Driverless AVs race to the future but new studies find flaws

There have always been conflicting views on how autonomous vehicles could help – or hurt – in the climate change fight, but there’s another problem too: Are the driverless (AV) vehicles smart enough?

Two new studies suggest the technology may still have a way to go before the vehicles can be trusted, and both of them indicate that it’s far too easy to trick the machines and their artificial intelligence.

The first comes from the scientists at Ben-Gurion University of the Negev in Israel, with support from an independent Tesla researcher; Tesla is, of course, a pioneer in producing the self-driving cars and trucks, although Volvo, BMW, Renault and other tech and automotive companies are working to create them.

What the Israeli team found, though, is that it’s far too easy to force driving errors by simply creating phantom images. By using an inexpensive image projector and commercial drone, they said, they were able to project “depthless” images that tricked camera sensors designed to tell the difference between 2D and 3D images.

The vehicles braked when they saw a phantom human in the road. They also responded to fake traffic signs that were beamed onto the surfaces of trees or billboards, and were deceived by fake lane markings beamed down from a drone. The flaws all point to security risks the researchers say aren’t being considered by system designers, and that could be used in targeted attacks.

“These are not bugs or poor coding errors, but fundamental flaws in object detectors that are not trained to distinguish between real and fake objects and use feature matching to detect visual objects,” says Ben Nassi, the lead author of “Phantom of the ADAS” and a doctoral student at Ben Gurion, as well as a former Google employee.

The second study comes from international cybersecurity firm McAfee, which looked at the same MobileEye camera technology used in the Advanced Driver Assist Systems (ADAS) of some 40 million vehicles, including some Tesla models. In one experiment shown in the video below, they stuck black electrical tape on a 35 mile-per-hour speed limit sign, merely extending the middle line on the number 3.

“Even to a trained eye, this hardly looks suspicious or malicious, and many who saw it didn’t realize the sign had been altered at all,” the researchers said. “This tiny piece of sticker was all it took to make the MobilEye camera’s top prediction for the sign to be 85 mph.”

Both studies focus on safety and security risks, and they acknowledge that fully autonomous driving systems remain an evolving technology field. Yet they also underscore the challenges in creating the connected vehicles, roadways and smart cities that advocates hope will one day help reduce carbon emissions.

On the one hand, that future could lead to 80 percent emissions reductions if the autonomous vehicles are used in shared electric-vehicle systems that lead to less traffic and less miles on the road. On the other hand, as the Union of Concerned Scientists pointed out last year, the driverless vehicles make matters worse if they simply lead to more cars, more trips and more inequality in access to clean transportation.

“Smart policies are critical for ensuring self-driving car technology ushers in a new era of clean, affordable, and efficient transportation rather than the zombie car apocalypse,” they warned. “AVs may be able to drive themselves, but it is up to us to steer them in the right direction.”

And that process, they all note, still demands a lot of work.