Jia Haonan and Yang Jing from Aofei Temple
Qubit report official account number QbitAI
Shanghai auto show, more restless than auto companies, is the automatic driving supplier.
In order to get a firm foothold in the great change, tier 1 and the car factory show their own skills.
But the excitement is all from the manufacturers, but the users are crazy:
There are so many autopilots. Who should I trust?
What should the real era of automatic driving be like?
At present, domestic car manufacturers and suppliers still use the definition and classification of automatic driving by American SAE
But on the road, it seems that the performance is not bad.
In 2021, automatic driving technology makes ordinary users more and more confused.
Now we know that at least the L5 level, that is, the automatic driving technology which does not need human operation at all and even cancels the steering wheel, has not yet been realized and is not allowed by relevant laws and regulations.
Secondly, the L4 level automatic driving experience, in addition to the road scene, in more detailed and complex garages and charging stations, must also be able to park and find the way independently, so that users can realize remote collection, return, automatic charging and other operations at the resident points (such as home and company) through mobile phones and other mobile terminals.
Finally, urban road navigation and terminal autonomous parking are integrated to give users end-to-end, full scene coverage of automatic driving experience.
At present, the main domestic automatic driving technology providers are Tesla and Xiaopeng. In the tier 1 camp, there are old players Baidu Apollo and Xiaoma Zhixing, as well as new entrants Huawei, Dajiang and didi.
At present, there are data available on the fundamentals of various automatic driving routes as follows:
Data source: Beijing automatic driving road test Annual Report
So to be fair, let's compare the data of several major domestic automatic driving suppliers (as of 2020)
Among them, the data of Xiaoma Zhixing and Wenyuan Zhixing are more than 2 million km and more than 1 million km respectively, and Baidu takes the lead in road test with 10 million km.
From the side view, the number of road test licenses obtained by enterprises can also be used as a reference for the development of automatic driving technology
Among them, 216 were publicly disclosed by Baidu, dozens by Wenyuan Zhixing and about 10 by Xiaoma Zhixing.
In terms of the scale of the test teams, there are generally 100-200 vehicles in China. Only Baidu has 500 test teams
Data sources: official disclosure and public information
Intuitively, in the field of civil vehicles, the car factory and several major tier 1 also have different routes in the selection of underlying sensor solutions
Among them, baidu Apollo and Dajiang's technical solutions support both pure vision and lidar, which are more comprehensive than vision based or lidar solutions.
Similarly, in terms of autonomous driving ability, each company has its own emphasis and difference
At present, Tesla's FSD beta is being tested on a small scale, and only abroad. FSD has been able to achieve the basic city internal navigation, and the performance is good under good road conditions in North America.
The uncle crossing the road directly leads to the exit of FSD
The latest version of Xiaopeng xpilot, including NGP function, can cope with high-speed and urban ring road, but can not adapt to the internal road conditions of the city. For example, in Beijing, NGP can only be activated outside the Fifth Ring Road.
Huawei and Dajiang, which promise full scene automatic driving ability, are just starting, and their real ability has yet to be tested.
Access to the roundabout:
Baidu Apollo's global autonomous driving experience has been commercialized, and the Weima W6, a mass-produced vehicle equipped with these technologies, has just been launched two days ago.
When visiting CCTV, Wima W6 used Apollo's ability to make Sabine exclaim.
Therefore, in general, so far, the global automatic driving function that can adapt to China's road conditions is still the Apollo scheme with the widest coverage and the largest accumulation. The most important thing is that the relevant L4 level function has been put on the market in front of mass production vehicles and accepted the market test.
In this case, take Baidu Apollo as an example to talk about what is needed behind an autonomous car?
Baidu's Apollo Lite (L4 level pure visual perception solution for urban roads) is the key to the entire L4 level automatic driving ability.
Not only in the parking domain (AVP) to achieve intelligent car, but also further extended to the driving domain (ANP), to achieve the freedom of driving in two scenarios.
Besides the algorithm, another factor of automatic driving is computational power.
There are more than 30 deep learning models in Apollo Lite, but they focus on algorithm optimization. A GPU graphics card can run, and the overall computing power is only 30tops.
In addition to computing power and algorithms, Baidu's high-precision map, cloud computing, and the industry chain partnership of Apollo platform are all the elements behind Apollo's automatic driving.
So in general, Huawei and Dajiang have their own emphasis in the current automatic driving scheme and layout. Tesla Xiaopeng is the main engine manufacturer that attaches the most importance to the self research of automatic driving.
Baidu Apollo is the most comprehensive, the strongest and the most complete. It has the global automatic driving ability optimized for China's road conditions. It also has a technology platform that takes into account lidar and pure vision solutions. Behind it is the support of the automatic driving industry chain.
The good thing is that Dajiang, Huawei and other companies have begun to enter the market. Although it will take time to catch up with Baidu Apollo, the market is more lively.
What ordinary users can directly feel is lower vehicle cost and more mature intelligent driving experience.