At the Baidu World Congress in September this year, the L3 Smart Car Division (hereinafter referred to as the “L3 Business Unitâ€) was announced. Unlike Baidu ADU, the new department is taking a gradual development route, aiming at the L3 autonomous driving that is more commercially viable at this stage.
At the Baidu Yunzhi Summit in 2016, Che Yunyin once again encountered the L3 business unit. At this time, it was three months since the establishment of the department. We have some new understandings about L3 product development and business planning.
Baidu L3 Business Unit Product Architecture
According to Gu Weijun, general manager of Baidu L3 Smart Car Business Unit, the product structure of L3 Business Unit is “1+3†mode. For L3 autopilot (positioning, perception, decision, control), car networking (CarLife, Map Auto, CoDriver, MyCar), automotive big data services (user analysis, CRM) three directions, the map is the basic support of the entire product architecture.
In L3's product architecture design, the map serves two roles, one is used as a sensing tool for self-driving cars, and is a map for self-driving cars; an interface used as a self-driving car to interact with people. It is a map that human drivers look at.
In autonomous vehicles, sophisticated high-precision maps can be used with sensors to achieve the vehicle's perception of the outside world. High-precision maps will write information such as lane lines, signal lights, signs, etc. that do not change frequently. If you fuse with other sensors while driving, you can reduce the burden of real-time operation of the sensor accordingly. For example, “there is a signboard with a speed limit of 60 on the right side of the road.†When driving to the vicinity, because the accurate position of the signboard is already in the high-precision map, the car camera can analyze the area with the signage in the picture in a targeted manner, confirming Speed ​​limit information, eliminating the need for full frame analysis to reduce the amount of calculation.
On the other hand, it is possible to use a mass-produced sensor such as a camera to match the high-precision map for precise positioning. Similarly, there is a signboard with a speed limit of 60 on the right side of the road. After the camera captures the speed limit card, it can match the position of the speed limit card in the map to determine the distance between the lane in which the vehicle is located or from the emergency stop. According to Gu Weijun, some car companies and the L3 business unit have already cooperated in this regard.
At present, Baidu's high-precision map coverage is more than 10,000 kilometers, including China's highways and urban expressways. The relative accuracy of 10-20 cm, the absolute accuracy of 60 cm, and the automation rate of up to 90% in the production process are the information given by Baidu to prove the high-precision map capability. The cooperation on high-precision maps is at the stage of Baidu providing sample data and assisting car companies to test and jointly develop. Baidu's high-precision maps are really applied to smart cars depending on the OEM's product planning, and the time may be in 2018.
Self-driving cars also need to communicate with people and need to give people a map. Zhang Hui, general manager of HMI & Ecology of Baidu L3 Business Unit, compares the experience of taking a self-driving car to a special car - enter the starting point and end, get a rest after getting on the bus, wake up to know where it is. As a function of high frequency use in the car, he believes that the map will become a human-machine interface. Our interaction with the car will be carried by the map, and voice will be the main way of interaction. Baidu car networking products are intended to use the mobile phone car interconnection program CarLife, the Baidu map car version into the car, through the car voice CoDriver to interact with the vehicle.
A new map production model extended by the Internet of Vehicles
One of the reasons to package car networking products with autonomous driving is to lay the Learning Map. When Gu Weizhen introduced the software service architecture, we can see that there is data exchange between the Internet of Vehicles product and the Learning Map in the cloud.
Learning Map is not a new concept, it is what we often call "crowdsourcing map." The road information is collected by the on-board sensor of the moving car, and the map data is supplemented and updated after the cloud is returned. The reason for this approach is that the cost of surveying vehicles is high. At present, there are only 10 vehicles in Baidu's 250 surveying and mapping vehicles that can be used for high-precision map data collection. The most important reason is high-priced equipment. On the other hand, in addition to the fineness of the high-precision map, it is also necessary to emphasize the "freshness" of the data, so that the map data update is contributed to the car that was originally running on the road, and the cost performance is the highest.
According to reports, Baidu has used the data returned by the mobile terminal to update the standard map. It has already been tried for two years or so. It can be used to return the picture to the mobile phone, and the background judges whether there is any change or not. Now the project in Baidu's experiment is to update high-quality high-precision maps with low-quality image data. Gu Weiqi gave an example of the road change image recorded by the driving recorder to supplement the 4K image of the collecting car with a 768 resolution image.
At this stage, the post-installation hardware manufacturers that cooperate with Baidu have become the data providers. The driving recorder itself has a camera function, which automatically powers up when the vehicle is started, which is more direct than the mobile phone. Baidu also listed the camera hardware of the vehicle itself as the data input port of the Learning Map, but because of the open data, it can only be the next step.
How does L3 autopilot work with the OEM?
Positioning, perception, planning decision-making, and control are the four core parts of autonomous vehicles. The auto-driving program provided by Baidu L3 Business Unit to the main engine factory also wants to penetrate into these four parts step by step.
At present, the cooperation between the OEM and Baidu L3 is still in its infancy. According to Gu Weijun, many OEMs have independent research and development capabilities, and only need local high-precision map data. The cooperation with Baidu L3 business unit only includes high-precision map and mining map solution, and some OEM cooperation is in sensor plus map. The extent of positioning.
On the eve of the 3rd World Internet Conference on November 14th, Baidu L3 announced the first unmanned truck in China to be launched jointly with Foton Motor. According to the on-site information, the vehicle uses Baidu high-precision map and Foton's automatic driving technology to achieve Level 3 automatic driving level.
Fukuda self-driving truck exhibited at the event
Baidu L3 business unit has just started, but not from scratch. This new department will be supported by basic capabilities such as Baidu's artificial intelligence, maps, cloud computing, and big data. The already-built car networking business will also help auto-drift business expansion. For how to impress the car companies, enter a more in-depth cooperation level, waiting for the L3 business unit to bring more information.
USB Charger
The Universal Serial Bus (USB) port on computers and laptops is a data port that also serves as a 5-volt power source.Virtually all portable, personal electronics feature a USB port, though the connector can vary between one of several standards. The charger plugs into an AC outlet, and a USB Cable plugs into the charger.
USB Charger,30W USB C Wall Charger,Fast Charging USB C Wall Charger,Ports 60W Wall Charger Home Charger
Pogo Technology International Ltd , https://www.wisesir.net