A setup using only a laser and a laptop or tablet could be used to stop self-driving cars in their tracks, according to security researcher Jonathan Petit. By firing the laser in pulses using a pulse generator or computer, Petit is able to fool the lidar self-driving cars use to detect obstacles. He can create a false image of a vehicle, pedestrian or wall.
How It Works
The lidar used by self-driving cars sends out pulses of light that bounce off objects around the vehicle. A sensor detects the returning pulses and uses them to construct a 3D image of the area. A laser emitting pulses at the same frequency can fool the sensor into thinking its own pulses are bouncing off an object. This makes the car “see” something that isn’t really there.
The laser can be cast in a wide beam, and it doesn’t need to precisely target the lidar sensor. However, the pulses would have to be programmed for that specific model of lidar, as different companies use different frequencies. With the right programming, Petit can simulate a wide variety of situations. A single object in front of the car could make it change lanes to avoid the obstacle. Petite can also place numerous objects surrounding the vehicle to prevent it from moving at all.
This tactic has yet to be tested in a real-world situation, but Petit has demonstrated the ability to fool an IBEO Lux lidar unit. It’s also worth noting the laser could be used to blind human drivers even more easily than automated ones.
Fortunately, the danger of physical harm is fairly minimal. The vehicle will simply come to a controlled stop, just like it would if a pedestrian walked out into the street. However, a stopped vehicle does present a potential hazard for other drivers. If another driver isn’t paying attention, it could easily lead to an accident.
On the whole, self-driving cars are still significantly safer than those driven by humans. Eighty-one percent of deaths from crashes are caused by human error, and despite the fact that Google’s autonomous cars have already driven more than some people will in a lifetime, the vehicles have yet to be responsible for a single accident.
However, current lidar systems are alarmingly easy to spoof. Petit’s device only cost $60 to make and was able to fool a scanner that costs thousands. The pulses of light used by lidar aren’t encrypted or protected in any way. Conversely, short-range radar operates on a specially licensed radiation band. Even obtaining a device that projects the right kind of energy is more difficult than assembling Petit’s laser pulse emitter.
How We Can Fix It
There are a number of potential solutions to this vulnerability. Lidar scanners could introduce constant variations in frequency that would be much more difficult to reproduce. Vehicles could also cross-check the lidar data with information from another type of scanner, such as radar. Fooling two detection systems at once is significantly more difficult, and with two scanners using entirely different technologies, the task becomes incredibly daunting.
Improvements in existing laser technology could also make lidar less vulnerable to attack. One new type of laser system combines a mechanical oscillator and a surface-emitting laser into a single unit. This would allow the laser to be modulated more quickly, with a difference of 100,000%.
For now, Petit is hoping to use this discovery to make the creators of lidar scanners and self-driving cars more conscious of potential vulnerabilities in their systems. He will be presenting his findings in full at the Black Hat Europe conference in November.
I usually send out a weekly mailer with a recap of blog posts but also some personal anecdotes. If you want to know about competitions or just catch up in a more personal setting then you might like to receive the mailer.