Simple road sign vandalism could confuse driverless cars

Simple road sign vandalism could confuse driverless cars
Driverless cars could be easily mislead into taking wrong turns and missing stoppings through basic tampering with road signs, researchers have warned.

Simple alterations – such as partially covering a sign with a sticker – could trick the vehicles’ visual recognition software into misreading information.

Based at the University of Washington, the researchers hope their discovery can lead to the development of better defensive systems in autonomous vehicles.

READ MORE: ‘Widespread scepticism’ over autonomous cars

Just yesterday the government issued new guidance ruling that engineers will need to develop more robust protection and ‘design out’ hacking to make cars ‘cyber secure’.

But it seems that even more conventional attacks could lead to the cars’ undoing.

The team demonstrated how simple alterations to signs – which would appear virtually the same to a human driver – had the ability to confuse driverless cars into potentially life-threatening mistakes, such as ignoring stop signs.

They warned that by playing games with learning algorithms, like those used in the recognition software, the cars could potentially “misbehave” in unexpected ways.

IN OTHER NEWS: Tech giants ‘struggling to build trust in driverless cars’

As a test case in the study, graffiti stickers reading “Love / Hate” were placed over a stop sign. This caused the vehicle in question to misread the sign as one instructing of a 45mph speed limit.

In another test, subtle changes to the colour of a right-hand turn sign caused it to be mistaken for a stop sign – an error which would not have been made by a human driver, the team added.

“Both of our attack classes do not require special resources – only access to a colour printer and a camera,” the researchers said.

While certain cars already on the market are able to read and detect road signs, including the Autopilot feature on Tesla’s Model S electric cars, no vehicles have yet been programmed to successfully react to them.

Copyright Press Association 2017. Motoring News articles do not reflect the RAC's views unless clearly stated.