Welcome to the WeChat subscription number of “Sina Technology”: techsina
Text | Yang Jing
Source: Qubit
Musk, a 100% pure visual believer, has now thrown away the last radar.
Today’s Tesla sensor solution has only 8 cameras. In other words, whether it is L2-level assisted driving or FSD with unmanned driving capabilities, all rely on these 8 cameras.
And what he threw away was the king of cost-effectiveness among the Three Musketeers of Radar – Ultrasonic Radar.
Its price is only tens of yuan, and the deployment of a complete set of autonomous parking system is only 500 yuan, which is insignificant compared with millimeter-wave radar and lidar.
Some people even put it bluntly: ultrasonic radar has only advantages and no disadvantages.
Combined with the last time Musk threw away the millimeter-wave radar, netizens couldn’t sit still:
The “ghost brake” is still unsolved.
But even so, Tesla’s pure vision solution is still advancing. And the radar “appendix” that Musk once denounced has now been cut clean.
Ultrasonic radar with a price of only tens of yuan
Let’s take a look at the protagonist that was thrown away this time – ultrasonic radar.
In fact, compared with lidar and millimeter-wave radar, which are more often mentioned in autonomous driving, the application of ultrasonic radar in automobiles is already very mature, and it can be called “the king of cost performance”.
Its working principle is very simple, as the name implies, it measures distance by calculating the time difference between transmitting and receiving ultrasonic waves. The detection range can be between 0.1-3 meters, the accuracy can reach 1-3 cm, and it is waterproof and dustproof, and it can be detected even if it is blocked by sediment.
Usually, ultrasonic radars are arranged on the front and rear bumpers or on the side of the car. The former is used to measure front and rear obstacles, so it is also called UPA; the latter is used to measure side obstacles, which is called APA in the industry. The distance is farther, usually between 0.3-5 meters.
In smart car applications, there are mainly scenarios such as autonomous parking, assisted driving, automatic emergency braking, and assisted lane change.
Taking autonomous parking as an example, a set of reversing solution needs to be equipped with 4 UPAs in the rear bumper of the car; autonomous parking is based on this, and then equipped with 4 APAs and 4 UPAs in the front bumper.
The reason why it is called the king of cost performance is because it has such an application and its cost is very low.
On the market, the price of a single ultrasonic radar is only tens of yuan, the reversing plan may be more than 200 yuan, and the whole set of autonomous parking plan is almost 500 yuan – and this is still the general price, and a large mass-producer like Tesla can also have it. Lower cost supply price.
Since Tesla started the delivery of mass-produced models, it was initially equipped with the fifth-generation ultrasonic sensor provided by Bosch, with a minimum range of 15 cm (accurate measurement) and a minimum target detection of 3 cm.
Since then, Valeo’s ultrasonic sensor has also been used, and it is said that the maximum detection distance can be achieved 8 meters. And the number of ultrasonic radars stabilized at 12 very early.
However, ultrasonic radar also has certain limitations. The most significant is the propagation delay of ultrasonic waves.
The propagation speed of ultrasonic waves is slow and the error is large. Unlike the other two radars whose wave speed is the speed of light, the ultrasonic radar’s propagation speed is about 340m/s, and it will change with the ambient temperature.
If this is placed in the process of high-speed driving of the car, it is impossible to perceive the change in the distance of the car in real time. Moreover, ultrasonic waves are mechanical waves, which require a medium for propagation, and are easily affected by the surrounding environment, especially the weather.
In addition, there is also the problem of poor directionality. A single radar cannot accurately describe the exact location of obstacles, requiring more equipment to cover the same area.
As far as the entire industry is concerned, according to statistics from China Business Intelligence Network, the penetration rate of reversing radars of models currently on sale in China is close to 100%; the penetration rate of automatic parking systems is expected to reach 50% in 2025, and the overall market size is expected to be ) will grow to 5.84 billion yuan.
At present, the mainstream solution on the market is a solution equipped with 12 ultrasonic radars. According to Guosen Securities data, the penetration rate of this scheme will increase from 9.6% in 2019 to 26.1% in 2025.
However, Musk, who believes that autonomous driving should be driven by eyes just like humans, has never been optimistic about the autonomous driving solution that relies on radar. He once said wildly that “everyone who uses lidar is a fool”, and compared them to the long appendix on the human body. , showing his disgust.
Now, the two “appendixes” of ultrasonic radar and millimeter-wave radar have finally been cut off by Musk.
Tesla is pure visual now
With only 8 cameras left, Tesla is really going to a pure vision solution.
According to the official announcement, starting this month, the Model 3 and Y manufactured and delivered to North America, Europe, the Middle East and Taiwan will adopt this solution, no longer equipped with ultrasonic radar, and the Model S and X will be delivered globally in 2023. The same is true.
For now, however, the status of the China-made Tesla is uncertain.
One possible reason is that with the strengthening of my country’s smart car-related supervision, actions such as removing inherent sensors need to be reported and approved, and can only be carried out after being approved by the regulatory authorities.
Tesla’s announcement also pointed out that vehicles upgraded to the “Tesla Vision” system will face a short transition period after the ultrasonic radar is removed.
During this period, the ability to park assist, autonomous parking, Smart Summon, etc. will be affected.
Of course, the ultrasonic radar of the sold vehicle and its corresponding functions will not be deactivated.
In response to this operation, many people may think of the many ghost braking incidents that Tesla added after removing the millimeter-wave radar last year, and they can’t help but worry about what will go wrong this time.
However, Tesla has specifically stated that the removal of ultrasonic radar will not change Tesla’s crash safety rating.
In this year’s CVPR, Tesla announced a new algorithm: Occupancy Networks, which is said to be able to understand how much space the objects around the body occupy, and to predict the next trajectory and occupied space, instead of radar perception to avoid the blind spots of pure visual solutions (Corner Case) and so on.
Multi-sensor fusion or pure vision?
Undoubtedly, compared to Tesla’s pure visual idea of ”sword taking a slant”, many domestic car companies still insist on multi-sensor fusion, and neither cameras nor radars are left behind.
For example, Xiaopeng, its latest flagship model G9 is equipped with 2 lidars, 12 ultrasonic sensors and 12 cameras.
Another example is the Weilai ET7, which has 1 ultra-long-range lidar, 5 millimeter-wave radars, 12 ultrasonic sensors and 14 cameras.
Now Musk has thrown away the last ultrasonic radar and realized the ultimate pure vision solution, which once again triggered discussions on two different route plans:
Pure vision or multi-sensor fusion?
There are still many opponents of the former. The most unavoidable is the security issue.
For example, in the face of a blemish on the camera, Zhihu netizen @高燕 took the “human eye” as an analogy, saying that although one human eye is blocked, one can still walk normally, but for a machine, this requires a huge consumption of The computing power is used to identify and judge, and there is a high probability that it will not be able to judge the conditions for exiting the automatic driving, or it may directly filter out the black spots as pollution or interference factors to ensure that the automatic driving can continue to execute.
He pointed out that this is about sacrificing safety for ease of use to ensure a better autonomous driving experience.
Or more directly, some people pointed out that Musk’s operation of “reducing allocations without reducing prices” is to save costs.
(As a rough estimate, this car can save at least 500 yuan.)
On this point, many Chinese and foreign netizens see the same thing:
Of course, there are naturally voices supporting Musk.
There is an opinion that:
Tesla is originally the main vision, and it is logical to remove the millimeter-wave radar and ultrasonic radar in turn, indicating that its algorithms and models have reached new heights.
Just talking about the ultrasonic radar this time, Zhihu netizen @EatElephant believes that it is only a relatively weak assisted driving sensor and cannot accurately determine the location of obstacles.
In addition, thanks to the excellent capabilities of the latest Occupancy Network, the latest version of FSD has been able to identify the distance and shape of surrounding objects very accurately, so it is okay to remove it, and it may not be the same as the last time when the millimeter wave radar was cancelled. A “run-in period” occurs.
Regarding the view that the pure visual solution is not reliable, some people put forward such a view, saying:
Domestic multi-sensor fusion solutions have not substantially proven their capabilities in mass production and large-scale push functions.
Of course, those who hold this idea still believe that the upper limit of the multi-sensor fusion scheme will be higher than the pure vision scheme.
However, who is better in the end, as the previous netizens said, this is a battle of lines, it is useless to argue, let’s look at the actual effect.
So here comes the question:
Who wants to be the first testers of the pure vision solution? (dog head)
(Disclaimer: This article only represents the author’s point of view and does not represent the position of Sina.com.)
This article is reproduced from: http://finance.sina.com.cn/tech/csj/2022-10-10/doc-imqmmthc0368453.shtml
This site is for inclusion only, and the copyright belongs to the original author.