Stanford engineers test autonomous car algorithms in quest for safer driving
Understanding how an autonomous race car adjusts its throttle and brakes and makes use of the friction of its tires at high speed could inform the development of automatic collision avoidance software for the situations at the speeds at which most car crashes occur.
Go to the web site to view the video.
When Stanford’s autonomous car Shelley nears speeds of 120 mph as it tears around a racetrack without a driver, observers’ natural inclinations are to exchange high-fives or simply mouth, “wow.”
Chris Gerdes and his students, however, flip open laptops and begin dissecting the car’s performance. How many g-forces did Shelley pull through turns 14 and 15? How did it navigate the twisty chicane? What did the braking forces look like through the tight turn 5?
For the past several years, Gerdes and his students have been testing their autonomous driving algorithms with Shelley, a custom-rigged Audi TTS, on the 3-mile track at Thunderhill Raceway in Willows, California. Although the speedometer needle sometimes flies past 110 mph, the car spends a good deal of the course maneuvering at speeds of 50 to 75 mph. This is closer to the speed at which most car collisions occur, Gerdes said, and understanding how the car adjusts its throttle, brakes and makes use of all the friction of its tires in these situations could inform the development of automatic collision avoidance software.
“A race car driver can use all of a car’s functionality to drive fast,” said Gerdes, a professor of mechanical engineering. “We want to access that same functionality to make driving safer.”
Through careful iteration, the car is almost as fast around the track as an experienced racer. Several graduates from Gerdes’ lab have landed innovative careers in the auto industry. He prioritizes student-inspired research on track days and grants his students significant autonomy to manage their research, which touches on a broad range of subject areas.
For instance, graduate student Nitin Kapania combines learning control and artificial intelligence to help the car improve its performance over time, particularly related to how it steers through different corners. Another graduate student, John Kegelman, works to find ways to convert the skilled controlled behaviors of drivers into computer algorithms. Recent PhD graduate Joseph Funke analyzed Shelley’s maneuverability on the track under normal conditions to develop emergency lane-change algorithms.
“The students organize themselves like a race team for our test days. They figure out tasks and schedule the time allotted for different tests together,” Gerdes said. “But once you get to the track, things can go differently than you expect. So it’s an excellent lesson of advanced planning, but also how to come together as a team and deal with changing priorities.”