Agility and Autonomy - UPenn's Nanocopters
6:09 PM on Mar 02, 2012
If you've seen the Youtube videos of swarming nano-UAVs dancing, or building a tower, then you might be interested in the story behind the technology, as told in a TED talk by Prof. Vijay Kumar of the GRASP laboratory at the University of Pennsylvania.
While the tiny quadrotors' antics are amusing, they have a serious purpose: to exploit the agility of small aerial robots by giving them the autonomy to cooperate to perform larger tasks - in the same way ants cooperate to move pieces of food many times their size.
When unmanned aircraft are scaled down the 8-inch diameter of the nano-quadrotors used by UPenn, their inertia reduces dramatically and they become extremely agile: able to maneuver through doors and windows and around obstacles indoors.
As Kumar explains, the nano-UAVs sense their environment 100 times a second, but calculate flight-control commands for the rotors 600 times a second. They are programmed to map out a path around obstacles and through apertures that results in a smooth "minimum snap" trajectory.
Indoors, where motion-capture cameras are used to track the vehicles and locate any obstacles, the UAV can fly itself through a hoop thrown into the air by a researcher - as Kumar's talk shows. The quadrotors also learn different manuevers, he says, and piece them together to perform difficult tasks.
Video: GRASP lab
The disadvantage of nano-UAVs is their tiny payload, so the GRASP lab has taken a leaf out of the ants' book and programmed its quadrotors to cooperate autonomously. Like the ants, each UAV can sense only its neighbors - there is no central coordination - monitoring their separation while flying in formation and staying within acceptable limits.
The nano-quadrotors can fly planar and three-dimensional formations and adapt their formation "on the fly", Kumar says (and shows), to fly through an obstacle. The extremely fast flight-control processing means they remain stable even when one UAV flies inside the flow field of the rotors of another.
This ability to cooperate autonomously, and anonymously, has allowed the GRASP lab to program its quadcopters to build cubic stuctures by picking up and positioning standard truss-like elements that click together. "They know what to pick up when and where. We just give them a blueprint of what to build," he says.
Video: GRASP lab
All this has happened in the lab, using the motion-capture cameras for guidance. Outside the lab, UPenn researchers have shown how the nano-UAV can navigate autonomously indoors, without GPS, using a laser scanner to map the building on the fly - including doorways, windows, furniture and people - all in a coordinate system relative to the robot itself. Here autonomy allows the UAV to look at the map, see where information is missing, and go collect it. Such vehicles could be used as first responders, Kumar says.
awt, unmanned, autonomy