2 minute read
Taking It to the Street
Graduate student Yongxin Wang (on the left) and research assistant professor Bo Yu working together on an autonomous vehicle on the Arlington Campus. Photo by Evan Cantwell
Autonomous vehicles that promise to decrease traffic and make driving safer won’t get far without sophisticated sensors to detect hazards. That’s why researchers from George Mason University, Virginia Tech, and Old Dominion University are collaborating on a $1 million program funded by the Commonwealth Cyber Initiative (CCI).
Driving three decked-out Toyota Corollas, researchers are taking to the busy streets of
Northern Virginia and Washington, D.C., to test sensors and cameras they’ve installed to simulate a self-driving car. “My number-one goal is to take on Dupont Circle, but we’ll see if we can overcome that crazy obstacle,” says computer science professor and team lead
Duminda Wijesekera.
Wijesekera is using artificial intelligence (AI) to help create smarter car sensors. His research seeks to improve an autonomous car’s ability to recognize potential hazards at night, in severe weather, and in construction zones, making the roads safer for everyone. CCI’s funding allowed Wijesekera to purchase three Toyota Corollas as testbeds. Prior to receiving the grant and vehicles from CCI, Wijesekera was volunteering his own car to test the AI technology on the streets of Washington, D.C. Traditionally, autonomous vehicles create obstacle-free driving paths using image recognition software. Sometimes, the software can fail to recognize obstacles in the road in low-light conditions, such as at dusk or dawn, when it is raining, or in construction zones. This can result in the software mistakenly creating driving paths with dangerous obstacles in the way. Wijesekera and his team are using AI to prevent such a scenario. “To combat this issue, first, I use two cameras—a color camera and an infrared camera,” Wijesekera explains. “Then I take data from both cameras and ‘fuse them’ using a deep-learning process.” Distracted driving claimed the lives of 1,730 drivers, 605 passengers, 400 pedestrians,
and 77 bicyclists in 2018 alone, according to the National Highway Traffic Safety Administration.
“Just like automated trains, an entirely automated-car world would be cleaner, safer, and greener in the long run,” says Wijesekera. “I’m hoping this is the first step of many toward making driving totally handled by machines.”
— Duminda Wijesekera, team lead and professor, Computer Science
—Aubrey Medina