This paper illustrates various aspects of technological advancements in designs and functioning of prosthetic hands. Schunk, a top supplier of robotic grippers, believes humanoid hands will make robots more flexible and eliminate tens of thousands of dollars’ worth of auxiliary equipment. Shadow Robot's hand can grip an egg or a pair of pliers. The company is building hands that mimic human motion for service robots, which will act like valets to perform a variety of tasks for their masters. With a hand, a robot could determine if it could get to a part and then configure its hand to get to the location and grasp the part. Touch Bionics’s i-LIMB hand combines an independently powered thumb and index finger with three fingers that move in unison. Muscle contractions in the forearm control its movement. It can peel a banana, lifting a credit card off a table, or holding a briefcase. It is expected that surgeons in hospitals may use robotic humanoid hands to perform delicate operations in remote locations.

## Article

Consider the hand. It can turn and lightly pinch a single grape off a stem, yet squeeze and twist hard enough to unscrew the top of a jar. It can brush away the tears of a crying child, or use a hammer to drive a nail into a wall.

With five digits that bend and an opposable thumb, the hand is a marvel of dexterity. According to Michael Pollitt, CEO of Shadow Robot Co. in the United Kingdom, the hand accounts for 26 percent of the human body 's movement potential. In the mechanical world, there is nothing like it-at least not yet.

A number of small firms like Shadow Robot-and some larger ones like Germany's Schunk GmbH & Co., one of the world's largest suppliers of robotic gripperswant to change that. They believe that flexible mechanical hands will unlock the potential of robots and replace lost limbs in humans.

Everything about hands makes that vision an uphill struggle. For most engineers- even those at universities and research centers- the installation of a machining center with six or seven degrees of freedom is an event. Yet, they applaud with hands that cram 21 degrees of freedom into the space of, well, a hand.

Nature has not created a simple template for mechanical engineers to follow. "The way the thumb moves in and out, and the way the muscles overlap and attach, is just a fearsome nightmare," said Shadow's technical director, Rich Walker, a 20-year veteran of the field. " If you get three professors of anatomy together to explain it and get them drunk, they'd have a fight."

Engineers seek to ape that combination of strength, compliance, and precision with exoskeletons, actuators, motors, gears, sensors, and electronics. According to Pollitt, "After artificial intelligence, developing a dexterous hand is the next hardest challenge in robotics."

With a hand, a robot could determine if it could get to a part and then configure its hand to get to the Location and grasp the part.

## Flexible Hands

So why bother? On the industrial side, an answer is' easy: Flexible hands make robots much more capable. Schunk Automation Group 's product manager, j esse Hayes, believes that the flexibility of robotic hands will earn them space on the factory floor over the next five to 10 years.

This will happen because of inherent problems with end effectors, products that Hayes knows intimately. End effectors are tools that fit on the end of robotic arms to weld or paint, or to grip par ts to pack or assemble them. The problem is that these tools are highly specific. They are designed and machined to pick up one specific part (or perhaps a family of parts with similar features) in one specific position. "If I have to load 15 parts, then I need five or 10 different effectors," Hayes said.

This can be an expensive proposition. Most mechanical grippers cost $200 to$1,500, plus $100 to$1,000 more for the metal fingers that actually pick up the parts. A truly flexible manufacturing setup needs a wide range of sizes and shapes. It also needs an automatic tool changer so it can swap end effectors in seconds, and this can run to thousands of dollars more.

Most robotic assembly or packaging lines pack parts on pallets in a specific order and orientation, so the robot can find and grab them easily. On advanced lines, though, robotic vision systems recognize unsorted parts in a bin and grab them-sort of. According to Hayes, "The robot is designed to open and close at a fixed distance, so they can only pick up parts in one certain area. Parts that are not in that area have to go around again."

Flexible hands would change everything. Instead of a rack of tools, the hand could reconfigure its fingers to pick up just about anything.

"With a hand, a robot could determine if it could get to a part and then configure its hand to get to the location and grasp the part," Hayes said.

Such flexible robots would find a place in high-precision custom and semi- custom manufacturing, where companies run a batch of 1,000 parts one day and 2,000 different parts the next. It might help consumer products companies keep up with product and packaging trends. If the price fell enough, even traditional manufacturers might embrace the technology, because it would enable them to eliminate changing tools or palletizing parts before sending them to the robot.

"It would help us expand our gripper business so that we could get away from fixed automation, where a tool does one job, to truly flexible automation," Hayes noted. First, though, someone has to make a hand that factory operators can afford.

## Breakaway Design

Hayes is the first to admit that the company's threefingered end effector is not that hand. It is primarily a research tool. It has seven independently operated servos and instrumented all over with pressure arrays to provide tactile feedback.

"The final version may be dumbed down to save cost, but right now, people want every bit of information and feedback they can get to understand the application," Hayes said. He estimates the cost at $75,000 to$80,000. So what do you get for the money? Unlike pneumatic or hydraulic grippers, Schunk's three-fingered hand is all-electric. It uses seven motors-two in the joints in each,of its three fingers and one that rotates two of the fingers so they either face the third finger to grip or parallel it to form a hook.

Schunk worked with Harmonic Drive LLC to build the tiny brushless motors and gears. The company's harmonic drives produce reduction ratios of 50:1 to 320:1 in very little space. They also put out a lot of torque.

"The gripper produces force similar to a human finger," Hayes said. "The guy who heads the project is big-six-feet-six tall-and when we were developing the hand, we tested it against the joints of his fingers."

The hand itself is encased in a nickel-plated, all-aluminum, dustproof, and water-resistant exoskeleton. It is designed to be maintenance-free. It contains everything needed to run it-motors, sensors, gearbox, controllers except the power. The entire three-fingered package weighs about 5.5 pounds.

In many ways, Schunk's design builds on many of the innovations in the original three-fingered robotic hand, the BHS, which was developed by Banett Technology Inc. of Cambridge, Mass., in the 1990s. It was also an electric hand, so it didn't require the tubing that makes pneumatically or hydraulically actuated robotic hands more complicated as their degrees of freedom increase.

Why three fingers? "We asked, 'What is the minimum number of joints and motors and actions we need to get the job done?' " Barrett's president and CEO, Bill Townsend, said. "We tried not to use human motion as a model. We got that idea from research at the University of Pennsylvania. We use two thumbs, and they rotate 180 degrees around the palm, much more than human thumbs. They can form a hook when they're on the same side, or spread out for a power grip."

Townsend also made a number of engineering compromises to keep the BH8 affordable. Well, relatively affordable. A hand costs about \$30,000, and at least some of the 50 or so hands he sells annually go into factories.

Engineering compromises start with actuation. Unlike human fingers, which have three joints, the BH8's fingers have two. One motor controls both joints through a device that Townsend calls a torque switch. The switch drives the extensions above both joints to curl forward at the same time. When one meets resistance, it stops, but the other keeps going until it, too, meets resistance.

A fourth motor rotates the two thumbs. This gives the hand eight degrees of freedom with only four motors. " It cuts the complexity in half, and that helps us keep the price down," Townsend said.

Barrett's fingers have an interesting feature. Their motors use a pair of worm drives to make sure that when the inner part of the finger (closest to the palm) contacts an object, the outer part continues to move until it has also come in contact with the object and then will stop. That allows the finger to curl over it. Barrett uses a cable running on the outside of the fingertip to reverse the process.

The reason for the cable has to do with the punishment robotic arms dish out to grippers and hands. Sure, robotsare very precise, but only after they have been optimized. During progranming, they regularly smash end effectors into parts, walls, I-beams, and sometimes people.

The impacts are harder than they look. "You see a robot moving, and the arm's the same size as a baseball bat, and you have a feeling that the impact's not going to be so bad," Townsend said. In reality, a robot's motor is running 100 times faster than the gearhead. When the arm crashes into an unmovable object, the only place that inertia can go is into the end effector.

"It's a real challenge," Townsend said. "A hand is as intricate as a watch. Imagine asking someone to design an intricate watch, and, oh, by the way, it has to withstand 1 ,000 pounds of pressure."

Townsend assumes his hands will get smashed. His solution is to build them like the motorized model airplanes he used to fly on the end of a tether when he was young. "They were held together with rubber bands. After they would crash, we'd find the wing here and the fuselage over there. So we'd just put the parts back together."

That's the function of the cables that run along the top of the fingers. They are sacrificial. When the hand crashes, the fingers snap off. Engineers then reattach them.

A hand is as intricate as a watch. Imagine asking someone to design an intricate watch, and, oh, by the way, it has to withstand 1,000 pounds of pressure.

## Not Moore's Law

Townsend has a tongue-in-cheek theory about robotic hands that he calls Archimedes' Law, after the Greek who elucidated many fundamental principles of mechanics. He contrasts it to Moore's Law, which states that micro processor power doubles every year or two. Graph Moore's law and it looks ready for parabolic takeoff.

"When it comes to the development of mechanical hands, Archimedes' Law has only a few data points ," Townsend said. "There's what Archimedes knew, the use of cables added by da Vinci, and what we know today. It forms a very shallow curve, so don't expect a lot of change."

Townsend is the first to admit that Archimedes' Law is not to be taken too seriously, but it does make an important point. Mechanical. design moves forward slowly, usually driven by integrating electronics and materials into slowly evolving mechanical devices. While the field jumps on advances like harmonic drives, even Archimedes would recognize the principles behind them.

Townsend's views serve as a counterbalance for Pollitt's optimism at Shadow Robot. Pollitt is betting that the world will move beyond industrial robots and embrace service robots. These mobile automatons will help in jobs and homes, or handle tasks that are too difficult or dangerous for people. Ultimately, he expects robots to shadow their owners like valets, providing services on demand.

Pollitt believes the industry is where computers were a few decades ago. Robots can do some limited tasks weldajoint, vacuum a room (2.5 million Roomba robot vacuum cleaners have been sold), remove a bomb-but nothing that requires true generalized intelligence. That will come with advances in such critical technologies as computing power, battery power, control software, vision recognition systems, and actuation.

"Like the IT industry, when the right applications are found, we'll see a big ramp-up," he said. "I think personal home robotics is only 10 to 15 years away."

Some of those applications are already apparent. One of the most widely discussed is personal care. Robots could nurse people who are disabled or chronically ill. For example, they could monitor someone with Alzheimer's disease, making sure the patient took medication and did not wander off. Robots could help people who fall regain their feet, and eventually cook or clean up as well.

That will require hands as well as artificial intelligence.

According to Rich Walker at Shadow Robot, "Hands will let robots manipulate the world around them."

Walker sees a broad range of applications for robotic hands, even before anyone develops the robotic intelligence needed for autonomous applications. With remote hands, surgeons could operate on a cruise ship or a battlefield from halfway across the world. Technicians could defuse bombs without risking their lives. Laboratories could manipulate and then dispose of hazardous materials.

A mobile hand might even make better sense than a prosthesis for the handicapped. "A prosthesis is no good unless you're already able-bodied," Walker said. "If you're in a wheelchair, the quality of your prosthetic hands does not solve your problems if you can't reach things. If you put a robotic arm and hand in that environment, you'd have a hand where you need it and that would make your life better."

Whether it is a service robot or mobile hand, the engineering challenge is creating something that mimics the ability of human hands well enough to do human tasks. "A hand can hold 5 to 10 kilograms in its grip, yet play a sonata on a piano," Walker said. "You can build hands that are that strong, but they are usually big. Or you can make them more fluid, but the strength is not there. We've tried to combine both."

Like Schunk, Shadow Robot worked from a human model. "The two guys who did most of the design worked at a bench opposite each other," Walker said. "Whenever they needed a measurement, one would take out the calipers and measure the movements and distances on the other. They were looking at how hands bent and how the relationship between their parts changed. Then they had to ask: Where can you put the axles and bearings to do that in a mechanism?"

The hand's innovations mimic human capabilities. Humans sense hand location and shape based on the tension of their muscles and the stretching of their skin. Shadow tries to mimic this by using magnetic sensors. "We measure the rotation of the magnet," Walker said.

Rather than try to mimic the complex motion of the thumb, the company opted for a design with two relatively large movement joints that mimic most, but not all, thumb movements.

Tendons, which connect muscles to bone in human hands, proved a stumbling block in Shadow's mechanical system. According to Walker, "The ideal tendon needs to be rigid when pulled, bend around a turn, and you have to be able to tie it." Tie it? It turns out that fasteners take up too much space, and they tend to rip up tendons unless they are thick. Tying, Walker said, is simpler and makes the tendons easier to adjust. "Besides, there's a huge body of naval literature on what knots are good for what tasks," he added.

Driving the hand is a device called the air muscle. It looks like a long rubber balloon with a criss- crossed mesh wrapped around it. When the balloon expands, it pushes the mesh outward. This causes the mesh to shorten. When the balloon contracts, the mesh expands. Walker said that finger-size air muscles can lift a few pounds. Shadow also makes industrial versions up to 3 yards long that can lift hundreds of pounds at a time.

Surgeons in hospitals may use robotic humanoid hands to perform delicate operations in remote locations.

## Mending Humans

Surprisingly, hands developed as human prostheses are generally less capable than those under development for service robots. This is more a matter of control than mechanics. Prosthetic hands are usually controlled by attaching myoelectric sensors to the forearm. The sensors pick up muscle contractions that would ordinarily guide the hand's motions.

In the past, mechanical hands used these signals to determine whether to open or close. The United Kingdom's Touch Bionics uses the same signals to manage a much more complex prosthetic hand, the i-LIMB. The iLIMB combines a rotating thumb with an independently operated index finger. The remaining three fingers are independently powered, but operate as a single group.

This enables the hand to form a number of configurations. The thumb can close down onto the side of the index finger to hold a key or credit card. The hand can surround an object, even something as fragile as a glass, and the fingers will automatically stall when they meet resistance. The company's Web site shows users picking up a flat CD from a table, lighting a match, lifting a briefcase, and even assisting in tying a tie.

The i-LIMB can lift 22 pounds, compared with about 130 pounds for a human hand. "A human hand doesn't need to grip with all that power because it takes advantage of friction generated by the skin," according to Hugh Gill, Touch Bionics' director of technology and operations. "We recently acquired a company that specializes in making high-friction artificial skin. That makes it more efficient to pick things up. Our patients can use our hands to drive cars and play golf."

The system uses many of the advances found in robotic hands, such as direct current brushless motors. "A control board delivers high-frequency pulse modulation to slow them down or speed them up, and the patient can control the amount of signal delivered," Walker said. "You can use a small force to pick up something fragile, or go faster for something heavier."

Instead of a harmonic drive, the hand uses a combination of planetary drives that achieve similar reductions in small spaces, combined with worm gears and cables, to control the fingers . The fingers themselves are modular and quickly replaced if needed.

In fact, while the five-fingered i-LIMB and the three fingered BH8 appear very different from one anothe rand from hands developed by Schunk and Shadow they share some core technologies. Slowly, engineers are defining the new mechanical hand.

These hands may not approach the complexity or functionality of nature's handiwork. Yet, increasingly, they are good enough to get the job done, whether it is picking parts out of a bin, performing remote surgery, or just reaching for a key to open a door.

One day, we may yet live in a world surrounded by service robots and responsive prosthetics. Mechanical hands will make it possible.