Cruise is testing computer vision and sound detection AI to help autonomous cars respond to passing emergency vehicles, a company spokesperson told VentureBeat in an email. More specifically, the systems are alert for the sound and sight of sirens to understand if a police car, fire truck, or ambulance is in motion.
The company is using emergency vehicle detection on all of the autonomous vehicles it’s testing on roads. It currently has nearly 180 autonomous vehicles in the field, primarily in San Francisco.
To train its emergency vehicle detection systems, Cruise used a combination of synthetic and real-world data. At the time of publishing, the company was unable to provide additional details about training data sets or more technical information about how an autonomous vehicle responds when it detects an emergency vehicle.
How an autonomous vehicle responds to an emergency vehicle could utilize systems Cruise currently uses to navigate chaotic double parking on the streets of San Francisco.
In a video provided to VentureBeat, Cruise senior engineering manager Sean Harris demonstrated the visual and sound detection systems in action. In the video, the vehicle pauses at an intersection and waits for an ambulance to drive by.
“For most vehicles, we don’t care about [whether[ they’re making noise or have lights flashing on their roof, but for an emergency vehicle that’s really important because we need to behave differently in this kind of active state,” Harris said in a presentation at the Computer Vision and Pattern Recognition conference (CVPR) held last week in Long Beach, California.
In California, drivers are legally obligated to yield to emergency vehicles, but with autonomous vehicles there’s more to consider than whether they pull to the side of the road if they hear sirens. Understanding the precise location of an approaching or receding siren can be challenging when sound bounces off tall buildings in a city, but it’s worth the effort required to solve this problem. As Harris notes — and many other autonomous vehicle companies surely know — emergency vehicle detection will be important for the safety of both emergency responders and passengers in cars.
In the years ahead, if autonomous vehicles become more prevalent on congested city roads and emergency vehicle detection systems are reliable, we could see a drastic decrease in emergency call response times. The Fire Department of New York City and New York University recently received a grant from Google.org’s AI for Good initiative to reduce emergency vehicle response times in congested urban areas.
Construction zones pose another complicated scenario for autonomous vehicles. To address this challenge, Waymo — and more recently Zoox — have been tuning their machine learning systems to recognize and navigate such sites.