Going with the Flow
Morning rush hour traffic on U.S. Highway 101 in San Francisco. (Time-lapse photography by Adam Lau / Berkeley Engineering)
Berkeley researchers are addressing the emerging era of self-driving vehicles on multiple fronts, including a tool that uses machine learning to manage traffic where autonomous, partially-automated and manual vehicles share the road. Their project, called Flow, is the first time deep reinforcement learning has been integrated with traffic-simulation tools.
“Flow solves large-scale, multi-vehicle problems by providing cloud integration of industrial simulation software with state-of-the-art machine learning libraries,” says Alexandre Bayen, professor of electrical engineering and computer sciences and director of the Institute of Transportation Studies. “We’ve made it open source so the development community can continue to build on it, and we’ve created leaderboards so groups around the world can compete on the best performance of their algorithms.”
A novel feature of the system is automated cars using data from nearby smart vehicles or infrastructure to manage traffic, effectively becoming mobile traffic-managing robots. For example, an automated car could use its speed and position to control nearby vehicles as they merge. Or it could pace its speed to help prevent the random, human-caused slowdowns that increase travel time and frustrate drivers.
The study specifies highly detailed scenarios that engineers can use to solve traffic challenges like bottlenecks and intersection control. The solutions become shared baselines — benchmark scenarios on which researchers can compete — that are critical to making progress.
Researchers aim to tackle increasingly complex scenarios, with the end goal of having the system manage traffic at citywide scale. They also plan to study potential downsides and unintended consequences of this technological approach.