Photo Courtesy of NASA
Robonaut’s grip on a rubber ball demonstrates that its humanoid hand could manipulate tools in deep space to make repairs on the International Space Station.

Issue: Summer 2003

Robotica

Watch your step. On any given day scurrying underfoot, hovering overhead and otherwise gadding about are a menagerie of robotic devices spawned by USC’s School of Engineering – in case you didn’t know, a major breeding ground for the artificially intelligent.

By Mark Ewing, Eric Mankin, Bob Calverley

Thirty years ago, Woody Allen gave us a hilarious vision of robots cooking, cleaning and even sexually gratifying their owners. Almost 20 years ago, James Cameron gave us a terrifying vision of robots that all but exterminate the human race. And two years ago, Steven Spielberg gave us a heart-rending vision of robots that can love eternally. Nothing resembling the robots of Sleeper, The Terminator or A.I. has arrived yet, but hidden in simple, unthreatening devices like washing machines, coffeemakers and microwaves, robotics is already a significant presence in our daily lives. And toys like Sony’s Aibo dog are bringing it out in the open, helping us adjust to the idea of robots scurrying in our midst.

Such acceptance is important, says computer scientist Maja Mataric, because the next quarter century could witness an unprecedented boom of robotic devices, not just in industry but in homes and offices. Mataric is co-director of the USC School of Engineering’s Robotics Research Lab and the director of its new Center for Robotics and Embedded Systems. The largest multidisciplinary robotics effort in Southern California, CRES has the pragmatic goal of bringing robotics out of the lab and into society.

When robots do enter the mainstream, don’t expect them to be sophisticated stand-alone systems. The promise of the early ’70s never materialized, in part because the focus was on building a single complex robot. Hollywood only reinforced this assumption. Recent research emphasis, however, has been on smaller units that can communicate with each other. An example of this is the wildly successful RoboCup competition, a worldwide academic contest that pits teams of robots in an enclosed-arena version of soccer.

These soccer bots are not to be confused with the angry and aggressive (and misnamed) remote-controlled BattleBots one sees on Comedy Central; those creations are not robotic, as a human hand is at the tiller controlling their actions.

So what exactly is a robot?

Mataric defines it as “an autonomous system that exists in the physical world, that can sense its environment, and can act on it to achieve defined goals.” The definition is broad enough to encompass microscopic modules scaled down to the size of bacteria.

Software gives a robot its ability to act autonomously, without constant human intervention and guidance. Little wonder, then, that today’s roboticists are mostly computer scientists.

“Robot brains use artificial intelligence, which is driven by software engineering,” says Mataric, an expert in AI control, cooperation, learning and cognitive neuroscience modeling.

As companies like Honda and Sony move toward wider production and commercial distribution of robots for consumers, who will create their software programming?

USC engineering graduates are poised to perform much of this work, according to Gerard Medioni MS ’80, PhD ’83, chair of the Department of Computer Science. “We have a long and productive track record in the area of robotics,” he says. With two institutes, CRES and the Institute for Robotics and Intelligent Systems (the latter directed by Ram Nevatia, an expert in machine vision), the USC School of Engineering figures prominently in the field. “Both in terms of size, research grants and reputation, USC is second only to Carnegie-Mellon,” Medioni says.

Don’t just take his word for it, though. Here are nine of USC’s most promising robotics projects, which could someday lead not just to toys, but to robotic servants and co-workers.

Deep Space Construction Workers

Fact: For every hour of work in space, a human astronaut needs 48 hours of preparation and recovery. This alone makes robots very attractive candidates for an extraterrestrial workforce.

Enter Robonaut, a robot being jointly developed by NASA’s Johnson Space Center and the USC Robotics Research Lab. NASA hopes Robonaut will one day be a space-walking construction worker able to perform dangerous, intricate maintenance tasks on the International Space Station. For now, the multimillion-dollar robot is a highly complex humanoid torso residing at NASA.

A USC team of researchers with access to Robonaut is now getting this prototype to “learn” from a human teacher. The instructor wears a “sense-suit,” wired with sensors that transmit movements to Robonaut. The space robot memorizes these movements just as a golfer might improve her swing when a golf pro lays his hands over her grip, demonstrating the proper technique. The human teacher will show Robonaut a task – such as opening a tool box, lifting a tool and using it to perform a repair. Ideally, Robonaut will be able to repeat that task even under different and unpredictable conditions, for example, in weightless space.

This kind of teaching is a particular strength of USC’s robotics program.

“When we teach robots to perform human-like tasks, they usually learn from some kind of human demonstration that the robot then mimics,” says Mataric.

Imitation is a fundamental learning process for humans, but it is rarely found in other species. Though it is often confused with simpler forms of mimicry displayed by various animals, in its true form, which involves learning a new skill purely by observing a teacher, imitation is found only in chimps, dolphins and humans.

Research into imitation at the Interaction Lab (a subgroup within the USC Robotics Lab) has two goals: to enable robots to learn new skills through imitation and use imitation to interact with humans in a natural, intuitive way; and to give human researchers a better understanding of this complex phenomenon in nature by modeling it on robots.

Mataric’s group at the Robotics Research Lab is developing a model of imitation inspired by neuroscience evidence of so-called “mirror neurons,” which allow animals to interpret behaviors they see in the way they would perform them.

The work is also inspired by neuroscience evidence for so-called “movement primitives,” which allow animals to produce sophisticated, flexible movements from a small vocabulary of primitive movement building blocks. The researchers have combined these ideas into an imitation model now being tested on a variety of robots, from the Sony dog (which is learning to play with humans, entertain and fetch) to the Robonaut.

Learning by Imitation

Monica Nicolescu’s robots look like vacuum cleaners crossed with coffeemakers, with a little bit of camcorder thrown in for good measure. A doctoral student who works in the USC Interaction Lab, Nicolescu MS ’99 has been teaching these appliance-like robots, called Pioneers, to recognize colors, distances and angles, and to detect a target, like a ball or box. First she leads a robot around an obstacle course. As the Pioneer follows her – like a duckling imprinting on its mother – it learns to navigate the course.

“My software does the mapping between what the robot senses and its own set of basic skills,” she explains.

Next it’s the Pioneer’s turn to play teacher. When the robot has learned the obstacle course, it can teach another robot by leading it around the course. “I’ve had as many as seven robots going around the course,” she says. “Each has learned from the robot in front of it, except for the lead robot, which learned the course from me.”

In one startling demonstration, Nicolescu teaches a Pioneer to find a small box, pick it up and place it somewhere else. The next time, she complicates the task by wedging the target box among larger boxes – visible but inaccessible to the robot. Then she leaves the lab. The Pioneer locates the box but can’t reach it. After a few futile attempts, it scans the room. Spotting a human, the robot rolls over to him, tries and succeeds in getting his attention and leads him to the inaccessible box. The robot is asking for help in fulfilling its task, and showing a rudimentary ability to reason out a solution to a problem.

“The model we use to teach robots is biology,” says Mataric. “This is the kind of thing your dog might do if you forgot to feed him.”

Scout Helicopters

On an athletic field minutes from his lab, assistant professor of computer science Gaurav Sukhatme MS ’93, PhD ’97 shows off the hovering capabilities of one of his robotic helicopters. If any human chopper pilots were around, their jaws would drop. Hovering is one of the most difficult skills for a helicopter pilot to master. That a machine can do it without human guidance or intervention is remarkable.

In research supported by a NASA/ JPL division specializing in aerial vehicles for planetary exploration, Sukhatme not only taught his helicopters – sturdy prototypes equipped with 6-foot rotors – to hover, but also to use their on-board cameras to locate, and then settle on, a landing pad.

With graduate students Srikanth Saripalli MS ’02 and Stefan Hrabar, Sukhatme is currently teaching the robo-chopper to interact with the lab’s ground-based Pioneer robots and to chase after moving targets and descend on them.

An expert in accurate measurement and benchmarking of mobile robot performance, Sukhatme directs the Robotic Embedded Systems Laboratory, which focuses on flying robots, distributed robotics and large-scale robotic sensor networks.


Gaurav Sukhatme with one of the robotic helicopters used in his research.
Photo by Irene fertik

Together with graduate student David Naffin ’91, Sukhatme is now working on two miniature helicopters small enough to fly indoors. He hopes to develop a robot helicopter so light and smart that a soldier could carry it in a backpack and launch it to scout dangerous terrain. This work, and the aerial-ground robot cooperation research, is funded by the Defense Advanced Research Projects Agency, which is developing technology for future combat missions.

Sukhatme’s robotic helicopters have civilian applications too. It isn’t hard to imagine future autonomous choppers patrolling Los Angeles freeways, reporting traffic conditions, carrying out hazard mitigation in case of an accident and even assisting in urban search-and-rescue operations in the aftermath of an earthquake or some other emergency.

Attack of the Robosapiens

The robotic head in Stefan Schaal’s lab has all the charm of a novelty store Halloween mask. But don’t be fooled. MAVARIC (as it’s called) has two cameras in each eye, one for close vision and one for distance. It can also hear and has a system for maintaining balance, duplicating the function of the vestibular apparatus in the human inner ear. Created by Utah-based robotics R&D firm Sarcos Design, MAVARIC is currently getting a sensory workout at USC in a project led by research assistant professor Sethu Vijayakumar.


Monica Nicolescu with a few of her educated Pioneers.
Photo by Michele A.H. Smith.

“People don’t just move. They perceive, react, plan a precise movement, move and stay balanced while moving,” says Schaal, who heads USC’s Computational Learning and Motor Control Laboratory, where MAVARIC is being fine-tuned. The lab concentrates on motor control and learning in humanoid robots.

Like other USC roboticists, Schaal draws inspiration from biology, particularly the primate brain. A neuroscientist as well as a computer scientist, he employs principles of neuroscience, not just traditional artificial intelligence, to create a robot that behaves like a human “simply because biological entities perform so much better than any available artificial intelligence algorithms,” he says. If all goes as planned, a version of MAVARIC will someday sit atop a new humanoid robot’s body.

For now, the research team is trying to understand and duplicate the behaviors of the ocular motor system and gain insight into how eye movements guide body movements. Building humanoid robots will ultimately require a complete understanding of human vision, Schaal believes.

Another project in Schaal’s lab centers around a life-sized, hydraulically powered robot arm. “Despite its slightly clunky appearance when it’s turned off,” says Schaal, “this robot arm is beautiful when it moves, with speed and strength that comes close to human performance.”

Schaal’s research group uses this masterpiece of engineering to study learning algorithms, imitation learning and virtual environments with force feedback to a human operator. Students in Schaal’s robotics course all have to implement a project on this robot and operate it themselves.

Schall also directs a group doing motor-control research on one of the few full-body humanoid robots in the world, located in Japan at Advanced Telecommunications Research International – that country’s equivalent of Bell Labs. “We are working on a revolutionary new humanoid robot that will have movement abilities close to human performance, and still be autonomous – unlike all the current humanoids that are rather slow, conventionally engineered machines,” he says.

“With a little luck in funding, one of these machines will soon walk around at USC.”

Robotic Home Builders

Meanwhile, back at home, here’s an idea that will give procrastinating contractors pause: “The ultimate goal for our project is to completely construct a one- or two- story, medium-size home in one day without using human hands,” says industrial and systems engineer Behrokh Khoshnevis.


Stefan Schaal with MAVARIC, a robotic head capable of vision, hearing and balance.
Schaal Photo by Michele A.H. Smith

The process, called Contour Crafting, has been under development for seven years, ever since Khoshnevis found inspiration while smoothing plaster on his house.

Contour Crafting builds up shapes in layers – using two movable, programmable trowel-like tools deployed around a nozzle. It is completely automatic (hence robotic), guided by computer controls.

The system can form plaster, concrete, adobe, plastic or even wood particles mixed with epoxy into a paste. Contour Crafting working prototypes can currently build three-dimensional items in any desired shape.

The system is a derivative of a construction technique called “rapid prototyping,” which uses a computer-controlled head to build up, selectively solidify or selectively remove layers of material. But those materials are specialized, expensive and not durable – and despite its name, rapid prototyping is slow, and therefore rarely used to manufacture products. Rather, it’s used in creating prototypes for die-casting, injection molding and other mass-production manufacturing processes.

Contour Crafting, however, can mold a wide variety of ordinary materials quickly. The double-trowels-around-
a-nozzle design allows much thicker layers, better control and, most important, the fabrication of much larger structures than any existing rapid prototyping system. Khoshnevis believes that an upsized version of his current Contour Crafting system could erect entire buildings layer by layer. He recently received an NSF grant to build a full-scale section of a house, which will be put under extensive civil engineering tests for conformity with building code requirements.

The technology’s future applications might include road construction work or even building structures on the moon, says Khoshnevis.

Warrior Softbots

Imagine a squadron of Marine fighter jets whose operations – maintenance, flight assignments, scheduling – are handled by what roboticists call softbots: software programs that interact autonomously to guide human organizations through complex sequences of difficult tasks.

Autonomous Negotiating Teamware software does exactly that. Created by scientists at USC’s Information Sciences Institute and Vanderbilt University’s Institute for Software Integrated Systems, ANT performs in minutes scheduling functions that normally require many hours of human labor.

“Creating schedules for a squadron involves balancing a huge number of factors,” explains Robert Neches, director of ISI’s distributed scalable systems division. “Pilots want to get the maximum number of flying hours to maintain their ratings and extend their skills. The airspace has to be clear, suitable and acceptably safe for intended operations. Policies and commitments from higher command have to be satisfied. And, of course, the weather changes.”


Behrokh Khoshnevis with a small model fabricated by his robotic home-building system.
Photo by Peter Menzel

The first operational schedule produced by ANT was accepted for use onboard a carrier 18 months ago.

The “teamware” consists of individual software modules, each representing a different interest or goal involved in managing a combat air squadron. The modules communicate with one another, sharing information, overruling or yielding according to a set of predetermined priorities. These structured exchanges of requests and counterproposals lead to agreements that become elements of a schedule.

ANT was extensively tested by Marine Air Group 13 both at its Yuma, Ariz., base and onboard carriers.

A novel feature of ANT is its ability to simultaneously balance tricky maintenance requirements against operational demands, while considering risks involved with each decision. ANT considers resource constraints, such as how many mechanics are available; and factors in risks, such as the additional stress of performing many complex procedures simultaneously. It can help the commanders and schedulers perceive choke points – places where everything is held up by one factor – and find ways around the difficulty.

All the variables for operations and maintenance and their many interactions add up to thousands of issues that must be settled to make a squadron’s schedule for a single day.

“It takes an experienced operations scheduler as much as six hours per day – and lots of time for a maintenance controller as well – to create daily schedules that balance all these variables,” says retired Marine Corps Col. Russ Currer, former commander of Marine Air Group 13. “This software lets schedulers do the job in four minutes.” (ANT doesn’t schedule operations until alternatives have been reviewed and approved by a human manager.)

Last spring, the system was presented to all general-rank Marine air officers. Col. Mark Savarese, Marine Air Group 13’s current commanding officer, requested operational deployment of the experimental software for all his squadrons, including units going out on operations with Marine Expeditionary Units, after the system scored outstanding marks in repeated evaluations by senior flight officers.


USC-developed teamware assigns softbots to manage Harrier squadrons’ intricate scheduling.

Many non-military planning tasks requiring complex coordination of a large number of variables could use similar software systems – commercial air, trucking and package-delivery operations being obvious examples, Neches says.

Elfin Softbots

Vehicle fleets aren’t the only possible beneficiaries of softbotic scheduling assistance. As people become increasingly reliant on computers, PDAs and cell phones to go about their daily tasks, robotic administrative assistants and executive secretaries are on the horizon for average workers. Milind Tambe of USC’s Information Sciences Institute has developed just such a program. Tambe and members of his research group in the intelligent systems division are testing a new system of computer software “agents,” whimsically called Elves.

The Elves arrange human meetings by consulting among themselves without human intervention. Using global-positioning system equipment carried by the humans, the Elves keep track of the location of each member of Tambe’s group, sending messages warning other participants that their human will be late when the GPS data shows them too far afield to reach a scheduled destination on time.


Diagram of a half-mile-long self-assembling space solar power station designed by roboticists at USC’s Information Sciences Institute.
Diagram courtesy of isiPolymorphic Robotics Lab


Recently, for example, when Tambe went to the airport to pick up a visitor, the GPS function picked up that he was mired in traffic and wouldn’t arrive in time for a meeting. With no human intervention, the Elf signaled its peers and the meeting was postponed.

Elves also sign their humans up for tasks, such as giving speeches, on the basis of who is ready (or should be ready) to present material, and who has gone the longest since a presentation. They can also get their human excused from a presentation. They even know what their humans like to eat and can send out for lunch.

The Defense Advanced Research Projects Agency, which is funding the project, is interested in the possible role of Elves in keeping track of individual soldiers and sailors during massive maneuvers involving thousands of troops, or even on the battlefields of the future.

But that future is now for Tambe’s research team members, who invented and programmed the system and have been using the system to guide their own interactions since June 2000.

“Our Elves have shown that they can handle the professional lives of our nine-person group over an extended period of time,” says Tambe, who is also on USC’s computer science faculty.

The Elves can interface with ordinary PCs, Palmtop computers, GPS systems and cellular phones. “I would expect that, given the potential advantages for organizations, systems like these will be in the marketplace and on the job soon,” Tambe says.

Orbiting Robotic Structures

Also at the USC Information Sciences Institute, two researchers are currently designing self-organizing robots controlled by “hormonal” software. These robots may soon be blazing off into space.

Computer scientist Wei-Min Shen’s project involves getting robotic pieces of the proposed half-mile-long Space Solar Power System satellite to assemble themselves without the help of astronauts.

Shen and research partner Peter Will are testing the hardware and software in the ISI Polymorphic Robotics Laboratory, which Shen directs. The two have developed modular, individual robot units – each with a computer chip programmed with software the researchers term “hormonal.” Such software allows “bifurcation, unification and behavior shifting by the modules,” says Shen, meaning the individual robotic units can link up, detach from one another and assume different behavior depending on how many units are linked together and where each unit is positioned in the assemblage.

The units can combine into larger wholes, or divide into smaller ones. “If a six-unit snake splits in half,” says Shen, describing an earlier prototype, “you get two smaller, three-unit snakes that function as the larger one did.” Separated modules communicate using infrared signals, maneuvering their coupling units into a lock. “Behavior shifting” means that the individual units exhibit different behavior according to their position in the assembly.

Shen and Will’s new SOLAR space station concept – funded by NASA, the National Science Foundation and the Electric Power Research Institute – expands the concept to a gigantic scale.

In the laboratory, Shen and Will have modeled the SOLAR robot modules in two-dimensional form, working on an air-hockey table. The components will learn to find each other by sensing infrared signals, to maneuver and align themselves using built-in fans, and to lock on and pull units together using a motorized cable.

Marine Nanobots

How about a swarm of nanoscale robots swimming our coastal waters and monitoring potentially dangerous microorganisms in the ocean? A team of USC engineers has laid the groundwork for just such a microscopic armada.

In theory, when these “nanobots” identify a problem in the water – say, Brown Tide or a toxic spill – they will relay that information back to shore to alert beachgoers not to dip into water that’s considerably less than pure blue.

“With increasing urban runoff, sewage spills and blooms of harmful algae off heavily populated coastal areas, it’s important to sense and identify particular ocean microorganisms quickly,” says computer scientist Ari Requicha. “The quicker we learn that a pathogen is present in the water, the sooner we can warn people and take action to correct the situation.”

Requicha directs the engineering school’s Laboratory for Molecular Robotics, where the team he directs has been experimenting with nanometer-scale structures for nearly seven years. They’re currently refining the nanomanufacturing process and developing the ultra-small robotic sensors and software systems to control the actions of nanobots.

To understand the nature of this work, you need a measuring stick: one nanometer is one-billionth of a meter. In other words, a nanometer is to a meter what a small grape is to the Earth.


Ari Requicha envisions robots so tiny they’revisible under a microscope.
Photo by Michele A.H. Smith

Requicha’s group has programmed a special atomic-force microscope to slide nanoscale particles of gold and silver balls into precise positions on tiny slabs of mica. They can chemically link the particles to form crude assemblies and make nanowires by depositing carefully positioned metal balls in strings.

In theory it should be possible to build a nanoscale device with electrical and mechanical components to propel itself, send electronic signals and even compute. While individual nanoscale devices would have far less computing power and capability than full-sized devices, Requicha plans to deploy vast numbers of them operating in concert, like an army of ants.

Requicha’s team is building small robots that move, sense and communicate while tethered in a tank of water. The researchers will gradually progress toward building and controlling greater numbers of increasingly tiny free-moving robots. The end goal of the project will be to create an armada of robots that are as small as the microorganisms they seek to monitor.

“Today we commonly do experiments with five or 10 robots,” says Sukhatme, a co-investigator on the project. “But we’ll need algorithms to coordinate a million or more robots, all of which are freely moving in the ocean. Each robot will have limited capability and will only communicate with other robots that are close by. That is a daunting problem, and we must start laying out the foundations for large numbers of robots long before they are a reality.”

Speculating on the far-reaching applications of his research, Requicha says, “I don’t think these robots will be confined to the ocean. We will eventually make robots to hunt down pathogens or repair cells in the human body, not unlike the story line in the 1960s movie Fantastic Voyage.”


Robo-Liberation

Maja Mataric with a few pet projects.

Robots have been feared, demonized and shunted aside for decades. Think of USC’s new Center for Robotics and Embedded Systems as the Great Emancipator, leading the charge to integrate robots seamlessly into our culture.

By Gia Scafidi

“People have this strange fear and misconception of robotics based on movie and sci-fi book portrayals,” says USC’s Maja Mataric.

“They see robots as beasts that will attack us, or as hyper-intelligent creatures that will replace us. All of these fears are natural, yet unfounded. The robots of today, and those being developed, are very simple, very practical and very controlled, relative to those of the imagination.”

OK, it’s not the Emancipation Proclamation, but it represents a major shift away from the last 30 years of robotics research and development, carried out and implemented largely in hiding.

A fair number of machines and gadgets now work in collaboration with robotic components – car parts with embedded “intelligence,” washing machines, coffeemakers, microwaves, toys and “smart” houses that autonomously adjust temperature and light levels. Yet consumers often aren’t even aware of a robotic presence.

“Until now, societal pressures and fear of robots in our lives have kept robotics at bay,” says Mataric, director of USC’s new Center for Robotics and Embedded Systems (CRES).

“Robotic devices are socially acceptable today because they don’t stand out, and they don’t scare people or appear as competitors.”

The key to fitting robots into society, she adds, will be gradual change.

For now, however, they’re outsiders, “thriving on society’s fringe, performing tasks that are dangerous for people – like cleaning up hazardous waste material,” says Gaurav Sukhatme, director of USC’s Robotic Embedded Systems Laboratory. The military uses reconnaissance robots to scout potentially dangerous areas and buildings. Archaeologists send robots into Egyptian pyramids to explore areas unseen by human eyes for 2,000 years. Robots were deployed in Sept. 11 search-and-rescue efforts. They can even assist doctors in complex surgeries.

Of course, robotics research is still at a rather primitive stage.

“Robots in homes are where computers were in 1978 and ’79,” says Rodney Brooks, director of MIT’s Artificial Intelligence Laboratory, who spoke at the CRES inauguration. Many puzzles remain. For example: Robots can sense the physical world, says Mataric, but “knowing things like where I am, who that is and what is that object is extremely tricky and largely unsolved.

”Even so, Mataric is sanguine about the future. “As robotic technology becomes more and more advanced, this field will have a huge impact on society,” she predicts.

Mataric and Brooks both believe, for example, that there will be a major convergence of man and robotics within this century. Humans are already becoming more robotic, while robots are becoming more human. People already live with artificial hearts and hips equipped with robotic components, while robots imitate human sounds and facial expressions and recognize the presence of individuals.

More robotics research and development could mean better ways to care for disabled persons, remotely check on elderly parents or children home alone or even replace underpaid and overworked factory workers, Mataric suggests.

That last carries a built-in fear factor: the economic impact that robotic technology could have in terms of replacing human labor. But roboticists aren’t about to hold up research over such scruples. Like society itself, the economy will just have to adapt.

“I definitely see large changes taking place in the next 50 to 100 years,” Mataric says.

“The future looks bright for robotics,” adds Sukhatme. “I work in this area because I believe it’s a technology that will impact the lives of many in the years to come.”


How I Started the Revolution

In early 1990s, George Bekey and Scott Cozy ’97 worked on Marscar, a robot designed to study mobility on Mars-like terrain.
Bekey photo by John Livzey


In the beginning, USC robotics research amounted to one professor indulging his intellectual curiosity. Growing it into one of the nation’s top programs took luck, pluck and bucks – as well as colleagues with a shared passion. By George A. Bekey

I think it’s fair to say that I started robotics research at USC by lifting myself up by my bootstraps. I became interested in the field around 1980, but my own research had been in other areas (system identification and biomedical engineering).

Two events got the ball rolling. One, I recruited Barry Soroka in 1981. He had worked on robot manipulators at Stanford University. The second event was that I obtained an NSF grant to purchase our first robot: a PUMA 560 industrial manipulator. Such manipulators are familiar: We see them on TV helping assemble automobiles (bringing a door into position, for example) or using a torch to weld part of the body together. With this machine, Barry (now at Cal Poly Pomona) and I started the first robotics lab at USC. We ambitiously called it the USC Robotics Institute, after the successful major program at Carnegie-Mellon University. The provost gave us a small start-up fund, and we were in business.

About this time we began teaching a course called “Introduction to Robotics,” which was concerned almost entirely with the mathematical description and control of multi-jointed robot manipulators. We purchased several small robot arms to be used in student experiments as part of the course. They could do little more than pick up an item and move it, sometimes stacking two blocks on top of each other. These little robots were controlled by early Apple II computers.

Our stock of robots increased rapidly thanks to several gifts. When a Northrop Corp. lab was closed, we received several large industrial manipulators. At this time we had three or four graduate students working in the lab.

The arrival in the late ’80s of dancer Margo Apostolos in the USC School of Theatre added an interesting aspect to our robotics program. Margo had written her dissertation at Stanford on programming manipulators to move gracefully to music. Imagine an 8-foot arm with its “shoulder” fastened to the ground, and the arm (including elbow, wrist and hand) moving in graceful curves. Margo developed “robot choreography,” in which human dancers would move with and around wiggling robot arms. Several such performances were given at USC.

The “Robotics Institute” merged into a major computer vision program directed by Ram Nevatia. The combined program was called Institute for Robotics and Intelligent Systems (IRIS), but our lab continued to operate quite independently.

From 1985 to 1990 I became interested in building hands for robots that weren’t just simple grippers but resembled human hands. A Yugoslavian colleague had built a five-fingered hand, but it lacked computer control and sensors. We obtained the hand, connected it to a computer and used it in a number of research projects. I received several grants to study robot hands, including their potential use as prostheses for people who may have lost a hand in an accident.

In the late 1980s and early ’90s we became interested in mobile robots. Since commercial mobile robots were very expensive and we had no funding for such research, my students and I built several interesting machines. “Rodney” was a six-legged walking machine (the name was inspired by Maja Matari´c’s advisor at MIT, Rodney Brooks). The unique aspect of this robot was that we developed the software to enable it to learn to walk, using a simulated form of evolution. We did not program the sequence in which the robot should move its legs, so it fell down frequently. Eventually it learned to walk with a “tripod gait” typically used by insects: the insect moves the front and rear legs on one side and the center leg on the other side at the same time, always maintaining a tripod of support to keep it stable.

We also built Marvin (named after robotics pioneer Marvin Minsky of MIT). This was a converted toy car which learned to roam about the lab without running into anything.

My first Ph.D. student in robotics graduated in 1986, followed by 20 others. By 1990 we had nearly 15 students participating in robotics work at various levels, including several very bright undergraduates.

Once we moved from manipulators and robotic hands to mobile robots, there was no turning back. We built a number of other robots, including one more walking machine – designed by Gaurav Sukhatme MS ’93, PhD ’97 as part of his doctoral research.

We also began struggling with robot helicopters in the early 1990s and won a national competition among universities to design and build autonomous flying machines.

In the late 1980s neural networks expert Michael Arbib joined us from the University of Massachusetts, where he had directed the Laboratory for Perceptual Robotics. His arrival broadened our computer science course offerings as well as our robotics research.

By 1996 we were achieving recognition for our research, presenting many papers at national and international robotics competitions. We had some 20 students in the program, and between $250,000 and $500,000 in research funding per year.

In 1996 I was appointed associate dean of the USC School of Engineering. I accepted the position on the condition that I would be authorized to recruit another faculty member in robotics. We began a nationwide search, and Mataric joined us in the fall of 1997. Sukhatme, who completed his Ph.D. in 1997, joined the faculty that same year.

Until his retirement in April, George A. Bekey was a University Professor as well as the Gordon Marshall Professor of Computer Science, Electrical Engineering and Biomedical Engineering. His current research interest is in cooperative behavior of multiple mobile robots and in highly distributed large networks of sensor-carrying robots.