In the Aerospace Robotics Laboratory in Howe Hall, Soon-Jo Chung is building a micro aerial vehicle (MAV) that utilizes flapping wings and is able to perform acrobatic maneuvers while expending minimal energy. The military uses MAVs for functions such as surveillance and scrutinizing potentially hazardous environments. The ability to avoid obstacles while moving in and around tight spaces is a key requirement.
Across the street in the Developmental Robotics Laboratory in Coover Hall, Alexander Stoytchev is working to make humanoid robots smarter. He wants to build a robot that can learn to adapt to new circumstances as it interacts with tools and its environment.
Both researchers are biomimeticists—that is, they are using ideas from nature to advance the development of new technologies. Nature, after all, has been evolving for millions of years and has come up with some very elegant solutions to perplexing problems.
Observing how nature has solved a problem, however, is a long way from understanding the complexities of the system and determining how to mimic it technologically. Nevertheless, both Chung and Stoytchev are working diligently to apply what they can glean from nature to projects in their respective labs.
‘We can learn a lot from bats’
Last fall, Chung, an assistant professor in aerospace engineering, received a three-year, $300,000 grant through the Air Force Young Investigator Research Program. He got the inspiration for his project, “Bio-Inspired Integrated Sensing and Control of Flapping Flight for Micro Aerial Vehicles,” from bats, and while these furry flying mammals won’t win any popularity contests with most people, Chung has nothing but praise for them.
“We can learn a lot from bats because they are so agile,” he explains. “They can change directions quickly and easily and do a lot of acrobatic maneuvers. Their membrane wings are very flexible and highly compliant.”
A control and robotics engineer who came to Iowa State in 2007, Chung has turned to biologists for facts about the characteristics of bats that make them so aerodynamic and agile. One of his collaborators is Kenny Breuer, a biologist at Brown University who has collected extensive data from studying bats.
“Biologists have learned that regardless of their size, bats are very efficient in flight,” Chung says. “They generate more lift and less drag than airplanes. They can also fly with damaged wings and with more than 50 percent of their proportional weight.
“These things all have implications for how we design airplanes,” Chung continues. “We’re not looking for strict mimicry, but rather to learn from bats in order to develop new mechanisms.”
A bat’s unique abilities lie in the structure of its membrane wings. Flapping the wings propels the bat through the air, of course, but the wings are also constantly adapting to the environment. Thousands of tiny hairs on the wing membranes provide sensory information that controls the shape and pitch of the wings, enabling the bat’s acrobatic movements. And with a minimum of 24 joints in each wing, there are countless configurations to provide the precise motion required for a particular maneuver.
The challenge of control
Chung’s challenge is to design a control system that will give his MAV, which is essentially a robotic bat, the same flexibility as a living bat. “It needs to work seamlessly,” he explains. “When you walk, you don’t have to think about one foot moving in front of another because signals coming from the brain make sure everything moves in a synchronized way.
To achieve a similar level of coordination in his MAV, Chung is devising a central sensing and control system that mimics the neuronal networks of bats in order to synchronize the wings’ flapping and joint movements, allowing them to respond appropriately to unique environmental conditions—in a word, the controller effectively serves as the MAV’s “brain.”
Chung’s project draws on his previous work on the synchronization of formation-flying satellites. As a graduate student at MIT, he worked extensively in the SHERES lab, a highly sophisticated test bed dedicated to developing autonomous formation-flight and docking-control algorithms. There, Chung was challenged to develop new theories and methods to ensure that the movements of individual satellites were synchronized across entire formations.
Similarly, Chung explains, all of the parts of the bat’s wings must work together in order to achieve the desired motion. “The MAV has to be able to maintain its flight in a robust and sustained fashion,” he says. “With its multiple joints, the wing has many more degrees of freedom than an airplane wing. It makes it very challenging to control. The wings have to flap at the same frequency, more or less, and all of these other parts need to be synchronized in their motion so you have the right pitch angle and face difference.”
Once he has a working model, Chung plans to begin experimental studies with the MAV in Breuer’s wind tunnel at Brown. Yet, while his immediate goal is to achieve highly maneuverable MAVs equipped with intelligent sensors, the work has more far-reaching possibilities.
“The successful reverse-engineering of flapping flight has implications for innovation in aircraft design,” Chung points out. “Aircraft have been dominated by fixed-wing design, but this early work indicates some advantages of flapping wings that should be investigated. This work is also pushing the frontier of our understanding of neurobiological mechanisms underlying animal flight and locomotion.”
Looking beyond R2-D2
Over in his robotics laboratory, Stoytchev, an assistant professor in electrical and computer engineering (ECpE), is taking a developmental approach in his efforts to create intelligent robot behavior, using human behavior as his model.
“Humans, and to some extent higher animals, are the only existent proof we have so far of natural intelligence,” he explains. “To build artificially intelligent robots, we must study how complicated biological systems work and try to emulate them.”
Designing robots with human-like intelligence is something Stoytchev has wanted to do since he was in elementary school. He was nine years old when he first saw the movie Star Wars, whose droids, R2-D2 and C-3PO, captured his interest.
“I remember asking my parents if such robots existed for real, and what would it take to build one,” he says.
Stoytchev’s childhood interests continued, and while robotics clubs weren’t part of the educational scene yet, he was drawn to the next best thing, namely computers. As a college student, Stoytchev majored in computer science and began to specialize in artificial intelligence. Yet it was not until he became a graduate student at Georgia Tech that he worked with robots for the first time.
Joining ECpE in January of 2008, Stoytchev’s work today is centered in developmental robotics, a new field that crosses the disciplinary boundaries of robotics, artificial intelligence, developmental psychology, developmental neuroscience, and even philosophy. The goal of his research, he says, is to build autonomous robots that are more intelligent, adaptable, and useful than traditional robots.
“Today, we have many industrial robots that are very useful for manufacturing tasks such as building cars,” Stoytchev says. “However, their actions are very carefully scripted by human programmers, so they are not truly intelligent. They can use complicated tools to perform cutting, painting, and welding operations, but only as long as their environment—i.e., the conveyor belt—is structured in a predictable way. If somebody replaces their tools, they would not notice the difference.”
Learning from one’s mistakes
Stoytchev wants to create robots that will notice that difference and says that the only way to achieve this goal is for the robots to be able to interact with the tools and learn from their experiences. “It is naïve,” he adds, “to expect that we can pre-program them with all of the knowledge that it takes humans decades to master.”
Stoytchev’s basic hypothesis is that robots must undergo a developmental period similar to that of humans and animals. Tool use, he says, is a prime example. Animals use tools for many different purposes to overcome limitations imposed by their anatomy and over time have learned how to adapt their available resources to meet specific needs—using sticks to reach out for food or to dig a hole, for example.One long-term project in Stoytchev’s lab addresses the problem of autonomous tool use in robots. Unlike the highly specialized approach used with industrial robots programmed for one specific task, Stoytchev’s robots autonomously interact directly with a tool in order to learn a representation of the tool that is grounded in the robot’s sensorimotor repertoire, thus reducing its dependence on human programmers to tell it what to do.
As with human learning, however, even robots must learn from their mistakes. Last October, for example, Stoytchev and his students were given the opportunity to design a robot to cut the red ribbon at the dedication of the ECpE building addition. It was a perfect opportunity to showcase his work, Stoytchev says, but the opportunity presented some anxious moments.
“For humans, it is common knowledge that scissors have two handles that form a plane,” he explains, “and pulling on the handles in opposite directions perpendicular to that plane is futile. Well the robot didn’t know that, so a week before the ceremony, it broke three of its fingers trying to do exactly that.”
By the time the ceremony rolled around, however, the fingers had been repaired, and, to the delight of the crowd, the robot had learned how to operate the scissors, successfully cutting the ribbon.
Windows of opportunity
In his quest to continually improve the intelligence of robots, Stoytchev is studying the fundamental processes and principles that drive human cognitive development. Language-learning studies conducted with eight-month-old infants, for example, have shown researchers that infants learn to segment the basic units of speech by extracting statistical information from the audio signal. Stoytchev’s team was able to replicate those results in code to apply to robots, work that won the team a best paper award.
One area of particular interest to Stoytchev concerns what are termed “developmental windows of opportunity” during which a given skill must be learned. It is known, for example, that human infants must learn to speak by the age of three, otherwise their language skills and other higher cognitive abilities will be severely affected. Accordingly, Stoytchev wants to understand the significance of the order in which these windows “open”—i.e., what other skills might be affected by a child who doesn’t learn to speak by the age of three?
Since finding the answers to these kinds of questions would disrupt the normal developmental flow in humans and animals, Stoytchev hopes that one day robots themselves may be capable of helping researchers find the answers. That means that the intelligence of robots must definitely increase, and Stoytchev is confident that day may not be so far off.
“Robotics today reminds me of the computer industry in the early 80s,” he says. “There were no standard computer components, and the field was driven by hobbyists and forward-thinking companies and universities. Once Apple unveiled the PC, it did not take long for somebody to invent the first killer app. The rest, as they say, is history.
“Robotics,” Stoytchev adds, “is waiting for its first killer app.”