A vision for the future
Computer vision researcher developing innovative techniques to help autonomous systems navigate
In the futuristic world of tomorrow, robots may be helping elderly patients, leading planetary exploration and manufacturing cutting-edge, nanoscale technologies.
But before they can do any of that, they have to be able to interact with the human-centric world around them. As such, research into computer vision ties into a number of interdisciplinary research areas the College is pursuing, including both unmanned autonomous systems (UAS) and advanced manufacturing.
"UAS and also advanced manufacturing rely a lot on cameras," said George Bebis, professor and chair of the Computer Science and Engineering Department. "This is a very important sensor for UAS to capture image data and then analyze the data to identify objects and detect obstacles. In the manufacturing environment, as far as computer vision is concerned, visual inspection is the major thrust."
Both UAS and robots working in busy manufacturing environments may also rely on vision to figure out where they are in an environment and then navigate appropriately. That problem is the topic of Bebis' current research, although he's focusing on these problems in a particularly challenging environment - outer space.
He is working on a project funded by NASA that helps robots up in space orient themselves so that they can move around and perform mundane tasks, freeing the astronauts up for more advanced work. In a zero-G environment, some of the traditional sensors that might be used for this problem won't work. Additionally, computing power is limited, so solutions that rely on bulky sensors or lots of processing power aren't an option.
"You might have a very brilliant algorithm that solves the problem very well. If it requires a lot of computational resources and power, it's not going to be applicable to those missions," Bebis said. "Same things when you fly these UAVs. There's going to be a trade-off between accuracy and computation resources."
Bebis has developed a solution that uses an artificial robot in a 3D model of the environment to determine position and orientation. Using the horizon line for reference, the process compares two images: one taken by the actual robot with one taken by the artificial robot. If the images match, the researchers know they've accurately placed the artificial robot in the environment.
However, robots in outer space may have one advantage over robots on the ground: a relatively predictable environment. By contrast, autonomous systems moving around on earth must be able to deal with a lot of variability in their environment.
"When we develop a system in a lab we have a lot of control," said Bebis. "We know what kind of objects are around. We can control the illumination. We can control a lot of things, but when you try to do something is some unrestricted environment there are major challenges."
Optimizing performance, whether that's enabling an autonomous aerial system to fly faster and still have enough time to analyze incoming data or allowing multiple robots to coordinate their actions, is key to developing systems that can truly be useful in a wide range of environments. Because accuracy and processing speed are two key elements to optimization, Bebis believes the long-term solutions will rely on multi-disciplinary expertise.
"If you try to identify and obstacle in a very cluttered area it's going to become much more challenging and it's going to require more processing," Bebis said. "So this is definitely a multidisciplinary field. You cannot only have computer scientists working on that, you have to have people from multiple areas working together. That's why we need mechanical engineers and people that do hardware design to design systems that are more efficient, lighter, that don't consume a lot of energy."
That's one reason Bebis is excited about the College of Engineering's latest academic offering—an interdisciplinary minor in unmanned autonomous systems.
"I expect that things will change in the future for those fields," he said. "There's no doubt that you need to have people that have expertise in multiple areas."