Opportunities and concerns of high-precision maps

For human drivers, the main function of electronic maps is navigation, including route planning from A to B, positioning and matching of vehicles and roads, POI retrieval, etc. So when cars in the future can achieve a certain degree of autonomous driving, and even do not need a driver and can be fully autonomous, what kind of maps will be needed?

The answer is high-precision maps.

Image source: NavInfo

How do high-precision maps help autonomous driving?

Accurate assessment of one's own position and perception of the surrounding environment are particularly important.

At present, sensors such as cameras, millimeter-wave radars, and lidars still have some defects when performing environmental perception, especially in extreme weather conditions such as dust, rain, snow, and dense fog, which are prone to misjudgment or even failure. Even if multiple sensors are integrated, this cannot be completely avoided. High-precision maps can not only still function in extreme weather conditions, but also have a field of view that is not restricted by occlusion, distance, or vision. They can complement the above sensors at the perception layer and provide more reliable perception capabilities for self-driving cars.

In terms of perception information, high-precision maps can provide high-precision static information for self-driving cars, such as road networks, road shapes, lanes, POIs, buildings, road signs, etc., as well as dynamic real-time traffic information. By integrating these two types of information, a virtual driving environment is formed for vehicles to perceive, recognize and understand the environment, and carry out path planning, avoid congestion and traffic obstacles. From this point of view, high-precision maps are actually equivalent to a super perception container. On the one hand, they can assist existing sensors, and on the other hand, they can serve as a platform to meet the needs of lane-level planning, ultimately achieving dual enhancements in perception and decision-making.

It is worth mentioning that based on the reconstruction of the 3D road environment, high-precision maps can help autonomous vehicles reduce their dependence on expensive sensors, significantly reduce system costs, and reduce computing pressure within the vehicle.

According to Lu Zheyuan of Shanghai Jingzhong Information Technology Co., Ltd., high-precision maps can not only help self-driving cars to perceive their locations and plan routes more accurately, and provide support for decision-makers, but also greatly benefit the development of smart transportation. For example, in the field of smart parking, they can be used for parking guidance and reverse car search, helping users find parking spaces quickly.

"High-precision maps can also serve as a supplement and enhancement to existing sensors for autonomous driving, strengthen the vehicle-side perception capabilities in the vehicle-road collaborative architecture, and thus improve intelligent connected applications. It can also help automakers, scientific research institutions, etc. conduct virtual testing of autonomous driving." Lu Zheyuan said.

Image source: NavInfo

Guo Panshi, deputy director of the NavInfo Research Institute of New Vehicle-Road Collaboration, believes that compared with traditional maps, the maps needed for autonomous driving have higher accuracy requirements. This does not mean that the absolute accuracy must reach a certain level, but that the accuracy should cover various scenarios required for travel. "For example, when we drive on the highway, what we need is not the absolute geographic coordinate accuracy of the vehicle, but the relative relationship between lane lines."

In addition, Guo Panshi believes that maps used for autonomous driving should also have the characteristics of complete elements, fast updates, and strong coordination. "What is complete elements? For example, when we use Didi to take a taxi, you need to know whether the pick-up point is the east gate, west gate, south gate or north gate of the community, and this should also be displayed on the map. As for updates, although updates have been achieved in seconds, it is not enough for autonomous driving, and it needs to be further achieved in milliseconds. Strong coordination means that the map can be fully coordinated with the perception system, computing system, communication system, etc."

However, given that there are multiple different levels of autonomous driving, not every stage requires such advanced technology.

Liu Bin from the Policy Research Center of the Automotive Technology and Information Research Institute of China Automotive Technology and Research Center believes that at the L1 and L2 stages, traditional electronic maps are still needed as a reference for travel, because the human driver is still in control of the vehicle. However, at the L3 level, since humans and systems share control of the vehicle, the introduction of high-precision maps combined with sensors will help reduce R&D costs and facilitate subsequent mass production.

Yang Diange, director of the Department of Automotive Engineering at Tsinghua University, also believes that for L1 and L2 ADAS systems, sub-meter ADAS maps are sufficient, and at L3, in addition to ADAS maps, high-precision maps may also be used, but they are not necessary. At the L4 stage, centimeter-level high-precision maps are necessary, and the same is true for L5, which must not only have them, but also be able to be updated in real time.

Current status of high-precision map development

Compared with traditional electronic maps, high-precision maps have higher accuracy and cover more information. They can provide vehicles with more detailed environmental information than sensors, and have become one of the core technologies of self-driving cars. However, since the data provided by self-driving maps is too detailed and involves spatial information security, current laws and policies have many restrictions on data collection, transmission, storage, use and expression, which to some extent restricts the development of high-precision maps.

Image source: momenta

Specifically, according to Zhang Wei, director of the General Office of the Department of Geographic Information Management of the Ministry of Natural Resources, the difficulties faced by high-precision maps at this stage are mainly in the following aspects:

First, the collection, use and expression of data are restricted. This is mainly for car companies and autonomous driving solution providers. Due to current regulatory restrictions, they do not have surveying qualifications and cannot collect, use and store these spatial location information. They can only cooperate with qualified map vendors. In addition, it also includes crowdsourcing collection, as well as road elevation, slope, curvature, height and weight limits of bridges and tunnels. According to current policies, there are actually clear restrictions. Car companies have a strong demand for this data, which indirectly affects the development of autonomous driving.

Second, there is no unified standard for autonomous driving maps in China. The current international standards are basically designed based on European and American roads, which is very different from the domestic scene.

Third, confidentiality technology needs to be improved. According to current regulations, autonomous driving maps are still a type of navigation electronic map and need to be processed before public use. However, autonomous driving has very high requirements for positioning, which is obviously inconsistent with regulatory requirements.

Fourth, there may be some problems with the current map review method. Currently, navigation electronic maps are subject to a licensing system and need to be reviewed by the geographic information authority before publication and distribution. However, autonomous driving maps use digital strings to express relevant information and have a high update frequency and cycle. The current map review model is difficult to meet these needs.

Fifth, there is no designated testing area. At present, domestic autonomous driving test sites are mainly focused on testing autonomous driving technology, and there is no testing and verification of map-related issues, which leads to a lack of assessment of geographic information security.

Sixth, there is a lack of a unified data management platform. Autonomous driving maps not only contain highly accurate static road information, but may also contain dynamic information such as traffic incidents and road construction in the future. Based on this feature, the cost of data collection and updating will be very high in the future. If there is a unified data management platform for autonomous driving data collection, diagnosis, and evaluation, more efficient data sharing can be achieved.

It can be seen that the road to industrialization of high-precision maps is also a long one. Despite this, the field has attracted a large number of companies to enter in the past few years. Looking at the market, in addition to traditional map vendors, technology giants such as BAT and traditional car companies such as BBA have entered the field of high-precision maps through acquisitions, investments or cooperation, and even a large number of start-ups have been born, such as Momenta, Kuandeng, Jingzhong, etc.

Among BAT, Baidu has received a high-precision map and self-positioning mass production order from Great Wall Motors, and has signed commercial fixed-point agreements with many brands such as BYD, Chery, Hyundai, BAIC New Energy , Han Teng Automobile, GAC Trumpchi, and Dacheng Automobile. Alibaba's AutoNavi has won two high-precision map commercial orders from Cadillac and Geely, and Tencent's NavInfo has won a mass production order from BMW China. Start-up companies are also constantly strengthening their cooperation with car companies to accelerate their integration into the market and gain greater development space.

With the competition from multiple parties, the market pace of high-precision maps has accelerated. According to the forecast of Gasgoo Automotive Research Institute, with the continuous development of autonomous driving, especially the intensive launch of L3 models starting from 2020, the high-precision map industry is expected to usher in a golden period of development. It is estimated that by 2025, the domestic high-precision map market will reach 8 billion yuan, and will grow rapidly in 2026, exceeding 10 billion yuan.