Computations Underlying the Execution of Movement: A Biological Perspective

Author(s): Emilio Bizzi, Ferdinando A. Mussa-Ivaldi and Simon Giszter

Source: Science , Jul. 19, 1991, New Series, Vol. 253, No. 5017 (Jul. 19, 1991), pp. 287-291
Published by: American Association for the Advancement of Science

Stable URL: https://www.jstor.org/stable/2879070

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

American Association for the Advancement of Science is collaborating with JSTOR to digitize,
preserve and extend access to Science

This content downloaded from

14.139.196.16 on Sat, 18 Jun 2022 11:48:51 UTC
All use subject to https://about.jstor.org/terms
32. G. F. Lipscomb, A. F. Garito, R. S. Narang, Appl. Phys. Lett. 75, 1509 (1981). 61. J. Hecht, High Technol. (Boston) 3, 55 (October 1983).
33. J. Zyss, J. F. Nicoud, M. Coquillay, J. Chem. Phys. 81, 4160 (1984). 62. Y. S. Abu-Mostafa, D. Psaltis, Sci. Am. 258, 88 (March 1987).
34. J. Zyss, D. S. Chemla, J. F. Nicoud, ibid. 74, 4800 (1981). 63. J. L. Horner, Optical Signal Processing (Academic Press, New York, 1987).
35. S. Okada and H. Nakanishi, Kagaku Kogyo 37, 364 (1986). 64. D. S. Chemla, Rep. Prog. Phys. 43, 1191 (1980).
36. R. J. Twieg, "Organic materials for SHG" (Final Report, LLNL-2689405, 65. D. Neher, A. Wolf, C. Bubeck, G. Wegner, Chem. Phys. Lett. 163, 116 (1989).
Lawrence Livermore National Laboratory, Livermore, CA, March 1985). 66. R. D. Miller and J. Michl, Chem. Rev. 89, 1359 (1989).
37. D. F. Eaton, in Materialsfor Nonlinear Optics: Chemical Perspectives, G. D. Stucky, 67. G. C. Bhar, U. Chaterjee, S. Das, J. Appl. Phys. 66, 5111 (1989).
S. R. Marder, J. E. Sohn, Eds. (ACS Symposium Series 455, American Chemical 68. A. Penzkofer, P. Qiu, P. Ossig, in Nonlinear Optics of Organics and Semiconductors:
Society, Washington, DC, 1991), p. 128. Proceedings of the International Symposium in Tokyo, July 25-26, 1988, T. Kobayashi, Ed.
38. B. A. Fuchs, C. K. Syn, S. P. Velsko, Appl. Opt. 28, 4465 (1989). (Springer Proceedings in Physics, vol. 36, Springer-Verlag, Berlin, 1989), p. 312.
39. D. S. Donald, L.-T. Cheng, G. Desiraju, G. R. Meredith, F. R. Zumsteg, Abstract 69. J. Wynne, Phys. Rev. 178, 1295 (1969).
Q9.2, Fall meeting of the Materials Research Society, Boston, 27 November to 2 70. Y.-C. Chang, J. Appl. Phys. 58, 499 (1985).
December 1989. 71. L.-T. Cheng, N. Herron, Y. Wang, ibid., 66, 3417 (1989).
40. C. C. Frazier, M. A. Harvey, M. P. Cockerham, E. A. Chauchard, C. H. Lee, 72. Y. Ohashi et al., in Nonlinear Optics of Organics and Semiconductors: Proceedings of the
J. Phys. Chem. 90, 5703 (1986). International Symposium in Tokyo, July 25-26, 1988, T. Kobayashi, Ed. (Springer
41. D. F. Eaton, A. G. Anderson, W. Tam, Y. Wang, J. Am. Chem. Soc. 109, Proceedings in Physics, vol 36, Springer-Verlag, Berlin, 1989), p. 81.
1886 (1987). 73. H. Nasu, Y. Ibara, K. Kubodera, J. Non-Cryst. Solids 110, 229 (1989).
42. W. Tam and J. C. Calabrese, Chem. Phys. Lett. 133, 244 (1987). 74. R. K. Jain and R. C. Lind, J. Opt. Soc. Am. 73, 647 (1983).
43. _ _, ibid. 144, 79 (1988). 75. N. F. Borrelli, D. W. Hall, H. J. Holland, D. W. Smith, J. Appl. Phys. 61, 5399
44. M. L. H. Green et al., Nature 330, 360 (1987). (1987).
45. A. G. Anderson, J. C. Calabrese, W. Tam, I. D. Williams, Chem. Phys. Lett. 134, 76. S. Friberg and P. W. Smith, IEEEJ. Quantum Electron. QE-23, 2089 (1987).
392 (1987). 77. 0. R. Olbright, N. Peyghambarian, S. W. Koch, L. Banyai, Opt. Lett. 12, 413
46. W. Tam, D. F. Eaton, J. C. Calabrese, I. D. Williams, Y. Wang, A. G. Anderson, (1987).
Chem. Mater. 1, 128 (1989). 78. B. Van Wonterghem, S. M. Saltiel, T. E. Dutton, P. M. Rentzepis, J. Appl. Phys.
47. S. Tomaru, S. Zembutsu, M. Kawachi, Y. Kobayashi, J. Chem. Soc. Chem. 66, 4935 (1989).
Commun. 1984, 1207 (1984). 79. Y. Wang, N. Herron, W. Mahler, J. Opt. Soc. Am. B 6, 809 (1989).
48. Y. Wang and D. F. Eaton, Chem. Phys. Lett. 120, 441 (1985). 80. Y. Wang and W. Mahler, Opt. Commun. 61, 233 (1987).
49. I. Weisbuch, M. Lahav, L. Leiserowitz, G. R. Meredith, H. Vanherzeele, Chem. 81. S. Gray, Photonics, 125 (September 1989).
Mater. 1, 114 (1989). 82. A. J. Heeger, D. Moses, M. Sinclair, Synth. Metals 15 (1986).
50. S. D. Cox, T. E. Gier, J. D. Bierlein, G. D. Stucky, J. Am. Chem. Soc. 110, 2986 83. M.-T. Zhao, B. P. Singh, P. N. Prasad, J. Chem. Phys. 89, 5535 (1988).
(1989). 84. N. Fukaya, A. Heinamaki, H. Stubb, J. Mol. Electron. 5, 187 (1989).
51. G. R. Meredith, J. G. Van Dusen, D. J. Williams in (9), p. 109; Macromolecules 85. S. A. Jenekhe, W.-C. Chen, S. Lo, S. R. Flom, Appl. Phys. Lett. 57, 126 (1990).
15,
1385 (1982); K. D. Singer, S. E. Sohn, S. J. Lalama, Appl. Phys. Lett. 49, 248 86. A. Ashkin et al., ibid. 9, 72 (1966).
(1986). 87. J. Feinberg, Phys. Today 39, 46 (October 1986).
52. J. D. Le Grange, M. G. Kuzyck, K. D. Singer, Mol. Cryst. Liq. Cryst. 150b, 567 88. R. S. Cudney, R. M. Pierce, J. Feinberg, Nature 332, 424 (1988).
(1987). 89. P. Giinter and H.-J. Huignard, Photorefractive Materials and TheirApplication, vols.
53. M. Eich et al., J. Appl. Phys. 66, 2559 (1989). I and II (Topics in Applied Physics 61 and 62, Springer-Verlag, Berlin, 1988).
54. M. Eich, B. Reck, D. Y. Yoon, C. G. Wilson, G. C. Bjorklund, ibid., p. 3241. 90. G. Roosen, J.-P. Huignard, M. Cronin-Golomb, Eds., J. Opt. Soc. Am. B 7
55. J. D. Bierlein, personal communication. (December 1990).
56. J. T. Lin and C. Chen, Lasers Optronics (November 1987), p. 59. 91. A. Gailis, A. D. Durandin, V. Skudra, Latv. PSR Zinat. Akad. Vestis. Fiz. Teh.
57. C. J. van der Poel, J. D. Bierlein, J. B. Brown, S. Colak,Appl. Phys. Lett. 57,2074 Zinat. Ser. 119, 122 (1983); Chem. Abstr. 100, 14630a (1984).
(1990). 92. K. Sutter and P. Giinter, J. Opt. Soc. Am. B 7, 2274 (1990).
58. H. Vanherzeele, Opt. Lett. 14, 728 (1989). 93. S. Ducharme et al., Phys. Rev. Lett. 66, 1846 (1991).
59. J. D. Bierlein and C. B. Arweiler, Appl. Phys. Lett. 49, 917 (1986). 94. M. S. Paley et al., J. Org. Chem. 54, 3374 (1989).
60. E. Abraham, C. T. Seaton, S. D. Smith, Sci. Am. 248, 85 (February 1983). 95. D. Eimerl, Ferroelectrics 72, 95 (1987).

Computations Underlying the Execution of

Movement: A Biological Perspective
EMILIO Bizzi, FERDINANDO A. MUSSA-IVALDI, SIMON GISZTER

To execute voluntary movements, the central nervous than that specified in the plan of action. Some of the
system must transform the neural representation of the mechanisms and circuitry underlying the transforma-
direction, amplitude, and velocity of the limb, represent- tion of motor plans into motor commands are described.
ed by the activity of cortical and subcortical neurons, A central feature of this transformation is a coarse map
into signals that activate the muscles that move the limb. of limb postures in the premotor areas of the spinal
This task is equivalent to solving an "ifl-posed" compu- cord. Vectorial combination of motor outputs among
tational problem because the number of degrees of different areas of the spinal map may produce a large
freedom of the musculoskeletal apparatus is much larger repertoire of motor behaviors.

T -* HE CENTRAL NERVOUS SYSTEM (CNS) CONTROLS THE motor goals are represented by the activity of populations of
events from the planning to the execution of movements. To neurons in different cortical and subcortical areas (1). Recordings of
specify a plan of action, the CNS must first transform sensory the electrical activity of single neurons from the parietal and frontal
information into motor goals such as the direction, amplitude, and cortices of monkeys show a correlation between neural activity and
velocity of the intended movement. In higher vertebrates, these the direction of the movement of the arm (1). Furthermore, on the
basis of these recordings, a number of investigators have argued that

The authors are in the Department of Brain and Cognitive Sciences, Massachusetts
the activity of cortical cells is represented in spatial coordinates
Institute of Technology, Cambridge, MA 02139. without any specification about how muscles are to be engaged to

19 JULY 1991 ARTICLES 287

This content downloaded from

14.139.196.16 on Sat, 18 Jun 2022 11:48:51 UTC
All use subject to https://about.jstor.org/terms
produce the forces necessary for the movement (1). If actions are planned number of possible solutions exist for the same image. In the study of
in spatial or extrinsic coordinates, then for the execution of movement motor control, ill-posedness arises from the imbalance between the
the CNS must convert the desired direction, amplitude, and velocity of degrees of freedom of a limb and the number of variables that are
the limb into signals that control the muscles. A large number of muscles sufficient to specify a task. In our blackboard example, any chalk location
are simultaneously active during the execution of even the simplest kind can be reached by a variety of joint angles. By the same argment,
of limb trajectory. because there are several muscles acting upon any articulation, the same
To convert a planned movement from extrinsic to musde (intrinsic) joint torque can be generated by a variety of different patterns of muscle
coordinates, the CNS must perform a number of transformations. forces.
Consider the simple task of writing one's name on a blackboard. The The excess or "redundancy" of degrees of freedom is not only a
planned motion in this case is completely specified by a set of coordinates source of complex computational problems but also the basis of the
that define the position and the pressure of the chalk on the blackboard. versatility of biological systems-that is, their ability to perform
different tasks in a wide range of environmental conditions. For this
These coordinates are sufficient to describe the final result of the action.
However, they are not sufficient to specify how this final result can bereason, the design and control of redundant manipulators is attract-
achieved by a coordinated motion of the arm. To this end, the CNS musting increasing interest in robotics research (4). In this article we
convert the trajectory of the hand into the corresponding motion of each present a theoretical and experimental framework that describes the
joint of the arm. Only after the desired motion of each joint is defined is way in which the CNS takes advantage of a limb's biomechanical
properties to transform planned movements into muscle activations.
it possible to establish the torque that must be generated at each joint in
order to provide the motion. The final step of this computational process To illustrate how the CNS achieves this transformation, we consider
consists of deriving the control signals to be delivered to each musde sothree points: (i) the key role played by the mechanical properties of
that the appropriate joint torques can be generated. the muscles; (ii) the way in which the circuits of the spinal cord
Researchers of motor control have become increasingly aware of theseexploit muscle properties to express motor outputs; and (iii) the
computational problems in the last decade, because the same problems spatial organization of the circuits of the spinal cord.
have become a focus of investigation in the field of robotics (2). The
simplest artificial manipulators are open chains of rigid bodies intercon-
nected by rotational or translational (or "prismatic") joints. For this class
Mechanical Properties of Muscles
of mechanisms, the task offinding a motion of the joints from the desired
motion of the end point is known as an "inverse kinematics" problem. The mechanical properties of muscles are important because the
Similarly, the task of deriving the forces to be delivered by the actuators features of the CNS have probably evolved as a result of its need to
in order to achieve a desired motion has been called an "inverse both control and take advantage of the viscous and elastic properties
dynamics" problem. (5) of the musculoskeletal apparatus. The tension developed by a
Both of the inverse problems mentioned above may be "ill-posed"; muscle depends on its length in a way that is reminiscent of a spring;
that is, an exact solution may be either not available or not unique. if we stretch a stimulated muscle, there exists a restoring force that
Ill-posed problems arise frequently in different domains of biologicaldepends on the amount of stretch (5). In addition, many muscles are
information processing (3). A well-known example in the study of visionarranged about the joints in opposition to each other. Hence, the
is the task of recovering shape, location, and orientation of a surface from position of a limb is maintained when the torques exerted by
its image over the retina. This problem is ill-posed because an infinite opposing muscle groups are equal and opposite.
The elastic behavior of the muscles implies that when the limb is
perturbed by an external force, the limb will be displaced by an
amount that varies with both the external force and the stiffness of
A . .2cm B .
the muscles. When the external force is removed, the limb returns to
its original position. When small displacements are applied in several
directions to the hand, elastic restoring forces are observed (6).
These forces are organized as a field, because a single force vector is
consistently associated with each position of the limb. The point in
the field at which the force vector is zero is an equilibrium point.
The observation that posture derives from the.interaction be-
tween the springlike properties of opposing muscles led to the
formulation of the equilibrium point hypothesis. According to this
CX D hypothesis, first proposed by Feldman (7), limb moVements result
from a shift in the neurally specified equilibrium point. Studies on
the movements of a single joint have provided experimental evi-
dence that supports the equilibrium point hypothesis (8). This
evidence shows that arm trajectories of moderate speed are gener-
ated by neural signals that specify a series of equilibrium positions
for the limb.
The equilibrium point hypothesis has implications both for the
control and for the computation of movements. With respect to
control, the elastic properties of the muscles provide instantaneous
Fig. 1. Procedure used to estimate a force field. (A) Locations in the correcting forces when a limb is moved away from the intended
workspace at which we recorded forces at the ankle after microstimulation trajectory
of by some external perturbation. With respect to computa-
a site in the premotor spinal cord. (B) Force vectors recorded at nine tion, the same elastic properties offer the brain an opportunity to
locations in the frog's workspace. (C) Partitioning of the tested workspace
deal with the inverse dynamics problem. Once the brain has
into a set of nonoverlapping triangles. Each vertex of a triangle is a tested
achieved the ability to represent and control equilibrium postures, it
point. Force vectors that were measured at each point at the same latency are
displayed as solid arrows. (D) Interpolated field. can master movements as temporal sequences of such postures. In

288 SCIENCE, VOL. 253

This content downloaded from

14.139.196.16 on Sat, 18 Jun 2022 11:48:51 UTC
All use subject to https://about.jstor.org/terms
this context, a representation in the CNS of the inertial, viscous, and The above expressions have six unknown paramet
gravitational parameters contained in the equations of motion is no the interpolation problem with three data vectors
longer necessary. The competence of the equilibrium point model in components) has a unique solution.
dealing with multijoint, goal-directed movements has been established In most instances, the spatial distribution of the forces induced by
by the simulation studies of Flash (9), who reproduced the kinematic the stimulation was structured in a well-defined pattern that we refer
features of human multijoint movements with such a model. to as a convergent force field (CFF) (Fig. 2). The field was
A set of experiments performed in frogs with spinal cords that characterized by a single equilibrium point to which the force
were surgically disconnected from the brainstem has provided vectors converged. This point is indicated by a filled circle in Fig. 2,
neurophysiological support for the equilibrium point hypothesis B and D, and represents the locus at which the leg would be at
(10). By microstimulation of the spinal cord, we have shown that steady state if it were free to move. The CFFs were qualitatively
this region is organized to produce the neural synergies necessary for similar for a variety of stimulus parameters. For instance, when we
the expression of equilibrium points. These experiments have indi- changed the duration of the train of stimuli from the standard 250
cated that the spinal cord contains circuitry that, when activated, ms to 500 ms, we found that a similar CFF was maintained for a
produces precisely balanced contractions in groups of muscles. longer period. Similarly, the expression of a CFF was not substan-
These synergistic contractions generate forces that direct the limb tially dependent on a specific frequency of stimulation. The area of
toward an equilibrium point in space. the spinal cord from which such single, well-defined equilibria were
observed was centered on the lateral neuropil region (Fig. 2).
The CFFs must derive at least in part from the activation of the
interneurons. The branches of these interneurons make synaptic
Generation of Force Fields connections with different pools of motoneurons and activate
When the spinal cord of the frog is surgically disconnected from groups
the of muscles. Because the motoneurons of the frog have
brainstem, the cord still retains significant motor skills (11). For example,
a frog that has been treated in such a manner is able to remove a noxious
stimulus from the skin by the coordinated motion of multiple limb
segments. We used these surgically altered frogs to explore the organi-
zation of the spinal circuitry through microstimulation of a number of
spinal cord structures. We sought primarily to determine the field of
A *,%& 2cm B
static forces associated with the stimulation of a spinal cord site. To this A~ ~ ~ B
end, we measured the isometric force produced by the musdes of the leg
at different ankle locations when subjected to the same spinal stimula-
tion. Typically, we recorded the force vectors at the ankle in a set from 9
or 16 locations forming a grid (three by three or four by four) in the
workspace of the limb. Here the term "workspace" indicates the range of
movement of the ankle in the horizontal (x-y) plane. At each grid
location, we recorded the force vector elicited by the stimulation of the
same spinal cord locus. The force vector varied as we placed the leg at
different workspace locations, denoted as r = (x, y). The change in force
through the workspace resulted from the interaction of a number of
factors, such as the lengtis, moment arms, and elastic properties of the
musdes and stretch reflexes.
C D
The -data recorded in our microstimulation experiment can be de-
scribed as a collecton of samples from a time-varying force field, F(r, t),
obtained at a number of tested sites. We set the instant of stimulation as
t = 0. Hence, F(r, t) denotes the force vector at the location r after a
latency t from the onset of the stimulus. The field at t = 0, FO(r), is the
"resting" field that characterized the mechanical behavior of the frog
before the stimulation. The field at steady state, which we indicate as
F(r), was measured when the forces induced by the stimulation had
reached maximum amplitude at all the test points. This "peak" field often
persisted for some time (100 to 500 ms, depending on the stimulus
duration) before the measured forces returned to their resting values. At Fig.
2. E
stimul
given latency, we used the measured force vectors to estimate the force Transver
dorsal ne
field across a broad convex region of the ankle's workspace. To this end
region; V
(Fig. 1), we implemented a piecewise-linear interpolation procedure region. (A
according to the Delaunay triangulation algorithm (12). The algorithm leg. (B) T
partitions the workspace into triangles as close to equilateral as possible. was prepa
figce.
roots 2.e
71,
The vertices of each triangle were the tested grid points. Within each
Stimulati
triangle, we applied a linear interpolation to the force vectors measured region;rV
microstim
at the corners. Thus, within each interpolation triangle the force com- with con
ponents were given as: spread of
Stmltion
each locat

FX= a1,1x + a1,2y + a1,3 (1)

Fy= a2,1x + a2,2y + a2,3 (2)

19 JULY 1991 ARTICLES 289

This content downloaded from

14.139.196.16 on Sat, 18 Jun 2022 11:48:51 UTC
All use subject to https://about.jstor.org/terms
 A B '
A tl \ B

A t t
- \ \ \\ \ s

B N ,~~~~ , , , \ \ \ \ \

,, t , I, \
\\\YSf
N\\ ~, "\\
~ ~_
_ . _ \ \ \ . /

t I~~~~~~

C D

tG~~~~~~~ \ D\/
t i \ \ SE\

Fig. 3. (Left) Regions of the lumbar spinal cord containing the neural
circuitry that specifies the force fields (A through D). Within each region,
similar sets of CFFs are produced. The diagram is based on 40 CFFs elicited
by microstimulation of premotor regions in three frogs with transected spinal
cords. (Right) Four types of CFFs. To facilitate comparison among CFFs Fig. 4. Combinations of multiple s
recorded in different animals, we subtracted the passive force field from the fields resuliting from sitimulaition at
force field obtained at steady state. The passive field is the mechanical of the lumbar spinal cord. The equili
behavior generated by the frog's leg (recorded at the ankle) in the absence ofis in flexion. (C) The computed field
any stimulation. The force field is at steady state when the forces induced bysummation of fields (A) and (B).- (D
the stimulation of the spinal cord have reached their maximal amplitude. of (A) and (B) together. The equilibr

A Coarse Map of CFFs

extensive dendritic arbors and some degree of electrical connectivity,
a CFF may result either from direct activation of the motoneurons.
or from random spread of excitation to a number of motoneural We investigated the spatial organization of the areas that specify
pools. Our experimental evidence is not, however, compatible with different equilibrium-limb positions. After mapping most of the
direct activation of the soma of the motoneurons. Our results premotor area of the lumbar cord, we reached the conclusion that
indicate that direct stimulation in the motoneuron (MN) area there are at least four areas from which four distinct types of CFFs
are elicited. These regions form stripes that are oriented rostrocau-
generates fields that are different from the convergent fields elicited
by the stimulation of the intermediate gray. When we placeddaily. our They extend dorsoventrally over a distance at least 300 p~m in
microelectrodes directly among the soma of the motoneurons, depth.we Within each region, a qualitatively similar set of x and y
forces are produced. This map of postures is shown in Fig. 3.
obtained force fields with divergent or parallel patterns (Fig. 2C).
In a preliminary series of experiments, we also tested the hypoth-Preliminary experiments have shown that the simultaneous stim-
esis that CFFs may result from the random activation of several ulation
turnMN of two different
genelubrsiate points
forefeds. Ifhe in some
asslbrume of the
thtths areassum
field(Ais shown in the(B
lxenike Fig. 3
pools. We investigated this question by combining the isometric results
CFs in a force bye
gnfexinerated field proportional to thepeithen
cmicrotefimuldAB>ion sum ofbthe fieldsofemtoria
a vaietye obtained
from the stimulation
responses of individual muscles in a simulation. First, we stimulated
pumatierns may riesltds(A delivered at each point
acnsequ.(DTenacua ofitel (Fig. 4). This result iso
terminaldibystribution
individual leg muscles through a pair of implanted electrodes.unexpected
We because of the complex nonlinearities that characterize
obtained muscle force fields by recording x and y forces at thethe interactions both among neurons and between neurons and
ankle
fibe ars. a fC F
that were produced at each location of the workspace by the muscles. The superposition apples to the vectors of the fields. In
electrical activation of individual muscles (10). Second, we simulatedcontrast, the equilibrium points do not add vectorially. If one of the
random combinations of the measured muscle fields. We modulatedtwo Mireostimulation of the pubremoord
fields has a stiffness aereashof
larger than that of thethe spinaluscord
other, thas
the summed
each measured field by multiplying it with a randomly selected field has an equilbrium that les closer to the equilibrium of the
coefficient representing a certain amount of muscle activation. We stiffer field. We view the superposition mechanism as the simplest
obtained sets of combined fields by adding together all these way to explain how the spinal cord may generate a vast number of
modulated fields. We found that, in a set of 20,000 simulated force fields from the limited variety of available fields (13).
combinations, the number of fields having an equilibrium point The observation that force fields sum vectorially suggests a way to
within the tested workspace was only 8.4%. Thus, random recruit- relate natural movements to the microstimulation results. Physio-
ment of motoneurons could not account for the CFFs observed in logical movements result from patterns of neural activity distributed
the majority of our experiments. by branching fibers throughout fairly wide regions of the spinal
Another explanation of our results is that CFFs result from the cord. These branches may stimulate local clusters of cells that, in
activation of afferent fibers recruiting portions of reflex pathways.
We investigated this issue in frogs that had been chronically
deafferented bilaterally and in which all sensory fibers had degener-
ated. Microstimulation of the spinal cord in these animals still
revealed the presence of CFFs (Fig. 2B).

290 SCIENCE, VOL. 253

This content downloaded from

14.139.196.16 on Sat, 18 Jun 2022 11:48:51 UTC
All use subject to https://about.jstor.org/terms
revealed a few zones that, when activated, produce distinct force of a very coarse map is a remarkable mechanism for producing a vast
fields. Through their mechanical and geometric properties, the repertoire of motor behavior in a simple fashion.
activated muscles specify force fields with equilibria. These zones are
ideal sites for the integration of signals coming from different REFERENCES AND NOTES
sources. Diverse neural signals conveyed by afferent inputs or
1. R. Caminiti, J. Neurosci. 10, 2039 (1990); A. P. Georgopoulos,
descending tracts may gain access to the premotor circuits, which, inB. Schwartz, ibid. 8, 2913 (1988); ibid., p. 2928; J. F. Kalaska, D. A. D. Cohen,
turn, specify the activation of sets of muscles. The premotor zones M. L. Hyde, M. Prud'homme, ibid. 9, 2080 (1989).
with their afferent inputs may represent one of the substrates 2. M. Brady, J. M. Hollerbach, T. L. Johnson, T. Lozano-Perez, M. T. Mason, Robot
Motion: Planning and Control (MIT Press, Cambridge, MA, 1982).
underlying transformation from extrinsic to intrinsic coordinates. 3. A. N. Tichonov and V. Y. Arsenin, Solutions of Ill-Posed Problems (Winston,
The idea that premotor spinal maps involving a few discrete types Washington, DC, 1977); T. Poggio, V. Torre, C. Koch, Nature 317, 314 (1984);
of movements are a key structure in sensory-motor transformations M. Bertero, T. Poggio, V. Torre, Proc. IEEE 76, 869 (1988); F. A. Mussa-Ivaldi,
P. Morasso, R. Zaccaria, Biol. Cybern. 60, 1 (1988); M. I. Jordan and D. A.
is consistent with investigations of the control of head movement in Rosenbaum, in Foundations of Cognitive Science, M. I. Posner, Ed. (MIT Press,
the owl. Masino and Knudsen (14) have postulated the existence of Cambridge, MA, 1989), p. 727; M. Kawato, Y. Maeda, Y. Uno, R. Suzuki, ATR
Technical Report TR-A-0056 (Advanced Telecommunications Research, Auditory
a few separate circuits for controlling horizontal and vertical head
and Visual Perception Research Laboratories, Kyoto, Japan, 1989).
movements. These structures, which are located in the brainstem, 4. D. E. Whitney, IEEE Trans. Man Mach. Syst. MMS-10, 47 (1969); C. A. Klein
receive inputs from the tectum and transform the tectal movement and C. Huang, IEEE Trans. Syst. Man Cybern. SMC-13, 245 (1983); T.
Yoshikawa, IEEE 1985 International Conference on Robotics and Automation (IEE
vectors into neck motoneural activation.
Computer Society Press, Washington, DC, 1985), p. 1004; J. Baillieul, ibid., p.
Our results described here imply that the spinal cord of the frog 722; J. M. Hollerbach and K. C. Suh, ibid., p. 1016; C. W. Wampler, IEEE 1987
implements position control and equilibrium point mechanisms. In International Conference on Robotics and Automation (IEEE Computer Society Press,
Washington, DC, 1985), p. 610; T. Shamir and Y. Yomdin, IEEE Trans. Aut.
contrast, on the basis of electrophysiological recordings, planning at Control AC-33, 1004 (1988); F. A. Mussa-Ivaldi and N. Hogan, Int. J. Robot.
cortical levels is generally assumed to involve other types of variables Res., in press.
5. A. G. Feldman, Biofizika 19, 534 (1974); ibid., p. 749; A. M. Gordon, A. F.
and coordinates. Nonetheless, evidence has accumulated that posi-
Huxley, F. J. Julian, J. Physiol. (London) 184, 170 (1966); P. M. H. Rack and D.
tion control occurs at the motor output stages of other CNS R. Westbury, ibid. 204, 443 (1969).
structures. This body of evidence is based on the use of stimulation 6. F. A. Mussa-Ivaldi, N. Hogan, E. Bizzi, J. Neurosci. 5, 2732 (1985).
techniques. 7. A. G. Feldman, Central and Reflex Mechanisms of Motor Control (Nauka, Moscow,
1979); Biofizika 11, 667 (1966).
In oculomotor research, there is evidence that activation of certain 8. E. Bizzi, N. Accornero, W. Chapple, N. Hogan, Exp. Brain Res. 46, 139 (1982);
circuits specifies the position of the eye in the orbit. Activation of a J. Neurosci. 4, 2738 (1984); E. Bizzi, A. Polit, P. Morasso, J. Neurophysiol. 39,
435 (1976); E. Bizzi, F. A. Mussa-Ivaldi, N. Hogan, Prog. Brain Res. 64, 345
number of CNS structures brings the eyes to a unique orbital (1986).
position irrespective of the starting position of the eyes. For 9. T. Flash, Biol. Cybern. 57, 257 (1987).
instance, the stimulation of the posterior portion of the superior10. F. A. Mussa-Ivaldi, S. F. Giszter, E. Bizzi, Symposia on Quantitative Biology 55, The
Brain (Cold Spring Harbor Laboratory, Cold Spring Harbor, NY, 1988), p. 827;
colliculus of the cat produces saccadic eye movements that bring the
S. F. Giszter, F. A. Mussa-Ivaldi, E. Bizzi, in Visual Structures and Integrated
eyes to a particular position in the orbit (15). Also, stimulation of Functions, M. Arbib and J. P. Ewert, Eds. (Plenum, New York, in press); S. F.
the most caudal portion of the colliculus in cats evoked head Giszter, E. Bizzi, F. A. Mussa-Ivalci, in Analysis and Modelling of Neural Systems,
F. H. Eeckman and C. D. Deno, Eds. (Kluwer, Moffett Field, CA, in press); S. F.
movements that bring the head to a particular location (15). In both Giszter, F. A. Mussa-Ivaldi, E. Bizzi, Exp. Brain Res., in press.
cats and monkeys, goal-directed saccades can be evoked by stimu-11. 0. I. Fukson, M. B. Berkinblit, A. G. Feldman, Science 209, 1261 (1980); S. F.
lation of the internal medullary lamina of the thalamus (16), the Giszter, J. McIntyre, E. Bizzi, J. Neurophysiol. 62, 750 (1989); D. J. Ostry, A. G.
Feldman, J. R. Flanagan, S. V. Adamovich, L. E. Sergio, Soc. Neurosci. Abstr. 15,
posterior parietal cortex (17), and the supplementary eye fields (16). 173 (1989); J. L. Schotland, W. A. Lee, W. Z. Rymer, Exp. Brain Res. 78, 649
Although there is a superficial resemblance between the CFFs and (1989); D. J. Ostry, A. G. Feldman, J. R. Flanagan, J. Neurophysiol. 65, 547
equilibria and the final orbital position of the eye, we do not know(1991).
12. F. P. Preparata and M. I. Shamos, Computational Geometry (Addison-Wesley, New
whether these oculomotor areas and mechanisms generating final York, 1988), p. 185.
eye position are in any way equivalent to the zones in the spinal 13. F. A. Mussa-Ivaldi and S. F. Giszter, in Proceedings: Fjfth International Conference on
cord. Advanced Robotics, Pisa, Italy, 20 to 22 June 1991 (IEEE, New York, in press).
14. T. Masino and E. I. Knudsen, Nature 345, 434 (1990)
Within the spinal cord are a number of potentially autonomous 15. A. Roucoux, D. Guitton, M. Crommelinck, Exp. Brain Res. 39, 75 (1980); D.
functional units underlying motor behavior. Examples of these units Guitton, M. Crommelinck, A. Roucoux, ibid., p. 63; J. T. McIlwain, J. Neuro-
include the spinal rhythm generators for locomotion (18) and physiol. 55, 97 (1986); D. Guitton, M. Crommelinck, A. Roucoux, Exp. Brain
Res. 39, 63 (1980).
scratching (19) and the propriospinal system for reaching (20). We16. J. Schlag and M. Schlag-Rey, J. Neurophysiol. 57, 179 (1987); M. Schlag-Rey a
do not yet know how the CFF units we have described relate to the J. Schlag, ibid. 40, 156 (1977).
spinal pattern generators or to the propriospinal system. 17. H. Shibutani, H. Sakata, J. Hyvarinen, Exp. Brain Res. 55, 1 (1984).
18. S. Grillner, Science 228, 143 (1985); _ and P. Wallen,Annu. Rev. Neurosci.
In conclusion, we have identified a discrete map of motor 8, 233 (1985).
behaviors: the convergent force fields. We believe that this map19.may P. S. G. Stein, Symp. Soc. Exp. Biol. 37, 383 (1983).
represent a mechanism whereby the CNS performs the transforma- 20. B. Alstermark, A. Lundberg, U. Norrsell, E. Sybirska, Exp. Brain Res. 42, 299
(1981).
tion from movement planning to execution. The vectorial combina-
21. Supported by the Office of Naval Research (N00014/88/K/0372) and NIH
tion of motor outputs derived from activation across different areas (NS09343 and AR26710).

19 JULY 1991 ARTICLES 291

This content downloaded from

14.139.196.16 on Sat, 18 Jun 2022 11:48:51 UTC
All use subject to https://about.jstor.org/terms