Is the millimeter wave radar cancelled by Tesla "chicken rib" or "necessary"?

Not long ago, Tesla released the FSD Beta V9 system. Although the driver still needs to hold the steering wheel to maintain a high degree of vigilance, it is an important step for Tesla to move towards fully automatic driving. At the same time, this is also an important step for Tesla to officially move towards a truly pure vision automatic driving route, because Tesla officially announced the cancellation of millimeter-wave radar and changed to rely entirely on the camera to achieve assisted driving.

Since the rapid development of autonomous driving, there have been many route disputes, especially whether to use lidar. That is, the route of pure vision and multi-sensor fusion, which was discussed in detail when we talked about lidar before. However, only on the purely visual route, everyone's plan is different. For example, Baidu's mass production models with cameras, millimeter-wave radars and high-precision maps are also called pure visual routes, while Tesla's previous pure visual routes also used millimeter-wave radars.

This time, Tesla decided to "purify" the pure visual route and completely abandon the millimeter-wave radar. Millimeter wave radar, as the most common sensing hardware in automobiles, is it a "chicken rib" or a "necessity"? Why did Tesla cancel the millimeter wave radar?

Millimeter-wave radar is popular in cars for the same reason that millimeter-wave radar does not exist, because it is a mature industry. In the previous LIDAR program, we talked about the third generation Debonair model introduced by Mitsubishi in 1992, which was the first in the world to introduce the ranging function and was equipped with a fixed line-of-sight LIDAR. On this basis, Mitsubishi developed ACC adaptive cruise function in 1995. But in this function, millimeter wave radar is obviously better.

1999, Mercedes-Benz equipped the S-class of W220 with adaptive cruise function, and replaced the laser radar with millimeter-wave radar. The large-scale application of millimeter-wave radar in automobiles began. With the popularity of ACC adaptive cruise, millimeter wave radar is more mature and the cost is more controllable. In recent years, with the rapid development of autonomous driving, millimeter-wave radar has become the most common but non-existent sensing hardware.

Then why did Tesla abandon millimeter-wave radar? Musk did not explain the detailed reasons. Probably many people will think of "cost control". Mature millimeter wave radar is not particularly expensive, but it is also a thousand yuan level. Cancelling millimeter wave radar is still very helpful to reduce the cost. But reducing costs is by no means the only reason.

In fact, there has always been a saying in the market: "Millimeter-wave radar, as an outdated technology, will eventually be eliminated in autonomous driving", because millimeter-wave radar does have many obvious shortcomings.

In radar, the basic electromagnetic wave bands are divided into meter band (HF, VHF, UHF band), decimeter band (L, S band), centimeter band (C, X, Ku, K band), millimeter wave band and laser band, among which millimeter wave band has the least application field. Compared with electromagnetic waves with longer wavelength, millimeter wave radar has lower radiation power and higher internal noise and meteorological clutter. However, when used in automobiles, the accuracy of millimeter-wave radar is far less than that of laser radar camera, and the problems of interference and noise also exist.

Our common millimeter-wave radar used for ACC adaptive cruise has a certain angular resolution, and vehicles in different lanes are distinguished by multiple transmitting and receiving antennas. However, due to the cost, only the angular resolution of the plane is generally designed, and no distinction is made in the vertical direction, so it is impossible to judge the height of the identified target from the ground.

Tesla had two sensational accidents and crashed into a white truck with assisted driving. The final investigation results show that the vehicle recognized the white truck as the sky and clouds, so it did not respond to avoidance or deceleration. These two accidents are obviously related to the inability of millimeter wave radar to judge the height of obstacles in front of the ground. As the pure vision algorithm of Tesla's camera recognition becomes more and more mature, the significance of millimeter wave radar to Tesla is also decreasing. Announcing the cancellation of millimeter-wave radar is enough to show Tesla's confidence in its pure vision algorithm.

But is millimeter wave radar really just a chicken rib hardware that will eventually be eliminated? Obviously not, the advantage of millimeter wave radar is the unique advantage in the current autopilot sensing hardware.

The first is all-weather. Millimeter-wave radar can work normally in bad weather such as rain, fog and sand, and is completely free from the interference of environmental factors such as sunlight and alternating light and dark. The former has a great influence on lidar, while the latter will seriously interfere with the normal work of the camera.

Before the release of Tesla FSD Beta V9 system, the small-scale test of FSD Beta version suddenly slowed down abnormally when the light and shade alternated. Tesla official said that this phenomenon has been improved in the FSD Beta V9 system. We don't know how Tesla did it, but human eyes will be temporarily blind when driving in and out of the tunnel. At present, the camera is far less tolerant than the human eye, and the camera with lower pixel is used for automatic driving.

On the other hand, millimeter wave radar is the only sensing hardware in the car that can detect the distance and speed of the target object at the same time. Millimeter-wave radar can calculate the speed of the target object while ranging by using Doppler effect. Doppler effect was put forward by Austrian physicist and mathematician Christian John Doppler in 1842. When the sound source moves towards the observer, the receiving frequency of the wave becomes higher, while when the sound source is far away from the observer, the receiving frequency becomes lower. Therefore, the Doppler effect can be used to measure the relative speed of the target object and the vehicle, and the current speed of the target object can be obtained according to the speed of the vehicle itself.

Of course, laser, as a wave, can also be used as laser Doppler radar, but at present it is mainly used for atmospheric measurement and has not been applied in automobiles. We have introduced the ranging principles of lidar, such as time-of-flight method, phase method and triangulation method, but these can only identify the distance of the target object. If you want to know the speed of the target object, you need to calculate it through multiple sets of data. In ranging, cameras are far less accurate than radars.

With the above two advantages, millimeter-wave radar is bound to gain a place in the hardware of automatic driving perception. Even if Tesla of running all the way announced the cancellation of millimeter-wave radar with pure visual route, it could not shake the status of millimeter-wave radar. However, the limitations of millimeter-wave radar mentioned above are also objective problems. Now the industry is also developing new imaging millimeter-wave radar to improve the accuracy, but it has just begun to be applied because of the increase in cost. In the next program, we will talk about the basic principles of imaging millimeter-wave radar and millimeter-wave radar, and understand what is going on with 24GHz and 77GHz millimeter-wave radar. (Text: Pacific Auto Network Micco)