Building better autonomous vehicles
Despite continued advances, many issues still remain as to how to best design and operate autonomous vehicles. Six Ohio State faculty from the departments of electrical and computer engineering, and computer science and engineering are working to integrate multidisciplinary advances in software, sensing and control, and modeling to address current weaknesses in autonomous vehicle design.
A key issue the researchers must address is the human component, including how humans make decisions and how to predict their future actions.
“The central bulk of our work is looking at specific scenarios where the existence of humans throws a wrench into your nice, automated world and figuring out how to design your way around that,” says Ümit Özgüner, principal investigator and professor, electrical and computer engineering.
Ashok Krishnamurthy, associate professor of electrical and computer engineering, has created a learning, probabilistic framework that analyzes vehicle data in order to deduce what the driver is going to do.
“I believe that the model is an important step in knowing how to understand driver behavior,” he says. “Right now it’s only looking at very restrictive situations, but within those it seems to have a good deal of predictive power.”
Füsun Özgüner, professor of electrical and computer engineering, leads the real-time and trust-aware computing portion of the project.
“Some of the vehicles will have many sensors to sense the environment,” she says. “I am responsible for developing a computer architecture that can quickly process all that information…It has to be small, fast, and use a reasonable amount of power.”
She is also researching wireless communication between vehicles to enable sharing of pertinent information, especially in the case of road obstructions or dangerous conditions. This, however, evokes the issue of cyber security and how to prevent malicious use of such networks.
“We developed a concept called trust aware computing,” Füsun Özgüner explains. “You sort of calculate the trust value for all the other cars around you, just like humans do in their relationships. And it’s dynamic, you keep updating it based on new data.”
“I was happy that we could test ideas on a semi-virtual world and reach meaningful results for the real world,” says Ümit Özgüner. “We transfer some of those results to cars that we test on campus.”
Ultimately, the researchers hope their work will lead to safer autonomous vehicles, including automobiles, service robots used in hospitals and rescue robots used after disasters.
This article is reprinted from the Electrical & Computer Engineering 2011-2012 Annual Report.