Perching Glider Takes a Cue from the Birds
Photo: Jason Dorfman, CSAIL photographer
It isn't everyday that a computer scientist wins one of the most coveted awards in aeronautics. But when Rick Cory started in Associate Professor Russ Tedrake's Robot Locomotion Group as a Ph.D. student, he wasn't out to conduct robotics as usual. "We were trying to think of a project that could push the limits of robot control," Cory explains. "And the idea came up of trying to build a robot that could fly like a bird. For me that was a very inspiring, fantastic idea."
Five years after that germ of an idea, Cory was named 2010's Boeing Engineering Student of the Year for his work with Tedrake on a robotic glider capable of landing on a perch like a bird. The relatively simple design represents a major step in the development of unmanned aerial vehicles (UAVs). That a couple of roboticists could make such a big impact in an arena ordinarily reserved for aeronautical engineers is a sign that the field of flight may be undergoing a sea change.
Tedrake and Cory developed their model by taking their cues from the natural world. Noting that your average pigeon is exponentially more maneuverable than a 747, they studied patterns of birds in flight. Aerodynamicists have no practical models to demonstrate the dynamics of avian flight, due to the sheer complexity and unpredictability of wind patterns around a bird's wings.
According to Tedrake, approaching the problem from a computer science perspective shed new light on the dilemma. "We thought that there was a chance that we could come in and do something because we're thinking about it a little differently, and we had some machine learning tools that could deal with that complexity," he says.
Still, they were far from experts in the fields of aeronautics and fluid dynamics. "It was literally a matter of picking up aerodynamics 101 books and learning as much as I could," Cory recalls.
Soon, they were studying hobby ornithopters and fixed-wing airplanes in search of a compelling experiment. They settled on a seemingly simple maneuver--landing on a perch. A bird accomplishes this action by flaring its wings to drag against the surrounding wind currents, slowing down enough in midair that it's able to land on a narrow point.
But stall, and the landing process post-stall, are fiendishly difficult to model. The airflow around a bird's wings is highly unpredictable, creating complex and ever-changing vortices of air. Lower angles of attack, such as those practiced by modern airplanes, are easier to control. But it also makes precision landing impossible.
Cory and Tedrake decided to simplify the problem. They set aside ornithopters in favor of a fixed-wing model--one whose fluid dynamics, while still complex, would be easier to understand. "It was really just a matter of trying out different things and asking the right questions. Stripping down the problem to its basics," says Cory.
Tedrake says that this type of approach is emblematic of the Robot Locomotion Group's work. "Simplicity is really key to experimental success," he explains.
The resulting model--the one that earned Cory the Boeing award--is about as simple as possible: a small, fixed-wing foam glider with a tiny motor to move the tail. An off-board control system is the key to the glider's on-board austerity. The system works by calculating a variety of possible trajectories and checking them against the actual path of the glider, tracked by motion capture cameras. It allows for deviation, and enables the glider to course-correct itself in order to hit its target.
This control system is designed to be robust enough to deal with high levels of model error. As Cory points out, there's no such thing as the perfect conditions when you're dealing with something as unpredictable as stall. "For these very complicated systems, you're never going to have a perfect model. And if your control system relies on having a perfect model, then it's doomed," he says. What's more, the technology is eminently adaptable. The group is implementing the same system for walking robots, and Cory believes it could be adapted for swimming as well.
The team has yet to take the glider outdoors, but that's the next step. Currently, unpredictable wind currents provide a serious obstacle. But Ph.D. student Joseph Moore, also of Tedrake's group, is working on an on-board sensing system that will hopefully improve the plane's chances in an open-air environment. "We want to get it to the point where you can just have somebody toss it and it would be able to land," Cory says.
Tedrake and Cory's control system has far-reaching applications. The United States Air Force hopes to create a bird-sized UAV by 2015, one with the ability to recharge itself by perching on a power line. Tedrake and company have been working closely with the USAF to achieve that goal.
And for Cory, who's moved on from MIT to build next-generation robots with Disney Imagineering, the system has other uses, too. "I visited the Air Force and I visited Disney, and they actually have a lot in common," he observes. "The Air Force wants an airplane that can land on a power line, and Disney wants a flying Tinkerbell that can land on a lantern. But the technology's the same."
Whether it's used to build spy planes or fairies, it's clear that Tedrake and Cory's work is helping to put computer science at the forefront of control theory. "It's not just computer science pushing that way; they're coming our way too. But this is just the first example of what we could try to do with a bird-sized UAV," says Tedrake. "There's a new, rich set of problems that open up when we start building things that are that size. And it's just the tip of the iceberg. We've got plenty to work on."
Jenna Scherer, CSAIL