BODYBUILDING: THE BIONIC HUMAN

VIEWPOINT

The Bionic Man: Restoring Mobility

William Craelius*

Bionicsengineersare makingincreasinglybold and successful use of their commands from their brains (or spinal
tools to restore mobility to persons with missing or nonfunctionallimbs. cords) are harnessed either noninvasively
These tools include the latest materials, minielectronicsand megacom- with the electroencephalogram (EEG) or
puters, advanced robotic mechanisms,and algorithms.With crucialhelp with implanted electrodes. Paralyzed pa-
from their pioneeringusers,they are learninghow and where the residual tients fitted with a brain-implanted chip
sensorimotor system can be tapped in order to transmit its intents to have learned to move cursors and select
replacementor reactivatedbody parts. letters on a computer (8), as well as to
direct robotic arm movements using tech-
When humans replace a missing body part to alternateregions (4). This proceduretrans- nology known as brain-computer interface
with an artificial one, they begin an inti- fers hand-controlsignals to healthy pectoral (BCI) (9). These systems operate by train-
mate relationship with a partnerthey barely muscles, where they can be conveniently ac- ing an algorithmic filter to associate specif-
know. The key to success for such a rela- cessed and deciphered. ic movement requests with specific neuro-
tionship may be no different than that The second requirementis more challeng- nal signal patterns, recorded either directly
found in marriage manuals: communica- ing owing to the complexity of humanmove- from neurons or indirectly from EEGs or
tion. Unfortunately, for bionic parts, com- ment control. Each action originates with a other noninvasive signals. The filter then
munication is the weakest link in the chain few neuronsin the motor cortex thattriggera directs the appropriate action after recog-
of components that includes electronics, large neural network that coordinatesthe ac- nizing, or decoding, the volition .
computing, actuators, mechanisms, and tivities of several effector muscles after re- How much more functioncan be restored
materials, all of which are adequate for the ceiving and processing feedback information by advancedbionic systems? Complex activi-
application. This situation is perhaps best from thousandsof tactile, positional, and vi- ties, such as walking, could probablynot be
illustrated by the problem of hand restora- sual senses. Transformingthis tangled mesh achievedby noninvasivemethodssuchas EEG,
tion. NASA, for example, has developed a of millions of electrical pulses into graceful owing to theirpoorresolutionof brainactivity.
robotic hand that approximates human movements is a routine accomplishment of The difficultyin extractingvolitionalrequests
dexterity, moving up to 22 joints indepen- our sensorimotor system that bionics engi- from the EEG is underscored by the limi-
dently (1). Available prosthetic technology, neers can only envy. Its artistrycan be appre- tations of present technology, which cannot
however, can control only one joint at a ciated by comparingthe elegance and adapt- decipher from the brain more than 25 bits
time, leaving amputees with little hope of ability of human motions with that of the of information (three characters) per
using the advanced hand. This glaring mis- most advanced walking robots, whose latest minute. This rate is many thousands of
match between machine and human capa- achievement is the slow climbing of stairs times too slow to control even the simplest
bilities clearly reflects the inadequate lines (5). Restorationof lost sensorimotorfunction movements. Finer resolution and hence
of communication. On the bright side, bion- by robotic assistance is correspondinglychal-
ics researchers are opening new vistas for lenging, but is steadily progressing,as exem-
restoring lost human functions as they plified by robotic devices such as the
steadily bridge the gap between human and RoboWalker, an active exoskeleton that as-
machine. sists walking by motor-impairedindividuals
Bionics can restore lost mobility to in- (6).
dividuals if (i) they can express cognitive The human sensorimotor system, though
control over relevant motor functions still far too complex to duplicate, is being
somewhere in their residual anatomy and invaded by increasingly more versatile
(ii) a device can pick up and decipher that bionic interfaces. Our bionic potential was
cognition. The first requirement is ade- recently demonstrated by a monkey in
quately satisfied by many individuals who Brooklyn, whose brain signals, monitored
have lost function either through paralysis by electrodes, controlled a three-dimen-
or amputation, and who can both sense and sional (3D) robotic arm located in North
imagine manipulating the nonfunctional or Carolina, while he watched it on the Inter-
absent joints. These individuals express net (7). This feat and similar ones were
motor control over their lost parts, includ- taught to the bionic animals (rats as well as
ing legs, joints, and individual fingers, primates) through the presentation of food
through nervous and/or muscular activity rewards, and did not require actual move-
directed to their residual limbs, and these ment of the animals' own limbs, because
expressions can be registeredby appropriate they quickly learned that movement was
technology (2, 3). Patients who do not meet unnecessary to be rewarded.Cognitive con-
this requirementbecause of damagedresidual trol of artificial limbs, at least for primitive Fig. 1. Bionicrestorationof handmobility.This
muscles may soon have a surgical option, but important motions such as grasping, user,havingnervedamageas a resultof spinal
wherebytheirhandmotornerves are rerouted thus can be achieved with a bionic brain- cordinjury,can graspobjectswhen his forearm
are stimulatedwith Bions.The control
machine interface (BMI) for individuals muscles in his shirt
with paralyzed or missing limbs. computer pocket reads his inten-
Departmentof BiomedicalEngineering,RutgersUni- tions to grasp and activates the wire coils
versity, Piscataway,NJ 08855-0909, USA.*E-mail: Humans with motor disabilities can al- aroundhis arm, to deliverelectricalpulses to
craelius@rci.rutgers.edu ready surpass animal performance when the Bions(brownrods).

1018 8 FEBRUARY
2002 VOL295 SCIENCEwww.sciencemag.org
 BODYBUILDING:
more functionality can be achieved by di-
rect recording with cortically implanted
electrodes if a sufficient number of them
can be permanently located in the brain.
The minimum number of individual neu-
rons required to transform thoughts into a
reasonable range of motions probably ex-
ceeds 1000 (10). This is achievable with
microwires in the brain; however, the re-
cording electronics for 1000 channels is
currently too bulky to fit in the cranium.
For now, the analog electronics of a versa-
tile BCI or BMI would need to be located
externally and receive brain signals by
wireless transmission. Near-real time oper-
ation of such a device is possible, because
earlier experiments showed that volitions
could be deciphered from monkeys by sim-
ply computing the weighted sum of firing
rates of several cortical neurons (7).
Whereas brain interfaces tap into the
head of the sensorimotor system, putting
Sel
M and unpredictablemanner, which is not yet
: Pneumaticsensors
computer interface
Prosthetichand
Fig. 2. Biomimetic Dextra hand prosthesis. The silicone "smart sleeve" fits snugly over the residual
limb and registers 3D forces produced by muscle activity within the hard socket. The pocket
computer allows the user to retrain the robotic hand for optimal performance. The hand can flex
and extend all five digits in response to commands from the natural motor pathways of the user.
[Figure
[Figure provided by D.
provided by D. Curcie]
Curcie]
THE BIONIC HUMAN
the brain in control of movement, alterna-
tive bionic approaches bypass the brain,
directly communicating with peripheral
nerves and muscles. These peripheral-ma-
chine interfaces (PMIs) operate by func-
tional electrical stimulation (FES) of mus-
cles and peripheral nerves and can be pro-
grammed for specific movement patterns.
A successful example is FreeHand, one of
the first commercially available and Food
and Drug Administration-approved FES
devices, pioneered by Peckham (11, 12).
FreeHand restores grasping to patients
with upper-limb paralysis or weakness by
giving them control over extrinsic hand
muscles through movements of their oppo-
site shoulder that generates radio waves to
activate electrodes in the forearm.
Bionic technologiescan be adaptedfor re-
storingsome degreeof almostany lost function.
Theirbroaderapplicationsare highlightedby a
paraplegicwho now ambulatesby operating
Malemanifold
(insidesleeve)
Pressuretransducers
and manifoldsiti
coning
- advantage of the RKI method is that it is
switcheson his walkerthatcontrola stimulator
chip implantedin his spinalcord (13, 14). The
strictlyperipheralapproach,however,requires
FES to micromanageall actions,muchlike that
of a puppet master. For spinal cord-injured
persons, this detail is a welcome remedy for
immobility;more naturaland gracefulcontrol
may eventually be possible through devel-
opments of hybrid brain-machineinterfaces
(HBMIs), whereby volitional EEG signals
recorded directly from the brain can control
the muscles (9, 10).
Bionic coordinationof muscle movements
has been advanced by a new device called
Bion (Fig. 1) (15). Bions are single-channel
stimulatorsabout the size of a long grain of
rice that can be injected into muscles with a
12-gauge needle and controlled by an exter-
nal radio-frequencycoil. Their excellent lon-
gevity and functionalityin vivo was demon-
stratedin clinical trials with users for over 1
year. Bions can independentlycontroleach of
the many muscles involved in coordinated
movements, given the appropriate motor
commands. Direct control over muscles is
desirablebecause humanmuscles, in contrast
to robotic actuators,respond to their natural
controller,i.e. neurons, in a highly nonlinear
understood.
When leaving the brainout of the control
loop, bionics engineers must somehow de-
code volition at the periphery.The most com-
mon approach is to train users to execute
specific muscle activities to produce surface
electromyographic(EMG) patternsrecogniz-
able by the decoder. This approach can re-
store a limited number of activities, such as
ID grasping;however, it does not adequately
resolve volition having more than one degree
of freedom (16). An alternativeto controlby
EMG, developed by Sam Phillips and others
in our laboratory,registers volitions by the
entire 3D pattern of forces in the residuum
(residual limb), known as residual kinetic
imaging (RKI) (17, 18). Such patternscan be
discriminated by using filters that can be
readily trainedand retrainedas needed. This
adaptibilitylessens the requirementfor pre-
cise placementof sensors, an importantprac-
tical considerationfor amputeeswhose resid-
uum is constantly changing and who must
don and doff their prosthesis daily. A key
biomimetic:The originalmotorpathwayscan
be used to control robotic replacementparts,
such as fingers (Fig. 2).
Each approach to bionic restoration of
movement is specialized for particularuser
characteristics (Table 1). For example, BCI
is invaluable for severely paralyzed pa-
tients, for whom simple communication
with the outside world is a primarygoal (8).
CONTINUEDON PAGE1021

www.sciencemag.org SCIENCE VOL295 8 FEBRUARY

2002 1019
 BODYBUILDING: THE BIONIC HUMAN

Table 1. Strategiesfor bionicrestorationof movement

Bionicappro (with electrod C Actuatortype (with example
either Implanted or xternal) references)
BMI(brain-machine
interface) Completelyparalyzedpersonswho can benefitfrommechanical Motors,i.e.,roboticarmscontrolled
assistdevices by monkeys(7)
BCI(brain-computer
interface) Completelyparalyzedpersonswho wishto simplycommunicate Computerscreen,i.e.,movinga cursor
(8)
PMI(peripheral-
machineinterface) Amputeesandpersonswith intactcentralnervoussystembut Motors,i.e.,prosthetichand(23)
weakmuscles
HBMI(hybridbrain-machine
interface) personswith intactlimbmuscles
Spinalcord-injured Muscles,i.e.,directbraincontrolof
FreeHand (9)
CBi (computer-braininterface) Parkinson'sdisease Muscles (19, 20)

CONTINUEDFROMPAGE1019
The converse of BCI, computer-braininter-
face (CBI), can treat conditions such as Par-
kinson's disease, wherein thalamic neuronal
activity is substantially impaired, causing
tremors.The first commerciallyavailable de-
vice for this purpose,Activa TremorControl
(19, 20), operates from a small computer,
located in the chest cavity, that rhythmically
stimulatesthe thalamusto simulate the oper-
ation of the diseased neurons.
Personswith an intactcentralnervous sys-
tem who have completely lost function of
specific muscles can be aided by robotic de-
vices controlledeither directly from the brain
with HBMI systems or from muscles or pe-
ripheral nerves with PMI systems. Strictly
peripheraldevices, such as the RoboWalker
mentioned earlier, can be controlled from
muscle activation or movement patterns.
Such interfaces,possibly in combinationwith
surgicalreinnervation(4), could benefit para-
plegics, amputees,and those with weak mus-
cles as a resultof stroke, spinal cord injury,or
neuromusculardisease.
Shouldprogresscontinueat its presentpace,
human-machinecommunication could soon
lose its distinctionas the numberone obstacle
to bionics.It is relevantthereforeto revisitother
technical problemsin bionics. First, as noted
earlier,the size of electronicdevices limits the
functionalitythat may be implantedinside the
brainor elsewhere.Analog recordingelectron-
ics alone for a minimal brain interface(125
electrodes) would occupy board space of at
least 60 cm2, even with very large scale inte-
grationelectronictechnology.Sufficientminia-
turizationcould be reached within a decade,
however, as long as "Moore'slaw" is not re-
pealed.The latteris a remarkablyaccuratepre-
diction from the year 1965 that transistorden-
sity on integratedcircuitswould double every
year (with a slightrevisionin 1975 to doubling
every 1.5 years).
A complication of shrinking size is the
difficulty of hermetically sealing small ob-
jects and protecting them from corrosion in
bodily fluids. Another issue for electronics
is susceptibility, especially of digital sys-
tems, to electromagnetic interference. This
risk is so severe that some bionic systems, tic sockets for limbs, are glaring anachro-
such as the Jarvik 2000 heart (21), have nisms relative to other bionic components.
eliminated all digital electronics. Such pre- Users who subject themselves to brain im-
cautions may not be an option for bionics plantation of hundreds of electrodes and
that restore mobility, wherein digital pro- wearing of transmitters may rightly expect
cessing is fundamental,and hence new anti- a more versatile and responsive attachment
interferencestrategiesmay be required.Bat- of their arm or leg. To help meet these
tery energy density and rechargingissues will expectations, much fruitful work is focused
become limiting especially as orthotic and on how to integrate prosthetic structural
prosthetic devices gain functionalityand de- components, such as the titanium pylon,
mand more power. Maximal energy density with bone, i.e., osseointegration (25). The
of implantablebatteriesis about 1.1 W-hours/ natural feel, or "osseoperception," of the
cm3 (lithium ion); however, external battery environment provided by such attachments
packs that can use anode air-cells have higher can complete a sensory link that is crucial
densities. The slight risk of explosion by the to bionic restoration of function.
latter cells is one willingly borne by users of
powered prosthetic devices. The need for References and Notes
1. "Compact Dexterous Robotic Hand," U.S. patent
more frequent rechargingmay require more
convenient options than radio-frequency 2. number 6,244,644 (12 June 2001).
R. Abboudi,C. Glass, N. A. Newby, W. Craelius,IEEE
transmission, such as optical recharging(22). Trans.Rehabil.Eng. 7, 121 (1999).
Computer requirements for a typical 3. V. S. Ramachandran,Proc. Natl. Acad. Sci. U.S.A. 90,
PMI hand controller include a relatively 10413 (1993).
4. T. A. Kuiken,D. S. Childress,W. Z. Rymer,BrainRes.
high analog I/O of
throughput many kilo-
676, 113 (1995).
baud but minimal CPU power and memory. 5. http://world.honda.com/robot/(accessed 15 Novem-
The main processing task is decoding: de- ber 2001).
riving motor commands from volitions. The 6. www.yobotics.com/
task involves a mathematical operation 7. J. Wessberg et al., Nature 408, 361 (2000).
8. P. R. Kennedy,R.A. E. Bakay,M. M. Moore, K.Adams,
known as pseudoinversion that decodes sig-
J. Goldwaithe,IEEE Trans.Rehabil.Eng. 8,198 (2000).
nals from several sources in near-real time, 9. J. R. Wolpaw et al., IEEETrans.Rehabil. Eng. 8, 164
all of which contain a portion of the code (2000).
for a complex motion, such as finger flex- 10. N. Nicolelis, Nature 409, 403 (2001).
11. B. T. Smith et al., IEEETrans.Biomed. Eng. 45, 463
ion (23). There are several alternative ap-
(1998).
proaches to decoding, none of which is 12. L D. Maloney,Design News 90 (6 March2000).
clearly superior, including neural networks, 13. P. Rabischong, Bull. Acad. Natl. Med. 183, 531
pattern-recognition algorithms, and hybrid (1999).
14. Reuters, "Chip helps paraplegic walk," 20 March
filters.
2000, Strasbourg,France.
Although actuators are not yet opti- 15. G. Loeb,R.A. Peck,W. H. Moore, K.Hood, Med. Eng.
mized for bionic use, several commercially Phys. 23, 9 (2001).
available motors and servos are adequate 16. B. Hudgins,P. Paarker,R.N. Scott, IEEETrans.Biomed.
for some advanced applications, including Eng. 40, 82 (1993).
17. S. L. Phillips,W. Craelius,Robotica, in press.
a computer-modulated knee (24) and a 18. S. L.Phillipset al., Sixth WorldBiomaterialsCongress,
multifinger hand (23). The large sizes and Hawaii,#1343, 15 to 20 May 2000.
power inefficiencies of present actuators, 19. A. Benabidet al., Arch. Med. Res. 31, 282 (2000).
20. J. Rosengren,IEEESpectrum 35, 62 (1998).
however, limit functional expansion of ro-
21. R. Jarvik,V. Scott, M. Morrow, E. Takecuchi,Artif.
botic assist devices.
Organs 23, 487 (1999).
A final challenge for bionics is estab- 22. K.Goto, T. Nakagowa, O. Nakamura,S. Kawata,IEEE
lishing a convenient physical interface be- Trans.Biomed. Eng. 48, 830 (2001).
23. D. J. Curcie,J. Flint, W. Craelius,IEEETrans.Neural
tween effector devices and the body. Al-
Syst. Rehabil.Eng. 9, 69 (2001).
though limb replacement as depicted by 24. J. Michael,Clin. Orthop. Relat. Res. 361 43 (1999).
Hollywood will likely remain a fantasy, 25. R. Branemark,P. I. Branemark,B. Rydevik,R. R.Myers,
current interfaces, consisting of bulky plas- J. Rehabil.Res. Dev. 38, 175 (2001).

www.sciencemag.org SCIENCE VOL295 8 FEBRUARY

2002 1021