On April 9th, an important document issued by the central decision-making level stamped the importance of "data elements" and became one of the production factors. Data resources have become a mine in Huang Jinbao that needs to be excavated urgently.
In fact, the importance and imagination of data elements are not limited to this. In the future, the car will no longer be an information island, but a mobile sensing terminal, which is interconnected with roads and clouds and realizes smart travel through technologies such as big data and artificial intelligence. Data is the core factor that links all this.
Among many self-driving players, building a "data closed loop" is the key path for Tencent to help the industry develop and achieve breakthroughs.
The deep logic behind the efficient use of data elements lies in industry understanding and infrastructure support. Tencent Cloud provides powerful cloud service capabilities and builds an efficient autopilot data service system based on this. Therefore, Tencent has its own understanding of how data elements drive autonomous driving.
Based on this, Tencent has produced three lethal products in the autonomous driving business: big data cloud platform, simulation test platform and high-precision map.
Lei Feng. Com Xinzhijia will talk to Su Kuifeng, general manager of Tencent Autopilot Business Center, trying to understand how Tencent uses data elements to drive autopilot and the logic of efficient linkage of the three businesses.
In the vast cyberspace, data is a brick to build a huge virtual building.
In terms of autonomous driving, data runs through the whole life cycle of R&D, production, testing and operation in various forms. At the same time, the data explosion has also increased exponentially. It is conceivable that players are faced with tsunami-like data.
Therefore, in the use of data elements, the player's ability in two dimensions is very important: first, the data is closed-loop, and without this ability, the validity of the data cannot be verified; The second is how to realize efficient data operation on the basis of data closed loop.
Su Kuifeng told Xinzhijia that the core competitiveness of autonomous driving lies in the low-cost acquisition and efficient use of data elements and computing resources. Efficient collection and utilization of data elements to improve the speed of data circulation is the key point of the whole iteration of autonomous driving technology.
To this end, Tencent has launched three major services: big data cloud platform, simulation test platform and high-precision map. In order to understand the efficient driving of closed-loop data by these three business platforms, it is necessary to trace the chain generated by autopilot data. Su Kuifeng gave an example to the new intellectuals:
Thus, a data circulation chain is formed around data collection, scene construction, test and verification, and operation update.
Su Kuifeng also believes that the improvement of the utilization efficiency of data elements is a matter of the whole chain, not a certain point.
For example, from the perspective of data collection, it is very important to know what kind of data to collect. After understanding the data requirements of day and night, congested and non-congested road sections, dynamic scenes can be automatically or manually calibrated and triggered, which improves the efficiency of data collection and avoids duplication.
After the vehicle is really on the road, it can automatically filter out some useful data and send it back to the cloud according to the corresponding trigger conditions; In addition, Tencent will also provide a data management system to tag the collected data, quickly clean up, screen and find the needed data for efficient circulation.
Tool chain covers the whole automatic driving link from data collection, data training, algorithm module evaluation (including model in the loop, software in the loop, vehicle in the loop, hardware in the loop) to actual road test.
"Every link of every tool in this closed-loop system is improving the circulation and utilization efficiency of data elements in order to solve problems quickly or accelerate research and development." Su Kuifeng said so.
For example, in case of corner cases, Tencent can find relevant cases from the database for marking algorithm training, or quickly collect corresponding data to ensure the stability and effectiveness of data closed loop.
At present, Tencent can provide cloud tool chains including scene classifiers and other production tools for autonomous driving systems above L2.5 level, and at the same time provide some standards for vehicles and clouds, and users can choose according to their own algorithm requirements.
Based on the understanding of data closed-loop and efficient operation, Tencent's purpose in autonomous driving is also obvious: to provide industry customers with software and services that can efficiently collect and utilize data elements, speed up data circulation, help industrial technology evolution and accelerate product landing.
In other words, it provides industry customers with a tool chain for autonomous driving cloud development, which integrates data collection, training, evaluation and update. This is also the tool attribute of Tencent in autonomous driving that Su Kuifeng has been emphasizing.
Tencent's big data cloud platform, simulation test platform and high-precision map constitute the product form of these tool chains.
Taking the simulation test platform as an example, it can be understood that TAD Sim, Tencent's simulation platform, is a large-scale role-playing game for self-driving vehicles, which combines professional game engines, industrial vehicle dynamics models, virtual and real traffic flow integration and other technologies.
Su Kuifeng told Xinzhijia that one of the core functions of simulation testing is to transform the collected data into useful test scenarios.
He stressed: "The tool itself requires us to use data, and we also have data element reserves. But for simulation, the core role lies in the tool rather than the data itself. With this tool, the data collected by vehicles can produce a large number of test scenarios. "
Tencent's TAD Sim simulation platform can also provide stand-alone and cloud versions.
The stand-alone version can edit the scene and do various tests and verifications. The cloud version provides high concurrent testing capability in the cloud, including scene cloud simulation and virtual city cloud simulation.
Scene cloud simulation generates hundreds of thousands or even millions of test scenarios through a large amount of data, and accelerates in parallel on a large scale in the cloud to realize efficient verification of autonomous driving algorithm.
Virtual city cloud simulation can load real or edited city-level high-precision maps, realize parallel acceleration of millions of traffic vehicles and thousands of self-driving main vehicles, and conduct 7×24 hours of uninterrupted testing.
By constantly looking for corner cases or accumulating scenes that are not well handled by the algorithm, the scene library of autopilot test is enriched.
Of course, behind this, you can clearly see the traces of Tencent's powerful game technology support.
"The simulation system can break the data link and then verify it in stages. At the same time, this chain will be integrated for verification. This is closer to the actual road test in a certain sense. However, it must be emphasized that real vehicle testing is always needed, and simulation can never replace real vehicle testing. " Su Kuifeng said.
Previously, Tencent cooperated with the National Intelligent Networked Automobile (Changsha) Experimental Zone in the Intelligent Networked Automobile Simulation Laboratory.
Using high-precision map and simulation technology, the geographical panorama of the experimental area is digitally modeled, and the safe and efficient smart car experiment is realized in the simulation environment.
In addition, the simulation test platform is inseparable from the combination of big data cloud platform and high-precision map.
Su Kuifeng said that cloudization is the general trend in the future. Not only is data stored in the cloud, but many services and terminal decisions of the client will also change with the strengthening of the cloud trend.
"In the future, with the enhancement of 5G communication links, the upgrading of software architecture and hardware architecture, and the enhancement of cloud capabilities, autonomous driving will gradually migrate from end distribution to the cloud." This is why Tencent built a big data platform.
In addition, Tencent said that TAD Sim provides a map editor, which can directly edit high-precision maps or directly import the actually produced high-precision maps. TAD Sim provides a universal high-precision map interface, which can load the information of road elements in the map and also import three-dimensional environmental information such as buildings and trees.
"Overall, regardless of the car or the cloud, this closed loop is a nested system. The final presentation form can be a single module, but if we want to improve the efficiency of data communication and development, we need to closely couple this system to maximize efficiency. " Su Kuifeng said.
In the closed-loop system, the higher the efficiency of algorithm and data flow, the lower the cost of autonomous driving and the stronger the core competitiveness.
There is a complete closed loop on the tool chain, but Tencent's business model is flexible in business strategy.
At present, the "whole family barrel" packaging obviously cannot meet the current OEM's demand for product customization.
You can fight alone or together. In other words, the three businesses can be modularized or integrated into a team battle. All in all, flexible assembly, and even a certain degree of customization according to industry needs.
At the same time, with the blessing of three core business capabilities, Tencent has been polishing its own autonomous driving solution.
Compared with other players, Tencent does not classify solutions by autopilot, but provides scenario-based autopilot solutions based on users' high-frequency needs, and gradually realizes autopilot landing according to scenarios and requirements.
Starting from 20 19, Tencent has targeted the two just-needed scenes of high-speed users and parking users, and will launch automatic driving mass production solutions for these two scenes.
Su Kuifeng told Xinzhijia that at present, Tencent's high-precision map team has completed the collection and drawing of high-precision map data of expressways and expressways nationwide, laying a good foundation for realizing automatic driving of high-speed scenes.
As for commercial landing, Su Kuifeng said that due to the intervention of Internet companies, the inherent cooperation mode of traditional OEMs will change, especially the trend of separation of software and hardware is becoming more and more obvious. "The mode and mechanism of cooperation are changing. At present, Tencent and OEM are also promoting the upgrade of the model through some cooperation. "
For example, Tencent can provide both the algorithm module of map positioning and the algorithm module of perceptual fusion for the vehicle-side solution. However, different manufacturers and different sensor configurations still need to be customized, and it is difficult for the universal module to adapt to all models.
As for the big data cloud platform, the self-driving high-performance data development platform jointly developed by Tencent and BMW China has been delivered.
Just like the auxiliary role in the game, under the concept of "auxiliary output", Tencent is integrating into the autonomous driving ecology of OEMs and players in the industry with a very flexible attitude.
As Tencent CEO Ma wrote in a circle of friends: "Help car companies develop their own autonomous driving AI algorithms and big data platforms." Tencent, with its understanding and practice of data closed-loop efficient operation, is expected to help bicycle companies reach the future of autonomous driving as soon as possible.
(Lei Feng Net) Lei Feng Net