In 1969, the world-class comicbook company DC Comics was in the red, on the fence of shutting down. It was so bad that its new corporate master nearly took the drastic, permanent action of shutting it down.
It seemed that after years of battle with Marvel, which involved ideas theft, poached employees, spies, and even price wars, the former 800-pound gorilla of the comicbook industry was losing millions and going nowhere.
In a 2000 interview, you can read this: "Marvel was doing very well. We knew it because DC — independent news — was handling Marvel at the time and their numbers were coming in. Marvel had books like Spider-Man coming in at 70, 80, even 85 percent sales. And we had books coming in at 40, 41, 42 percent. Something was wrong, and [DC's executives] didn't know how to fix it."
So why did DC Comics survive? The answer lies in the rivalry. There is never a single winner, and Marvel and DC being rivals also boosted and helped create an entire industry. Along the years, Marvel and DC conceived this industry, through wars, but also collaborations like Spiderman vs Superman, and even price agreements. This rivalry helped DC not only survive, but it forced it to innovate.
When looking at something, it's easy to see it as black and white. But when you look at things from another angle, you start seeing an entire spectrum. Would DC have survived all these years alone if Marvel didn't exist? Watching from this angle, the answer is... probably not.
It's similar with the LiDAR industry. In the recent years, we've seen a lot of talking around Light Detection And Ranging, but it's often been one direction: Solid-State LiDARs are replacing mechanical LiDARs. FMCW LiDAR technology is the future, and replaces the traditional Time Of Flight. LiDAR Point Clouds are processed with Deep Learning, and no longer with traditional approaches.
The reality is much more nuanced, and this is what I'd like to show you in this article. Just like for Marvel and DC, things aren't black or white. There's an entire spectrum, and we'll learn all about it in this article. In particular, you'll learn:
- The different Scanning Methods of LiDAR instruments
- The different Dimensions and resolutions a LiDAR can produce
- The types of Modulation a LiDAR system can use to measures distances
So let's begin:
The LiDAR Types by Scanning Method
There aren't just 3, but I think 3 is a good start. We're going to see:
- Mechanical Scanning (Rotating/Spinning)
- Solid-State (fixed)
1) Mechanical Scanning —— Rotating
The most popular and widely adopted LiDAR has been for a long time the Velodyne HDL-64E, that uses a spinning 64-layer laser detector (64 horizontal planes) to cover a vertical angle of 27°. It spins at 360°, at a frequency of 10 to 30 Hz. This LiDAR can see at around 120 m range.
Let's take it from the beginning. LiDAR technology uses light to measure the world and generate a point cloud. For that, a LiDAR typically has an emitter, that will send a light wave, and a receiver, that will receive the bounced wave.
But how do you get a 360° point cloud? Well, you do that by rotating your emitter super fast, in a 360° way, to get a Point Cloud.
Did you see how fast it's going? The first time I saw this, I was amazed, and even more when you see how it uses this process to produce entire point clouds. We'll come to this in a minute, but if you arrived here and have no clue what a point cloud looks like...???... just go read this first.
You may have noticed, this LiDAR is moving, and this introduces a risk of failure, as well as a higher cost (back in 2017, these types LiDAR systems could be sold at 100k$). With that, they operate at 905 nm, which is good to see at long distances, but has a risk of damaging the human eyes.
Which is why we introduced a solid-state LiDAR systems, with no moving parts.
2) Solid-State lidar systems
In a rotating LiDAR, a laser beam is steered at 360° by a mechanical system. But in a solid-state LiDAR we don't rotate, so we must introduce a different way to cover the entire scene.
The most commons are:
- MEMS (Micro-electromechanical system) — A MEMS based mirror moves to “scan” the environment. This is what the self-driving car industry uses in replacement of rotating LiDARs these days.
- OPA (Optical Phased Array) — The truly solid-state solution; an array is adjusted and deliver light pauses to different directions. This is the least deployed technology, and hasn't been tested in self-driving cars at scale (to my knowledge).
Still, below is an image of the different types we just covered:
You can find a more hardcore explanation in this article where I found the image, but I think you can already get the main ideas.
- In rotating systems, we have horizontals laser beams, and we rotate around a vertical axis to cover a 360° scene.
- In MEMS (micro-electromechanical systems), we use mirrors to deflect light, and thus cover a bigger angle.
- In Optical Phase Arrays (OPA), we have arrays of closely spaced (around 1µm) optical antennas that radiate light in a broad angular range. By adjusting the "phase" of the light, we can then control the direction of the beam and thus "fake" the rotation.
A Note — While MEMS and OPA offer a solution to the rotating LiDAR, they can't cover a 360° view, which is why you'll often see self-driving cars with several LiDARs mounted on them.
This brings us to the last type of LiDAR, the Flash LiDAR.
3) Flash LiDAR systems
Finally, one of the most interesting types of LiDAR is the Flash LiDAR. In effect, it operates like a flash camera, by expanding a laser beam to illuminate the scene. Unlike a scanning LiDAR, we don't construct the point cloud point by point, nor layers by layers, but by generating the entire 3D point cloud in one flash.
It's basically functioning like a digital camera: the target scene is divided into squares (like pixels), but rather than measuring the light intensity, we measure the time of flight.
The idea is fantastic, but because of the nature of laser, it doesn't really work like a camera. The resolution is (today) very poor, and it doesn't really work at more than 10 meters.
Here is the result of a company named LeddarTech and their product (named Pixell): (source)