Could Street-Sign Graffiti Interfere with Self-Driving Cars?

Did you know that the Pre-Collision Assist with Pedestrian Detection feature, available on many new Ford models, is based on research into self-driving cars? The quest for a Knight Rider-like autonomous vehicle has been ongoing for over a decade, and has resulted in amazing advances in technology.

But to make a vehicle reliable enough for the driver to read a book or take a nap behind the wheel, many potential issues must be addressed first.

What happens, for instance, when a kid thinks they're being funny by spray-painting a 30 MPH speed limit sign into 80 MPH? That sort of run-of-the-mill graffiti might be something machine learning can compensate for, but researches from University of Washington, University of Michigan, Stony Brook University, and UC Berkeley have created an algorithm that generates adversarial perturbations—markings that look like graffiti—specifically designed to interfere with a self-driving car's perception.

Automakers like Ford can now use this data to build more secure systems that can't be confused.

Categories: News


Nothing posted yet.