Which LiDAR is the best? A must-read guide for buying new cars with built-in LiDAR!

Mass-produced LiDAR cars are here! Recently, several mass-produced cars with pre-installed LiDAR have been released, which has left many users confused - why do we need to install LiDAR on cars? What is its use? How to determine whether the LiDAR is good or not? Is it safe for people outside the car? And so on. In order to give everyone a reference when choosing a new energy vehicle equipped with LiDAR, Zhichepai ​​will popularize the things about LiDAR with everyone today.

Livox HAP invested by DJI

Why install lidar when we already have millimeter-wave radar and cameras?

As autonomous driving is getting closer and closer to us, more and more sensors that assist autonomous driving are being used in cars. For a mature L2.5 ADAS, the standard sensor configuration is based on 12 ultrasonic radars, 7 millimeter-wave radars, 4 surround-view cameras, and 1 front-view camera. If you want to reach L3 and above, you need 12 ultrasonic radars, 9 millimeter-wave radars, 5 surround-view cameras, 2 front-view cameras, and 5 lidars.

Various sensors work together to achieve autonomous driving

With so many sensors, why do we need to add LiDAR? The reason is that the existing sensors cannot cope with emergencies caused by driving in all weather and all scenarios. Millimeter wave radar is all-weather, but its image formation ability is too weak; ultrasonic radar has a short detection distance; the camera cannot see when the light is bright and dark... In actual road conditions, any small mistake will cause losses. So we need to make up for what is missing. Since the existing sensors cannot cover all scenarios, we can add another one - LiDAR.

What are the advantages of lidar?

The biggest advantage of LiDAR over cameras and millimeter-wave radars is that they can be touched. To use an analogy, you see the bread with your eyes, and the sensor in the car is the camera, which can see it no matter how many sensors it has. Although you see it, is the bread hard or soft? At this time, you have to touch it with your hand, and this hand is the LiDAR. In fact, millimeter-wave radar and ultrasonic radar are both hands, but these hands have shortcomings, perhaps too short or not strong enough.

LiDAR can detect and sense objects in front through laser beams, and can achieve fast composition, especially in terms of the accuracy of constructing the outline of obstacles and performing better in low-light scenes. For example, when a car enters a tunnel under the sun, the camera will be "blind" for a short time, even if the time is short, it is dangerous enough. LiDAR will not have this problem. When facing the same scene switching, it can effectively minimize the risk (the risk is 0 and cannot be determined).

How to evaluate the quality of LiDAR?

There are many parameters of LiDAR. Generally speaking, it is enough to know the following parameters: number of beams, FOV, detection distance, angular resolution, and point cloud density. So what do these parameters mean?

The number of beams is how many laser beams the LiDAR emits. We can use a face mask to understand the benefits of the number of beams for the LiDAR.

Compare the entire mask to all the light emitted by a laser radar. What will you find after applying it? The outline of the face is revealed. Although you can't see what it looks like, you know it is a face. But what if the entire mask is cut into 1 cm wide strips? Whether it is applied horizontally or vertically, only the outline of the part where it is applied can be seen. There is no outline in other places, so it is not clear that it is a face.

The more lines you have, the more accurately you can depict the outline of an object.

This makes it easy to understand. LiDAR constructs the outline of obstacles by scanning light, so one laser beam is definitely not enough. Only multiple laser beams can restore the outline of obstacles. However, the higher the number of laser lines, the more expensive it is. Currently, there are not many cars equipped with LiDAR on the market. The most recent ones are ARCFOX αS HBT and Xiaopeng P5. The former has a LiDAR of 96 lines, and the latter has a point cloud density equivalent to 144 lines (it does not directly say how many lines), which means that the transmitter can emit 96 laser beams.

FOV: FOV is the range of your eyes that can be seen from top to bottom and left to right. The larger the FOV, the wider the field of view. For example, the Huawei 96-line laser radar built into ARCFOX αS HBT has a horizontal FOV of 120°, which means that the horizontal direction adds up to 120°; the vertical FOV can reach 25°, and the same principle applies to horizontal direction.

Small angle and narrow field of view

The number of lines and the size of FOV directly affect the accuracy of contour depiction, that is, the resolution. Of course, there are more than just these two factors that affect the depiction accuracy. The scanning frequency is also one of the key factors. I will talk about this later. Next, let’s talk about resolution.

Angular resolution: As mentioned above, in theory, the more line bundles there are and the larger the field of view, the better the accuracy and breadth of contour depiction (limited by technology and price), which determines the resolution.

Angular resolution is divided into horizontal resolution and vertical resolution. For example, the Livox HAP laser radar on the Xiaopeng P5 has a horizontal angular resolution of 0.16° and a vertical angular resolution of 0.2°. If this parameter is not marked, you can calculate it yourself. Divide the number of lines by FOV to get the angular resolution, but the result you get will definitely not be this. Let's use Livox HAP to calculate. The vertical FOV is 25°÷144, and the result is 0.174°. The nominal vertical angular resolution is 0.2°. Who ate the difference of 0.026°?

The answer is no one eats it, because this number is also related to the density. Our algorithm is based on the uniform distribution of 144 laser beams, but this is not the case in reality. The actual situation is that the laser beam density in a certain area is higher. Manufacturers mark the part with the highest density when they mark the nominal value, so the uniform 0.174° becomes 0.2°. Even so, it does not affect the comparison. If you know the algorithm, you can calculate it yourself when the time comes.

Detection distance: There is no need to use an analogy for this, and even the smart car group is quite tired of it. Here I want to share with you a "mine" in the parameter nominal. You must pay attention to it when asking or listening to others' introduction.

The detection range of Huawei's 96-line laser radar built into ARCFOX αS HBT is 150m @10% and 200m @50%. What does this mean? That is, when the reflectivity of an object is 10%, the farthest detection distance is 150 meters. For example, if there is a large piece of black sponge in the middle of the road (black has a low reflectivity), the car can depict this obstacle within 150 meters. If there is a large piece of white sponge on the same road, the car can depict this obstacle within 200 meters.

However, the parameter tables of some laser radars do not indicate how many meters the range is under a certain reflectivity. Similarly, when purchasing a car with a laser radar, if the salesperson says that the laser radar can "touch" as far as 500 meters, you must ask him what the reflectivity is for 500 meters. If the salesperson cannot answer, then he is unqualified and thinks he can make sales by fooling people. Ignore him.

Point cloud density: also called the periodic acquisition number, in layman's terms, it is the number of points that depict the contour in one period. The higher the density, the more accurate it is.

How is the point cloud density calculated? For example, a 64-line LiDAR has a horizontal FOV of 120° and a horizontal resolution of 0.2° at a scanning frequency of 10 Hz (10 scans per second). The point cloud density of this LiDAR is 64×120÷0.2×10=384,000. The result of this number depends on the scanning frequency at which the corresponding resolution is generated.

The higher the density, the more accurate the depiction of the object.

We have all seen machine gun fire in war movies, right? This point cloud density is similar to machine gun fire. The faster the fire, the more holes are made in the target. It's the same principle.

There are a few points to remind you:

With some understanding of the above data, you will have a better idea when choosing a car equipped with laser radar, but you should pay attention to a few points:

Xiaopeng P5's LiDAR location

1. The salesperson may tell you that the horizontal FOV can reach 150°. At this time, you should ask whether it is 150° for a single laser radar or the horizontal FOV of two laser radars added together. Zhichepai ​​will clearly tell you that it is most likely 150° for the two radars added together;

2. It doesn’t mean that a car equipped with a LiDAR is necessarily powerful, because if the car’s computing power is not enough and the algorithm is not strong, even if it is equipped with a LiDAR, it is just a decoration;

3. Automotive-grade lidar must at least meet Class 1 standards, so there is no need to worry about safety.