Those early Greek theatrical simulations of living beings conveyed a sense of life through motion, and motion remains a hallmark of artificial beings. We feel that motion means vitality, and hence is essential for lifelike synthetic creatures; in fact, the Nursebot, a robot developed at Carnegie Mellon University, which is designed to assist the elderly, not only has facial features (cartoonish ones) but blinks its eyes at regular intervals so that observers understand that something is “alive” in there, even when the robot is at rest. There are also practical reasons to incorporate mobility: What would be the use of a household robot that could not clean the floor, or an industrial robot that could not move its arms to weld an automobile door panel? While an artificial mind is an essential part of a useful robot or android, it is the addition of motion along with sensory interaction that turns a mind into a full being.

But what kind of motion? That depends on the goal: Is it to make a robot that is functional, but might not look at all human, such as Roomba, a robotic vacuum cleaner? Or is it to make a robot that, although it could not be mistaken for human, is sufficiently humanoid to operate in everyday environments, from homes to offices and

factories—the approach taken by the Honda Corporation. The goal might be a true android that, among other humanlike behaviors, walks on two legs and grasps and holds objects, and does so with human-appearing body parts, as might be important for entertainment robots. Or the aim might be to make a being that surpasses human capabilities with enhanced strength or speed, a feature that interests the military, or is designed from an utterly different premise than matching the human body, such as changing its shape to suit the task at hand.

For designers of artificial beings, there is an obvious appeal in selecting the simplest appearance and physical behavior that will do the job. For instance, there are easier means of locomotion than walking. As a kind of controlled fall endlessly caught and repeated, walking requires the ability to sense and maintain bodily balance, which requires in turn appropriate sensors and cognitive ability. But no matter what the design or form of its limbs, specific types of sensing and cognition are necessary if the being is to operate in the real world.

For instance, it would be valuable if the Nursebot robot, assisting in caring for the elderly, could follow a human’s instructions such as, “Go to Mr. Smith’s room, ask him if he would like lunch, and if he says yes, guide him to the dining room.” To successfully navigate its way to a specified location, whether externally determined or self-selected, the being needs vision or other means to examine and map its surroundings, knowledge of its present location in that environment, and the ability to plan a workable route from here to there while recognizing and avoiding obstacles, making proper use of door-ways and so on, along the way.

Similar considerations apply to artificial arms, hands, and fingers. They can be correctly positioned to perform their functions only if the artificial being has what is called in humans the kinesthetic sense; that is, the ability to determine where the body part is in space relative to the body itself. And to ensure that a hand is set correctly to safely hold a fragile object, or that an arm exerts enough power to lift a heavy load, the being needs a sense of touch and the ability to sense how much force is required to hold or move an object.

Hence two things are required to give artificial beings the ability

to move and to handle objects. One is mechatronic engineering design, which couples the mechanical principles of wheels, gears, and joints to electrical and electronic devices such as computer chips and servos. A servo is a type of actuator that animates artificial limbs. It is similar to an electric motor, with a shaft that can be rotated to any desired angle as set by an electrical signal. The units are powerful for their small size, and because they can be positioned just as desired, are good choices for versatile control of artificial body parts. The characteristic whine that always seems to accompany robotic movement in the movies is the sound of a servo being driven to a specific position.

Equally important, and generally more difficult to construct, are the bodily senses and cognitive abilities that enable the being to work out, for example, where and how its legs must be set to walk in a given direction, and then issue the necessary commands to its servos. That’s a complicated task. To make it even more demanding, the computation and the mechatronic response must occur quickly enough to allow the being to function in real time.

Depending on the bodily details, different levels of cognitive abilities are needed. As we will see below, wheeled robots, for instance, and robots that walk on four or six legs need less brainpower than those that walk on two legs.

Like a tricycle resting firmly on its three wheels, a wheeled being rolling on more than two wheels is inherently stable, and needs minimal cognitive capability to maintain its balance. Wheels offer a relatively simple mode of locomotion that is a good choice for robots when humanoid appearance is neither necessary nor desired. Nevertheless, wheels are not right in all circumstances. They work well on smooth surfaces, but rough terrain or a profusion of obstacles—such as occur in nature, at disaster sites, or in warfare—can defeat them. Under those conditions, continuous treads or tracks with cleats, like those on a bulldozer or military tank, serve better and still provide stable underpinning.

Because wheeled units are comparatively simple to design, they are used as test-beds by those who develop robots. Students and faculty in Carnegie Mellon University’s Robotics Institute, for instance, have gotten used to seeing wheeled cylindrical robots rolling through their building, testing the robots’ ability to map the environment and navigate through it.

Also for the sake of simplicity, present-day commercially available robots—other than those that simulate four-legged pets—roll on wheels. At least one, the vacuum-cleaning robot called Roomba, placed on the market in 2002, is a consumer item. It is made by the iRobot Corporation whose co-founder, chairman, and chief technology officer is Rodney Brooks, the MIT roboticist responsible for Genghis and Cog. Roomba is a 6-pound, disc-shaped unit, slightly more than a foot across, that rides on two wheels and an agitator brush. Unlike conventional vacuum cleaners that draw their electrical power from a cord plugged into a wall socket, Roomba operates from rechargeable batteries and hence has complete freedom of motion.

Put Roomba down in the center of a room, and it begins to cover the floor in a pattern of broad spiral sweeps (which give Roomba its name) until the robot encounters a wall. Then its behavior changes to track along the wall and clean the wall base with side brushes, until it makes another foray across the room. If Roomba encounters an obstacle such as a chair leg, it stops and goes off in a new direction, and it is also intelligent enough to halt and turn around at the top of a flight of stairs. Eventually the robot cleans the entire area, and then it stops, ready to be placed in another room or recharged. While the robot has sensors and a level of intelligence called “heuristic learning logic” that was first developed for units used to clear minefields, the fact that it does not need to maintain balance as it rolls simplifies its design and lowers its price.

The iRobot company has also developed tracked robots in the form of small units called PackBots. These were designed to function in difficult environments, and so were well suited to enter the wreck-age of the World Trade Center after the terrorist attack of September 11, 2001. Equipped with video cameras, they sent back images as they

clambered over debris, and helped human searchers recover a number of bodies.

In 2001, in the first actual appearance of robots in warfare, the U.S. military used PackBots in the Afghanistan campaign to search caves for Al-Queda operatives and to locate hidden weapons and mines. These 42-pound, backpack-size units were not truly autonomous because they required human operators; but if they and other military robots are developed further as envisioned, we might one day see a single soldier monitoring several self-directing units, thus achieving the military dream of force multiplication: “You could have 10 people on the battlefield doing what once took 40 soldiers,” notes Ronald Arkin of Georgia Tech, who has written software for military robots.

Treads are good for tackling steep grades, but less so for stairs. In fact, for typical human environments that can include stairs or varied surfaces such as a deep-pile rug or a tile floor, neither treads nor wheels work as well as legs and feet, although not necessarily only two of each. Robots that walk on four or more legs remain balanced even if not all the legs are on the ground. That inherent stability can be valuable for the broken terrain to be found in the field, or in NASA-sponsored expeditions of planetary exploration, and requires fewer cognitive resources to maintain balance.

Rodney Brooks’s Genghis robot illustrated that multilegged walking on flat surfaces and over obstacles can be achieved with only limited computational power. My I-Cybie robot dog is not the most advanced robot toy on the market and operates at a low level of artificial intelligence; nevertheless, it manages to shuffle along nicely on four legs (it can even right itself to stand firmly on those legs if you tip it over, and can also work its way around obstacles, walking off in a new direction when it bumps into something). One multilegged robot that has dealt with extremely rough terrain is the eight-legged Dante II. In 1994, this unit semiautonomously walked and rappelled

its way down into the crater of an Alaskan volcano, which it explored for the better part of a week. Many six-legged robots have been developed as well.

To develop general approaches to designing robots with legs, however, especially those that walk on two legs as people do, is a complicated business that has occupied researchers for years. One well known establishment, the MIT Leg Laboratory, is devoted to the study of locomotion and the construction of legged robots. Other research on two-legged beings is carried out internationally at universities and corporations, with the leading efforts in Japan at Tokyo University, the Honda Corporation, and elsewhere. These efforts are central to the widespread proliferation of artificial beings. If they are to work along-side people and interact with them, they will have to be two-legged and generally humanoid so that they can operate in regular human environments and use human devices such as screwdrivers and door-knobs. Another advantage of humanoid robots is the versatility of the human frame. Cheetahs run faster, dolphins swim faster, and chimpanzees exert more strength than humans, but the multipurpose human frame is moderately good at all these things. This adaptability defines the approach chosen by many researchers to build a humanoid robot with similarly broad functionality.

That versatility begins with the ability to walk on two legs, as shown, for instance, by the robots developed by Hirochika Inoue, of the Department of Mechano-Informatics at the University of Tokyo. His latest units, called perception–action integrated humanoid robots, are named H6 and H7. They are human-shaped with head, torso, and limbs but lack facial features and are recognizably mechanical. Conditioned as we are to seeing inscrutable movie robots stalk off to carry out their plans, like the faceless Gort in The Day the Earth Stood Still, we might expect H6 and H7 to look menacing—but they are neither hulking nor brutish. They are white in color, and not very big: H7 stands 1.4 meters (4’ 7”) tall, with a mass of 55 kilograms (120 pounds, including 9 pounds of batteries to power it).

It is illuminating to see H6 and H7 in action because their walking style is almost tentative. When either unit walks, it is accompanied

by human minders, placed so as to catch the robot and keep it from damaging itself if its balancing abilities fail and it topples. Moreover, the walking pace is slow. Most telling of all, H6 and H7 do not swing their arms while walking, as people do, but hold their arms at their sides, bent at the elbow. This makes the robots seem slightly nerdy, as if they were too tightly buttoned up to stride freely down the street.

But even that cautious, somewhat geeky walk, though a far cry from the menacing lurch with extended arms featured in horror movies, is highly significant. Its achievement has required considerable mechanical and computational development. Successful humanoid robots need mechanical frames that are highly flexible, which is defined in terms of “degrees of freedom.” Each degree of freedom means the capability to move a limb or other appendage in a given direction about a joint. H7 can move each leg in any of six directions, corresponding to six degrees of freedom, plus a seventh that comes from an adjustable toe joint in the foot. There are 23 additional degrees of freedom built into the robot’s body, and all its joints are moved by electric motors that drive gears.

For H7 to walk in a given direction, its legs must be set correctly and their movements coordinated in space and time. At the same time, the robot’s body must be constantly adjusted to maintain its balance as it walks. Moreover, the robot cannot be allowed to self-collide—that is, have one moving part strike another, such as the legs becoming entangled. With the robot’s humanoid shape and many degrees of freedom, there is a multitude of possible bodily configurations, each of which must be examined to ensure safe and successful walking. This is an extremely demanding computational task that must be performed in real time as the robot advances through the world. The necessary calculation power is provided by the equivalent of two powerful laptop computers built into the robot, with more power coming from other remotely linked computers. Even this computational armory would not be enough without clever algorithms that minimize computation time.

Similar considerations apply to what are probably the best-known walking robots in the world, which have been undergoing develop-

ment since 1986 at the Honda Corporation in Japan. The earliest unit in this series could indeed have come from a science fiction film. It consisted of a pair of legs attached to a large squarish box the size of a microwave oven, and resembled the walking battledroids seen in the Star Wars movies. If this inhuman-looking robot moved with any facility and speed, it would be a fearsome thing to see bearing down on you. Fortunately, this early model was not a very impressive walker: it took all of 5 seconds to calculate the leg position and foot placement for each step, and it could walk only in a straight line.

Further development produced refinement after refinement, but it took 10 years for Honda to unveil its first humanoid walking robots, called P2 and P3, followed by an improved version called ASIMO (advanced step in innovative mobility), announced in 2000. P3 and ASIMO have appeared in Honda’s corporate advertising and are available for public events and expositions. In early 2002, ASIMO rang the opening bell at the New York Stock Exchange to celebrate the twenty-fifth anniversary of Honda’s listing on the exchange. In fact, the robot is being groomed as a general-purpose unit: According to Honda, “In the future, we anticipate ASIMO developing capabilities in areas such as household assistance and tasks dangerous for humans—like firefighting.”

Both P3 and ASIMO resemble a person in a white spacesuit or suit of armor, topped by a helmet with a dark visor; nothing like a face is visible. Each robot carries a sizable backpack, which houses its on-board computer and its batteries. The differences between P3, and ASIMO, developed only a few years later, illustrate the rapid pace of robotic improvement. P3 is the size of a very small adult, standing 1.6 meters (5’ 1”) tall, but despite the use of the light metal magnesium in its construction, comes in at a hefty 130 kilograms (285 pounds). ASIMO, however, has been pared down to child-size, standing 1.2 meters (3’ 10”) tall and weighing only 43 kilograms (95 pounds) Yet this smaller robot is smarter and more able than its older brother, although it walks slightly slower, at 1.6 kilometer per hour (1 mile per hour). One battery charge keeps it going for 30 minutes.

ASIMO is more flexible than P3, sporting 26 degrees of freedom

including six in each leg. In addition, it incorporates a “predicted movement control.” Like a walking man seeing a corner coming up and shifting his stance to accommodate a turn, ASIMO looks ahead to the next stage of its motion and shifts its center of gravity accordingly. With its greater flexibility and improved cognitive skills, ASIMO is a smoother and better walker than P3.

ASIMO can maintain its balance standing on a steeply tilting seesaw, using a telescoping knee joint rather than bending a knee as a human would. It can neatly execute a turn (which P3 does only awkwardly in a series of shuffling steps), balance on one leg, and climb confidently up and down stairs, although it must know the stair height in advance. In fact, to a human observer, both ASIMO and P3 radiate a certain self-assurance in walking that H6 and H7 do not. The reason is that the Honda robots swing their arms in human fashion while walking, giving the impression of a confident robot that knows what it’s doing and where it’s going—one more illustration of the effectiveness of humanlike clues built into artificial creatures.

Although humanoid robots offer great versatility, and walking on two legs is an important achievement, there are reasons to consider other body shapes and ways to move. Many living creatures progress by crawling, slithering, or skittering rather than walking. For all its versatility, a humanoid body cannot efficiently emulate these motions, which might prove to be the best choice for certain applications. Hence some roboticists are developing robots with nonhumanoid body shapes and means of locomotion.

The robots made by Shigeo Hirose of the Tokyo Institute of Technology, for instance, are not humanoid. His ACM R-1 (active cord mechanism, revised model) is long, skinny, and snakelike and slowly slithers along at about 40 centimeters a second (just less than one mile per hour). Hirose was, in fact, inspired by studying the movements of snakes, and his choice of body type also fits into an engineering philosophy that takes the simplest solution to be best. Instead of trying to

make a multipurpose humanoid robot, this approach chooses from among the multitude of possible bodily designs the best one to do a specific job. ACM R-1, for instance, is ideally shaped to explore underground pipes, if not for much else.

While Hirose’s robots are not humanoid, they maintain a fixed form. But there is a more radical approach to locomotion and bodily design: robots with no permanent legs or arms, no fixed bodily configuration, that dynamically change their shape and means of locomotion to meet the needs of the moment. One version of these reconfigurable robots, called PolyBots, is being developed by Mark Yim, Ying Zhang, and David Duff of the Palo Alto Research Center. These researchers contrast the fixed assembly-line world of industrial robots with the shifting demands and terrain of the real world and envision a robot that could

The key to this flexibility is to construct the robot from individual modules of only one or two types, but numbering in the hundreds and potentially in the millions.

One PolyBot under study, dubbed G2, has modules, each a cube 5 centimeters (2 inches) on a side, that can automatically connect with each other to form long strings. There are two types of modules: motion units, which use a hinge moved by an electric motor to inch along the floor, and node units, which have multiple attachment points so that other modules can branch off in different directions. Each module houses a powerful computer-processing chip with a lot of memory, giving the robot an intelligence that distributes instructions and data throughout its structure, including commands that synchronize the motion hinges so that the entire PolyBot moves.

With its many motors, sensors, and computer chips, Polybot requires a good deal of electrical power, and battery life is a problem. Also, programming PolyBot becomes a struggle, according to the researchers, when many modules (units have been made with up to 100

modules) are involved. Nevertheless, the robot successfully demonstrates different styles of locomotion, including earthwormlike (the modules alternately squeeze and stretch), Slinky-like (end-over-end, like the child’s toy), and caterpillarlike (many small feet). A PolyBot can also reconfigure itself, and not only for motion. It can create armlike limbs to deal with small objects.

A video designed to display PolyBot’s versatility shows it making its way through a military-style obstacle course while a bemused U.S. Marine watches (the military connection is that the project is partly funded by the Department of Defense through DARPA). Another video, this one computer-simulated, shows PolyBot successively adopting different shapes and motions for different terrains: a rolling tank-tread-like loop for a flat surface, an earthworm to crawl down a step, and a multilegged spider for rough terrain. The configuration change from worm to spider is especially striking because it dramatically illustrates the effective operation of a robot utterly different from anything human.

Walking on two legs, as humans do, might seem a disadvantage compared to the flexibility of a PolyBot or to the stability enjoyed by four-legged creatures. But walking on two legs rather than four frees our arms, hands, and fingers to carry out the complex functions of grasping, holding, and feeling. This extraordinary versatility is not only an essential part of being human; it is one reason we are human, because our flexibility in manipulating the world has improved our thinking capacities.

That flexibility comes largely from the opposability of the human thumb—that is, it can be brought into contact with the tips of any of the four fingers. Although some other primates can also do this, the crucial point is that humans have a large area of contact between the thumb and the sensitive skin of the fingertips. This gives enormous dexterity in dealing with objects and their textures, which has strongly affected the development of our species. Benjamin Franklin was

among the first to realize that the ability to manipulate is essential to human culture; it was he who defined humanity as “the tool-making animal.” The physician and primatologist John Napier enlarges on this comment in his book Hands, explaining that the opposable thumb

But the power of a hand cannot be brought to bear without immense flexibility from shoulder to wrist, which in humans draws on six degrees of freedom. Two of these degrees represent rotation at the shoulder, raising the arm higher or lower, and rotating it backward and forward. A third is found at the elbow joint, and three more operate at the wrist—rotation around the axis of the forearm, movement of the hand up or down, and movement left and right. Designing so versatile a jointed system for an artificial being is an engineering challenge, and designing it to move properly is a geometric and computational one, as complex as programming a robot to walk on two legs, and sometimes requiring advanced mathematics. But the humanoid robots H6, H7, P3, and ASIMO all have arms and wrists as multijointed as human ones and are programmed to carry out some manipulations. H7 can reach under a table and grasp an object on the floor, and P3 can turn a handle to open a door.

However, what passes for hands on these units are poor substitutes for human hands. Rather than flexible thumb and fingers, they have a gripper design that grasps an object between stiff tines, with a squeezing action like a big pair of tweezers. Moreover, the robotic hands lack any sense of the weight or texture of an object. This limited design appears in prosthetic hands as well; Paolo Dario and his colleagues of the Mitech Laboratory, Scuola Superiore Sant’Anna, in Pisa, Italy, note that “current prosthetic hands are simple grippers with one or two degrees of freedom, which barely restores the capability of the thumb–index [finger] pinch.”

Nevertheless, even limited capability is worth a great deal in a prosthetic replacement and is not necessarily a problem for many

robotic applications. An assembly-line robot that manipulates automobile parts needs brute strength, not sensitivity, and works perfectly well with powerful arms terminating in pincers rather than fingers and thumb. Universally useful artificial beings, however, need functioning fingers-and-thumb hands that include appropriate sensory feedback. A differently shaped or unfeeling hand could not operate electrical switches and valves, or properly use screwdrivers and hammers, all of which are designed for the human hand.

A more profound issue is summarized in John Napier’s comment “a lively hand is the product of a lively mind…. When the brain is empty, the hands are still.” The inverse might also be true—limited artificial hands might ensure a limited artificial mind because the hands extend the abstract power of the brain into the real world. Even if the brain is made of processed silicon rather than living neurons, the addition of hands turns an isolated artificial intelligence into an interactive artificial being. Such a being could be designed to grow in knowledge and capacity as it explores the world with its questing fingers, just as the human brain has grown through a two-way feedback with the body’s own searching fingers.

Writing in the late 1970s, Napier expressed little hope that an artificial hand could be built. “There is nothing comparable to the human hand outside nature,” he said, “for all our electronic and mechanical wizardry, we cannot reproduce an artificial forefinger that can feel as well as beckon.” Although the human hand is indeed difficult to copy, since Napier’s pessimistic statement, we have come a long way toward building robotic and prosthetic hands that work like natural ones. One leading example is under development at NASA’s Johnson Space Flight Center in Houston, in conjunction with DARPA. NASA has considerable expertise in building wheeled robots, like the Sojourner rover that explored the Martian surface during the 1997 Pathfinder mission, but the artificial hand is different. It is the critical part of Robonaut, a robotic astronaut designed to stand in for a human when work must be carried out in space on Earth satellites, or on the International Space Station now taking shape in orbit.

Astronauts cannot breathe in a vacuum and need considerable time to suit up before venturing into space, time that could be critical in an emergency, The nonbreathing Robonaut can be ready to go at a moment’s notice while the astronaut remains safely inside the spacecraft. Because Robonaut is designed to work in space where it will float weightlessly, it does not need a full body but might eventually be outfitted with a single support leg. At present it consists simply of a torso with attached arms. Covered in a protective white synthetic material and surmounted by a sleek helmet, the robot resembles nothing so much as an Olympic fencer clad in the classic white jacket and protective headgear.

Robonaut is not autonomous. It is remotely controlled by a human who sees three-dimensionally through two video cameras in its head that send their images to the operator’s eyes through special goggles. The operator’s hand is inserted into a data glove that faithfully converts movements into electronic signals, which the robotic hand obeys to mimic the operator’s hand movements. The artificial hand matches a natural one in size and shape, but it is less flexible. It has 12 degrees of freedom, and two more in the up–down and side-to-side movements of the wrist (all operated by electric motors), compared to the 22 degrees of freedom in the human hand. Robonaut’s thumb is opposable only to the pointing and index fingers, and this triad is used where dexterity is needed, whereas the remaining two fingers bend back and forth toward the palm to aid in grasping. While this is nominally a limitation, Robonaut’s designers note that it is still more dexterous than a human astronaut constrained by a spacesuit.

Operating the robot is an immersive experience that Robonaut’s creators call “tele-presence” even without feedback of force or tactile information from the hand (which is now under development). The operator tends to sense the robot’s body as his own, illustrated in one incident where the operator moved his feet when Robonaut dropped a tool, although the robot has no feet, and so the human’s own well-developed bodily intuitions help him control the remote hand. Robonaut has proven adept at handling an electric drill, giving injections for medical emergencies, working with hardware components

used in the International Space Station, and, in one demonstration with implications for antiterrorism operations, opening a backpack, sorting through its contents, and choosing and removing a particular item—demonstrating how Robonaut might be used to search for a weapon or explosives while its operator remains at a safe distance.

With its human-appearing outline but different joint construction and texture, Robonaut’s hand combines organic and machine looks, and there is a cyborglike aspect to the entire robot. The operator, although in no way neurally plugged into the robot, functions as the partly merged cognitive center of a piece of machinery, like a brain transplanted into a mechanical body. Robonaut is not the only remotely operated robot. The Utah-based Sarcos Corporation, for instance, has demonstrated a humanoid, human-size unit whose arms and legs follow the actions of a human operator (plugged into what amounts to a full-body data glove) well enough to dance, although not very gracefully.

Robonaut and the Sarcos robot represent halting first steps toward beings as capable as the fictional RoboCop and Deirdre. But researchers are getting closer to those sophisticated imaginary cyborgs, driven primarily by the desire to make better prosthetic devices, with new tools to accomplish that. Gerald Loeb and Frances Richmond, of the University of Southern California, note that although two centuries have elapsed since Galvani observed that electricity makes muscles twitch, only in the last three decades have roboticists and neurophysiologists begun seriously addressing the problem of how to make an artificial limb move under neural control. Now, they say, we have reached the point where it “appears feasible to graft robotic and electrophysiological instrumentation onto a biological system to repair it,” but also note that this requires

The possibilities, and the problems, for direct neural control of limbs are illustrated by a unit proposed by Paolo Dario and his colleagues. This effort aims at a prosthetic or “biomechatronic” hand

designed to replace a missing hand with one that looks real and copies natural motion and sensing abilities. This design also lends itself to bionic enhancements, as well as use in artificial beings—a prime example of the fruitful crossovers between medical applications and robotics.

Like Robonaut, the biomechatronic hand gives priority to the thumb and first two fingers, which move around an object and fit themselves to its shape to grasp it. The three digits are driven by electric motors and linkages tiny enough to be embedded in the palm and the fingers without making the hand look unnatural; however, the remaining two fingers are not motor-driven. Dario’s group has not yet built an entire hand, but they have made a prototype plastic finger, with two degrees of freedom corresponding to the two joints of a human finger. Despite the smallness of the motors in the artificial finger, the force it exerts as it bends is comparable to that from a real finger, and so the artificial hand seems capable of fine manipulation at least.

As far as its mechanical design goes, the biomechatronic hand in its present form could serve as a robotic appendage. But Dario’s project also includes efforts to link the mechanical device to the human nervous system to make an advanced prosthesis, and so enters the area of “neurorobotics.” As defined by John Chapin of the State University of New York and Karen Moxon of Drexel University, “ ‘neurorobotics’ seeks to obtain motor command signals from the brain and transform them into electronic signals suitable for controlling a robotic device.” Those commands could come from the cortex of the brain or from the peripheral nervous system.

The cortex is the thin, wrinkled layer of “gray matter” that covers much of the brain. It performs higher intellectual functions and deals with sensory information, speech, and motor activities. The peripheral nervous system is the part of the nerve network that carries outside stimuli to the brain and returns appropriate responses. Even when a hand or limb has been amputated, the neural signals that once controlled the appendage are still generated in the cortex and sent to the appropriate nerves. The goal is to extract these neural pulses at the

cortex or in the peripheral system with an interface that converts them into electrical signals that can control a prosthetic hand (or any device), and, working in the opposite direction, that accepts electrical pulses from the hand (for instance from tactile sensors) and converts them into meaningful neural signals for the brain.

The technique explored by Dario’s group is one of several used or proposed for such electroneural interfacing and relies on an interface implanted into the peripheral nervous system. This approach uses a so-called regeneration type of neural interface, which blends techniques from electrophysiology and from the well-established nanotechnology used to manufacture silicon computer chips. Silicon is a good material for implantation because it seems to be nontoxic and has the necessary mechanical and electrical characteristics. Also, it can be readily manipulated by using state-of-the-art chip technology at the small scale of the human nerves.

The interface unit begins with a minute square of silicon 1.5 millimeters (0.06 inches) on a side. A set of metal electrodes is deposited on the surface of the chip, and further processing produces an array of tiny square holes, measuring only thousandths of an inch across, that pierce the chip. Then the whole chip, now called a die, is placed within a small conduit made of a nontoxic plastic that will not deteriorate inside a living body. The electroneural connection is accomplished by cutting a specific nerve and letting it regenerate within the plastic conduit. Nerve fibers reconnect themselves through the holes in the die, bringing them near the electrodes, so that current can flow in both directions between nerve and electrodes. Tests in which the interfaces were installed in rabbits showed that the interchange of current works: Signals originating in the nerve were detected in wires attached to the electrodes on the die, and external electrical signals sent to the electrodes affected the nerve, making the rabbit’s leg twitch.

For all the preliminary success of the rabbit experiments, serious obstacles stand in the way of extending the technique to humans. The neural signals are small and full of extraneous noise. They require complex processing to interpret correctly, and it has not yet been shown that the signals could actually control an artificial hand. On the

biomedical side, we don’t know if there would be harmful long-term effects on the body from the implants or if the combination of regenerated nerve and silicon chip will hold up over a long period. And most important, cutting a nerve is not something to be done lightly. Such a risk might be acceptable for an amputee who wants to replace a lost limb with an artificial one, but it might not be so for an uninjured person seeking only to be bionically enhanced.

Nevertheless, setting risks aside for the moment, in principle the regeneration-type interface dramatically illustrates how neural science combined with nanotechnology enables the transmission of internal neural signals to and from external digital electronics. The intrusive nature of this approach is an issue, but less invasive methods, which I will describe in Chapter 8, might be more acceptable. Electroneural interfaces have also made it possible to create living–nonliving hybrids where the living part consists of neurons or low-level animal systems, also considered in Chapter 8.

While the successful merging of the biological with the artificial is still under development, in one respect artificial beings have clear biological roots: In gross outline, artificial beings largely resemble natural living beings—if not humans, then animals such as snakes. Even the formless multiunit PolyBot moves somewhat like an earthworm or spider. The reason is easy to see. Many contemporary roboticists would agree with the engineer in Karel Capek’s R.U.R. who says, “[T]he product of an engineer is technically at a higher pitch of perfection than a product of Nature…. God hasn’t the slightest notion of modern engineering.” Nevertheless, evolution has produced, in the billions of years of earthly life, varied and workable solutions to the problems of motion and manipulation, from insect legs and mammalian limbs to the ribs of the snake and the tentacles of the squid.

This does not mean that roboticists cannot find solutions that go beyond what nature provides. There are natural limits to the strength and speed of animals. But some natural solutions, such as the

extremely rapid skittering motion of a cockroach, offer unique possibilities for artificial beings. However, nature does not move by means of wheels, gears, or servos, and so artificial beings do not move like animals.

Roboticists sometimes deliberately copy animal locomotion, as in Shigeo Hirose’s snakelike ACM R-1 unit. Examples like that reflect a general perception that if we study biological systems, we can use the principles we learn to improve our own creations. Applied to locomotion, this biomimetic approach draws on biomechanics, the science of how animals move. One of its leading practitioners is Robert Full of the University of California at Berkeley. The name of his Poly–P.E.D.A.L. Laboratory reflects what he does, which is to study the performance, energetics, and dynamics of animal locomotion in multifooted creatures.

Full’s main goal is a biological one: to understand the mechanical performance of animals in quantitative terms, building toward a general theory of how animal bodies have developed and how they work. The significance of this is to better grasp the enormous diversity we observe in animals and relate it to their evolutionary history and ecological significance. But as Full learns in detail how animals walk, run, climb, jump, and bounce around the world, his findings provide remarkable inspirations for the designers of robots. Full’s studies of lizards, centipedes, and other creatures have contributed to the design of, for instance, legged robots that maintain balance with fewer cognitive resources than conventional designs and a crablike unit that can move on land or underwater, and in creating Mecho-Gecko, a robot based on studies of gecko lizards and cockroaches that can climb walls.

Another aspect of biomimetics is to understand how to build machines that look more human or animal-like, more natural, as they move. A walking robot, even one with many degrees of freedom in its legs, simply does not walk like a person. Our limbs move quietly and smoothly as our muscles contract and extend, whereas a robot’s movements are defined by rigid mechanical linkages driven by precision servos. Hence a robot’s motion does not look quite right when it performs large-scale movements such as are involved in walking, or

fine movements such as the changes in eyes and mouth that animate human faces.

Kismet, the robot built by Cynthia Breazeal at MIT to explore human–robot interactions, shows to great effect the importance of small-scale movements. Much of Kismet’s impact on people comes from the humanlike motion of its head and face. Kismet, says Breazeal, “could actually make eye contact with you. It’s night and day when something looks into your eyes versus at your face or just at you. Eye contact is profound.” Kismet’s head movement and changes in expression are accomplished by approximately two dozen servos. They drive, for instance, its lips, which are sufficiently flexible to show different expressions, but are not meant to reproduce the versatility of natural human lips.

What is important about Kismet is its demonstration of the power of a social component in human–robot relationships, even though the robot is not a human replica. The social connection might be enhanced, however, with robots whose facial and bodily movements appear natural, and this objective might call for a different system than motors and mechanical linkages. An alternate approach that shows promise for simulating natural movement comes from new classes of “smart materials,” which change their properties according to external stimuli or environmental conditions. For example, the plastics called electro-active polymers (EAP) change length or shape under electrical voltage, mimicking what natural muscles do under neural control. EAP materials date to the late nineteenth century, but they came under serious study only in the 1990s. In one type, electrons in the material set up an electric field that causes stretching or shrinkage. In another, which usually operates in a liquid environment, the motion of charged atoms—ions—makes the material bend. According to the Jet Propulsion Laboratory’s Yoseph Bar-Cohen, a leading researcher in the area, “the main attractive characteristic of EAP [materials] is their operational similarity to biological muscles, particularly their resilience.”

Along with Bar-Cohen, researchers around the world are exploring these materials. One dramatic example of these efforts can be seen

in action at the Artificial Muscle Research Institute (AMRI) at the University of New Mexico, where a skeleton called Myster Bony has been fitted with artificial leg muscles. Perched on a fixed exercise bike, with its muscles hooked up to electrical power, Myster Bony pedals away indefinitely—or at least as long as the electricity holds out. Another example at AMRI is an artificial fish slowly swimming through the water as its tail, powered by an artificial muscle, switches gently from side to side. AMRI, where researchers study several types of artificial muscles, involves both engineers and medical researchers; some of its projects provide hope for eventually using synthetic muscles to help people with muscular dystrophy.

Further uses of EAP and similar materials can be seen in Bar-Cohen’s laboratory in Pasadena, every corner of which contains a different device that illustrates the versatility of smart materials. For example, NASA is constantly seeking reliable, lightweight materials for use in space, so Bar-Cohen designed a tiny EAP-driven wind-shield wiper that sweeps back and forth to dust off a small glass observation window. As Bar-Cohen puts it, this is a device where “suddenly the material is everything,” meaning it performs without conventional moving parts such as gears and bearings, a tremendous advantage for space applications. NASA approved the wiper for use aboard a small, wheeled robot called Nanorover. With a mass of 1.1 kilograms (2.5 pounds), Nanorover was designed to explore and send back data from a small asteroid, much as the Sojourner rover explored the surface of Mars.

Also in Bar-Cohen’s laboratory is a model of a human head sculpted by David Hanson, who has designed animatronic entertainment robots for the Walt Disney Company. Its facial details, texture, and coloring are persuasive, down to the slightly bloodshot eyes: but the most fascinating feature is that its eyes and mouth are moved by artificial muscles of Hanson’s own design. Although these muscles do not employ EAP materials, they show what synthetic muscles can do: With the power turned on, the mouth smiles and the eyes move (one winks), a powerful example of the power of muscles over gears in robot animation.

Bar-Cohen, along with Cynthia Breazeal, has edited a book called Biologically-Inspired Intelligent Robots to further explore the possibilities for artificial muscles. He notes, however, that artificial muscles have a long way to go to become effective in robots. For instance, the relatively weak forces that EAP materials exert limit the strength of an artificial limb and need to be enhanced. He also admits to some doubts about the creation of highly advanced artificial beings. “I am concerned,” he says, “because once you release a technology you never know which way it is going to be developed.”

But he feels entirely confident about designing prosthetic devices that use artificial muscles, and holds out some dazzling speculations for the future. One type of material now under study can change both its configuration and its color under electrical stimulus, which might lead to artificial faces that not only smile and frown but also blush. And Bar-Cohen imagines putting EAP materials into a form that can be sprayed out from a special printer much as ink droplets emerge from inkjet printers. The result might be an EAP-operated butterfly printed flat onto a sheet of paper, ready for shipping, that flaps its wings and flies off when released from its box.

It’s a far cry from the ponderous walk of Gort the robot to ASIMO’s confident stride and on to the smaller motions of Kismet’s expressive face. These movements are important in the usefulness and acceptance of artificial beings, with small motions as meaningful as large when it comes to eliciting human reactions.

Successful artificial bodies, however, require more than just the right facial expression. Witness Bar-Cohen’s arm-wrestling challenge. In 1999, to stimulate researchers, he set them the task of building an arm driven by artificial muscles that could defeat a human arm wrestler. Although this challenge has yet to be met, it is significant as a kind of Turing test for machine physicality. Turing’s original test for machine intelligence depended on verbal ability, to be judged by a machine mind’s response to queries. Isaac Asimov’s story “Robbie” presents a kind of Turing test for machine feelings, as little Gloria judges her robot to be as kind and patient as any human, through its responses to her needs. Bar-Cohen’s challenge adds another Turing-

like dimension, for we might ask a person to wrestle an arm protruding through a curtain without seeing what the arm is attached to. After wrestling, and perhaps losing, we ask the human to judge whether the arm’s owner is human or artificial—and not only by the strength of the arm, but also by the strategy and tactics employed by the brain controlling the arm.

Building persuasively human artificial limbs and bodies is more than a matter of mechatronics or the faithful modeling of the human form. If an artificial arm is to do well at arm-wrestling, it must include sensory abilities that determine what the opposing arm is doing, such as judging the strength and direction of the forces it exerts, and it must draw on cognitive abilities to interpret that information and work out effective defensive and offensive movements. Just as a chess-playing computer has to weigh the moves made by its human opponent and develop an opposing mental strategy, an arm-wrestling artificial being must assess its opponent’s physical behavior and work out a physical strategy. This kind of intelligent response requires artificial sensing and artificial thinking, the subjects of the next chapters.

This page intentionally left blank.