Posted May 22nd, 2014 by Engineering

Teaching flying robots to learn

Angela Schoellig

“What I really like about robotics is that you can make a machine that
extends human capabilities and the possibilities that we have. ” – UTIAS
researcher Angela Schoellig (Photo: Roberta Baker).

A helicopter drone zooms along the shores of Boston’s Charles River, carefully flying back and forth to photograph algae growth for nearby researchers. And who’s at the controls? Nobody – thanks to autonomous algorithms developed by Angela Schoellig, an assistant professor at the University of Toronto Institute for Aerospace Studies (UTIAS).

Next month, Schoellig – an expert in robotic controls – shares her latest research for science communicators from across the country at the upcoming Canadian Science Writers Association annual meeting at U of T.

In her research, Schoellig applies mathematics to engineering systems to develop the algorithms that systems need to regulate themselves and move autonomously.

“These algorithms are used in such items as a car that drives autonomously or in a home air-conditioner that regulates the temperature in the house,” said Schoellig. In addition to monitoring ecosystems in Boston, her algorithms are also used in the unmanned aerial vehicles that many farmers employ to monitor plant health and soil moisture.

Shoellig’s current work looks broadly at robotics, trying to understand how robots can learn from experience, collaborate and share information. It builds on her PhD research at ETH Zurich – the Swiss Federal Institute of Technology – where she applied new controls to aerial vehicles to help them fly autonomously.

“Robots work really well in controlled environments where you can give them all the information that they need to work efficiently and effectively ahead of time,” Schoellig said.

It’s trickier, however, in environments that are unknown or include people. Since humans aren’t predictable, the robots must learn to react to unexpected events.

“If one robot has learned about the environment, I want to know if that information is useful to transfer and understand how a robot could share that with its peers,” she said. “Learning is a complex task and it might take 1,000 trials. If the robot could share its experience, other robots could learn 1,000 times faster.”

Using algorithms that she has worked on, Shoellig can teach multiple flying robots to interact with each other in new ways – even having them learn a complex dance to the Pirates of the Caribbean theme song:

[youtube http://www.youtube.com/watch?v=NPvGxIBt3Hs]

“What I really like about robotics is that you can make a machine that extends human capabilities and the possibilities that we have,” she said. “They can give us a view on things that we would never have otherwise.”

Angela Schoellig will take part in the CSWA’s ‘Better Living Through Technology Panel’ on Friday, June 6 at 8:30 a.m. Read the entire CSWA Annual Meeting program.