Research and Discussion on Autopilot Map Technology and Standard System

The rapid development of electronic maps is an indispensable support for the intelligent services of the Internet of Vehicles. In vehicle navigation, electronic maps are the basis of in-vehicle navigation data. In the Internet of Vehicles service, electronic maps are the key link of vehicle networking services. Electronic maps are the core technology of driverless. The autopilot technology route changes from weak map mode to strong map mode, so smart cars and autopilot maps are also changing. Different levels of autopilot have different content and accuracy requirements for maps. For high-level autopilot L4 and unmanned auto Driving an L5, an automatic driving map is a must.

In the "China Autopilot High-precision Map Industry Development and Innovation Conference", we invited Mr. Jiang Kun, an assistant researcher from the School of Vehicle and Carrier Studies at Tsinghua University, to explain the research progress and exploration of the autopilot map technology and standard system.

First of all, Jiang explained the progress of autopilot maps at home and abroad. Foreign autopilot map progress: Foreign countries collect and update high-precision map data through laser radar + camera, 3D image, artificial intelligence and other methods. They have begun testing unmanned vehicles. The commercial cooperation model shows a steady growth trend, but for high-precision Map map standardization also needs constant exploration. Domestic autopilot map progress: Domestic mainstream map collection equipment self-produces and uploads high-precision maps through self-made or purchased mature equipment. The auto-driving map format has not been unified yet, and a cooperative collection platform has not yet been established. The progress of the acquisition on the mainstream map is basically the same, and the acquisition of the 300,000 km highway is basically completed. It is expected that the urban road collection will be completed around 2020. Start-up companies focus on crowdsourcing.

The development of high-precision map technology has promoted the transformation of surveying and mapping technology. Traditional maps are human and capital-intensive, mainly manpower processing, and high-precision maps are mainly automated processing with artificial intelligence.

Then Jiang explained the difference between traditional maps and autonomous driving maps: it needs more richness in data content. Traditional maps provide navigation information for drivers, while autopilot maps are aimed at driverless cars. More precision is needed in data accuracy. Traditional maps are synopsis information with low accuracy requirements, while automatic autopilot maps require three-dimensional reconstruction with high precision requirements. More dynamics are needed in data aging. Traditional maps are based on static map information, while autopilot maps require dynamic information such as weather, traffic, roads, and roadblocks. In data applications, it needs to be more intelligent. Traditional maps only participate in path planning, while autopilot maps are involved in perception, decision, and control. I then showed us the technical process of autopilot maps.

Finally, Mr. Jiang explained the mainstream technical routes and practice cases of automatic driving and the current national policies on autonomous driving maps. In the mainstream driving route of automatic driving, it is mainly divided into three technical routes: vector map technology route, perceptual map technology route and vector + sensing technology route. Vector map technology route: from data collection, processing to data release, map data is stored as a complete spatial database. Perceptual map technology route: Produce maps through crowdsourcing of sensors (visual, laser) + algorithms. The corresponding sensor maps are generated based on different sensors, which can be quickly understood by the car. It plays a key role in high-precision positioning of automatic driving, and is mainly divided into three categories: laser radar, camera and millimeter wave radar. Vector + Perceptual Technology Route: Maps cooperates with smart sensor vendors, and the perception layer is used as a supplement to vector maps, and serves as the main data source for vector map crowdsourcing updates.

Due to national security considerations, China's current regulations and policies have made many provisions for automatic driving map data collection, element expression data encryption, data update, review, release, and confidentiality of results. In encryption: China's automatic driving map is managed according to the regulations of navigation electronic maps, and it needs to be encrypted according to regulations. In terms of confidentiality: the map data of the automatic driving technology test and road test shall not be contacted by personnel outside the scope. In terms of expression restrictions: road slope, radius of curvature, plane coordinates, elevation, etc. must not be expressed. In terms of qualification restrictions: the drawing of high-precision maps must be undertaken by the navigation electronic map qualification unit.