Sign In

Communications of the ACM

ACM TechNews

Driverless Cars May Struggle to Spot Children, Dark-Skinned People

View as: Print Mobile App Share:
People may approach intersection with different levels of apprehension, depending on their age and skin color.

Some pedestrians may be less reliably detected by driverless cars than others.

Credit: Dorothy Alexander/Alamy

Scientists in the U.K. and China evaluated eight artificial intelligence (AI)-based pedestrian detectors used in driverless car research, and found they may have difficulty detecting children and dark-skinned people.

The researchers learned the detectors' accuracy identifying adults was nearly 20% higher than it was for children, and 7.5% higher for light-skinned pedestrians versus those with dark skin.

Jie Zhang at the U.K.'s King's College London said while automakers' software details are confidential, they are usually based on existing open source models, which "must also have similar issues."

Carissa Véliz at the U.K.'s University of Oxford said these problems must be corrected before deploying AI systems in cars on real roads, although engineers must ensure their remedies do not intentionally harm overall safety.

From New Scientist
View Full Article


Abstracts Copyright © 2023 SmithBucklin, Washington, D.C., USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account