimenter (Beek ized motor program (Keele, 1968; Schmidt, 1975) and
& van Santvoord, 1992; Haibach et al., 2004; Huys coordination dynamics (Kelso, 1995, 1997).
et al., 2004a, 2004b). Juggling is a space-time problem but k has not been
The formula for juggling engages the interaction with extended to accommodate directly the spatial demands of
the environment in the form of the space-time trajecto- juggling or the coordination of the two hands in this jug-
ries of the hands and balls. The Shannon equation stipu- gling task (Van Santvrood & Beek, 1996). Both of these
lates that the hand cycle time (V þ D) is a fixed ratio properties may be implicit in the k ratio but do they
(determined by the ratio N/H) to the ball cycle time need to be accommodated explicitly in a candidate col-
(F þ D). It follows that juggling has the characteristic lective variable? Perhaps the richness of the dynamical
whereby the performer can adjust within bounds the indi- structures of juggling also makes the task a candidate for
vidual times of balls in air/hand as long as the collective more than one collective variable?
of time in the hands and time in the air satisfies
Equation (5).
Closing Comments
Beek and colleagues through a series of experiments
have produced a comprehensive decomposition of the Collective variables, in terms of order parameters of
learning and performance of juggling, particularly 3-ball macroscopic movement phenomena, were given a formal
cascade juggling (Beek, 1989; Beek & van Santvoord, and explicit place in the coordination dynamics of the
1992; Van Santvrood & Beek, 1996). They have shown HKB model (Haken et al., 1985). Yet, in spite of the
the multiple time scales and multiform dynamics of jug- many experiments inspired by the dynamical systems
gling over a substantial period of practice (e.g., Huys approach to the study of movement in action over the
et al., 2004a, 2004b). Furthermore, and central to the last 35 years, there has not been a proportionate theoret-
theme of this section of the paper, they have identified a ical or experimental examination of collective variables,
ratio k (Equation (6)) that is the time that the hand is and their proposed informational character (Kelso, 1994;
loaded with a ball relative to the complete hand cycle Mitra et al., 1998). Moreover, and relatedly, the postula-
(Beek & van Santvoord, 1992). tion of reciprocal causality (Kelso, 1995), and the associ-
ated notion of time scale separation, have not been
k ¼ D=ðV þ DÞ (6)
experimentally tested in the movement domain in spite
k reflects then, how a juggler accommodates the of its centrality to the coordination dynamics framework.
Shannon equation within the different extreme expres- Our intuition is that the notion of collective variables is
sions of low dwell ratio (hot potatoe juggling) or high a relevant theoretical postulation in regard to the degrees
dwell ratio (delayed juggling). k can take on a range of of freedom problem (Bernstein, 1967) but that it is diffi-
values but for standard 3-ball juggling it is on the order cult to implement experimental tests of its implications
of .75 (Beek & van Santvoord, 1992). in the context of motor learning and control.
k is a macroscopic variable of the timing of balls and A considerable proportion of the studies on collective
hands in juggling but is it a candidate collective variable variables either comes from or were motivated by the
for the juggling task? The juggling experiment of Beek research program of Kelso. The resulting experiments
and van Santvoord (1992) was not set up to test for a were set-up and interpreted within the HKB synergetic
collective variable using the sequence of experimental framework, with relative phase as the collective and
procedures of coordination dynamics (Kelso, 1995 – see coherent variable in both brain and behavior dynamics.
appendix here). Rather, the juggling experiments were This work has been the most extensive and formal of the
set up under the related approaches to dynamical systems studies in identifying the collective variable and the add-
of Farey Tree Ratios and nonlinear oscillator theory. itional emphasis on the integration of theory, model
Although the terms order parameter or collective variable and experiment.
were not used by the Beek group with respect to k in We have reported a number of experimental studies
juggling, a key hypothesis in this work is that k is set-up to determine the collective variable of tasks

790 Journal of Motor Behavior


Collective Variables

beyond the 2 DF bimanual control tasks. Most of these protocol and to more directly examine the longstanding
experiments have a more informal identification protocol postulation of Bernstein (1967) that skill is the mastery
than outlined in the coordination dynamics validation of the redundant DF.
sequence (Appendix A) and, also at this point in time,
limited replicability. Nevertheless, the studies have iden-
Author notes
tified the potential of several types of collective variable
in movement coordination other than a relative phase. We would like to thank the reviewers for their most
These findings are consistent with the position of the helpful comments and critiques on earlier versions of
task relevant nature of the collective variable the paper.
(Kelso, 1995).
The role of collective variables has been masked by
Disclosure statement
the task selection for experimental studies of motor
learning and control (Newell, 1985) that historically has No potential conflict of interest was reported by
tended to err on the side of either single DF motor tasks the author(s).
or in 2 or more DF tasks that the participant has engaged
in previously (Newell et al., 1989). As a consequence, in
Notes
most experiments the participants can or have already
produced a movement pattern coherent with the task 1. We thank an anonymous reviewer for this suggestion.
relevant collective variable at some point prior to the 2. We thank an anonymous reviewer for this suggestion.
testing leaving the study of learning, retention and trans- 3. In an informal presentation at Carnegie Mellon
fer as largely a movement scaling problem to the exter- University July 1983 (from Beek, 1989 through
nal task demands. This narrow experimental framework Raibert, 1986, p. 198).
has restricted the breadth of investigation of the dynam-
ics of perceptual-motor skill learning as reflected in the
prevalence of monotonic functions of performance out- ORCID
come in learning curves (Newell et al., 2006). Yeou-Teh Liu http://orcid.org/0000-0002-0559-1896
The ecological theory of perception and action and the
related domain of coordination dynamics have advanced
REFERENCES
the proposition that the collective variable is likely to be
an abstract informational variable and not a standard bio- Adams, J. A. (1971). A closed-loop theory of motor learning.
mechanics physical variable (Haken, 1996; Kelso, 1994, Journal of Motor Behavior, 3, 111–149. https://doi.org/10.
Mitra et al., 1998). This postulation reflects a shift in the 1080/00222895.1971.10734898
theoretical emphasis away from the centrality of the Adolph, K. E., & Hoch, J. E. (2019). Motor development:
dimensions of joints and muscles in the long held but Embodied, embedded, enculturated, enabling. Annual Review
of Psychology, 70, 141–164.
still ongoing debate on ‘what’ is controlled in movement
Alexander, R. Mc. (1992). The human machine. Natural
(Latash, 2008). Ecological psychology and coordination
History Museum.
dynamics provides a theoretical framework to consider Bardy, B. G., Marin, L., Stoffregen, T. A., & Bootsma, R. J.
relational (informational) constraints as well as physical (1999). Postural coordination modes considered as emergent
constraints to the acquisition of perceptual-motor skills phenomena. Journal of Experimental Psychology: Human
(Kelso, 1984; Kugler & Turvey, 1987; Turvey, 2002; Perception and Performance, 25, 1284–1301.
Warren, 2006). In this view, the collective variable will Beckmann, H. (2003). MA Thesis: Vergleich von
be to a large degree task dependent because each task Techniktrainingsansatzen im Kugelstoßen (in German)
has its own confluence of physical and informational University of M€unster, M€unster.
constraints (Newell, 1986; Warren, 2006). Beckmann, H., & Sch€ollhorn, W. (2003). Differential learning
Finally, a renewed emphasis on the place of task in shot put. In Sch€ollhorn, W. I., Bohn, C., Jager, J. M.,
decomposition from a dynamical view (Beek, 1989; Schaper, H., & Alichmann, M. (Eds.), European workshop
Newell et al., 2001; Saltzman & Kelso, 1987; Warren, on movement sciences (p. 68). Sport & Buch Strauß.
Beek, P. J. (1989). Juggling dynamics. Free University Press.
2006) would allow us to address further the influence of
Beek, P.J., & van Santvoord, A. A. (1992). Learning the cas-
task type on attractor dynamics, together with the emer-
cade juggle: A dynamical systems analysis. Journal of Motor
gence of the collective variable in the early stage of per- Behavior, 24, 85–94.
ceptual-motor skill learning. Task decomposition from a Bernstein, N. (1967). The co-ordination and regulation of
dynamical framework could contribute to understanding movements. Pergamon.
tasks for study that engage more than 2 DF in joint Bernstein, N. A. (1996). On dexterity and its development. In
space. This would offer a richer set of conditions to M. Latash & M. T. Turvey (Eds.), Dexterity and its develop-
build on the framework provided by the bimanual ment (pp. 3–244). Lawrence Erlbaum Associates.

2021, Vol. 53, No. 6 791


K. M. Newell & Y.-T. Liu

Bernstein, N. A. (1947/in press). In M. L. Latash (Ed.). Stelmach (Ed.), Information processing in motor control and
Bernstein’s construction of movements. Routledge. learning (pp. 1–40). Academic Press.
Chen, H-H., Liu, Y-T., Mayer-Kress, G., & Newell, K. M. Frank, T. D., Michelbrink, M., Beckmann, H., & Sch€ollhorn,
(2005). Learning the pedalo locomotion task. Journal of W. I. (2008). A quantitative dynamical systems approach to
Motor Behavior, 37, 247–256. differential learning: Self-organization and order parameter
Cesari, P., & Newell, K. M. (1999). The scaling of human grip equations. Biological Cybernetics, 98, 19–31.
configurations. Journal of Experimental Psychology: Human Frank, T. D., Richardson, M. J., Lopresti-Goodman, S. M., &
Perception and Performance, 25, 927–935. Turvey, M. T. (2009). Order parameter dynamics of body-
Cesari, P., & Newell, K. M. (2000). Body scaled transitions in scaled hysteresis and mode transitions in grasping behavior.
human grip configurations. Journal of Experimental Psychology: Journal of Biological Physics, 35, 127–147.
Human Perception and Performance, 26, 1657–1668. Fuchs, A., & Jirsa, V. K. (2008). Coordination: Neural, behav-
Chow, J. Y., Seifert, L., Herault, R., Chia, S. J. Y., & Lee, ioral and social dynamics. Springer.
M. C. Y. (2014). A dynamical system perspective to under- Fuchs, A., & Kelso, J. A. S. (2018). Coordination dynamics
standing badminton singles game play. Human Movement and synergetics: From finger movements to brain patterns
Science, 33, 70–84. and ballet dancing. In S. Mueller et al. (Eds). Complexity
Connolly, K. J. (Ed.) (1998). The psychobiology of the hand. and synergetics (pp. 301–316). Springer-Verlag.
Cambridge University Press. Fuchs, A., Jirsa, V. K., Haken, H., & Kelso, J. A. S. (1996).
Daffertshofer, A., Huys, R., & Beek, P. J. (2004). Dynamical Extending the HKB model of coordinated movement to
coupling between locomotion and respiration. Biological oscillators with different eigenfrequencies. Biological
Cybernetics, 90, 157–164. Cybernetics, 74, 21–30.
Daffertshofer, A., Lamoth, C. J. C., Meijer, O. G., & Beek, Garcia-Rubio, J., Gomez, M. A., Canadas, M., & Ibanez, S. J.
P. J. (2004). PCA in studying coordination and variability: A (2015). Offensive rating-time coordination dynamics in basket-
tutorial. Clinical Biomechanics, 19, 415–428. ball. Complex systems theory applied to basketball. International
Davids, K., Button, C., & Bennett, S. (2008). Dynamics of skill Journal of Performance Analysis in Sport, 15, 513–526.
acquisition. Human Kinetics. Gel'fand, I. M. & Tsetlin, M. L. (1962). Some methods of con-
Deutsch, K. M., & Newell, K. M. (2004). Intra-limb segmental trol for complex systems. Russian Mathematical Surveys, 17,
influences on random like movements. Neuroscience Letters, 95–116.
367, 218–223. Gentile, A. M. (1972). A working model of skill acquisition
Dutt-Mazumder, A., & Newell, K. M. (2017). Transitions of with application to teaching. Quest, 17, 3–23. https://doi.org/
postural coordination as a function of oscillatory platform 10.1080/00336297.1972.10519717
dynamics. Human Movement Science, 52, 24–35. Gentile, A. M. (1987). Skill acquisition: Action, movement,
Dutt-Mazumder, A., & Newell, K. M. (2018). Influence of skill and the neuromotor processes. In J. H. Carr, R. B. Shepard,
level on acquisition of dynamic cyclical postural stability as A. M, Gentile & J. M. Hind (Eds.), Movement science:
a function of practice. Scandinavian Journal of Medicine Foundations for physical therapy in rehabilitation (pp.
and Science in Sports, 28, 1604–1614. 93–154). Aspen.
Edelman, G. M., & Gally, J. (2001). Degeneracy and complex- Gesell, A. (1952). Infant development: The embryology of early
ity in biological systems. Proceedings of the National human behavior. Greenwood Press.
Academies of Sciences USA, 98, 13763–13768. Gibson, J. J. (1979). The ecological approach to visual percep-
Ericsson, K. A., Krampe, R. T., & Tesch-Romer, C. (1993). tion. Houghton Mifflin.
The role of deliberate practice in the acquisition of expert Gulick, D. W., & O’Reilly, O. M. (2000). On the dynamics of
performance. Psychological Review, 100, 363–406. Dynabee. Journal of Applied Mechanics, 67, 321–325.
Fink, P.W., Foo, P. S., & Warren, W. H. (2009). Catching fly Haas, R., Haken, H., & Korndle, H. (1995). Movements on a
balls in virtual reality: A critical test of the outfielder prob- pedalo: An analysis based on synergetics. Unpublished paper.
lem. Journal of Vision Research, 9:14.1–8. Haibach, P. S., Daniels, G., & Newell, K. M. (2004).
Fitts, P. M. (1964). Perceptual-motor skills learning. In A. W. Coordination changes in the early stages of learning to cas-
Melton (Ed.), Categories of human learning (pp. 243–285). cade juggle. Human Movement Science, 23, 185–206.
Academic Press. Haken, H. (1983). Synergetics: An introduction: Nonequilibrium
Fleishman, E. A., Quaintance, M., & Broedling, L. A. (1984). phase transitions and self-organization in physics, chemistry,
Taxonomies of human performance: The description of and biology (3rd rev ed). Springer-Verlag.
human tasks. Academic Press. Haken, H. (1991). Synergetic computers and cognition. Springer.
Foo, P., Kelso, J. A. S., & de Guzman, G. C. (2000). Functional Haken, H. (1996). Principles of brain functioning: A synergetic
stability of unstable fixed points. Human pole balancing using approach to brain activity, behavior, and cognition. Springer.
time to balance information. Journal of Experimental Psychology: Haken, H., Kelso, J. A. S., & Bunz, H. (1985). A theoretical
Human Perception and Performance, 26, 1281–1297. model of phase transitions in human hand movements.
Forner-Cordero, A., Levin, O., Li, Y., & Swinnen, S. P. Biological Cybernetics, 51, 347–356.
(2005). Principal component analysis of complex multi-joint Haken, H., Peper, C. E., Beek, P. J., & Daffertshofer, A. (1996).
coordinative movements. Biological Cybernetics, 93, 63–78. A model for phase transitions in human hand movements dur-
Fowler, C. A., & Turvey, M. T. (1978). Skill acquisition: An ing multifrequency tapping. Physica D, 90, 179–196.
event approach with special reference to searching for the Higbie, J. (1980). The physics of “Dyna Bee”. Physics
optimum of a function of several variables. In G. E. Teacher, 18, 147–148. https://doi.org/10.1119/1.2340452

792 Journal of Motor Behavior


Collective Variables

Hoenkamp, E. (1978). Perceptual cues that determine the label- Kelso, J. A. S. (2012). Multistability and metastability:
ling of gait. Human Movement Studies, 4, 59–69. Understanding dynamic coordination in the brain. Philos
Hong, S. L., & Newell, K. M. (2006). Practice effects on local Trans R Soc Lond B Biol Sci, 367, 906–918. https://doi.org/
and global dynamics of the ski-simulator task. Experimental 10.1098/rstb.2011.0351
Brain Research, 169, 350–360. Kelso, J. A. S., & Jeka, J. (1992). Symmetry breaking dynamics of
Huys, R., Daffertshofer, A., & Beek, P. J. (2004a). Multiple time human multi-limb coordination. Journal of Experimental
scales and multiform dynamics in learning to juggle. Motor Psychology: Human Perception and Performance, 18, 645–668.
Control, 8, 188–212. https://doi.org/10.1123/mcj.8.2.188 Kelso, J. A. S., & Engstrom, D. A. (2006). The complementary
Huys, R., Daffertshofer, A., & Beek, P. J. (2004b). Multiple nature. MIT Press.
time scales and subsystem embedding in the learning of jug- Kelso, J. A. S., & Fuchs, A. (2016). The coordination dynam-
gling. Human Movement Science, 23, 315–336. https://doi. ics of mobile conjugate reinforcement. Biological
org/10.1016/j.humov.2004.08.009 Cybernetics, 110 (1), 41–53.
Huys, R., Perdikis, D., & Jirsa, V.K. (2014). Functional architec- Kelso, J. A. S., Buchanan, J. J. & Murata, T. (1994).
tures and structured flows on manifolds: A dynamical frame- Multifunctionality and switching in the coordination dynamics
work for motor behavior. Psychological Review, 121, 302–336. of reaching and grasping. Human Movement Science, 13, 63–94.
Jirsa, V. K., Fuchs, A., & Kelso, J.A. S. (1998). Connecting Kilby, M. C, Molenaar, P. C. M., & Newell, K. M. (2015).
cortical and behavioral dynamics: Bimanual coordination. Models of postural control: Shared variance in joint and
Neural Computation, 10, 2019–2045. https://doi.org/10.1162/ COM motions. PLos One, 10(5):e0126379. https://doi.org/10.
089976698300016954 1371/journal.pone.0126379
Jirsa, V. K., Fuchs, A., Haken, H., & Kelso, J. A. S. (1994). A Ko, J-H., Challis, J. H., & Newell, K. M. (2014). Transition of
theoretical model of phase transitions in the human brain. COM-COP relative phase in a dynamic balance task. Human
Biological Cybernetics, 71(1), 27–35. PMID 8054384. Movement Science, 38, 1–14.
Jirsa, V. K., & Kelso, J. A. S. (Eds.), (2004). Coordination Kostrubiec, V., Zanone, P. G., Fuchs, A., & Kelso, J. A.
dynamics: Issues and trends. Springer. (2012). Beyond the blank slate: Routes to learning new
Jones, L. A., & Lederman, S. J. (2006). Human hand function. coordination patterns depend on the intrinsic dynamics of the
learner – experimental evidence and theoretical model.
Oxford University Press.
Kay, B. A. (1988). The dimensionality of movement trajectories Frontiers of Human Neuroscience, 6, 222.
Kugler, P. N. (1986). A morphological perspective on the ori-
and the degrees of freedom problem. Human Movement Science,
gin and evolution of movement patterns. In M. G. Wade &
7, 343–364. https://doi.org/10.1016/0167-9457(88)90016-4
H. T. A. Whiting (Eds.), Motor development in children:
Kay, B. A., Saltzman, E. L., & Kelso, J. A. S. (1991). Steady
Aspects of coordination and control (pp. 459–525). Martinus
state and perturbed rhythmical movements: Dynamical mod-
Nijhoff.
eling using a variety of analytic tools. Journal of
Kugler, P. N., & Turvey, M. T. (1987). Information, natural
Experimental Psychology: Human Perception and
law, and the self-assembly of rhythmic movement:
Performance, 17, 359–369.
Theoretical and experimental investigations. Erlbaum.
Keele, S. W. (1968). Movement control in skilled motor per-
Kugler, P. N., Kelso, J. A. S., & Turvey, M. T. (1980). On the
formance. Psychological Bulletin, 70, 387–403. https://doi.
concept of coordinative structures as dissipative structures: I.
org/10.1037/h0026739 Theoretical lines of convergence. In G. E. Stelmach & J. Requin
Kelso, J. A. S. (1981). On the oscillatory basis of movement.
(Eds.), Tutorials in motor behavior (pp. 1–49). North–Holland.
Bulletin of the Psychonomic Society, 18, 63. Kugler, P. N., Kelso, J. A. S., & Turvey, M. T. (1982). On the
Kelso, J. A. S. (1984). Phase transitions and critical behavior control and coordination of naturally developing systems. In
in human bimanual coordination. American Journal of J. A. S. Kelso & J. E. Clark (Eds.), The development of
Physiology: Regulatory, Integrative and Comparative, 15, movement control and co-ordination (pp. 5–78). Wiley.
R1000–R1004. Kuramoto, Y. (1984). Chemical oscillations, waves, and turbu-
Kelso, J. A. S. (1990). Phase transitions: Foundations of behav- lence (Vol. 19). Springer Series in Synergetics.
ior. In H. Haken & M. Stadler (Eds.), Synergetics of cogni- Lafe, C. L., Pacheco, M. M., & Newell, K. M. (2016).
tion (pp.249–268). Springer-Verlag. Bimanual coordination and the intermittency of visual infor-
Kelso, J. A. S. (1994). The informational character of self- mation in isometric force tracking. Experimental Brain
organized coordination dynamics. Human Movement Science, Research, 234, 2025–2034.
13, 393–413. https://doi.org/10.1016/0167-9457(94)90047-7 Lamoth, C. J., Daffertshofer, A., Huys, R., & Beek, P. J.
Kelso, J. A. S. (1995). Dynamic patterns: The self-organization (2009). Steady and transient coordination structures of walk-
of brain and behavior. MIT Press. ing and running. Human Movement Science, 28, 371–386.
Kelso, J. A. S. (1997). Relative timing in brain and behavior: Latash M. L. (2012). The bliss (not the problem) of motor
Some observations about the generalized motor program and abundance (not redundancy). Experimental Brain
self-organized coordination dynamics. Human Movement Research, 217, 1–5. 10.1007/s00221-012-3000-4.
Science, 16, 453–460. https://doi.org/10.1016/S0167- Latash, M. L. (2008). Synergy. Oxford University Press.
9457(96)00044-9 Latash, M. L. (Ed.). (1947/in press). Bernstein’s construction of
Kelso, J. A. S. (2009). Coordination dynamics. In R. A. movements. Routledge.
Meyers (Ed.). Encyclopedia of complexity and systems sci- Lee D. N. (2009). General Tau theory: evolution to
ence (pp. 1537–1564). Springer. date. Perception, 38, 837–858.

2021, Vol. 53, No. 6 793


K. M. Newell & Y.-T. Liu

Lee, I-C., Liu, Y-T., & Newell, K. M. (2017). Learning to ride developmental systems theory and methodology (pp.
a unicycle: Coordinating balance and propulsion. Journal of 316–342). Guilford Publications.
Motor Learning and Development, 4, 287–306. Newell, K. M., & Morrison, S. (2016). The evolving dynamical
Lipsitz, L. A. (2002). Dynamics of stability: The physiological landscape of movement forms. A degrees of freedom per-
basis of functional health and frailty. Journal of Gerontology spective. Kinesiology Review, 5, 4–14.
A, Biology, Science, Medicine, 67, B115–B125. https://doi. Newell, K. M., & Vaillancourt, D. E. (2001). Dimensional
org/10.1093/gerona/57.3.B115 change in motor learning. Human Movement Science, 4-5,
Liu, Y-T., & Newell, K. M. (2015). S-shaped motor learning 695–716.
and non-equilibrium phase transitions. Journal of Newell, K. M., Challis, S., & Morrison, S. (2000). Dimensional
Experimental Psychology: Human Perception and constraints on limb movements. Human Movement Science,
Performance, 41, 403–414. 19, 175–201.
Liu, Y-T., Chuang, K-L., & Newell, K. M. (2019). Mapping Newell, K. M., Liu, Y.T., & Mayer-Kress, G. (2001). Time
collective variable and synergy dynamics to task outcome in scales in motor learning and development. Psychological
a perceptual-motor skill. PLoS One, 14(4): e0215460. https:// Review, 108, 57–82.
doi.org/10.1371/journal.pone.0215460 Newell, K. M., Liu, Y-T., & Mayer-Kress, G. (2003). A
Liu, Y-T., Mayer-Kress, G., & Newell, K. M. (2006). dynamical systems interpretation of epigenetic landscapes for
Qualitative and quantitative change in the dynamics of motor infant motor development. Infant Development and Behavior,
learning. Journal of Experimental Psychology: Human 26, 449–472.
Perception and Performance, 32, 380–393. Newell, K. M., Liu, Y-T., & Mayer-Kress, G. (2009). Time
Liu, Y-T., Mayer-Kress, G., & Newell, K. M. (2010). Bi-stabil- scales, difficulty/skill duality, and the dynamics of motor
ity of movement coordination as a function of skill level and learning. Advances in Experimental Medicine and Biology,
task difficulty. Journal of Experimental Psychology: Human 629, 457–476.
Perception and Performance, 36, 1515–1524. Newell, K. M., Mayer-Kress, G., & Liu, Y-T. (2006). Human
Liu, Y-T., Luo, Z-Y., Mayer-Kress, G., & Newell, K. M. learning: Power laws or multiple characteristic time scales?
(2012). Self-organized criticality and learning a new coordin- Tutorials in Quantitative Methods for Psychology, 2, 66–76.
ation task. Human Movement Science, 31, 40–54. Newell, K. M., van Emmerik, R. E. A., & McDonald, P. V.
Magill, R. A., & Anderson, D. I. (2014). Motor learning and (1989). On simple movements and complex theories (and
control: Concepts and applications (10th ed.). McGraw Hill. vice-versa). Behavioral and Brain Sciences, 12, 229–230.
Mitra, S., Amazeen, P. G., & Turvey, M. T. (1998). Motor Newell, K. M., Broderick, M. P., Deutsch, K. M., & Slifkin,
learning as decreasing active (dynamical) degrees of free- A. B. (2003). Task goals and change in dynamical degrees
dom. Human Movement Science, 17, 17–65. of freedom with motor learning. Journal of Experimental
Molenaar, P. M. C., & Newell, K. M. (2003). Direct fit of a Psychology: Human Perception and Performance, 29,
theoretical model of phase transitions in oscillatory finger 379–387.
motions. British Journal of Mathematical and Statistical Newell, K. M., Kugler, P. N., & Emmerik, R. E. A. van, &
Psychology, 56, 199–214. McDonald, P. V. (1989). Search strategies and the acquisi-
Molenaar, P. C. M., Wang, Z., & Newell, K. M. (2013). tion of coordination. In S. A. Wallace (Ed.), Perspectives on
Compressing movement information via principal components coordination (pp. 86–122). North Holland.
analysis (PCA): Contrasting outcomes from the time and fre- Newell, K. M., Mayer-Kress, G., Hong, S. L., & Liu, Y-T.
quency domains. Human Movement Science, 32, 1495–1511. (2009). Adaptation and learning: Characteristic time scales
Newell, K. M. (1985). Coordination, control and skill. In D. of performance dynamics. Human Movement Science, 28,
Goodman, I. Franks & R. Wilberg (Eds.), Differing perspec- 655–687. https://doi.org/10.1016/j.humov.2009.07.001
tives in motor learning, memory and control (pp. 295–317). Newell, K. M., Scully, D. M., Tenenbaum, F., & Hardiman, S.
North-Holland. (1989). Body scale and the development of prehension.
Newell, K. M. (1986). Constraints on the development of Developmental Psychobiology, 22, 1–13.
coordination. In M. G. Wade & H. T. A. Whiting (Eds.), Newell, K. M., Scully, D. M., McDonald, P. V., &
Motor skill acquisition in children: Aspects of coordination Baillargeon, R. (1989). Task constraints and infant grip con-
and control (pp. 341–360). Martinies NIJHOS. figurations. Developmental Psychobiology, 22, 817–832.
Newell, K. M. (1991). Motor skill acquisition. Annual Review Nourrit-Lucas, D., Zelic, G., Deschamps, T., Hilpron, M., &
of Psychology, 42, 213–237. https://doi.org/10.1146/annurev. Delignieres, D. (2013). Persistent coordination patterns in a
ps.42.020191.001241 complex task after 10 years delay: How validate the old say-
Newell, K. M. (2020). What are fundamental motor skills and ing: once you have learned how to ride bicycle, you never
what is fundamental about them? Journal of Motor Learning forget. Human Movement Science, 32, 1365–1378.
and Development, 8(2), 280–314. https://doi.org/10.1123/ Phinyomark, A., Hettinga, B. A., Osis, S., & Ferber, R. (2015).
jmld.2020-0013 Do intermediate- and higher-order principal components contain
Newell, K. M. (1996). Change in movement and skill: Learning, useful information to detect subtle changes in lower extremity
retention, and transfer. In M. Latash & M. Turvey (Eds.), biomechanics during running? Human Movement Science, 44,
Dexterity and its development (pp. 393–429). Erlbaum. 91–101. https://doi.org/10.1016/j.humov.2015.08.018
Newell, K. M., & Liu, Y-T. (2014). Dynamics of motor learn- Palut, Y., & Zanone, P.-G. (2005). A dynamical analysis of ten-
ing and development across the lifespan. In P. C. M. nis: Concepts and data. Journal of Sport Sciences, 23,
Molenaar, R. Lerner & K. M. Newell (Eds). Handbook of 1021–1032.

794 Journal of Motor Behavior


Collective Variables

Polanyi, M. (1958). Personal knowledge: Towards a post-crit- ground walk-to-run transition. Journal of Experimental
ical philosophy. University of Chicago Press. Biology, 216, 3047–3054.
Poulton. E. C. (1957). On prediction in skilled movements. Sorensen, V., Ingvaldsen, R. P., & Whiting, H. T. A. (2001).
Psychological Bulletin, 54, 467–478. https://doi.org/10.1037/ The application of co-ordination dynamics to the analysis of
h0045515 discrete movements using table-tennis as a paradigm skill.
Profeta, V. L. S., & Turvey, M. T. (2018). Bernstein’s levels Biological Cybernetics, 85, 27–38.
of movement construction: A contemporary perspective. Sternad, D., Duarte, M., Katsumata, H., & Schaal, S. (2001).
Human Movement Science, 57, 111–33. Bouncing a ball: Tuning into dynamic stability. Journal of
Raibert, M. (1986). Legged robots that balance. MIT Press. Experimental Psychology: Human Perception and
Renshaw, I., Davids, K., Chow, J. Y., & Button, C. (2015). Performance, 27, 1163–1184.
Nonlinear pedagogy in skill acquisition: An introduction. Stewart, I. (1989). Does god play dice: The mathematics of
Human Kinetics. chaos. Addison-Wesley.
Saltzman, E. L., & Kelso, J. S. (1987). Skilled actions: A task Strogatz, S. (2003). Sync: How order emerges from chaos in
dynamic approach. Psychological Review, 94, 84–106. the universe, nature, and daily life. Hachotte Books.
Saltzman, E. L., & Munhall, R. G. (1992). Skill acquisition and Strogatz, S., & Stewart, I. (1993). Coupled oscillators and bio-
development: the roles of state, parameter, and graph dynam- logical synchronization. Scientific American, 269(6),
ics. Journal of Motor Behavior, 24, 49–57. 102–109.
Schmidt, R. A. (1975). A schema theory of discrete motor skill Teulier, C., Nourrit, D., & Delignieres, D. (2006). The evolu-
learning. Psychological Review, 82, 225–260. https://doi.org/ tion of oscillatory behavior during learning on a ski-simula-
10.1037/h0076770 tor. Research Quarterly for Exercise and Sport, 77, 464–475.
Schmidt, R. A., & Lee, T. D. (2012). Motor control and learn- Thelen, E. & Smith, L. B. (1994). A dynamic systems approach
ing: A behavioral emphasis (5th Ed.). Human Kinetics. to the development of cognition and action. MIT Press.
Schmidt, R. C., Fitzpatrick, P., Caron, R., & Mergeshe, J. Thelen, E., & Ulrich, B. D. (1991). Hidden skills: A dynamic
(2011). Understanding social motor coordination. Human systems analysis of treadmill stepping during the first year.
Movement Science, 30, 834–845. Monographs of the Society for Research in Child
Sch€ollhorn, W. I., Mayer-Kress, G., Newell, K. M., & Development, 56, 1–98. Discussion 99–104.
Michelbrink, M. (2009). Time scales of adaptive behavior Thelen, E., Kelso, J. A. S., & Fogel, A. (1987). Self-organizing
and motor learning in the presence of stochastic perturba- systems and infant motor development. Developmental
tions. Human Movement Science, 28, 319–333. Review, 7, 39–65.
Scholz, J. P., & Sch€oner, G. (1999). The uncontrolled manifold Thelen, E., Skala, K. D., & Kelso, J. A. S. (1987). The
concept: Identifying control variables for a functional task. dynamic nature of early coordination: Evidence from bilat-
Experimental Brain Research, 126, 289–306. https://doi.org/ eral leg movements in young infants. Developmental
10.1007/s002210050738 Psychology, 23, 179–186.
Sch€oner, G. (1989). Learning and recall in a dynamic theory of Turvey, M. T. (1990). Coordination. American Psychologist,
coordination patterns. Biological Cybernetics, 62, 39–54. 45, 938–953. https://doi.org/10.1037/0003-066X.45.8.938
https://doi.org/10.1007/BF00217659 Turvey, M. T. (1992). Ecological foundations of cognition:
Sch€oner, G., & Kelso, J. A. S. (1988a). A synergetic theory of Invariants of perception and action. In H. L. Pick, P. van
environmentally-specified and learned patterns of movement den Broek & D. C. Knil (Eds.), Cognition: Conceptual and
coordination. I. Relative phase dynamics. Biological methodological issues (pp. 85–117). American Psychological
Cybernetics, 58, 71–80. https://doi.org/10.1007/BF00364153 Association.
Sch€oner, G., & Kelso, J. A. S. (1988b). A synergetic theory of Turvey, M. T. (2004). Impredicativity, dynamics, and the per-
environmentally-specified and learned patterns of movement ception-action divide. In V. K. Jirsa & J. A. S. Kelso (Eds).
coordination. II. Component oscillator dynamics. Biological Coordination dynamics: Issues and trends (pp. 2–20).
Cybernetics, 58, 81–89. https://doi.org/10.1007/BF00364154 Springer.
Sch€oner, G., & Kelso, J. A. S. (1988c). Dynamic pattern gener- Turvey, M. T. (2007). Action and perception at the level of
ation in behavioral and neural systems. Science, 239, synergies. Human Movement Science, 26, 657–697. https://
1513–1520. https://doi.org/10.1126/science.3281253 doi.org/10.1016/j.humov.2007.04.002
Sch€oner, G., Haken, H., & Kelso, J. A. S. (1986). A stochastic Turvey, M. T., & Kugler, P. N. (1984). An ecological approach
theory of phase transition in human hand movement. to perception and action. In H. T. A. Whiting (Ed.), Human
Biological Cybernetics, 53, 247–257. motor actions: Bernstein reassessed (pp. 373–412). North-
Sch€oner, G., Zanone, P. G., & Kelso, J. A. S. (1992). Learning Holland.
as change of coordination dynamics: Theory and experiment. Turvey, M. T., Shaw, R. E. & Mace, W. (1978). Issues in the
Journal of Motor Behavior, 24, 29–48. theory of action: Degrees of freedom, coordinative structures
Segers, V., Aerts, P., Lenoir, M., & De Clercq, D. (2007). and coalitions. In J. Requin (Ed.), Attention and performance
Dynamics of the body centre of mass during actual acceler- VII. Erlbaum.
ation across transition speed. Journal of Experimental Turvey, M. T., Shaw, R. E., Reed, E., & Mace, W. (1981).
Biology, 210, 578–585. Ecological laws of perceiving and acting: In reply to Fodor
Segers, V., De Smet, K., Van Caekenberghe, I., Aerts, P., & and Pylyshyn (1981). Cognition, 9, 237–304. https://doi.org/
De Clercq, D. (2013). Biomechanics of spontaneous over- 10.1016/0010-0277(81)90002-0

2021, Vol. 53, No. 6 795


K. M. Newell & Y.-T. Liu

Van der Kamp, J., Savelsbergh, G. J. P., & Davis, W. E. Zanone, P. G., & Kelso, J. A. S. (1992). Evolution of behav-
(1998). Body-scaled ratio as a control parameter for prehen- ioral attractors with learning: Nonequilibrium phase transi-
sion in 5- to 9-year-old children. Developmental tions. Journal of Experimental Psychology: Human
Psychobiology, 33, 351–361. Perception and Performance, 18, 403–421.
Van Santvrood, A. A. M., & Beek, P. J. (1996). Zanone, P. G., & Kelso, J. A. S. (1997). Coordination dynam-
Spatiotemporal variability in cascade juggling. Acta ics of learning and transfer: Collective and component levels.
Psychologica, 91, 131–151.
Journal of Experimental Psychology: Human Perception and
Vereijken, B. (1991). The dynamics of skill acquisition.
Performance, 23, 1454–1480.
Unpublished doctoral dissertation, Amsterdam: Free University.
Zatsiorsky, V. M. (1998). The kinematics of human motion.
Vereijken, B., Emmerik, R. E. A., Whiting, H. T. A., &
Newell, K. M. (1992). Free(z)ing degrees of freedom in skill Human Kinetics.
acquisition. Journal of Motor Behavior, 24, 133–142. Zhang, M., Kelso, J. A. S., & Tognoli, E. (2018). Critical diver-
Vicinanza, D., Williams, G., Smith, L., Irwin, G., & Newell, sity: Divided or united states of social coordination. PLoS One
K. M. (2018). Limit cycle dynamics of the gymnastics long- 13, e0193843. https://doi.org/10.1371/journal.pone.0193843
swing. Human Movement Science, 57, 217–226. Zhang, M., Beetle., C., Kelso, J. A. S., & Tognoli, E. (2019).
Wang, Z., Ko, Ji-Hyun., Challis, J. H., & Newell, K. M. Connecting empirical phenomena and theoretical models of
(2014). The degrees of freedom problem in human standing biological coordination across scales. Journal of the Royal
posture: Collective and component dynamics. PLoS One, Society Interface, 16, 20190360. https://doi.org/10.1098/rsif.
9(i), e8544. https://doi.org/10.1371/journal.pone.0085414 2019.0360
Warren, W. H. (1984). Perceiving affordances: Visual guidance
of stair climbing. Journal of Experimental Psychology:
Human Perception and Performance, 10, 683–703. Appendix A
Warren, W. H. (2006). The dynamics of perception and action.
Psychological Review, 113, 358–389. https://doi.org/10.1037/ An experimental strategy for determining collective
0033-295X.113.2.358 variables (adapted from Kelso, 1995)
Wickstrom, R. L. (1977). Fundamental motor patterns. Lea &
1. Define the system and the joint motions used to
Feibeiger.
Wimmers, R. H., Beek, P. J., & van Wieringen, P. C. W. perform the task. Consider relevant variables other
(1992). Phase transitions in rhythmic tracking movments: A than joint motions.
case of unilateral coupling. Human Movement Science, 11, 2. Make an ‘ansatz’ as to the system dynamics of the
2117–226. interactions of environment, task and individual and the
Wimmers, R. H., Savelsbergh, G. J., Beek, P. J., & Hopkins, B. likely collective variable(s) and control parameter(s).
(1998). Evidence for a phase transition in the early development 3. Determine the relative stability of the patterns within
of prehension. Developmental Psychobiology, 32, 235–248. a phase transition experiment. The changes in the
Winter, D. (2009). Biomechanics and motor control of human
control parameter should induce phase transitions
movement. Wiley.
Witte, K., Schobesberger, H., & Peham, C. (2009). Motion pat- from less stable to more stable patterns. Phase
tern analysis of gait in horse back riding by means of princi- transitions are critical for the identification of
pal component analysis. Human Movement Science, 28, patterns with different stabilities. Critical fluctuations
394–405. should occur before the phase transition itself and
Witte, K., Bock, H., Storb, U., & Blaser, P. (2003). A syner- possibly hysteresis effects if the scaling is reversed.
getic approach to describe the stability and variability of The time scale of the experiment should be long
motor behavior. In W. Tschacher & J.-P. Dauwalder (Eds.),
enough to observe the phase transitions. Identify the
The dynamic systems approach to cognition: Concepts and
empirical pardigms based on self-organization, embodiment, slowest moving relations or patterns among joint
and coordination dynamics. Studies of nonlinear phenomena motions (potential collective variables).
in life science (Vol10) (pp. 133–144). 4. Scan the dynamics of the slower moving
Witte, K., Ganter, N., Baumgart, C., & Peham, C. (2010). relationships and identify all possible relations
Applying a principal component analysis to movement (patterns) among the components.
coordination in sport. Mathematical and Computer Modeling 5. Identify the preferred and not-preferred patterns.
of Dynamical Systems, 16, 477–488. 6. Mathematically model your system. Play with control
Viviani, P., & Terzuolo, C. A. (1980). Space-time invariance in
parameter and see how the system reacts to changes.
learned motor skill. In G. E. Stelmach and J. Requin (Eds.).
7. Test the model.
Tutorials in motor behavior (pp. 525–533). North-Holland.
Zaal, T. J. M., & Michaels, C. F. (2003). The information for
catching fly balls: Judging and intercepting virtual ball in a Received November 12, 2019
CAVE. Journal of Experimental Psychology: Human Revised October 7, 2020
Perception and Performance, 29, 537–555. Accepted October 8, 2020

796 Journal of Motor Behavior


Copyright of Journal of Motor Behavior is the property of Taylor & Francis Ltd and its
content may not be copied or emailed to multiple sites or posted to a listserv without the
copyright holder's express written permission. However, users may print, download, or email
articles for individual use.

You might also like