You are on page 1of 5

Exp Brain Res (2012) 217:15

DOI 10.1007/s00221-012-3000-4

MI NI-R EVIE W

The bliss (not the problem) of motor abundance (not redundancy)


Mark L. Latash

Received: 18 July 2011 / Accepted: 2 January 2012 / Published online: 14 January 2012
Springer-Verlag 2012

Abstract Motor control is an area of natural science Preamble


exploring how the nervous system interacts with other body
parts and the environment to produce purposeful, coordi- Over ten years ago, a brief Editorial was published in
nated actions. A central problem of motor controlthe Motor Control with the title There is no motor redundancy
problem of motor redundancywas formulated by Nikolai in human movements. There is motor abundance (Latash
Bernstein as the problem of elimination of redundant 2000). In that paper, it was claimed that the formulation of
degrees-of-freedom. Traditionally, this problem has been the classical problem of motor redundancythe MR prob-
addressed using optimization methods based on a variety of lem(Bernstein 1967) was misleading. There are good
cost functions. This review draws attention to a body of reasons to re-visit the redundancy-abundance issue. First,
recent Wndings suggesting that the problem has been formu- many researchers are still focusing their eVorts on the MR
lated incorrectly. An alternative view has been suggested as problem in its original formulation. Second, much experi-
the principle of abundance, which considers the apparently mental data have been collected supporting the claim that
redundant degrees-of-freedom as useful and even vital for the apparently redundant design of the human neuromotor
many aspects of motor behavior. Over the past 10 years, system is not a source of computational problems for the
dozens of publications have provided support for this view central nervous system (CNS). For the purpose of this brief
based on the ideas of synergic control, computational review, I would like to deWne motor control as an area of
apparatus of the uncontrolled manifold hypothesis, and the natural science (physics, including physiology) exploring
equilibrium-point (referent conWguration) hypothesis. In how the nervous system interacts with other body parts and
particular, large amounts of good variancevariance in the environment to produce purposeful, coordinated move-
the space of elements that has no eVect on the overall ments. Motor control is not a subWeld of control theory and/
performancehave been documented across a variety of or engineering trying to decipher software in the brain that
natural actions. Good variance helps an abundant system has to control the poorly designed body plagued with the
to deal with secondary tasks and unexpected perturbations; hosts of complex interactions among body parts and
its amount shows adaptive modulation across a variety of between the body and the environment. So, this mini-
conditions. These data support the view that there is no review accepts as an axiom that the goal of motor control is
problem of motor redundancy; there is bliss of motor to Wnd a set of laws of physics that bring about observed
abundance. patterns of physiological and behavioral variables.

Keywords Motor redundancy Principle of abundance


Synergy Referent conWguration Classical formulation of the MR problem

At any level of analysis of the system for movement pro-


M. L. Latash (&)
Department of Kinesiology, The Pennsylvania State University,
duction, there are more variables produced by elements
Rec.Hall-267, University Park, PA 16802, USA (elemental variables) than constraints associated with typi-
e-mail: mll11@psu.edu cal tasks (Bernstein 1967). For example, for the task of

123
2 Exp Brain Res (2012) 217:15

reaching a target in the three-dimensional space, the num- tion, a whole sequence of MR problems becomes unavoid-
ber of arm joint rotations is typically more than three. For able. These problems are inherent, for example, to the Weld
the task of producing a certain magnitude of joint torque, of internal models based on an idea that neural structures
there are typically several muscles that span the joint. For compute/predict eVects of the interactions among parts of
the task of producing a certain level of muscle activation, the body and between the body and the environment
there are numerous motor units that can be recruited at (reviewed in Kawato 1999). A diVerent view on internal
diVerent frequencies. So, the CNS seems to always con- models assumes that these models represent neural struc-
front problems of choice from an inWnite number of possi- tures that map the realized eVector system changes on
bilities, similar to solving n equations with m unknowns, expected perceptual consequences (Hommel et al. 2001).
n < m. This deWnition is close in spirit to the notion of neural rep-
The most common approach to the MR problem has resentations and does not allow the degree of formalization
been looking for a solution that optimizes (typically, mini- aVorded by the former deWnition.
mizes or maximizes) a cost function (reviewed in Nelson
1983; Prilutsky and Zatsiorsky 2002). Cost functions have
been selected rather arbitrarily, based on engineering prin- Early facts that spoke against the MR problem
ciples or reasonable considerations from physiology, psy-
chology, mechanics, and other areas. Best known examples One of the Wrst classical studies by Nikolai Bernstein pro-
include minimum jerk (Flash and Hogan 1985), minimum duced results that were already suYcient to claim that no
torque-change (Uno et al. 1989), minimum eVort (Hasan unique, optimal solution is found for typical MR problems.
1986), minimum discomfort (Cruse and Bruwer 1987), as Bernstein (1930) studied professional blacksmiths using his
well as more complex cost functions (Rosenbaum et al. original motion analysis system (kimocyclograph) and dis-
2001). Recently, there have been a few attempts at recon- covered that high variability in the joint angle space across
structing unknown cost functions based on sets of experi- repetitive trials was associated with low variability of the
mental observations using inverse optimization (Bottasso hammer trajectory. It is noted that the subjects in that study
et al. 2006; Terekhov et al. 2010). were as well trained as one could possibly imagine for the
Among more recent approaches to the MR problem is task they performed: hitting the chisel with the hammer. So,
optimal feedback control (Todorov and Jordan 2002). Opti- if there were such a thing as an optimal solution for the
mal control is a branch of mathematics developed to Wnd problem of kinematic redundancy, these subjects were in
ways to control a system, which changes in time, such that the best position to discover it. Nevertheless, they showed
certain criteria of optimality are satisWed. The model of repetition without repetition (Bernstein 1967), that is,
Todorov and Jordan minimizes the weighted sum of the repetitive solutions of the task with variable means. A num-
squared diVerence between a function of the eVector out- ber of later studies provided more evidence for variable
puts and its required value and the eVort deWned as the vari- solutions used by humans (and animals) in tasks that
ance of the control signals. This approach does not view involved motor redundancy, even following extensive prac-
MR as a problem (Diedrichsen et al. 2010). It suggests a tice (Jaric and Latash 1999; Latash et al. 2002; Yang and
particular method of producing Xexible behaviors leading Scholz 2005).
to a desired goal, which is one of the trademarks of volun-
tary movements. The approach is based on assuming com-
putations within the central nervous system, an assumption The principle of abundance
that goes against the main axiom formulated in the Wrst par-
agraph. Consider the following problem: When an action potential
is generated on the membrane of a neuron, a huge number
of Na+ ions cross the membrane. There are many more ions
Roots of the MR problem in the vicinity of the site where the action potential is being
generated. How does the CNS decide which ions should
The classical formulation of the MR problem is deeply cross the membrane? This is a problem with zillions of
rooted in Newtonian mechanics and control theory. This unknowns. Likely, 100% of the readers would agree that
formulation is based on an implicit (sometimes explicit) the CNS does not micromanage at such a level. It does not
assumption that, to produce a movement, the controller has care which particular ions participate in the process. The
to make sure that all the involved elements produce certain laws of physics make sure that about the right number of
outputs, in particular that all muscles produce requisite ions cross the membrane.
force time proWles (Hinder and Milner 2003; for review see Consider now the aforementioned problem of deWning a
Shadmehr and Wise 2005). If one starts with this assump- motor unit recruitment pattern for a desired level of muscle

123
Exp Brain Res (2012) 217:15 3

activation. Does the CNS care about speciWc patterns of problems. VGOOD emerges in optimal feedback control
recruitment or does it allow laws of physicsincluding the schemes that allow numerous solutions to emerge; these
size principle (Henneman et al. 1965)solve the problem? approaches look for both optimal solutions and a feedback
In this context, physics implies classical physics, chemis- law around them (Todorov and Jordan 2002).
try, and physiology; however, it does not imply computa- VGOOD is not simply a by-product of the imperfect sys-
tion. Does the CNS care about speciWc solutions for the tem design, neuromotor noise. Several studies have
problems of sharing joint torque among the muscles and shown that the CNS increases the relative amount of VGOOD
sharing Wngertip trajectory among the joints of the limb? (and sometimes even its absolute magnitude) in conditions
There will be less agreement among researchers with of possible changes in the target location (de Freitas et al.
respect to these questions. 2007), practice in an unusual force Weld (Yang et al. 2007),
An alternative view on the apparently redundant design practice of a novel task (Latash et al. 2002; Yang and
of the neuromotor system was suggested as the principle of Scholz 2005), and fatigue of one of the elements (Singh
abundance (Gelfand and Latash 1998). According to this et al. 2010). VGOOD has been shown to help an abundant
principle, the CNS facilitates families of solutions equally system to deal with secondary tasks (Zhang et al. 2008) and
able to solve the task. It does so by imposing (time-varying) unexpected perturbations (Gorniak et al. 2009; Mattos et al.
constraints on the system body + environment (cf. Hu 2011).
and Newell 2011) and allowing solutions to emerge given
the actual state of this system (which never repeats exactly
over repetitive trials). In other words, laws of physics The equilibrium-point (referent conWguration)
deWne behavior of the neuromotor system at any level just hypothesis
like they deWne motion of Na+ during the action potential
generation. The phrase CNS imposes constraints should According to the opinion of the author, currently there is
also be viewed as a reXection of our current lack of knowl- only one theory of motor control that is based on the physi-
edge of physical processes underlying decision making by cal approach to movement production and is compatible
the brain. For remarkable reviews and insights on this issue, with the known physiology of the neuromotor system, the
see Kugler and Turvey (1987), Strepp and Turvey (2010). equilibrium-point hypothesis (Feldman 1966, 1986). Origi-
nally this hypothesis was developed for one-muscle and
one-joint systems. It suggested that the very complex sys-
Evidence that motor abundance is a bliss tem of reXex pathways controlling a muscle (which
involves many apparent MR problems) can be adequately
Over the past ten years, dozens of publications have pro- described with setting just one parameter, threshold () of
vided support for the view that there is no MR problem, but muscle activation to length. This method of control is an
instead there is bliss of motor abundance. One such direc- example of applying the principle of abundance to a partic-
tion of research has been associated with the uncontrolled ular subsystem within the body. The controller imposes a
manifold (UCM) hypothesis (Scholz and Schner 1999; constraint on the system (sets a value of ), which facili-
Latash et al. 2007; Martin et al. 2009). The UCM hypothe- tates numerous patterns of neuronal activation, including
sis assumes that physical processes associated with per- motor unit Wring patterns, for any task, for example, mov-
forming a movement by a redundant system can be ing a constant external load to a new location.
adequately expressed as creating a sub-space (UCM) in the Recently, the equilibrium-point hypothesis has been
space of elemental variables corresponding to a desired generalized for multi-joint action in the form of the referent
value of an important performance variable to which all the conWguration (RC) hypothesis (Feldman 2009). The RC
elements contribute. It has to be noted that elemental vari- hypothesis may be viewed as reXecting a hierarchical sys-
ables are not controlled in any sense. It is important that tem, where at the upper level of the hierarchy, subthreshold
there is theoretical availability and actual use of numerous depolarization of a neuronal pool deWnes a RC of the body
(inWnite number of) solutions in typical tasks. Variance (given external force Weld) in terms of task-related vari-
along the UCM does not aVect performance and has been ables. RC is a conWguration at which all muscles are at
addressed as good (VGOOD), while variance orthogonal to thresholds of their activation. RC may not be attainable due
that space is bad (VBAD). Large amounts of VGOOD have to external and anatomical constraints; as a result, an equi-
been documented in studies of multi-joint kinematic tasks, librium state may emerge with non-zero levels of muscle
multi-digit kinetic tasks, and multi-muscle tasks (reviewed activation reXecting the diVerence between the actual con-
in Latash et al. 2007). VGOOD, by deWnition, has no eVect on Wguration and RC. After a RC is established, a sequence of
performance and, therefore, cannot be predicted by any few-to-many mappings leads to inputs into -motoneuronal
approach looking for single optimal solutions for the MR pools that deWne thresholds of the tonic stretch reXex for

123
4 Exp Brain Res (2012) 217:15

the involved muscles (Latash 2010). At each step of the on experimental observations, inverse optimization (Terekhov
hierarchy, the system functions according to the principle et al. 2010), have been promising, but interpretation of the
of abundance: No MR problems are being solved, but task- reconstructed cost functions is still at its infancy and does
speciWc constraints are imposed by the controller thus not go beyond rather simplistic biomechanical consider-
aVording the system Xexibility of getting to acceptable ations (Park et al. 2011a, b).
solutions.

Generalization of the principle of abundance


Two aspects of synergies: sharing and variable solutions
While the MR problem explicitly refers to motor redun-
Recent developments of the principle of abundance have dancy, such problems are inherent to other aspects of the
led to the idea of synergies as neural organizations that functioning of the CNS. For example, the idea of sensory
ensure task-speciWc co-variation of elemental variables abundance and sensory synergies has been discussed
providing for desired stability properties of an important (Latash 2008). Indeed, information on the persons own
output (performance) variable. Imagine a simple task body and external world is delivered by receptors of diVer-
involving a redundant system, for example pressing with ent modalities, and the task of the brain in to construct a
two Wngers to produce a speciWc total force magnitude, single, coherent picture of the body in the world based on
FTASK. Any point that belongs to the line F1 + F2 = FTASK all these signals (and preconceptions). This picture is stable
solves the task. Imagine now that a subject performs this if one or a few of the sensory signals become unavailable,
task many times and the data for each trial are plotted as for example when a person closes the eyes or puts on ear-
points on the forceforce plane (Fig. 1). Two major charac- phones. Recent studies on sensory re-weighting suggest
teristics of the data distribution can be introduced: (1) the that the CNS uses sensory signals of diVerent modalities
location of the center of the distribution; and (2) the shape with weights that depend on the reliability and availability
of the distribution. For simplicity, in Fig. 1, it is assumed of those signals (Maurer et al. 2006). Regularities in lin-
that the distribution forms a perfect ellipse and its center is guistic tasks involving picture description have been dem-
on the line of perfect performance. onstrated despite the variability across individual trials
The method developed within the UCM hypothesis can (Latash et al. 2011). In general, any decision making is typ-
be used to quantify the amount of VGOOD (along the dashed ically based on abundant information; the recent idea of
line in Fig. 1) and VBAD (orthogonal to this line). The other Feldman (2011) that control with thresholds for activation
characteristic, the location of the center of the data distribu- of neuronal pool may be involved in all cognitive processes
tion, may indeed be formally described as following an is compatible with the main message of this mini-review:
optimization principle. It is noted that in this context, opti- We are blessed with abundance; so, let us not waste time
mization is used not to deWne a single optimal solution for trying to eliminate it.
the task (in Fig. 1, the task is always the same, but the solu-
tions obviously vary) but the center of a distribution of Acknowledgments Preparation of this paper was in part supported
by NIH grant NS-035032.
solutions. Recent attempts at computing cost function based

References

Bernstein NA (1930) A new method of mirror cyclographie and its


application towards the study of labor movements during work on
a workbench. Hyg, Saf Pathol Labor 5:39, and 6:311 (in Rus-
sian)
Bernstein NA (1967) The co-ordination and regulation of movements.
Pergamon Press, Oxford
Bottasso CL, Prilutsky BI, Croce A, Imberti E, Sartirana S (2006) A
numerical procedure for inferring from experimental data the
optimization cost functions using a multibody model of the neuro-
Fig. 1 a An illustration of the notions of good and bad variance musculoskeletal system. Multibody Syst Dyn 16:123154
for the task of producing a magnitude of total force by pressing with Cruse H, Bruwer M (1987) The human arm as a redundant manipula-
two index Wngers on individual force sensors. The dashed line shows tor: the control of path and joint angles. Biol Cybern 57:137144
the sub-space corresponding to perfect task execution. b For the same de Freitas SM, Scholz JP, Stehman AJ (2007) EVect of motor planning
task, diVerent data distributions are illustrated corresponding to diVer- on use of motor abundance. Neurosci Lett 417:6671
ent sharing patterns (diVerent locations of the centers of the distribu- Diedrichsen J, Shadmehr R, Ivry RB (2010) The coordination of move-
tions, cf. D1 and D2) and diVerent co-variation (diVerent relative ment: optimal feedback control and beyond. Trends Cogn Sci
amounts of good and bad variance, cf. D2 and D3) 14:3139

123
Exp Brain Res (2012) 217:15 5

Feldman AG (1966) Functional tuning of nervous system with control Mattos D, Latash ML, Park E, Kuhl J, Scholz JP (2011) Unpredictable
of movement or maintenance of a steady posture. II. Controllable elbow joint perturbation during reaching results in multijoint
parameters of the muscles. Biophysics 11:565578 motor equivalence. J Neurophysiol 106:14241436
Feldman AG (1986) Once more on the equilibrium-point hypothesis Maurer C, Mergner T, Peterka RJ (2006) Multisensory control of hu-
(-model) for motor control. J Mot Behav 18:1754 man upright stance. Exp Brain Res 171:231250
Feldman AG (2009) Origin and advances of the equilibrium-point Nelson W (1983) Physical principles for economies of skilled move-
hypothesis. Adv Exp Med Biol 629:637643 ments. Biol Cybern 46:135147
Feldman AG (2011) Space and time in the context of equilibrium-point Park J, Zatsiorsky VM, Latash ML (2011a) Finger coordination under
theory. Wiley Interdiscip Rev: Cogn Sci 2:287304 artiWcial changes in Wnger strength feedback: a study using ana-
Gelfand IM, Latash ML (1998) On the problem of adequate language lytical inverse optimization. J Mot Behav 43:229235
in movement science. Mot Control 2:306313 Park J, Sun Y, Zatsiorsky VM, Latash ML (2011b) Age-related chang-
Gorniak SL, Feldman AG, Latash ML (2009) Joint coordination dur- es in optimality and motor variability: an example of multi-Wnger
ing bimanual transport of real and imaginary objects. Neurosci redundant tasks. Exp Brain Res 212:118
Lett 456:8084 Prilutsky BI, Zatsiorsky VM (2002) Optimization-based models of
Henneman E, Somjen G, Carpenter DO (1965) Excitability and inhib- muscle coordination. Exerc Sport Sci Rev 30:3238
itibility of motoneurones of diVerent sizes. J Neurophysiol Rosenbaum DA, Meulenbroek RJ, Vaughan J, Jansen C (2001) Pos-
28:599620 ture-based motion planning: applications to grasping. Psychol
Hinder MR, Milner TE (2003) The case for an internal dynamics mod- Rev 108:709734
el versus equilibrium point control in human movement. J Physiol Scholz JP, Schner G (1999) The uncontrolled manifold concept: iden-
549:953963 tifying control variables for a functional task. Exp Brain Res
Hommel B, Msseler J, Aschersleben G, Prinz W (2001) The theory of 126:289306
event coding (TEC): a framework for perception and action plan- Shadmehr R, Wise SP (2005) The computational neurobiology of
ning. Behav Brain Sci 24:849878 reaching and pointing. MIT Press, Cambridge, MA
Hu X, Newell KM (2011) Modeling constraints to redundancy in Singh T, SKM V, Zatsiorsky VM, Latash ML (2010) Fatigue and
bimanual force coordination. J Neurophysiol 105:21692180 motor redundancy: adaptive increase in force variance in multi-
Jaric S, Latash ML (1999) Learning a pointing task with a kinemati- Wnger tasks. J Neurophysiol 103:29903000
cally redundant limb: emerging synergies and patterns of Wnal Stepp N, Turvey MT (2010) On strong anticipation. Cogn Syst Res
position variability. Hum Move Sci 18:819838 11:148164
Kawato M (1999) Internal models for motor control and trajectory Terekhov AV, Pesin YB, Niu X, Latash ML, Zatsiorsky VM (2010) An
planning. Curr Opinions Neurobiol 9:718727 analytical approach to the problem of inverse optimization: an
Kugler PN, Turvey MT (1987) Information, natural law, and the self- application to human prehension. J Math Biol 61:423453
assembly of rhythmic movement. Erlbaum, Hillsdale, NJ Todorov E, Jordan MI (2002) Optimal feedback control as a theory of
Latash ML (2000) There is no motor redundancy in human move- motor coordination. Nat Neurosci 5:12261235
ments. There is motor abundance. Mot Control 4:257259 Uno Y, Kawato M, Suzuki R (1989) Formation and control of optimal
Latash ML (2008) Synergy. Oxford University Press, NY trajectory in human multijoint arm movement. Biol Cybern
Latash ML (2010) Motor synergies and the equilibrium-point hypoth- 61:89101
esis. Mot Control 14:294322 Yang JF, Scholz JP (2005) Learning a throwing task is associated with
Latash ML, Kang N, Patterson D (2002) Finger coordination in per- diVerential changes in the use of motor abundance. Exp Brain Res
sons with Down syndrome: atypical patterns of coordination and 163:137158
the eVects of practice. Exp Brain Res 146:345355 Yang JF, Scholz JP, Latash ML (2007) The role of kinematic redun-
Latash ML, Scholz JP, Schner G (2007) Toward a new theory of mo- dancy in adaptation of reaching. Exp Brain Res 176:5469
tor synergies. Mot Control 11:276308 Zhang W, Scholz JP, Zatsiorsky VM, Latash ML (2008) What do
Latash ML, Sun Y, Latash EM, Mikaelian IL (2011) Speed-diYculty synergies do? EVects of secondary constraints on multi-digit
trade-oV in speech: Chinese vs. English. Exp Brain Res 211:193 synergies in accurate force-production tasks. J Neurophysiol
205 99:500513
Martin V, Scholz JP, Schner G (2009) Redundancy, self-motion, and
motor control. Neural Comput 21:13711414

123