See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/247843393

Neural Control of Locomotion

Article in BioScience · December 1989

DOI: 10.2307/1311186

CITATIONS READS

51 1,419

1 author:

Gerald E Loeb
University of Southern California
402 PUBLICATIONS 17,571 CITATIONS

SEE PROFILE

All content following this page was uploaded by Gerald E Loeb on 06 February 2015.

The user has requested enhancement of the downloaded file.

 Neural Control of Locomotion
How do all the data fit together?

Gerald E. Loeb

L ocomotion is a particularly
richly studied but frustrating
aspect of biolpgy. Much is
known about individual components
of the process, but there is.no general
N o machines
successfully walk,
properties of the entirely conceptual
entities called central pattern genera-
tors, which are thought to control the
sequential recruitment of these mus-
cles. And it confronts the almost be-
hypothesis for how it works. swim, or fly using wildering complexity of the few neu-
Vertebrate locomotion is distin- the principles ral circuits that have been identified
guished from invertebrate locomotion to date, circuits that presumably con-
by the conjunction of two features. animals employ tribute to both levels of control.
The skeletal movements are complex The amount of data available is
but they are usually regular, repeating impressive and the neurophysiologi-
sequences (see Hildebrand page 766 We might be able to convince our- cal research involved in obtaining" it
this issue), and most of the organisms selves that all of these descriptions, in has been arduous, even heroic. But
are massive enough that the amount toto, constitute an understanding of the general biologist will likely and
of energy and force invested in animal locomotion, except for one rightfully be uneasy with how it all
moving the whole animal constitutes telling failure: there have been no might fit together. The last sections of
a significant percentage of its meta- successful machines that walk, swim, this article look to comparative, de-
bolic economy and is a significant or fly using any of the principles that velopmental, and evolutionary biol-
determinant of the details of its body have served animals so well. It is not ogy -for a perspective that is often
plan (see Biewener page 776 this is- for lack of trying. Although boats missing in the current rush to reduc-
sue). with fins and planes with flapping tionism.
At the phenomenological level, we wings are now reminiscent of Jules
have good descriptions of the move- Verne, walking machines with legs Reflexive control of muscles
ments and even the mechanics in- remain an active interest of industries
volved in walking, swimming, and such as forestry, mining, and defense. Each muscle is endowed with its own
flying. At reductionistic and compar- Modern materials and electronics ensemble of sense organs that report
ative levels, we have catalogued and can provide mechanical assemblies, on mechanical conditions within the
differentiated the properties of skele- actuators, and sensors that are stron- muscle, including its length and veloc-
tons (see Gordon page 784 this issue), ger, lighter, faster, and simpler than ity of stretch and contraction (muscle
muscles (see Weeks page 791 this the corresponding bones, muscles, spindles) and its force output (Golgi
issue), and sensory and motor neu- and receptors of living organisms. Yet tendon organs). In most limb muscles,
rons. We even know that the critical even a child knows that robots "walk the neurons from these sense organs
neural circuits for generating locomo- funny," characterized by graceless carry signals to other neurons, called
tor patterns are, for the most part, and inefficient gaits on regular terrain alpha motoneurons, which activate
confined to a relatively small and and prone to disaster in the face of the muscle fibers. This connection
primitive part of the central nervous unexpected obstacles. may be direct (via a single synapse) or
system, the spinal cord. The problem is that we do not through neurons within the spinal
really have a theory of control for cord (called interneurons). These
Gerald E. Loeb is the director of special locomotion. This article considers the pathways, or projections, give rise to
projects in the Bio-Medical Engineering relationship between the epiphenom- reflexes such as the clinical tendon
Unit, Queen's University, Kingston, On- ena known as reflexes and the mo- jerk, in which a sudden stretch of the
tario K7L 3N6, Canada. O 1989 Ameri- ment-to-moment regulation of indi- muscle spindles results in a rapid con-
can Institute of Biological Sciences. vidual muscles. It describes the traction from reflexively activated

BioScience Vol. 39 No. 11

muscle fibers (reviewed by McCrea
1986). In addition, each mammalian
spindle contains several different
types of intrafusal muscle fibers under
the independent control of special
gamma motoneurons. The output of
the spindle afferents represents a com-
plex sum of extrinsic muscle length
changes and intrinsic contractions of
the intrafusal muscle fibers (reviewed
bv Loeb 1984).
*These arran'gements of propriocep-
tive sensory feedback are reminiscent I I
of the feedback loom used in the ser- length follow-up servo a-y coact~vat~on stlff ness regulator
vocontrol of industrial torque motors. - Merton's hypothes~s servo-assist
Thus it is not surprising that servocon-
trol loops have figured prominently in Figure 1. Previously proposed servocontrol systems. The muscle spindle afferents (Ia)
hypotheses for motor control. Figure 1 can be driven by intrinsic activity of gamma motoneurons (y) as well as extrinsic length
shows a historical progression of such changes; their synaptic projections excite the alpha motoneurons (a;y-shaped ending)
controlling the extrafusal muscle fibers. a. Merton (1953) originally proposed that the
schemes that were designed to account length of a muscle could be commanded by activating the gamma motoneurons alone,
for vatterns of connection inferred with consequent alpha motoneuronal excitation via the Ia pathway, until the muscle
from' large numbers of animal and shortened sufficiently to off-load the spindles. b. Recordings of spindle activity during
human experiments. voluntary contraction in humans (Vallbo 1974a, b) led Granit (1975) to propose that
More recently, interest has shifted the alpha and gamma rnotoneurons were coactivated to form a servoassisted system
from the control of individual mus- under direct descending control. c. Houk (1979) incorporated the negative feedback
cles to the control of individual joints (ball-shaped ending) from the Golgi tendon organs in proposing that the spinal cord
through the balance of forces in an- regulated the muscle's stiffness (the ratio of force and length changes in response to a
tagonistic muscle pairs. By specifying perturbation). (Adapted from Loeb 1984.)
an operating point for the stiffness of
each muscle, the central nervous sys-
tem effectively could specify a target efficiency depend on the anticipation three reasons:
position at which the system is in of limb trajectories, which are deter-
equilibrium (Berkinblit et al. 1986, mined at least as much by inertia as Movement at any particular joint
Bizzi et al. 1982). Servocontrol for by muscular effort. has an effect on foot position, for
such a push-pull system requires re- For example, the motion of the example, that depends on the orien-
civrocal connections. in which the shank and foot during the swing tation of the proximal segment de-
effects of feedback signals from any phase of the cat hind limb depends on fining that joint, which in turn de-
given source are opposite for the two translational accelerations of the pel- pends on the angles of all the joints
antagonistic muscles. Figure 2 shows vis that propagate through the pen- proximal to it.
known projections of reciprocal inter- dant limb like the oscillations of a Intersegmental coupling causes mo-
neurons that could subserve such con- marionette (Hoy and Zernicke 1985). tion at any joint to generate torques
trol (Baldissera et al. 1981). This mechanical coupling saves en- at all joints (consider the mario-
ergy, because these rapid movements nette). These effects can be surpris-
Limb mechanics would be inefficient to produce by ingly large and even counterintui-
contraction of the individual muscles tive (Zajac and Gordon 1989).
Unfortunately, all of the above are operating across each joint. If the Many muscles cross more than one
basically static control schemes. In motion is perturbed (for example, by joint. Their torques at each joint
principle, continuous movements contact with an obstructing object), cannot be set independently by sin- 1

such as locomotion could be repre- the response is not an attempt to gle-joint controllers. The force out-
sented by a dynamic trajectory of return to the .desired course, but put of such muscles in response to
equilibrium positions. However, for rather a complex "stumbling correc- the commands of their motoneu-
most animals the goal of locomotion tive reaction" (Forssberg 1979) de- rons depends on their net length
is not a sloth-like gliding through a signed to get around the unseen ob- and velocity, which is influenced by
series of stable postures.1 Rather it is ject. Both of these strategies appear to motion at all joints that they cross.
to maintain a dynamic equilibrium in reflect a control scheme that is based
which postural stability and energy not on individual muscles or joints spinal circuiny
but rather on the mechanics of the
whole limb. If the control of any single muscle
'This scheme has been the strategy of most The dynamic mechanical properties depends on postural and dynamic
legged robots, but see Raibert (1986) for some
surprisingly simple and effective strategies for of a multisegmented limb are not a conditions throughout the limb, then
achieving dynamic equilibrium in high-speed simple combination of the properties we might expect rather more complex
movements of hopping machines. of the individual joints for at least control circuits than those shown in

December 1989
 whose force is sensed by the tendon of each muscle's natural activity may
organ. Many other equally complex be controlled individually (Grillner
interneurons undoubtedly remain to 1986). Furthermore, the reflex path-
be discovered. Methodological prob- ways projecting from sensory to mo-
lems have generally precluded system- tor neurons may be turned on and off
atic identification of interneurons that (gated) at different points in the step
synapse on other interneurons instead cycle so as to change the behavioral
of projecting directly to identifiable response to a given mechanical per-
motoneurons. turbation (Abraham et al. 1985). This
evidence suggests that the CPG must
Central pattern generators have more than two internal states.
Alternative models to the simple flip-
Although all of this sensory feedback flop include ring-cycle oscillators and
is undoubtedly important, much of systems of loosely coupled but inde-
the initiation and timing of muscle pendent oscillators, such as may con-
action seems to be the work of a trol the segmented axial muscles used
Figure 2. Reciprocal control of antagonist self-oscillating network of interneu- by fish for swimming (Grillner 1985).
muscle pair. A joint tends to reach an rons in the spinal cord that produces
equilibrium position in which equal a cyclic pattern of signals. The most
direct demonstration comes from pat- Scaling of motor control
torques are contributed by the antagonis-
tic muscles acting on that joint. Because of
the length-dependency of force output of terns of electrical activity recorded Comparative biologists have a long-
muscles (arising from both intrinsic prop- from muscle nerves in cats that have standing interest in the scaling (allom-
erties of their myofilarnentq.andextrinsic had the brain severed from the spinal etry) of musculoskeletal structure in
organization of stretch reflexes), they may cord surgically and the muscle con- animals of different sizes. Consider
be modeled as springs whose stiffness is tractions blocked pharmacologically. the strength-to-weight ratio of two
established by the motor commands (Bizzi Despite the complete absence of similar structures, one of which has
et al. 1982; illustration adapted from movement to produce sensory feed- all of its linear dimensions ten times
Kandel and Schwaitz 1984). Here the back, the spinal cord can produce greater than the other's. Because
alpha motoneurons (a)are shown con- sequences of motoneuronal activity strength is related to cross-sectional
nected to pull on ratchet mechanisms in
series with springs:.Sensory feedback in that are similar to those recorded area (x2)whereas mass depends on
such systems appears to be reciprocally during normal walking (Perret 1983). volume (x3),the larger structure has a
organized. The spindle affeients of one Recently, it has been shown that strength-to-weight ratio that is only
musde (la) directly exate the local alpha isolated sections of the spinal cord as 10% that of the smaller structure.
motoneurons as well as inhibitory inter- short as two or three vertebral seg- Therefore, large animals appear
neurons (la In) that act to reduce the ments produce cyclical outputs (Grill- to be built much more sturdily than
output of alpha motoneurons of the an- net 1985), suggesting that locomotor scaled-up versions of small animals
tagonist musde. These inhibitory inter- patterns are generated by a simple, (McMahon and Bonner 1983).
neurons also turn the self-inhibition of localized, and presumably primitive
motoneurons via the Renshaw cells (RC) Dimensional changes affect neuro-
into net excitation (actually disinhibition) circuit in the spinal cord, the so-called physiological as well as mechanical
of the antagonist motoneurons. central pattern generator (CPG). function. One of the limiting factors
However, there is little agreement on in achieving stable control in any mo-
- -
the nature of this oscillator or the tor system (robotic or biological) is
Figures 1 and 2, with motoneurons identity of the interneurons from delay in the feedback circuits. If cor-
receiving sensory feedback not only which it is formed. rective responses are initiated on the
from their own muscles, but also At first, it seemed that gaits such as basis of outdated information, the
from many other muscles and joints. walking and trotting could be divided system will tend to overcorrect; if the
In fact, that is what neurophysiolo- into just two reciprocating phases of strength (gain) of the feedback circuit
gists have discovered when such cir- muscle action in each limb (Engberg is high enough, the system may oscil-
cuits have been traced through the and Lundberg 1969), corresponding late wildly. In the neuromuscular sys-
spinal cord using electrophysiological to the stance phase (requiring mostly tem, significant delays occur at every
and histological techniques. extensor, load-bearing muscles) and stage, including the transmission of
Figure 3 shows a more realistic the swing phase (requiring mostly impulses along sensory axons to the
picture of the input-output connec- flexor, foot-lifting muscles). This spinal cord, processing of sensory in-
tivity of one neuron, called the Ib- model suggested that the CPG formation through interneuronal cir-
inhibitory interneuron (Harrison and worked like a flip-flop (Lundberg cuits, transmission back to the muscle
Jankowska 1985), which appears 1980), a common circuit used to con- along motor axons, spread of action
only as a ball-shaped ending signify- trol timing in computers and one eas- potentials along muscle fibers, and
ing a logical invertor gate in Figure 1. ily realized by reciprocally connected release and diffusion of calcium ions
This neuron converts excitatory syn- interneurons. to catalyze the formation of force-
apses from Golgi tendon organs into However, more recent EMG stud- producing cross-bridges. In animals
inhibitory synapses onto the mo- ies of a wider range of muscles have of different sizes, the various delays
toneurons supplying the muscle fibers noted that the amplitude and timing scale in quite different ways, suggest-

802 BioScience Vol. 39 No. 1 1

ing that the animals must use rather
different control strategies. a b
At a simple level, consider just the
Joint Corticosp.
control of large versus small muscles
having simple architecture (parallel
fibers). The larger girth is achieved
with a larger number of muscle fibers,
with relatively little increase in indi-
vidual fiber diameter. Electromechan-
ical properties of the fibers also tend
to remain fairly constant, including
the velocity of action potential con-
duction along the fiber and the rate at
which tension can be developed.
However, this simple constancy
poses a threat of instability. If muscle
fibers are more than a few centimeters
long, the time taken for action poten- -
1
-<
-
---
-
-7-Z'
z-.

tials to spread from the neuromuscu-

lar synapse (usually near the midpoint
of the fiber) to the ends of the fiber
tends to be greater than the delay in '
the onset ofv contractile force in. the Figure 3. a. Negative feedback from force-sensing afferents in the Golgi tendon organs
cross-bridges. Ther'efore. the middle (Ib) arises through the action of inhibitory interneurons (Ib In), which project to the
of the fib; would be abie to contract motoneurons (a)that control the muscle fibers that affect the tendon organs. However,
at the expense of lengthening the still- these interneurons also receive direct and indlrect input from a variety of other sensory
passive ends. Once lengthened past modalities including muscle spindles (Ia) and from higher control levels (corticospinal,
their optima'l length, these distal sar- rubrospinal, and reticulospinal pathways).b. In addition to inhibiting the motoneurons
comeres could not generate enough that control their parent muscles, the Ib interneurons inhibit motoneurons controlling
force to pull themselves back into many other muscles (Het a),and they send sensory information to higher control levels
homogeneity with the rest of the mus- (e.g., via the dorsal spinocerebellar tract, d.s.c.t.). (Modified from Loeb 1987.)
cle when the action potential finally
reached the ends. power of linear dimension. isfactory performance according to a
This ~ r o b l e mhas been overcome The synaptic propagation delays mix of criteria, including energetic
by a complex (and generally over- through the interneurons are proba- economy, peak strength, stability,
looked) architectural specialization. bly quite similar in large and small and safety. Only then could it wait for
In the long muscles of larger animals, animals. Therefore, interneuronal further evolutionary refinement.
muscle fascicles are composed of sev- computation tends to represent a This problem' suggests a general
eral short muscle fibers in series (Loeb much lower percentage of total feed- constraint on viable motor-control
et al. 1987). The motor axons serving back delay in large animals. Alterna- strategies for living organisms. A suc-
such fascicles conduct action Doten- tively, large animals might use more cessful controller for a robot needs to
tials much faster than the muicle fi- complex chains of interneurons to deal only with a fixed mechanical
bers. Branches of these motor axons achieve responses that are more care- apparatus; a neural controller must
innervate short muscle fibers distrib- fully computed rather than hastily be capable of adapting gracefully to
uted over the length of the fascicle. composed. This,strategy would seem mechanical changes during the evolu-
which can thereby be activated all practical given the much more drastic tion of the species, plus the growth
most synchronously. cons,equences of falls in large versus and development of individuals.
Differential scaling problems such small animals. These considerations may prove to be
as this must abound in the neural more instructive in developing and
control of large limbs. Conduction Evolution of testing hypotheses for biological mo-
delay in peripheral nerves depends on neuromuscular control tor control than the more direct ap-
axon length
u divided bv its conduction proach of circuit tracing in mature
velocity. Because axon diameters are The solutions to scaling problems en- specimens, in which the details may
relatively similar among animals, this tail parallel changes in both muscu- obscure such general principles.
delay tends to rise proportionately loskeletal (mesenchymal) and neu- Recently, we have begun to apply
kith the lengthu
of the limb. Fortu- ronal (ectodermal) subsystems. Yet in engineering tools to design distrib-
nately, larger animals may need to all likelihood, most of the individual, uted controllers that improve perfor-
update their motor programs much random mutations involved in each mance of a given sensorimotor system
less freauentlv than small animals be- evolutionary step have affected only (described mathematically as a com-
cause o l the Stabilizing effect of their one of the two subsystems. To sur- puter model) for particular perfor-
relatively much larger rotational in- vive, the modified organism must im- mance criteria (Loeb et al. in press).
nertias, which rise as the fourth mediately have had a reasonably sat- The underlying hypothesis is that dur-

December 1989 8 03
ing locomotion, the spinal cord inter- multidisciplinary approach, because 41: 99-114.
neurons route proprioceptive feed- it is basic to the lives of almost all Hoy, M. G., and R. F. Zernicke. 1985. Modu-
back to obtain the desired stability of animals, assuring a highly con- lation of limb dynamics in the swing phase of
locomotion. J. Biomech. 18: 49-60.
the limb as a whole, considering both strained and orderly evolution. Kandel, E. R., and J. H. Schwartz. 1984. Prin-
the intrinsic mechanical properties of ciples of Neural Science. Elsevier, Amster-
the musculoskeletal system and the References cited dam.
range of perturbations that are ex- Loeb, G. E. 1984. The control and responses of
pected on the basis of prior experi- Abraham, L. D., W. B. Marks, and G. E. Loeb. mammalian muscle spindles during normally
1985. The distal hindlimb musculature of the executed motor tasks. Exercise Sport Sci.
ence. Given the above-noted mechan- cat. 11. Cutaneous reflexes during locomo- Rev. 12: 158-204.
ical differences between species and tion. Exp. Brain Res. 58: 5 9 4 6 0 3 . . 1987. Hard lessons in motor control
during development, this approach Baldissera, F., H. Hultborn, and M. Illert. from the mammalian spinal cord. Trends
should be able to ,generate testable 1981. Integration in spinal neuronal systems. Neurosci. 10: 108-113.
predictions of differences in spinal Pages 509-595 in V. B. Brooks, ed. Han'd- Loeb, G. E., J. He, and W. S. Levine. In press.
book of Physiology. Section 1, The Nervous Spinal cord circuits: are they mirrors of
cord circuitry. However, assembling System, vol. 111. Motor Control, part 1. musculoskeletal control? 1. Mot. Behav.
and validating the necessary model of American Physiology Society, Washington, Loeb, G. E., C. A. Pratt, .C. M. Chanaud, and
-,-
the musculoskeletal system for just UL. F. J. R. Richmond. 1987. Distribution and
the adult cat has been a major task, Berkinblit, M. B., A. G. Feldman, and 0. I. innervation of short, interdigitated muscle
still in progress after seven years. Fukson. 1986. Adaptability of innate motor fibers in parallel-fibered muscles of the cat
patterns and motor control mechanisms. Be- hindlimb. 1. Morphol. 191: 1-15.
hav. Brain Sci. 9: 585-638. Lundberg, A. 1980. Half-centres revisited.
Conclusions Biewener, A. A. 1989. Mammalian terrestrial Adv. Physiol. Sci. 1: 155-167.
locomotion and body size. BioScience 39: McCrea, D. A. 1986. Spinal cord circuitry and
Both the musculoskeletal mechanics 766-783. motor reflexes. Exercise Sport Sci. Rev. 14:
Bizzi, E., N. Accornero, W. Chapple, and N. 105-141.
and the neural control circuitry are Hogan. 1982. Arm trajectory formation in McMahon, T. A., and J. T. Bonner. 1983. On
sufficiently complex that intuitive monkeys. Exp. Brain Res. 46: 139-143. Size and Life. Scientific American Books,
guesses about how locomotion works Engberg, I., and A. Lundberg. 1969. An elec- New York.
are unlikely to be useful. Quantitative tromyographic analysis of muscular activity Merton, P. A. 1953. Speculations on the servo-
analytical tools ,are needed just to in the hindlimb of the cat during unre- control of movement. Pages 247-255 in
strained locomotion. Acta Physiol. Scand. G. E. W. Wolstenholme, ed. The Spinal
compile all of the currently available 75: 6 1 4 6 3 0 . Cord. Churchill, London.
information about each subsystem Forssberg, H. 1979. Stumbling corrective reac- Perret, C. 1983. Centrally generated pattern of
and to direct and interpret future ex- tion: a phase-dependent compensatory reac- motoneuron activity during locomotion in
perimentation. Beyond simply coping tion during locomotion.]. Neurophysiol. 42: the cat. Pages 405-422 in A. Roberts and B.
with this mountain of heterogeneous 936-953. I Roberts, eds. Neural Origin of Rhythmic
Gordon, K. R. 1989. Adaptive nature of skele- Movements: Symposia of the Society for
data, such tools offer the hope of tal design. BioScience 39: 7 8 4 7 9 0 . Experimental Biology. Cambridge Univer-
developing formal links between the Granit, R. 1975. The functional role of the sity Press, Cambridge, UK.
subsystems. muscle spindles: facts and hypotheses. Brain Raibert, M. H. 1986. Legged Robots that Bal-
To accomplish this, the tradition- 98: 531-556. ance. MIT Press, Cambridge, MA.
ally separate experimental disciplines Grillner, S. 1985. Neurobiological bases of Vallbo, A. B. 1974a. Afferent discharge from
rhythmic motor acts in vertebrates. Science human muscle spindles in noncontracting
of biomechanics and neurophysiology 48: 143-149. muscles. Steady state impulse frequency as a
need to develop much ,closer ties so . 1986. The effect of L-DOPA on the function of joint angle. Acta Physiol. Scand.
that motor control ~roblemscan be spinal cord: relation to locomotion and the 90: 303-318.
attacked simultaneiusly from both half center hypothesis. Pages 269-277 in . 1974b. Human muscle spindle dis-
perspectives (Loeb 1987). Theoreti- Neurobiology of Vertebrate Locomotion. charge during isometric voluntary contrac-
Wenner-Gren International Symposium Se- tions. Amplitude relations between spindle
cians in both areas must address not ries vol. 45, section 111. MacMillan, London. frequency and torque. Acta Physiol. Scand.
only the problem of motor control in Harrison, P., and E. Jankowska. 1985. Sources 90: 319-336.
a given organism, but also how their of input to interneurones mediating group I Weeks, 0. 1989. Vertebrate skeletal muscle:
proposed solutions would fare in a non-reciprocal inhibition of motoneurones power source for locomotion. BioScience 39:
world governed by ontogeny and in the cat. J. Physiol. 361: 379-401. 791-799.
Hildebrand, Milton. 1989. The quadrupedal Zajac, F. E., and M. E. Gordon. 1989. Deter-
phylogeny. gaits of vertebrates. BioScience 39: 766-775. mining muscle's force and action in multiar-
Locomotion is a particularly ap- Houk, J. C. 1979. Regulation of stiffness by ticular movement. Exercise Sport Sci. Rev.
propriate motor behavior for such a skeletomotor reflexes. Annu. Rev. Physiol. 17: 187-230.

.BioScience Vol. 39 No. 11

View publication stats