"Can this car drive itself? I heard that Huawei intelligent driving is very powerful. A few days ago, in a Huawei store in Guangzhou, a man who looked around the M7 asked the sales staff for advice. Later, the reporter heard the sales reply like this: "More than seventy percent of customers who place a deposit for this car will choose the intelligent driving version of the model, and the fancy is the intelligent driving assistance system (NOA) that can realize the city and high-speed point-to-point..."

The topic of urban NOA is hot this year. In order to achieve low-cost and large-scale urban NOA popularization, many enterprises represented by Huawei and Xpeng have even begun to promote the "no map" technical route, that is, "high-precision map". However, this technical route was quickly "shelled". Some insiders said: intelligent driving "no map" is impossible, and the drawing of high-precision maps is still very important. As the most important direction of intelligent driving technology routes, the two technical routes of "no map" and "with map" have stalemated for a while. So which is more reliable? Is there a third option?

Text, photo/Guangzhou Daily all-media reporter Deng Li

Explanation of terms

Urban NOA, full name Navigate on Autopilot, literally translated as navigation assisted driving, but some car companies combine it with related functions to integrate into the large system of intelligent driving products.

According to the "Classification of Automobile Driving Automation" document, urban NOA is still L2 level autonomous driving, and in some cases can also be regarded as L2+. It can meet the needs of car owners in urban scenes such as following, overtaking, changing lanes, turning and recognizing traffic lights, and then independently arriving at the target location set by the owner.

Status quo:

The smart driving function has a significant impact on consumers' car purchase decisions

"In terms of user experience, point-to-point driving assistance capabilities such as NOA have made some navigation assistance systems and adaptive cruise in the past fundamentally different, and models such as the new M7 and Xpeng G6 with this as the core selling point have gained a lot of attention and orders after their launch." Cinda Securities pointed out in a recent research report that this year's intelligent driving function has had a significant impact on consumers' car purchase decisions.

Under this consumption trend, the technical route affecting the future iteration speed of intelligent driving has also diverged. More and more car companies and intelligent driving solution providers have begun to promote the "no graph" (that is, pure vision) technology roadmap, trying to get rid of the dependence on high-definition maps. In their view, going to high-precision maps and lidar is the right path for future intelligent driving products, especially the update speed of high-precision maps is slow, and it is simply unable to adapt to frequent changes in urban road conditions.

All media reporters noticed that in the "no picture" technical route, Tesla was the first to propose and firmly support internationally, and then Toyota also chose this route in 2022; In China, Huawei, Xpeng, and Lili have publicly displayed pure visual technology route products, and said that they have been advancing. On October 10, Jiyue Automobile, which is backed by Baidu's Apollo technology, announced its pure vision autonomous driving technology plan, becoming the latest "no picture" technology route supporter. The company said that the "pure vision" technology solution based on BEV+Transformer has been "highly mature", and the rapid upgrade of its underlying AI vision algorithm capabilities has made its system no longer need to use lidar, and gradually reduce the dependence on high-definition maps. In addition, some intelligent driving solution suppliers such as Milli and DJI have also chosen the "no picture" technical route, such as on October 17, Milli Mobility released 10 "no picture" intelligent assisted driving products below 11,3 yuan.


The "no picture" technical route is actually a pseudo-concept?

Before 2023, almost all domestic pilotage assistance systems need the assistance of high-precision maps, why will the thinking shift in 2023? "Recently, the call for the 'no map' technical route is very loud, and its essence is the enhancement of real-time mapping capabilities on the cloud brought about by algorithm and model upgrades, not the real 'no map'." A few days ago, NavInfo's new CEO Cheng Peng pointed out at the industry alliance of the intelligent networked vehicle autonomous driving map basic platform: "The map has space-time information in it, so that the vehicle is linked to the road, so from this point of view, the 'no map' technical route itself is a false proposition." ”

Some insiders pointed out that the so-called "no map" intelligent driving solution launched by some car companies is not completely without high-precision maps. "If a car company says 'not relying on high-definition maps', 'not dependent' means that it will still be used, rather than completely 'de-mapped'." The industry insider pointed out that for example, Xpeng Motors' urban OTA "no graph" XNGP intelligent driving (that is, NOA) function, but the actual operating logic is "it is good to have a high-definition map, and it can be used without a high-definition map". There are also some models that adopt Huawei's high-end intelligent driving solution, which is also based on high-definition maps.


Car companies can reduce costs by making efforts to "no map" routes

The choice of "no picture" or "with picture" technical route, from the current point of view, in fact, each has its own advantages, and the dispute between the parties has not yielded results. However, there is a point of view, whether it is to go or stay lidar and high-precision maps, the most important target trend is to promote the reduction of intelligent driving costs.

On the whole, the advantages of "no map" just make up for the shortcomings of high-definition maps: first, it can be quickly popularized in non-pilot cities. An intelligent driving system engineer working for a new energy vehicle company said that choosing the "no map" technical route, under the logic of heavy perception, the map can be generated in real time by algorithms to assist in regulation, and the promotion will be less limited by the city. Second, the cost is significantly reduced. According to the latest products released by Milli Zhixing, one of them only needs 3000,5000 yuan level "ultimate cost performance" high-speed no picture NOH, which can realize the integrated intelligent driving of driving and parking. It is cheaper than the <>,<> yuan low-cost pure vision solution released by DJI before. From the perspective of price, the "no map" version has great advantages over the high-precision map intelligent driving solution that costs tens of thousands of yuan. This can drive down the price of smart cars from the cost perspective.

Point of view PK

"There is a picture" Fang said:

(1) Is "no picture" a pseudo-concept?

Cheng Peng said that in the high-speed NOA scenario, theoretically based on existing sensors, Orin chips, large models, etc., the car does not need to be loaded with a complete high-definition map. However, it does not mean that the "map" can really be abandoned, but the "map" is placed in the "cloud", and when encountering complex intersections, it is called from the "cloud" to the car, and the problem is solved in a low-cost way. In his opinion, intelligent driving "no map" is impossible, and the drawing of high-precision maps is still very important.

From a map perspective, different maps support different levels of autonomous driving and requirements. Cheng Peng said: "Now many mainstream car companies have the ability to carry high-speed NOA. In this scenario, in theory, after the existing sensors, plus NVIDIA Orin chips, large models, and cloud training, the terminal (car) does not need to install a complete high-precision autonomous driving map, but it still needs to use high-precision maps. ”

(2) It is not technically realistic

Some insiders pointed out that from the perspective of the vehicle, no matter how strong the sensor is, it can not "see" something two kilometers away, but in terms of driving speed, driving two kilometers may be only more than a minute, such a short calculation speed will be very demanding on the system, "At present, even if there is LiDAR, there is a very good BEV+Transformer technology, but you can see 200~300 meters away." ”

On this issue, Cheng Peng also analyzed that traffic accidents, restriction changes, lane forks and other situations in a little longer distance need to be shared by a platform, and maps are natural carriers.

(3) Lidar is also reducing prices, which is conducive to more vehicle applications

The industry also pointed out that it is also difficult to completely do it with lidar. Because, at present, the price of lidar is constantly declining, promoting the rapid popularization of NOA intelligent driving in cheaper models, such as 20,6 yuan products. For example, the newly launched Zhiji LS7 is equipped with lidar as standard, and provides free intelligent driving for life; Rui Lan 20 includes a high-end intelligent driving option package for LiDAR, and the price is also within <>,<> yuan.

The "no picture" party put it this way:

(1) The number of pilot cities with "maps" is temporarily limited

The reporter noted that the Ministry of Natural Resources announced in August last year the first batch of high-precision map application pilots for intelligent networked vehicles in six cities: Beijing, Shanghai, Guangzhou, Shenzhen, Hangzhou and Chongqing. No seventh city has yet opened. This means that if car companies want to use NOA as a selling point for vehicle promotion, they will be slightly passive in pilot cities without high-precision map applications.

(2) The acquisition cost of "with maps" is high and the update time is long

Yu Chengdong, executive director of Huawei, CEO of Terminal BG, and chairman of Intelligent Vehicle Solutions, once complained that Huawei has invested 1~2 years in Shanghai and has not collected high-definition maps of urban areas. According to NavInfo's new CEO Cheng Peng, "Up to now, NavInfo has spent 8 billion yuan in 30 years to only complete the collection of high-precision maps of national highways and more than 120 urban arterial roads, with a total mileage of about 30,100 kilometers, and it takes 200 billion ~ <> billion yuan to cover it all." ”

Therefore, it is the industry consensus that high-definition map update frequency is low, maintenance costs are high, qualifications are difficult, and the map review process is complicated. According to the "White Paper on High-Precision Map of Intelligent Connected Vehicles" released by the China Intelligent and Connected Vehicle Industry Innovation Alliance, the surveying and mapping cost of decimeter-level maps is about 10 yuan per kilometer, while the surveying and mapping cost of centimeter-level maps can reach 40,50 yuan per kilometer. Cheng Peng also said: "The annual maintenance and update of the map also costs <> billion ~ <> billion yuan. ”

(3) High-definition maps require high synergy

With the rapid landing of high-end intelligent driving, the market requirements for maps are getting higher and higher. At present, the typical requirements for high-definition maps are: fresher freshness and better accuracy. That is, in terms of freshness, from two editions a year to the expectation of achieving "daily change"; In terms of accuracy, HD maps require decimeter-level accuracy, and even some high-speed scenes require 2~3 centimeters to truly meet the requirements of public safety. Based on this, the definition of map products for automotive intelligent driving has also changed from low-confidence sensors to high-confidence, real-time sharing platforms. As a result, a complete HD map is difficult to complete by a single company. From a closed-loop data perspective, industry collaboration is also required.

Observation: The all-round technical competition between car companies kicked off

The reporter noticed that Tesla, which actually supports the "no picture" technical route, also has uncertainty about the development of intelligent driving. At the earnings conference in the third quarter of this year, Musk admitted to being too optimistic about the progress of FSD. Xpeng Motors, which also supports the "no map" technical route, is also "love-hate" for high-definition maps. In March this year, He Xiaopeng, chairman of Xpeng Motors, made a suggestion: he hopes to further optimize and simplify the high-precision map review process, shorten the period from collection to use of high-precision maps, accelerate the process of high-precision map infrastructure, and escort the development of China's autonomous driving industry. Pony.ai CTO Lou Tiancheng said that getting rid of high-precision maps is not only a challenge to perceive the plate, but also the simultaneous improvement of all modules such as prediction and regulation.

Therefore, it is still widely believed in the industry that if there is a ready-made high-precision map in front of car companies, it is likely to be difficult to reject.

In fact, with the current capabilities of intelligent driving systems, it is not enough to ensure 100% driving safety, and even the development of enterprises is very different. Therefore, most traditional car companies and some leading overseas autonomous driving companies such as Waymo and Cruise still attach great importance to maps. An engineer from a traditional German car company told reporters that there are still many long-tail problems in intelligent driving that cannot be solved, which is why BBA and others are still very "conservative", in fact, these foreign car companies have shown L2+, and even L3 automatic driving many years ago, but finally stranded.

For how to quickly reduce costs and maintain rapid iteration of technology to achieve multi-city and large-scale urban NOA, it is clear that the whole industry is still looking for a suitable new method. The all-round technical competition between car companies has just begun.

Guangzhou Daily, October 2023, 10, version A26