NOA is making great progress, but where will the smart driving map go?

This year is a big year for mass production of assisted driving, especially high-end intelligent driving, which has led to an explosion in upstream demand.

Some typical technology tracks, such as lidar and high-power computing chips, have benefited from this. Hesai Technology and Horizon have recently announced that their product shipments have reached new milestones. The shipment volume of Hesai AT series products in the first half of the year exceeded 70,000 units, more than the whole of last year. Horizon Journey 5 chips recently shipped more than 200,000 pieces.

The same is true for the cloud service market. Ambitious car companies are buying large amounts of computing power this year and building new supercomputing centers. Liu Shuquan, vice president of Tencent Smart Travel, said directly at an industry conference not long ago that "the demand for smart car cloud is not an emergence, but a blowout."

The demand for mass production by manufacturers directly triggered Tencent to open its second smart car cloud zone in Huailai, Hebei Province to meet the needs of OEMs for capacity expansion and disaster recovery backup.

According to statistics from Canalys, in the first quarter of this year, the wholesale volume of China's new energy passenger vehicles equipped with standard L2 assisted driving reached 826,000 units. It is conservatively estimated that this figure may be in the range of 3.2-3.5 million units for the whole year.

HiEV's rough estimate, only counting new cars with high-end intelligent driving (including high-speed NOA and urban NOA), its domestic sales this year may be between 800,000 and 1 million units.

The mass production of high-end smart driving has brought a wide range of business opportunities. At the same time, the national expansion of urban NOA and high security requirements have also inspired the reconstruction of part of the technology chain, one of which is maps.

A person in charge of perception with 6 years of R&D experience in the field of autonomous driving told HiEV that under the technology stack of Urban NOA, the biggest impact and change is high-precision maps.

1. High-precision maps in the mass production of NOA

The industry's use of high-precision maps began with Cadillac's Super Cruise system debuted on its flagship sedan CT6 in 2017. At the time, Super Cruise had a fairly robust ability to maintain lane centering, supported automatic lane changes with the lever, and the product definition allowed users to "take their hands off".

Behind Super Cruise's bold functional definition at the time, there was a key supporting link, which was high-precision maps.

High-precision maps provide the intelligent driving system with over-the-horizon environmental information input, including the lane lines, curvature, slope, road topology, important road signs, etc. of the road ahead, thereby helping the system's prediction, decision-making, and planning control.

In the second half of 2018, the high-precision map collected and produced for the Cadillac Super Cruise in China received its first map approval number.

Subsequently, from 2019 to 2022, dozens of brands such as Xpeng, Weilai, Ideal, BAIC Jihu, Changan Avita, SAIC Zhiji/Feifan/Roewe, Geely, Great Wall, GAC and other brands have mass-produced high-speed pilot assist. Driving function, commonly known as "high-speed NOA". High-precision maps are also being installed on a large scale as models of these brands.

The pilot-assisted driving function at this stage mainly covers highway networks and urban expressways. Since the total mileage of this part of the road is limited, approximately within 400,000 kilometers, high-precision maps of these areas were initially collected and updated mainly through professional fleets of map dealers.

This is the 1.0 stage of high-precision map development.

In 2021, as Tesla FSD continues to expand the scale of internal testing of autonomous driving in urban areas in North America, companies such as Huawei, Xpeng, and NIO are the first to follow up on the urban pilot assisted driving function, commonly known as "urban NOA."

So far, almost all domestic mainstream OEMs have announced their entry into the urban NOA field, including: Xpeng, Weilai, Ideal, Great Wall, Changan Avita/Deep Blue, SAIC Zhiji, Geely Ji Krypton/Jiyue/Route Si, Chery Xingtu/Jietu, BYD Tengshi/Zhangwang, Dongfeng Warrior/Lantu, Xiaomi, Nezha, Leapao and so on.

But when it comes to actual functions, currently only a few car companies have opened functions and internal testing to users in less than 10 cities.

There is an important policy basis for the opening of urban NOA, which is the "Intelligent Connected Vehicle High-Precision Map Application Pilot Notice" issued in the second half of 2022, which clarifies that in six cities: Beijing, Shanghai, Guangzhou, Shenzhen, Hangzhou, and Chongqing Launch a pilot project on high-precision maps, referred to as the "6-city pilot project."

The 6-city pilot project is also the 2.0 stage of high-precision map development.

At this stage, high-precision maps have shifted from highways to main roads and capillary roads in urban areas, and the mileage required for collection and production has increased significantly. At the same time, because urban NOA requires a larger number of road elements, the complexity and cost of high-precision map collection have increased significantly. .

According to HiEV, the cost of HD map subscription for urban NOA is roughly an order of magnitude higher than that for high-speed NOA. HiEV heard another statement that a new power brand with an Internet background has invested in a surveying and mapping fleet of about 40 vehicles in a super first-tier city in China to collect and update urban NOA maps. The average cost of a bicycle is 600,000 yuan. .

Urban NOA continues to move forward and must expand from 6 cities to the whole country.

This year, Wenjie (Huawei), Xiaopeng, Ideal and Great Wall (Haomo) have each set goals for opening NOA cities within the year, which are 45 cities, 50 cities, 100 cities and 100 cities respectively. Wenjie has also recently upgraded its Kaicheng target for ADS 2.0: it will be available nationwide by the end of December.

By expanding the urban NOA beyond 6 cities, the collection of urban roads across the country suddenly becomes a huge amount of work, and the consequent update costs required also increase significantly. From the perspective of collection and production costs and update cycles, it is not commercially feasible at this stage to expand high-precision maps nationwide using the same model as the "6-city pilot".

Therefore, the large-scale development of urban NOA and how to build maps are what we call the 3.0 stage of high-precision map development.

2. Changes in maps: light map, no map, or cloud map?

Whether it is how to help OEMs achieve multi-city and large-scale urban NOA faster, or how high-end intelligent driving solutions can support OEMs to reduce costs, or how to quickly update maps to ensure the stable and safe operation of urban NOA systems, traditional elements are rich and collected. Fine high-definition maps are already difficult to apply.

The entire industry is looking for a suitable new approach.

Discussions about high-precision maps (HD Map), light maps (HD Air/HD Lite/SD Pro) and no maps (SD Map) have become the hottest topics in the smart driving industry and even the entire automotive industry in a short period of time.

The so-called light map solution refers to distinguishing layers and elements based on the original high-precision map architecture. According to the needs of the downstream intelligent driving system, by reducing layers and elements, the cost of map collection and update is reduced, and the update cycle is accelerated. .

The mapless solution does not rely on high-precision maps, but uses standard navigation map (SD Map) input to achieve high-end intelligent driving.

The map-free solution has better map cost, but currently there are big challenges in some complex road conditions.

Wu Xuesong, deputy general manager of Changan Prospective Technology Research Institute and general manager of Chongqing Changan Zhitu Technology, gave an example: In the 8D city of Chongqing, the connections between viaducts and roads are intricate. In these scenarios, relying solely on perception for decision-making and planning control is not the optimal solution, especially the increased hardware cost of enhanced perception needs to be considered. Therefore, Changan internally believes that "a map to some extent" will exist as an over-the-horizon sensor in the future and be used in conjunction with real-time perception.

In addition to light images and no images, Tencent recently proposed a new solution at the Global Digital Ecology Conference: cloud images, or car-image-cloud integration.

In the HD Map 1.0 and 2.0 stages mentioned earlier, the HD map was delivered as a software and hardware integrated component, and then updated map data was released in cycles of several months or quarters. However, this method can easily lead to untimely updates of changes in the road environment.

Zhang Shaoyu, vice president of Tencent Maps and head of Tencent’s digital twin business, told HiEV: In current high-speed scenarios, 50% of takeover problems may be caused by untimely map updates.

Tencent's Smart Driving Cloud Map completely cloudifies the map, and can configure the required layers and elements according to the downstream R&D needs of the OEM, and can provide element updates in minutes at the fastest.

The core difference between cloud map and light map is that light map means that the map has fewer elements, while cloud map emphasizes that the update speed should be faster.

"There are many opinions on light maps in the industry. Some say that the cost is lower, and some say that the amount of data per unit mile is smaller. We believe that to make maps truly lightweight, element updates must be fast enough. Element-level updates are the smallest particles. degree."