The success of automated driving technology hinges on rapid and accurate awareness of the surrounding environment to assess risk and anticipate people’s behavior. The still-developing field of automotive perception is about to jump into the fast lane with the application of machine-learning technology called deep learning. Trevor Darrell, professor of electrical engineering and computer science and the new co-director of California PATH, has formed the PATH-based Berkeley DeepDrive Industry Consortium, a multidisciplinary research alliance focused on applying state-of-the-art computer vision and machine-learning technologies for automated and assisted driving systems.
“When computers can recognize their surroundings, it makes everything safer,” says Darrell. “This research makes cars safer for people on the inside and for pedestrians on the outside.” Darrell is an early leader in the field whose code Caffe is one of the most widely used deep-learning frameworks for vision and has been broadly adopted by major Internet companies, startups, and in academia. The consortium will focus on deep-learning research using minimal computation resources to quickly and accurately recognize pedestrians, anticipate pedestrian behavior, and detect and classify objects, surfaces, and signage. The group’s emphasis will be on research, implementation, and real-world demonstrations.
In addition to PATH, the consortium includes faculty and researchers from the Department of Electrical Engineering and Computer Science, the Center for Information Technology Research in the Interest of Society, and the Berkeley Vision and Learning Center. Berkeley DeepDrive also has support from numerous automotive industry manufacturers and suppliers, who will have early access to the technology.
Deep-learning expert Trevor Darrell was appointed co-director of PATH in 2015. He’s a professor of electrical engineering and computer science, with a robotics and artificial intelligence focus.