Visual automatic driving has "accidents", which is expected to accelerate the loading of lidar on the road

Core Intelligent Driving -- A collection of the views of famous experts in production, education, research and enterprises, a comprehensive analysis of THE opportunities and challenges of AI chips, third-generation semiconductors and other opportunities and challenges in the era of "great deformation" of automobiles!

Jiwei network news, intelligence is becoming one of the selling points of new car-making forces, but also because of automatic driving caused by accidents, so that brands including Tesla, Weilai Automobile, Xiaopeng Automobile and other brands have been questioned by the market. A few days ago, a Xiaopeng P7 could not identify the faulty car in front of it, resulting in a crash, which once again pushed automatic driving to the cusp of the storm.

Shortly after the accident, Tan Xuguang, chairman and CEO of Huaichai Power, publicly said, "Unmanned driving is flickering, don't listen to them flickering, what L1 is OK, L2 is also so-so, L3 is impossible." As a veteran in the industry, Tan Xuguang's remarks undoubtedly poured cold water on automatic driving, and the market once again debated the "automatic driving" plan.

Visual autopilot is in trouble again

In recent years, with more and more L2-level and L2+ self-driving cars entering the market, when bringing a new scientific and technological experience to the public, there are occasional driving accidents caused by technical problems, as a new thing, new applications, accidents caused by ADAS are often hotly discussed in the market.

Tesla, the world's leading self-driving company, still has a number of ADAS lawsuits to date. Its first self-driving fatality in China occurred in 2016, and according to the disclosure information, the vehicle activated the autonomous driving function at the time of the incident, but did not identify the road sweeper that was carrying out the operation in front of it, resulting in the accident.

In August 2021, the representative of the new local car-making force, WEILAI Automobile, also had an autonomous driving accident, and an ES8 crashed into a stone pier in a NOP-led state. After the accident, the new domestic car-making forces collectively reiterated that assisted automatic driving is not equal to automatic driving, and called for the renaming of different levels of automatic driving.

On March 13 this year, another new local car-making force also encountered a similar situation. Xiaopeng Automobile's P7, which turned on assisted automatic driving, failed to identify the faulty car in front of it during the highway, resulting in a collision of vehicles, fortunately, only the driver was slightly injured in the accident. According to the owner of the accident, the car's ADAS perception configuration is: 3 high-precision millimeter-wave radar, 12 ultrasonic sensors, 4 surround view cameras, and 1 high-precision camera.

Afterwards, Xiaopeng Automobile said that the vehicle instruction manual clearly stated that the car had difficulty identifying stationary objects, and the driver did not observe the road conditions in front of him in time and took over the vehicle during the start of ACC+LCC (Adaptive Cruise & Lane Centering Function) driving, resulting in the accident.

In fact, the ADAS of the above 3 autonomous driving accidents are equipped with cameras and millimeter wave radar sensors for environmental perception redundancy, and in the accident, the object of the car is either a stationary object or a low-speed vehicle, which also reflects the bottleneck of the difficulty of static object recognition by the two sensors mentioned by Xiaopeng Motors.

Imaging industry experts told the author, "Although imaging technology is applied to the field of automatic driving, artificial intelligence technology still has certain deficiencies in detecting objects in the video, and the clarity is not enough, and the object and the environmental color are similar to the detection failure, which depends on the continued optimization of the algorithm." Another ADAS business leader also believes that Tesla's unswerving promotion of pure visual self-driving routes is due to its powerful algorithms.

Even so, Tesla autopilot accidents still occur occasionally, it is worth noting that from 2016 to 2021, Tesla had several accidents hit by white cars, according to accident reports, the reason is that the ADAS system identified the white body as the sky.

It should be noted that in these accidents, millimeter-wave radar did not play a warning role. According to supply chain news, the Xiaopeng P7 uses Bosch's fifth-generation millimeter-wave radar, with a detection range of 250 meters, mainly used for automotive collision avoidance systems. However, industry insiders believe that due to the low resolution of millimeter-wave radar, the ability to recognize objects with a small area will be greatly reduced, "for static objects such as road signs and even stopped vehicles, millimeter-wave radar is difficult to identify." ”

The "chicken rib" performance of millimeter-wave radar has been abandoned by Tesla, and its senior director of AI, Andrej Karpathy, said that the vision system built by Tesla has left other sensors far behind.

Lidar will accelerate on the road for ADAS "bottom".

Still, more OEMs aren't following Tesla's aggressive approach. Not only that, some Tesla users also believe that having radar will improve driving safety, and a Model 3 owner said that he once relied on millimeter-wave radar in a misty weather to detect the truck in front of him in advance.

At a time when millimeter-wave radar is becoming increasingly marginalized in autonomous driving, ADAS needs a stronger perception solution, of which the lidar solution has been realized to be carried on the car, and the Xiaopeng P5, which was launched at the end of last year, became the world's first intelligent driving car equipped with high-beam lidar.

For the Xiaopeng P7, which recently had an accident, the supply chain news said that in its latest redesigned model, it is planned to be equipped with 2 laser radars to improve the safety of the car in automatic driving. Gu Jianmin, CTO of Valeo China, also believes that lidar can detect stationary or moving objects well, whether it is day or night.

This scheme has also won the favor of most domestic OEMs, as of now, Xiaopeng P5, Xiaopeng P7, WM M7, Great Wall Mocha, Great Wall Mech Dragon, Jihu αS, Love To Eat U6, GAC Ean LX, Weilai ET7, SAIC Zhiji L7, Ideal X01, Nezha S and other models have been equipped with or planned to be equipped with lidar. In addition, BYD, one of the leading enterprises in new energy vehicles in China, has also reached cooperation with lidar companies such as Sagitar Juchuang.

A person in charge of the lidar marketing department said that for most OEMs, there are still some natural defects in pure vision algorithms, such as night imaging problems, which need to be combined with a variety of perception technologies to improve the safety of automatic driving, and lidar can make up for the lack of vision.

However, lidar on the car is not smooth sailing, on the one hand, the cost is higher, has become one of the factors restricting the purchase of OEMs; on the other hand, the above-mentioned person said, "At present, although lidar has been on the car, but the algorithm fusion is still optimized, which has also led to some lidar models having to be postponed from 2021 to this year's listing." ”

An industry insider said, "The cost of the lidar program (compared to the visual solution) is much higher, but the domestic car companies prefer to choose the lidar solution, which is a conservative scheme to improve safety, which is the bottom of the automatic driving." Dr. Yu Yuan, chairman of the Standards Council of the IEEE Consumer Technology Association, further pointed out, "Whether it is pure vision, lidar, or other technical routes, the development of artificial intelligence technology will certainly promote the popularity of autonomous driving, and the autonomous driving solutions that ultimately win in the market must be a good balance between cost and safety." ”

In the context of repeated setbacks in visual routes and millimeter-wave radar difficult to meet the needs of higher-level automatic driving, lidar has undoubtedly become one of the indispensable solutions for high-level automatic driving, Li Yifan, co-founder and CEO of Hesai Technology, recently said, "Lidar can provide another dimension of information, greatly supplementing the safety of existing sensors." ”

The market is also more aggressive towards lidar. It is understood that at present, Hesai Technology semi-solid-state lidar AT128 has obtained millions of fixed points, Liangdao Intelligence, Tucson Future, Jidu Automobile, etc. are accelerating the landing of lidar solutions; Weilai, Xiaopeng, ideal 3 local car-making new forces in the semiconductor field layout, are invested in lidar or photoelectric sensing field enterprises.

For the problem of high cost of lidar, industry insiders said, "I think the unit price of lidar will drop rapidly from about 7,000 yuan now, and it is expected that the unit price will reach 1,200 yuan by 2025, and may be as low as 600 yuan in the future." In the future, lidar will become the standard for high-level (L3 level and above) autonomous driving. ”

(Proofreader/James)