How does the computer in a driverless car that understands only 1’s and 0’s ‘see’ a pot hole on the road, or the bumper of a car in front of it, or an unexpected obstruction?
Over the last few years autonomous cars have become the most coveted invention and manufacturers are racing against time and spending billions to roll out a perfect working model. As the complexity of technology in an autonomous car grows, newer breakthroughs emerge.
One such technological marvel is LIDAR: LIght Detection And Ranging. Now LIDAR isn’t a new technology. For instance, it is already being used in spacecrafts to calculate distance and in archaeology to map out topography. But LIDAR turned out to be a game changer in autonomous cars and yes is one answer to ‘how do cars see?’ In fact LIDAR technology is the reason Google’s Wyamo and Uber are at legal war: Google claims Uber stole their LIDAR system designs.
LIDAR’s operation is similar to that of a RADAR system, only difference being LIDAR uses light pulses instead of radio waves. A LIDAR system transmits laser pulses and designs a 3-d map of its surroundings, which along with GPS and cameras helps the autonomous car navigate. LIDAR systems can be mounted on the roof of a car (something you have definitely noticed if you’ve seen a driverless car), where it rotates to get a 360 degree view.
LIDAR trumps other technologies?
LIDAR, RADAR and Cameras are currently the three technologies that manufacturers prefer to render 3-d maps of the autonomous car’s surroundings. LIDAR is the preferred choice of technology for most manufacturers, like Google’s Waymo, Toyota and Uber. Tesla Inc. is an exception that uses only cameras and RADAR to map its surroundings.
There however isn’t a correct answer for which is the best technology. For example Tesla’s self-driving car crashed into a truck in 2016 killing the driver, because the on-board cameras failed to differentiate between the truck and the bright sky. Cameras work by rendering complete images of its surrounding (much like your selfie camera) and then differentiates objects. If colours of an obstacle and a safe surrounding blend, like it did in Tesla’s case, the computer cannot identify the threat.
Similarly researchers claim a LIDAR sensor can be ‘blinded’ or tricked into believing an obstacle isn’t really there. Engineers at Korea Advanced Institute of Science and Technology have experimented and demonstrated two types of attacks a LIDAR system is susceptible to: Spoofing attack, and a saturation attack.
In the spoofing attack, researchers tricked the LIDAR system into miscalculating the position of an obstacle. A LIDAR system does not capture and image entire objects like a camera. Instead it reads the returning laser beams to get point images of an obstacle and the computer decides the next course of action. By introducing an object that could emulate these point images and by exploiting the LIDAR systems glass housing, researchers were able to trick the driverless car into misjudging the location of the obstruction.
In a saturation attack, by shooting the LIDAR with light waves of the same wavelength as its laser beam, researchers were able to ‘wipe out’ the LIDAR’s existing rendered map, effectively blinding it.
So, back to square one?
Not really. While neither one technology is perfect, when used together they become a formidable team. By coupling input from the three sensors (LIDAR, RADAR and cameras) along with high quality GPS systems, a car can become truly autonomous, where the driver can take a nap or better, doesn’t even have to be present!