Giving Legs to Robots

 By Michael J. Schwager

You can learn a lot from the cockroach. Lowly yet durable, primitive yet agile, the six-legged arthropod is attracting the attention of some astute engineers.

Dan Koditschek of Penn Engineering, for example, works closely with biologists who watch cockroaches scamper across the ground, videotape their movements, see how they run. Koditschek, who came to Penn from the University of Michigan, is the newly installed Alfred Fitler Moore Professor and Chair of Electrical and Systems Engineering. So why is he conducting these studies? Because his prime area of interest is legged robots. The movements of cockroaches help bolster the understanding of locomotion on legs.

“We would like nothing better,” he says, “than to approach the capability of the humble cockroach.”

One robot that Koditschek and his students have built juggles balls. Another, which looks something like an oversized mechanical cockroach, can propel itself across a grassy field, traverse the boundary between grass and gravel, and hop over a bump in the road.

Of course Koditschek isn’t aiming to build artificial cockroaches. Eduardo Glandt, Dean of the School of Engineering and Applied Science, calls Koditschek’s work “bioinspired. It’s not copies of nature. He’s a man who sat down with biologists to see how nature evolved the way hexapods walk. He looks at principles—he does not copy cockroaches. The result, having gone through fundamental concepts, is the most successful hexapod robots.When you look at the videos, it’s amazing how well they walk.”

When people think of robots, they often picture machines with human form and human or even superhuman capabilities. The Czech writer Karel Capek introduced the word in a 1921 play, but the idea had stimulated imaginations for centuries. Given how long robots have been around and how deeply they’ve insinuated themselves into the culture, Koditschek says people would be surprised at “how crummy they are.They’d be surprised at how hard it is to get a legged machine to move across a field or a streambed. It’s shocking how imperfect they are.” Robots can’t beat the fastest human runner in a race, Koditschek says. “They can’t climb stairs as well as a three-yearold. They can’t swing from trees the way orangutans can. They can’t hop on mountainsides like goats.”

And robots can’t yet conquer their environments the way even cockroaches can. For small animals, Koditschek says, cockroaches are impressively fast—racing at about 10 body lengths a second. “That’s like a human running 60 miles per hour. And they’re not just fast but amazingly maneuverable. They don’t slow down a lot, and they don’t destabilize.”

Koditschek has long been fascinated with biologists’ discovery that animals run like pogo sticks—hopping along the ground with surprising stability. “They harness simple spring dynamics,” Koditschek says. “Even millipedes, when they go fast, show some pogo stick-like exchange of potential and kinetic energy. “You wouldn’t think of a hopping pogo stick as particularly stable, but it is. There’s an enormous range of gaits in animals’ locomotion. But when they pick up speed and need agility, it seems to be an empirical fact that they allow the inner pogo stick to come out.”

Why investigate cockroaches? “They’re fabulous runners and convenient animals to study,” Koditschek replies. “Invertebrates don’t evoke the same response as vertebrates do; you can study them without offending the social order.” Cockroaches have existed for at least 350 million years and were among the first legged animals.

Koditschek acknowledges the fascination with robots that have humanlike form. But, he asks rhetorically, “Does human form bestow human capability? No.

“We don’t copy nature but learn from nature. People have long been confused on this point—such as building early airplanes with flapping wings. In the future, when we have more sophisticated materials that can more directly match the combination of strength, adaptability, weight, and power density that animal limbs display, it’s possible to imagine that robots’ morphologies will resemble the ones we’re used to.”

To build a successful legged robot, engineers must meld disparate technologies. Control circuits, springs, gears, wheels, treads—robots blend the logical and the mechanical. “Robots have to do work in the physical world,” Koditschek says. “We don’t yet have a strong conception of programs in the physical world. So we’re studying how to build programmable work. Nimble, adaptable locomotion is one of the most obvious examples, yet it remains a fundamental and unsolved problem.

“We’re trying to build things that work and to prove theorems about why they work.We’re trying to articulate the more fundamental concepts of the field.” Koditschek mentions Alan Turing, who in the early 20th century first envisioned machines capable of methodical, goal-directed computation. “Turing clarified the question, ‘What’s an algorithm?’ In robotics, we’re in the pre-Turing era.We don’t exactly understand what the fundamentals of our field are—how to formalize an analogous notion of methodical, goaldirected work upon the environment.”

So far, most successful autonomous robots have had treads rather than legs and look a bit like small army tanks. These “tracked robots,” Koditschek says, “are selling in increasing numbers to the military and to police departments. They were brought to the World Trade Center site after the September 11, 2001, attacks with some success.

“Many robots you can buy now have tracks or wheels,” Koditschek says. “None have harnessed the secrets of the animals.” The mechanical aspects of robotics present one chief hurdle. Consider animated cartoons, which came out decades ago and offer the illusion of physical motion. But they’re images, not machines—they don’t do any work.

“Real motors have torque limits that simulations don’t,” Koditschek says. “They undergo mechanical losses. Their appendages encounter complicated ground mechanics.When you walk over a gravel path, your interactions with it are astronomical. The world is a much more intricate, complicated place than cartoons.

“It’s even hard to build a laptop computer that doesn’t crash. If we can’t get that right— just exchanging bits—think how much harder it is to interact with mechanical objects.When my laptop crashes, it’s one thing.When an avionics system or an elevator control system crashes, it’s something else.” Koditschek says he entered the field because he wanted “to understand why robotics was so hard. For a while, you could blame computers—if only they were faster or more powerful. Then you could blame sensors, but they have recently made huge advances in capability and affordability. You can still blame the inadequacy of available materials (although seemingly not for much longer, the way technology is accelerating). But we still don’t have machines that can compete with my kids when they were three.”

Just as robotics has shown progress over the past decades, the coming years will bring even more impressive advances. “Factory automation has been going on for at least 30 years,” Koditschek says, “and machines are improving. But getting robots out of the factories is proving very difficult.

“Robots have captured a lot of interest in the toy industry. But it’s a treacherous place to compete—the profit margins are so small. There’s no question, though, that entertainment and the toy industry hold huge potential markets for our technology.” At a reception for Koditschek in late February, Dean Glandt spoke of Alfred Fitler Moore, the founder of the Moore School of Electrical Engineering and the namesake of the honorary chair that Koditschek holds. “For many, many years,” the dean says, “the Moore School was the most preeminent name in electrical engineering.What Alfred Fitler Moore did in creating the school was one of the most significant things in the birth of contemporary electrical engineering. Holding a chair carrying the name of Alfred Fitler Moore is historical.”

“It’s quite an honor,” Koditschek says. “The Moore School is the place where computing was invented. The GRASP Lab has been one of the leading institutions for robotics in the country. We want to make Penn one of the places where the robotics revolution occurred.”

Affable, thoughtful, and highly articulate, Koditschek speaks with the authority of a professor and the gusto of an excited freshman. The words pour forth logically, clearly, and with a cadence that bespeaks a scholar who studies motion. “What makes him so compelling,” says Dean Glandt, “is that he’s a man of tremendous personality. He has great energy and big dreams.What we needed for the chair was big dreams and vision and someone who has the energy and the academic taste to bring the dreams about. You can see how energetic he is and how visionary.”

The search committee that recommended Koditschek was headed by David Pope, Professor of Materials Science and Engineering and University Ombudsman. Pope, says Dean Glandt, “is the elder statesman in the school. He went around the country interviewing people, talking to the wisest minds to sift through and find which person would be the golden nugget. And he found it.

“Under Dan’s leadership, the department is already experiencing a renaissance. It’s starting to add faculty. His presence here and his vision is another positive development in our plans for a nanotechnology building on Walnut Street. Electrical engineering will be a major occupant of that building. The fact that we have Dan Koditschek on board gives traction to that undertaking.”

As a teacher, Koditschek longs to “create students who are more capable than I was at putting together these disparate pieces of human knowledge.We’re much closer than we ever were.We’ll have wonderful machines.When the motor revolution arises, it will make the Internet seem like small potatoes.”