How ironic: Camera-clad self-driving cars can be fooled by photographs

Cameras are a key piece of the technology used in self-driving cars and human-driven cars with features like auto braking – but just how hard is it to fool a camera-clad car? Ironically enough, self-driving car cameras can be fooled by photographs themselves, as YouTube and engineer Mark Rober demonstrated in a crash test inspired by the cartoon antics of Wile E Coyottee.

In the crash test, Rober creates a series of tests designed to determine what a self-driving car can and cannot see, putting a Li-DAR based car against one that uses cameras to power the self-driving features. The series of tests that included everything from fog to bright headlights culminated in a test inspired by Wile E Coyote himself: Will a self-driving car crash into a wall if a picture of the road is printed on it?

Here’s what happened when the Li-DAR and camera self-driving cars come to a photograph of a road stretched across the road:

Can You Fool A Self Driving Car? - YouTube

Watch On

The likelihood of such a photograph being placed over the road in a real life situation is slim to none, but the crash test illustrated important differences in self-driving cars that use cameras rather than LiDAR. The LiDAR technology sends out pulses of light to measure distance rather than using cameras as “eyes” on the road.

Wile E Coyote may be a fictional scenario, but in Rober’s tests, the LiDAR-based car also passed the fog and heavy rain tests that the camera-based car did not. The LiDAR was able to detect the wall as a wall, ignoring the picture printed on it, but also saw through the fog and rain better

In the first half of the video, Rober used LiDar to map out the in-the-dark Disney World ride Space Mountain, so in addition to making safer cars, LiDAR can apparently also be used to take away some of the magic and mystery of Disney.

You may also like

Browse the best dash cams or the best cameras for YouTube.