In conjunction with the car enterprise to achieve the unmanned timetable in stages, the map merchants also define the map products in stages.
Recently, DFC New DTO Dai Donghai updated the progress of high-precision maps in an interview: In June of this year, the company had the ability to support Level 2 national highway network coverage data and products to achieve mass production. The Level 3 map is being promoted according to the commercial plan in mid-2017. The Level 4 high-precision map has been pre-verified in two cities in Beijing and Shanghai, and some work has already been done. By the beginning of 2019, Map4 data of major cities will be provided.
After giving a specific time plan, how to provide unmanned products for car companies, how unmanned driving and car networking services work together, what changes in NavInfo's business processes, and Dai Donghai talked about some details information.
Four-dimensional map new CTO Dai Donghai
Chip-level map schemeIn May of this year, NavInfo acquired MediaTek, a subsidiary of MediaTek, a provider of in-vehicle infotainment systems solutions. The two companies will cooperate to develop and produce low-power and cost-effective chips. The main function is to perform map and sensor related operations in the driverless process, including processing the sensory data from the vehicle-side sensors for high-precision matching and decision making. Extract road attributes, feature information, and more.
Specifically, in the driverless process, high-precision maps need to communicate with the sensor at all times. On this chip, the data of the sensor and the high-precision map are merged to complete the real-time sensing task. At the same time, when the sensor crowd collects the map data, the deep learning algorithm written in the chip can analyze and process the data collected by the camera and the laser radar, and extract relevant information of the map, such as traffic and traffic facilities, to improve the unmanned map. .
The deep learning algorithm involved in the chip, the deep learning laboratory of the NavInfo New Basic Technology Research Institute is being developed and improved by itself. Taking the identification of traffic signs as an example, Dai Donghai said that the recognition rate achieved by the algorithm has surpassed that of human beings, and it is basically fully automated.
How to store high-precision mapsCompared with the high-precision map participation in the fusion data operation, the industry said that the more challenging is how the high-precision map is stored in the car. The high-precision map includes the basic navigation layer, high-precision road network layer, real-time dynamic layer, and many related environmental influence factors. There are crossovers on the layer and the traditional navigation map, but it will add a lot of fine elements, some of which require update. Non-high basic data will be stored locally. As the level of driverlessness increases, on the basis of ensuring the clarity, we must also find ways to control the size of the map data within a reasonable range.
The answer to Dai Donghai is that the storage capacity of the car end can support the storage of high-precision maps. NavInfo's new Level2 and Level3 driverless data is not large. As long as it increases by 1/3 of the traditional navigation map data volume, it can cover the current highway network nationwide. After Level 4, the amount of data will increase, but the amount of data is not very large.
Including data collected by the camera and lidar sensor crowdsourcing, the algorithm also performs vectorization to reduce local storage and operations. Although the data types collected by the lidar and the camera are different, because of the process of data vectorization, if the depot adds a laser point cloud based on the original camera solution, the logic and solution of the entire architecture can be continuously extended.
Driverless and car networking will mergeAccording to Dai Donghai, in the Level 2-Level3 driverless, this chip will be placed on the car end, which is also used to support the operation of the NavInfo new car networking products. At the Level 4 level, as the amount of computing increases, a separate chip can be provided for unmanned sensing data processing according to the needs of the vehicle enterprise.
It is worth noting that when Dai Donghai introduced this unmanned map service, he classified it as the car network version 3.0. The data collected by the vehicle-side sensors will be uploaded to the cloud after being processed locally. Some confirmed updates will be sent from the cloud to the local update map, and the maps will be used to support unmanned missions.
Driverless and car networking will merge, and a chip-level mapping solution will better connect the NavInfo new car network with the unmanned business. Throughout the process, NavInfo has also evolved from a simple map data supply to a data service for the entire process.
Tool chain innovation facing graphicsThe reason for mentioning this is because Dai Donghai believes that this is as important as carrying out the driverless business.
He introduced, in particular, the development of support for Level 4 driverless, the use of laser radar equipment and crowdsourcing methods to collect more and more data. It is not realistic to rely on manual processing for large amounts of data. It requires an automated intelligent system to participate.
In the process of electronicization of some physical worlds, some automated methods of operation have gradually been added. For example, a deep learning algorithm can be used to identify information such as POI position points and signs in the collected material, so that the machine can replace the manual mark. If the application of deep learning in the driverless car is to liberate the driver, the application in the field of maps is "Take the cartographer away from the mark."
We can also see some changes from the company structure. In the new research institute of NavInfo, four laboratories of intelligent driving, deep learning, future navigation engine core and cloud service have been established. In addition to the algorithm for developing the perceptual part, the deep laboratory also uses the various aspects of deep learning and map operations for the intelligent upgrade of the tool chain.
However, Dai Donghai also admitted that the change and update of the operation process is not a one-step process, but a process of gradual reform. After the deep learning recognition sign is recognized, it is necessary to further understand the meaning of the sign. The innovation of the tool chain inside the map is also gradually realized. "This is the same as the development of driverless."
EvoTec special generators mainly include 4 Pole,Evotec Special Generator,4 Pole Alternator,Ip22 Generator,Explosion-Proof Generator, high-efficiency special type generator (350kVA efficiency up to 96%), IP55 Generator (delivered more than 20 units to European Countries ), and medium-speed generator , Generator with auxiliary winding ,which are widely used in industrial fields such as mining, metallurgy, oil fields, etc.
EvoTec has several senior experts in the field of generators, as well as large-scale owned factory. with strong independent research and development capabilities, EvoTec builds customized alternators adapted to specific power plant applications.
Some features for evotec special generator :
Low losses magnetic steel
reinforced construction for high linear vibrations
flanged shaft for direct coupling with engine flywheel (in case of single bearing solution)
neutral point terminals in separate terminal box
increase protection degree up to IP 54, IP 55
insulated bearing and back shield
digital AVR
excitation PMG mounted on generator
lubrication system for sleeve bearing
other options available on request
EvoTec Special Generator
4 Pole,Evotec Special Generator,4 Pole Alternator,Ip22 Generator
EvoTec Power Generation Co., Ltd , https://www.evotecgen.com