[This column is jointly produced by Tencent Auto and Kung Fu AUTO]
2547 words Duration to read 4 minutes.
Autonomous driving is beginning to move towards the era of large-scale outbreaks.
Today, Beijing issued a notice of unmanned manned demonstration application, and Baidu and Xiaoma Zhixing became the first batch of approved enterprises.
This also means that the "no one behind the steering wheel" autonomous driving service is opened up for the first time in China's megacities.
With this sensational news, Xiaoma Zhixing also obtained China's first self-driving taxi license last week.
Previously, on Weibo, the bigwigs of the new car-making forces were still in the passionate debate about "where to put the lidar".
It is not difficult to see from all dimensions that the discussion on automatic driving is becoming more and more enthusiastic, and the heat is also increasing.
For a long time, the field of automatic driving has been coveted by car companies, but it is extremely difficult to gnaw on the "hard bones".
With the good news coming from the front, the traditional car companies and new car-making forces in China may once again set off a vigorous scramble for the position.
"Wei Xiaoli" is the first to bear the brunt of it, taking the lead in launching a "highest-equipped" intelligent car in history, and will also have a highest level of "arms race" in various dimensions such as hardware, software, and data.
After all, in order to grab the ticket to the decisive battle, the most technically representative dimension of automatic driving must be the forefront of the battlefield.
(1) Hardware "arms race": Standing still means being backward and being beaten
Mancun, the father of modern economics, once said that it is supply and demand that determines prices, not value itself.
For car companies, to survive needs to meet people's scarce needs, need to have a market, so that people are willing to pay for it.
Therefore, how to meet the needs of consumers is a key part of the landing of automatic driving.
Among them, "hardware" is the foundation of the foundation and the starting point of the "arms race" of car companies.
To put it bluntly, the type, quantity and performance of the hardware used in smart cars will directly affect the landing degree of automatic driving.
Starting from the computing chip of automatic driving, Wei Xiaoli took the lead in the hardware "arms race" to shoot the first shot.
The ET7 just delivered by Weilai took the lead in carrying Nvidia's latest Orin chip, surpassing the previous star chips NVIDIA Xavier chips equipped with Xiaopeng P7 and P5.
The chip hash rate of 254 TOPS per piece, far exceeding Xavier's 30 TOPS.
At the same time, the ET7 is also equipped with four at a time, and the computing power is as high as 1016TOPS, far exceeding all the models of Xiaopeng and Ideal, and directly pulling the hardware configuration full.
Under the pre-emptive strike of Weilai's computing chip, Xiaopeng and Ideal are not far behind, and they are also riveting in the next generation of models.
For example, Xiaopeng's latest model G9, as well as the ideal next-generation model X01, also chose the Orin chip, but it may be slightly slower in mass production time.
Not only that, but the hardware-level race also happens on the perception device.
In the process of achieving automatic driving, environmental perception is the first step, and the carriers are cameras, ultrasonic sensors, millimeter-wave radar, lidar and so on.
Last year, with the mass production of lidar equipped with Xiaopeng P5, more and more car companies joined the lidar camp.
However, in the planning of car companies to carry lidar, there are two routes: one is the standard of the whole series, and the other is optional (only high-end models are equipped, similar to optional).
Weilai ET7 belongs to the standard of the whole series, and it is rumored that all models of the NT2.0 platform replaced by Weilai this year are standard, and the Xiaopeng P5 is optional.
Among the models to be delivered, there are also Zhiji L7, Ideal L9, Xiaopeng G9, etc., Gaohe, GAC Aian, JiKr, Jihu and other brands have plans to launch models equipped with lidar.
It is not difficult to see that in this hardware "arms race", whether it is a new car-making force or a traditional car company, it is accelerating iteration, and no one dares to relax in the slightest.
(2) Software "ahead of the competition": a full-dimensional game of data and algorithms
Of course, the competition in the field of autonomous driving, relying only on hardware "stacking" is far from enough.
Under the hardware, there is also a need for sufficiently intelligent algorithms and data support, which is the key to really opening up the gap between car companies.
Tesla, for example, has been continuously training "bicycle intelligence" through computer vision technology.
In principle, Tesla Autopilot uses neural network chips to simulate the operation of the human brain through software, based on the analysis and learning of existing data, to predict the current road environment.
Such a technical path is quite dependent on road test data, and "long tail problems" are emerging in an endless stream, but its advantages are also obvious.
To put it bluntly, it is to save hardware space and configuration with the help of software, rather than relying on pure hardware to pile up accuracy.
At the same time, the software can provide sufficient functional "imagination" without adding and replacing hardware.
In Wei Xiaoli, Xiaopeng is the first player to publicly claim to be full-stack self-developed.
In January last year, Xiaopeng launched a high-speed scene NGP based on the NVIDIA Xavier platform, becoming the first car company in China to realize self-development and mass production of automatic driving full-stack.
Compared with Xiaopeng, other car companies lag behind a lot.
For example, Ideal, a full-stack self-development of automatic driving only after being equipped with the Horizon J3 chip.
It wasn't until December last year that Ideal provided users with a navigation assistance driving system NOA, nearly a year later than Xiaopeng's NGP.
On the next generation of models X01, the ideal narrows the gap in hardware, but whether it can make good use of the powerful computing power of NVIDIA Orin chips depends on the ideal software self-research level.
Weilai's self-developed software landed the latest among the three new forces, and the three models on sale, ES8/ES6/EC6, were equipped with Mobileye EyeQ4 chips.
However, WEILAI is trying to overtake forcefully, and its first delivery of ET7 will directly cross the NVIDIA Xavier currently used by Xiaopeng, equipped with NVIDIA's latest Orin chip.
Compared with the closed Mobileye, NVIDIA's platform is more open and can meet the needs of self-research of perception algorithms.
Based on NVIDIA's computing platform, car companies can fully independently develop full-stack autonomous driving technologies for positioning and high-precision map fusion, perception algorithm and sensor fusion, behavior planning and control.
In addition to self-developed algorithms, the importance of data for autonomous driving is also beyond doubt.
As we all know, automatic driving is a technology based on machine learning algorithms, data is the basis for algorithm modeling and software landing, and a large amount of data collection is the premise of automatic driving technology development.
From this dimension, Wei Xiaoli's accumulation is leading other domestic car companies, but there is still a gap compared to Tesla.
After all, Tesla has the most mass-produced models in the field of automatic driving, which also continuously delivers real road condition information to the background algorithm for iterative optimization and upgrading.
Earlier and more mass-produced models landed "started", which also means greater and stronger data advantages and algorithm advantages.
However, for domestic car companies, China's intricate real road scene will still be the best "algorithm testing ground" and "data accumulation field" for autonomous driving.
In general, the "advanced competition" of automatic driving software is completely a long-distance run in the true sense of car companies.
(3) Kung Fu shooting
From the small wave in 2018 to the big wave in 2021, the entrants in the field of automatic driving have entered the product landing stage from the start-up stage of barbaric growth technology after years of precipitation.
Whether it is a traditional car company or a new car-making force, it is "new car generation".
Ten years of grinding a sword, the frost blade has not been tried.
The autonomous driving competition of the masses is still unknown, who can become a prince in the industrial chaos of "five generations and ten countries".
But in 2022, China's self-driving players have come to a moment before the decisive battle.
In order to grab the ticket to the decisive battle, this year's autonomous driving industry must be short and fierce.