Chapter 4
AI and Robotics

In 1958 Joseph Engelberger created the first robots—called Unimates—for factory work. They looked nothing like the fictional robots in movies or books; they looked more like giant flexing arms. The robots were hydraulically powered and programmed to move in a repeated pattern with exacting precision. Automobile manufacturers soon replaced assembly line workers with Unimates, which could perform the most exacting and physically demanding tasks.

A person who picks up a hammer and pounds in a nail does not think about the distance between his hand and the hammer or what force is needed. That mathematical and geometric calculation is immediate and unconscious. But that process must be hard-wired into industrial robot computers so that mechanical movements are precise.

These giant arms are dangerous machines that swivel, swing, hammer, and weld with great force as partially assembled cars move past them. The welding arm must create strong, perfect seams to connect large metal panels on the assembly line. First it must analyze the alignment with its laser vision system. If it detects a misalignment, the robotic arm adjusts the pieces so that they line up correctly. Then it welds them to create the strongest bond possible. This kind of repetitive and precise work is perfectly suited for an AI machine that does not cough, sneeze, blink, or get tired and make a mistake during welding.

Today robotic arms are used in almost every type of industry. They are especially useful in places where humans risk injury to themselves or their product. Smaller arms, for example, are used in the sterile environment of a silicon chip factory, where human workers wear special head-to-toe coveralls so that dandruff, flaking skin cells, and dirt do not contaminate the product. Industrial robotic arms work safely with toxic materials and perform their tasks in nuclear reactors and other dangerous places humans cannot go.

But these AI arms are not versatile. Hand a robotic welding arm a screwdriver and it is useless. Unless reprogrammed, the industrial arm cannot adapt to another tool or task, but that versatility may come in the future. In the 1960s AI experts promised a robotic maid in every home. That has not happened, but AI experts are optimistic that it will. "A hundred years ago, we couldn't even fly, and now we have spacecraft exploring our solar system and beyond," says Tucker Balch, a computer scientist at Georgia Institute of Technology. "I think we're going to see just as radical a change in robots in the next century."¹² Already robotics and AI researchers are moving beyond the robotic arm to create intelligence in moving machines of all shapes and sizes. So far, some of the most successful and intriguing designs mimic nature.

Mechanical Animals on the Move

Using living creatures as inspiration for machines is called biomimicry. One of the leading scientists in the field is Robert J. Full of the University of California at Berkeley. He has never designed or built a robot, but he provides robot researchers with blueprints that come from the animal world. These blueprints may later be incorporated into various robot prototypes. Full gets his ideas from ghost crabs, ants, cockroaches, centipedes, and geckos.

As a biologist, Full studies the principles of animal locomotion. While cockroaches run on miniature treadmills, Full measures the electrical impulses in the insects' muscles with tiny electrodes and videotapes their movements with high-speed cameras. He details the locomotion movement by movement and documents the forces and dynamics involved. The information that he uncovers is then used to create some of the most mobile robots in the world that will be used for both civilian and military purposes. His work with the gecko, a sticky-toed lizard that can scurry up vertical surfaces, has led the company iRobot to create the Mecho-gecko, a small sensor-laden device that can cling to a wall and walk across the ceiling.

And for a robot, it seems that the more legs it has, the better. Full's analysis of cockroach movements has been used by researchers at Stanford University to build insectoid robots that scurry along the ground. Like the real insect, the insectoid robot moves along using two alternating sets of legs (two legs on one side; one leg on the other) as springs. And a crablike robot called Ariel wades sideways into the water and walks along the bottom of a pond.

In a giant tank of water at MIT, a robotic fish swims freely. Its creator believes it can be used to detect contamination in water reservoirs. If its sensors detect chemicals that should not be there, the robot can surface and transmit an alarm signal to the control center. Knowing that a fish with its slim muscles and small fins can accelerate at a rate of eight to twelve g's—as fast as a rocket—scientists hope to make better underwater propulsion systems too.

NASA is working on a robot shaped like a snake that will be able to slither into crevices, crawl into holes, and climb over the most rugged terrain without toppling like many wheeled vehicles. The snakebot, first developed by Mark Yim at Xerox Palo Alto Research Center, is made up of thirty hinged modules linked together in a chain. A central computer is located in the snakebot's head module, and smaller sensors and computers operate in each module throughout the body. The snakebot can sidewind, slither, and crawl like an inchworm. It can also coil up to grasp a tool or flip over obstacles. NASA hopes this style of robot will help on future missions to Mars. Added sensors and adaptive learning software will make it one of the more self-sufficient robots to go into space.

In Japan, scientists at Nagoya University have moved robots up the evolutionary ladder to create a monkeylike robot. Shaped like a gangly gibbon, this robot can hang and swing from bars suspended from

the ceiling. Video cameras track its movement, and every time it makes a mistake, it has to learn how to correct it and try again. Called Brachiator III (brachiate means "to swing"), it is almost as agile and quick as a real monkey.

Motor Intelligence

Each complex movement that a robot makes to mimic the swish of a fin, the swing of an arm, or a slither of a snake involves some form of AI. These movements can be made only with the right kind of programming. Joseph Ayers of Northeastern University says, "The same basic organizational units of the nervous system are involved in the motor systems of all animals. If you know how to use that architecture, you should be able to build a robot to operate in any environment on the planet."¹³ Understanding the motor intelligence of animals helps researchers build more sophisticated robots.

Many robotic animals operate with low-level behaviors. When such a robot perceives a situation through its sensors, it responds with a fixed set of behaviors. For example, when a robot encounters a rock, it is programmed to crawl over it. If the rock is large

and the robot tries to crawl over it as usual, the robot might topple over instead. With the help of motor intelligence, AI robotic experts are layering on adaptive learning behaviors so that the responses are no longer fixed. With higher level behaviors, a robot would sense the large rock and respond based on a repertoire in its memory. If it attempted to crawl around the rock and was successful, that connection between the sensors and the response would be strengthened. The next time the robot encountered a similar-sized object, it would know to crawl around it. AI software is the connection between what is perceived by the robot and the robot's response.

Robot Explorers

Why mimic nature? Researchers hope these designs will be able to go where humans cannot go and do what humans cannot do. Robots do not need oxygen to breathe; they do not get claustrophobic or feel pain. Crab robots can blend into the seabed and patrol underwater for hours. The military hopes they can be used for detecting underwater mines. Insectoid robots can crawl into the tiniest spaces, like into small pipes or inside a wall, and snakebots can slither into cracks and crevices. But not all robotic explorers look like animals. Some resemble tiny tanks or golf carts.

In 1994 an eight-legged robot was sent clambering into the steaming caldera of Mount Spurr in Alaska. Called Dante II, the robot explorer took air samples and allowed scientists to view, for the first time, the active volcano via a remote-controlled video camera. A similar Dante robot made by Red Whittaker at Carnegie Mellon Institute also hunted for meteorites in frigid Antarctica.

These machines also protect human workers from toxic levels of radiation. When the Chernobyl nuclear power plant blew up in April 1986, it released four hundred times more radioactive waste than the U.S. bombing of Hiroshima. The radioactive building and all of its contents has been sealed off in a cement sarcophagus ever since. In 1996 a robot named Pioneer was sent in to map the interior, collect samples, and inspect for leaks. Less than four feet tall, Pioneer was narrow enough to fit through doorways, strong enough to bulldoze through piles of debris, yet agile enough to pick up samples of radioactive waste and water.

A shoebox-sized robot with tanklike treads helped archaeologists explore previously unknown parts of the Great Pyramid at Giza in Egypt. It climbed a tiny airshaft of the forty-five-hundred-year-old pyramid and inched along more than two hundred feet before coming to a door made of plaster. When the robot poked its camera though the plaster, it discovered…another door!

Into tiny airshafts, hot lava, or freezing ice, robots are capable of going anywhere, even into outer space. On January 3, 2004, the first of two NASA rovers landed safely on Mars and began to explore the surface of the Red Planet. These two rovers, Spirit and Opportunity, are identical robotic vehicles about the size of golf carts. Riding on six wheels, the rovers' sturdy bodies contain all they need to survive for three months. Their computer systems monitor their health to make sure the insides do not get overheated during the hot Mars day and do not freeze during the cold nights. Each is equipped with X-ray, infrared spectrometry, microscopes, and other geological instruments to collect soil and rock samples and test for evidence of water. Spirit and Opportunity are not the first robots on Mars. The smaller Sojourner explored Mars in 1997. It was the first time an intelligent robot was able to react to unplanned events on the surface of another planet. Sojourner had been on the planet only a few days when its hazard avoidance system switched on and it had to make its own decisions about where to go without any internal mapping information.

Search-and-Rescue Robots

When the twin towers of the World Trade Center were destroyed on September 11, 2001, robotics expert Robin Murphy and three colleagues from the University of South Florida drove eighteen hours to New York City to help search through the rubble for survivors. They brought eight different search-and-rescue

robots with them. The most successful was a miniature tank with treads that crawled into voids thirty feet deep. Although the robots did not discover any survivors, they did prove their usefulness. They could get into spaces that men and rescue dogs could not, and the thick dust in the air did not hamper their olfactory sensors, as it did some of the rescue dogs.

Since the terrorist attack, robotics experts and rescue workers from the Federal Emergency Management Agency (FEMA) have collaborated on designing better search-and-rescue robots. Recently FEMA crews blew up an empty building in order to test different models of robots and learn what works and what does not. Even small adjustments, like making the joystick controls large enough to manipulate with gloved hands, made a difference in the robot's effectiveness. "Until you experience a rescue operation, it's hard to understand the real rescue questions that must be addressed for this technology to be transferred to search-and-rescue users,"¹⁴ says Robin Murphy, the director of the Center for Robot-Assisted Search and Rescue at the University of South Florida. These robots are equipped with infrared sensors that detect body heat, night vision, and cameras that search for colors distinctive from the gray dust that blankets a site. Any speck of color, such as that of blood or fabric, is easily identified.

Another important part of rescue missions and worker safety is knowing secure routes to take. In 2002 nine miners became stranded in a flooded mine in Somerset, Pennsylvania, by following a flawed map that did not show how dangerously close they were to a flooded mine shaft. In response to this crisis a new kind of robot was created to assist miners in mapping out safe routes in real time. The mapmaker is a small robot called Groundhog. Scientists at Carnegie Mellon equipped Groundhog with cameras; gas, tilt, and sink-age sensors; laser scanners; and a gyroscope. The robot can map out mine shafts as it moves through them using sensor data processed through AI software called Simultaneous Localization and Mapping, or SLAM. Groundhog successfully mapped a thirty-five-hundred-foot Pennsylvania mine shaft in May 2003.

Warbots

Many search-and-rescue robots and animal robots are products of research and development programs sponsored in part by the U.S. military. The Defense Advanced Research Projects Agency (DARPA) hopes that in the future whole schools of robotic fish will patrol shipping channels looking for sunken mines, and a patrol of Mecho-geckos equipped with cameras and audio equipment can be sent up the side of a skyscraper to peek into windows and assess the situation inside. According to military robotics pioneer Scott D. Myers, "Military robots are being developed and fielded to do three things: perform the dull, the dirty and the dangerous."¹⁵

With an increased focus on counteracting terrorist attacks, the Pentagon is paying closer attention to machines that perform dangerous missions and keep people out of harm's way. In 2002 U.S. soldiers in Afghanistan used robots to explore caves before the troops were sent in. The Packbot, which looks like a miniature flattened tank, runs on treads and has flipperlike arms that allow it to crawl over anything. It is sturdy enough to be thrown through a window, climb up stairs, and fall from the balcony and still be fully operational. During Operation Iraqi Freedom in 2003, a squad of Packbots searched tunnels under the Baghdad airport to look for enemy soldiers thought to be hiding out and examined equipment left on an airfield that was believed to be booby-trapped.

A state-of-the-art unmanned reconnaissance airplane called Predator can cruise at twenty-five thousand feet for more than five hundred miles and can stay aloft for up to twenty minutes. Controlled by a pilot on the ground using a joystick, the small Predator can relay back information about the location of enemy troops and weather conditions, leaving troops safe on the ground.

According to Colonel Tommy Dillard, Airforce Battlelab commander, "One of the things we have asked [the robotics] industry to do is to be able not only to detect [intruders] with robots, but to start a neutralization phase before we get response forces out there."¹⁶ Robotics experts have created robots that can detonate suspect packages in an airport or on a crowded street in Baghdad. Small robotic guard units equipped with a range finder, motion detectors, a communication system, and weapon can locate and track an intruder, call for backup, and even fire a gun.

The Race for Unmanned Vehicles

In 2001 the U.S. Congress mandated that one-third of military ground vehicles be unmanned by the year 2015. That means they should be able to navigate, steer, and respond to various situations without the help of a driver. The parts needed for such a vehicle are advanced laser, radar, and sonar sensor systems to help the vehicle navigate, and Global Positioning System (GPS) to plot its location. So far all the components have not been assembled to create a workable machine that can maneuver across rocky terrain on its own. One prototype, called Navlab II, is a U.S. Army truck driven by a computer called ALVINN (Autonomous Land Vehicle in a Neural Network). Its computer steering algorithms allow it to drive without a driver, and basic pattern recognition systems allow the vehicle to recognize and follow the lines of a road. But an off-road situation is completely different. Dodging a boulder in the path would be fairly easy for Navlab II, but having the vehicle determine whether the obstacle was a boulder or a tumbleweed is more difficult.

To help the military achieve a truly autonomous vehicle, DARPA organized a contest to build the world's first truly autonomous robotic land vehicle. Teams of professional robotics engineers, individual garage

tinkerers, and one high school science club vied for the $1 million prize. The crew from Palos Verdes High School called themselves the Road Warriors, and they reconfigured a sports utility vehicle donated by Honda. One lone graduate student tried to make a motorcycle that could get around roadblocks, zip down tight alleyways, and be parachuted into a city. A team of college graduate students worked on a giant Humvee. To win the million-dollar prize, the vehicle had to be able to cross a specific 210-mile course through the desert from Los Angeles to Las Vegas within ten hours. The army's Future Combat Systems program was hoping that the robotics contest would produce a winner, but so far no vehicle has finished the course. The contest will be held again in 2005.

However, large corporations like Lockheed and United Defense Industries may cross the finish line first. United Defense Industries is working on an armed robotic vehicle with missiles and gun turrets to provide targeting information for other weapons and to drop miniature sensors onto the battlefield. Another vehicle still in the research stage is called the Soldier Unmanned Ground Vehicle (SUGV). Operated by remote control, it will be small enough to climb stairs and use its sensor devices to see around corners during urban combat. The SUGV is also being fitted with a grenade launcher, directional microphones, and motion detectors, which would allow it to stand sentry while soldiers sleep.

All of this AI robotic technology is meant to keep soldiers out of harm's way. Many people suspect that the idea of a platoon of robotic soldiers, reminiscent of a Star Wars movie, is the inevitable next step.