LAWRENCE — Computer science students at the University of Kansas will test their skills throughout the remainder of the semester with a series of races using self-driving model cars developed by Amazon Web Services.
The students in Andrew Williams' Introduction to Artificial Intelligence class have spent the semester programming their cars and virtually testing the results using AWS' DeepRacer, a cloud-based 3D racing simulator. For the races underway throughout the remainder of the spring semester, though, students have built a real track in the atrium of the School of Engineering's LEEP2 building and are testing their AI programs using 1/18th-scale race cars.
“It’s a method for us to teach deep reinforcement learning — an artificial intelligence technique that allows cars to learn to drive by themselves, using their video cameras and other sensors they have," said Williams, the KU Engineering associate dean for diversity, equity and inclusion and the Charles E. and Mary Jane Spahr Professor in Electrical Engineering and Computer Science.
"It's going to be pretty exciting," said Amanda Nelson, a junior in computer science from La Cygne.
In artificial intelligence, "reinforcement learning" uses reward functions — essentially, points for achieving a task — to help a machine to learn. In the case of the DeepRacer cars, students deploy code, then the car receives points for staying on the track or for completing a lap quickly. The car responds to those points and adjusts its behavior accordingly.
“I’ve seen my car drive off the track a thousand times at this point," Nelson said. “But as the car learns, it will get better and better.”
There will be two tiers for the KU DeepRacer competition. The first event, held April 14, pitted four-person student teams against each other. Later in the semester, the students will compete individually in a series of socially distanced races, with a champion to be crowned during the final class of the semester.
KU students aren't just racing against each other. Using the 3D simulator, programmers from around the world compete in AWS' monthly DeepRacer time trials. The top 10% of finishers in those races can advance to the company's "Pro Division," where qualifiers have an opportunity to compete for the AWS DeepRacer League Championship Cup, a live event that will be held in Las Vegas in December.
“Every single person on my team has qualified for the pro division," Nelson said.
Williams said the DeepRacer program is giving students hands-on experience they can use when they enter the job market.
"It’s interesting, different and fun, so I think they’re more engaged," he said. "The other thing I look at — you have companies like Tesla, Google and Amazon working on cars and trucks that can drive by themselves. This is real practical experience they can put on their resumes."
Photo: Amanda Nelson, a junior in computer science.