LiDAR. Automakers are buying into it with huge sums of cash, Telsa is shunning it, Ford has invested all kinds of money in it, and it’s generally considered to be one of the more important parts of the self-driving car picture. But what the h*ck is it? And how does it work?
It helps, though, to talk about radar first. If you’ve ever been on a plane, or walked through an automatic door, or gotten a speeding ticket, you’ve experienced radar. Think of radar as the grandfather of lidar. It stands for RAdio Detection and Ranging, which is a complicated way to say that it uses radio waves to determine where things are. Radio waves bounce off an object, and the time it takes for them to come back, plus the frequency they’re oscillating at when they arrive tell you where that object is and how fast it’s moving. Or how fast you’re moving.
LiDAR, on the other hand, doesn’t stand for anything at all. Not originally, at least, although you’ll hear people now say that it stands for LIght Detection and Ranging. Instead of radio waves, LiDAR uses, you guessed it, light.
Radar is great for detecting big objects, like the wall behind you when you’re backing up. But it’s not good with smaller objects and higher resolutions. Seeing a cyclist from far away or picking out an arm from the rest of a person is too tricky for radar. S automakers are learning how to tell computers to look at a picture from a camera to “see” what is in front of it, but it’s a long and slow process to turn a 2D into the 3D that a car needs. LiDAR, on the other hand, is very precise and gives an exact point measurement to everything around it. Detailed enough for the computer to tell what it is seeing.
Now that you’ve got the general idea, let’s dive into how it actually works.
LiDAR uses a laser to generate pulses of light. Lots of them. The laser fires out hundreds of thousands of light pulses every second (but don’t worry, these lasers use a wavelength that is safe if it happens to catch you in the eye. You won’t even notice it). Those pulses of light are shot out in every direction by a rapidly spinning mirror. That’s why you’ll see the spinning cans mounted high up on an autonomous vehicle. Those are the LiDAR sensors.
And once that light hits something it bounces back. Kind of like radar. Even more usefully, though, much like the colours you see, light bounces off of stuff and you can see it. Along with the laser, there is a photodetector. That’s the device that reads the bounced-back light and records it.
Self-driving cars use several of these sensors, sending out massive numbers of laser pulses. The computer calculates each one of the pulses to find out how far away it was, and from which direction it came. Each one of those becomes a dot in a 3D picture. Every dot is accurate to within millimeters, letting you know precisely where everything is located around you.
A bunch of dots (called a point cloud) doesn’t sound like it would tell you where anything was. Much like a Seurat painting, though, a bunch of dots fills in nicely to make a complete picture.
In that point cloud, it’s easy to see the road, the curbs, the signposts, the buildings, the pedestrians, and the dogs that are in or around the street. When you compare multiple point clouds, your picture turns into a video. The car already knows how fast it is moving, but now it can tell how fast objects are moving. And how the objects are moving relative to the car.
It can tell that the bundle of dots on the shoulder is a pedestrian. By monitoring how that shape is moving, it can tell if the pedestrian is moving toward the road or away from it. It can tell if they’ll intersect with the car’s path, or if they’ll maintain a safe distance.
Two of the biggest challenges for developing autonomous driving are making the sensors cheaper, and then figuring out what to do with the data. The first is well underway, and the LiDAR system has dropped from what was $100,000-a-piece to somewhere around $4,000 each today. And that’s expected to keep on falling. The sensors can see further and further away, too. Initially, distance was somewhat limited, but new sensors can see hundreds of meters.
What to do with the picture is the hardest part. If you look at a picture, even one that’s just dots, you can probably tell what’s happening. What’s in the picture, where everything is, and with multiple pictures, you can even tell where everything is moving.
Telling the computer how to do that is still tough, though. That’s the part that automakers are spending tens of billions of dollars to figure out. And since they haven’t figured it out, I’m not going to take a stab at it.
The tl;dr here is that LiDAR uses lasers to fire beams of light everywhere. Then it uses a sensor and a computer to turn those spots of light into a picture. And that’s what the h*ck LiDAR is.