Preface: spring recruit season again! As a national travel service platform, Autonavi’s business is developing rapidly, and a large number of school/social recruitment positions are open. You are welcome to submit your resume, details are at the end of the article. In order to help you know more about Autonavi technology, we planned a series of articles **# Spring recruitment column, and organized senior students of each business team to make relevant introductions with business science popularization + technology application practice ** as the main content.
This article is the second in the series of “High precision Map and Collection and Production System for Autonomous Driving”, which was shared by Xiang Zhe, general manager of Autonavi High Precision Map business, AT the TECHNICAL forum of AT. A slight abridgment of the content without compromising the original meaning.
AT Technology Tribune (Amap Technology Tribune) is a technical exchange activity initiated by Autonavi. Each issue focuses on one theme, and we will invite experts from inside and outside Ali Group to make technical exchanges with everyone through speeches, QA and open discussions.
Xiang Zhe mainly shared two aspects this time:
1. What is the high-precision map for autonomous driving?
2. Current construction status and thinking of production system of high-precision map data acquisition.
High-precision map is an indispensable core condition for autonomous vehicles. It needs to accurately express the spatial position and relative relationship of various elements in the real world. Therefore, the production of high-precision map requires high accuracy of data collection.
Industrial-grade classification of autonomous driving
Xiangzhe began to talk about the application of automatic driving in industrial level.
At present, industrial-level automatic driving can be roughly divided into two categories. The first category is represented by intelligent cars with “auxiliary driving function” produced for users by tesla, Xiaopeng and other new car-making forces. When using these autonomous driving features, users are expected to take over the wheel at any time, and the responsibility for legal problems that arise while driving lies mainly with humans.
Autonavi has deep cooperation with the mainstream new power manufacturers in the high precision map. Xiaopeng, for example, uses autonavi’s high-precision mapping capabilities for its assist driving function. What kind of assisted driving capabilities are currently available to users? It has basically realized point-to-point automatic driving on expressways.
For example, driving from Beijing to Guangzhou on the freeway will encounter a number of freeway switches, from one freeway ramp to another, as well as lane changes to overtake cars on the freeway. Xiaopeng’s NGP assisted driving capability has the above two capabilities, and basically has the automatic driving function of the whole journey from Beijing to Guangzhou.
However, in the process of driving, the driver should always keep an eye on the road condition. If there is a risk of rubbing when automatic lane changing, the driver should take over the driving manually and continue to complete the lane changing manually. This means that the driver should be ready to take over while driving.
In addition, the driver will have to manually take control of the car when passing toll booths, as there is no lane-level information in the toll booths yet. When approaching a toll booth, the voice assistant will alert the driver that there is no high-precision map of the road ahead of the toll booth and the driver needs to take over manually. The above autonomous driving capabilities will be applied to all xiaopeng P7 models.
The second type of industrial automation is typically L4. For example, Google is doing self-driving taxis in the city, logistics line trucks, and so on. Compared with the first type of autonomous driving, this type of L4 autonomous driving theoretically does not have a driver on the car. Although we’re in the verification phase, there’s still a driver in the driver’s seat. According to Xiang Zhe, it will take four to five years for L4 self-driving taxis to enter the life of ordinary people.
Both types of autonomous driving rely heavily on high-precision maps.
High-precision maps and autopilot
High-precision maps are maps in the “brain” of autonomous vehicles that let them know what the road will look like “out of sight”. The four key functions of autonomous driving are perception, high precision positioning, decision planning and vehicle control. At least three of these features rely heavily on high-precision maps.
Perception: when driving a car, humans should observe the surrounding lane lines, traffic signs, poles and other information. Sensors in the smart car sense information about objects around the road. High precision maps provide a meta-visual perception of god’s perspective. In particular, high-precision map data can inform the vehicle of the road ahead, especially when there is a big truck in front of the vehicle and the human eye and sensor cannot see the information such as the lane line in front of the vehicle.
High-precision positioning: autonomous vehicles need to know exactly where they are on the map, based on the base map provided by high-precision maps. Autonomous vehicles need to know where they are on a map based on two capabilities. One is absolute location information provided by GPS, inertial navigation, and qihiro. The absolute position information can be matched with the longitude and latitude coordinates of the map to determine the specific position of the vehicle in the map (it depends on the absolute position positioning ability of the sensor).
But only absolute positioning is not enough, in a particular area, such as high-rise buildings, canyons, etc. Will keep out signal occurs, absolute positioning accuracy will become worse, lane line around the autopilot rely on observation, signal, the rod of relative positioning to assist, to high precision of map data matching judgment. In the actual project, Autonavi through in-depth cooperation with mainstream automakers, together to determine which technologies can obtain more accurate relative positioning ability.
Decision planning: Autonomous driving must comply with driving rules, so it is highly dependent on lane lines, traffic restrictions, traffic lights and other road elements.
These functions support each other.
High precision map for autonomous driving
Several key elements: road layer, lane layer, positioning object, dynamic layer.
Road tier: HD (high precision map) and SD (normal map) data are closely matched. At present, almost all autonomous driving starts with the user telling the intelligent system that I want to go from one place to another, and the driving route planning between the two places is supported by SD road data. HD data is not isolated, but connected to SD data. SD data capability is autonavi’s traditional strength, coupled with industry-leading HD capability, this matching autonavi must be the best in the industry. This is also a key consideration for automakers when choosing a mapping service.
Lane level: All autonomous vehicle control at the bottom relies on high-precision map data.
Positioning objects: Autonavi cooperates closely with the auto factory, which technologies are used for relative positioning, which reference objects are selected, and what level of accuracy should be achieved, etc. The two sides communicate and develop together.
Dynamic layer: Future high-precision maps will definitely contain dynamic layer, real-time data, and what dynamic traffic events are happening in a particular lane at any given moment.
High precision map in the city of ordinary road challenges
At present, Autonavi’s high-precision map has completed the acquisition of more than 300,000 kilometers of highways and urban fast sections, and is entering a stable state of regular update. Compared with the high-speed city fast, the more difficult problem in the city ordinary road.
One of the key challenges of urban high-precision map data is intersections, many of which lack ground traffic mapping. When a self-driving car turns between intersections and crosses the ground with no lines of traffic (paint), it relies on pre-compiled data from high-precision maps. Of course, it’s not just the ground traffic at the intersection that needs to be considered, but a number of other traffic elements as well. But autonomous driving on ordinary urban roads is certainly one of the key scenarios in which map services and new car makers will devote a lot of energy in the future.
Collection and generation of high precision maps
Conventional high-precision map production can be summarized into three stages: “acquisition”, “production” and “transition”.
The acquisition vehicle is a mobile acquisition system that is precisely integrated with a variety of advanced measurement sensors, generally including Lidar, inertial navigation, camera and other equipment. It is equipped with different types of sensor equipment according to different acquisition scenarios. After years of deep cultivation, gaode high precision team developed the high precision acquisition vehicle system, which has the characteristics of high precision, fast speed, short data generation cycle, high degree of automation, high security and information integrity.
After the acquisition equipment collects the data with precision in the external real world, it can “become” high-precision map data that can be used through image recognition, precision processing and manual processing.
High precision map of “fresh”
Road data in real life is in constant change, how to achieve “fresh” high precision map.
First of all, relatively expensive professional cabs were used to measure and collect high-precision map data on the country’s roads at the beginning. This kind of acquisition should ensure both relative and absolute accuracy. Then, a relatively inexpensive professional update vehicle is used to collect local changes (relative changes) in road information, such as repainted ground signs, newly erected signs, poles, etc. At the same time, we are using cheaper crowdsourcing equipment to do faster collection updates.
In order to achieve rapid update of existing data, improve the freshness of data. Autonavi’s high-precision team has set up a three-level collection system of professional bottom car, professional update car and crowdsourcing update ability to solve the problems of accuracy and freshness together. In real business scenes, it is necessary to find a balance between precision and freshness, repeated iterations.
In order to solve the challenges of “high precision”, “large scale” and “sufficient freshness” of high precision map data, we need to make breakthroughs in many technical points.
For example, how to design and manufacture acquisition and mapping equipment with different costs, different precision and different deployment capabilities;
How to cooperate with different kinds of equipment to collect and meet the product requirements of accuracy and freshness;
How to design and apply algorithms to improve the absolute accuracy and relative accuracy of the collected data, and ensure the alignment of the collected data for multiple times;
How to comprehensively apply image and point cloud to identify and improve the automation level of production.
Students in auTONavi Map team focus on different directions, accept business challenges with an open mind, discuss and design solutions together, and have achieved a lot.
With this high-precision map with the highest accuracy and the widest coverage in China, Autonavi successfully won commercial orders from a number of mainstream automakers at home and abroad, and began to provide intelligent driving models with high-precision positioning, beyond-line-of-sight perception, lane-level navigation and other services. As a key breakthrough area of Autonavi in autonomous driving ecology, The business of Autonavi High-precision Map develops rapidly and has many opportunities. We hope you can join us.
About the high precision map business center
Autonavi maps is one of its most innovative businesses, dedicated to measuring the world with sensors, understanding the world with algorithms, and redefining the world with data. We cover almost the hottest frontier disciplines, high precision mapping and autonomous driving are multi-disciplinary applied engineering systems. Automatic generation of high precision digital 3D map based on perceptual understanding, 3D reconstruction, fusion positioning, computational geometry technology. Using edge computing, big data processing, cloud services, real-time mass data map reconstruction. Through 5G/V2X information exchange, data exchange between map objects is realized to build a living map. We are not only data producers, but also the definers of a new life. Join us and the future is yours.
