NOA, a live test of the Great Wall in Wei Jianjun： ＂The speed is nothing, the city really has no map＂, and there are 3,000 people in the intelligent driving team, so there is no anxiety at all.

The head of Great Wall MotorWei Jianjun, also live in person!

Getting up is the hardest of all subjects.NOA, a city without graph.

What is more challenging is that the route selected by Lao Wei is not only the main road of the city, but also the narrow road, construction road or extremely complicated irregular intersection in the old city of Baoding, a "second-tier city".

Without the rules and regulations of the North, Guangzhou and Shenzhen, "chaos" is better, but the more chaotic it is, the more intelligent it is.

Wei Jianjun directly put words:

Juan Zhi drove to Hebei and welcomed friends to challenge our NOA line.

The system configuration measured in Wei Jianjun is based on Orin-X high computing power domain control platform, with 1 laser radar, 3 millimeter-wave radars, 11 high-definition cameras and 12 ultrasonic radars.

The time is 10:30-12:00 in the morning.

The live broadcast route starts from the Haval Technology Center of Baoding Great Wall and passes through a short section of expressway:

The high-speed scene is not difficult, and the up and down ramps are all completed by the system.

Wei Jianjun comments:

NOH, the high-speed highway of the Great Wall, was mass-produced several years ago.At that time, it was the industry leader, and now it is no challenge.

The old friends who refer to smart cars know that the high-speed NOH in Lao Weikou is the autonomous driving company of Great Wall.Hao mo zhi hangProducts,As early as 2021,It is mass-produced in Wei brand, tank, Euler and so on.

At that time, the mainstream of the industry was still competing whether the L2 family bucket was standard.

After getting off the expressway and entering Baoding City, the challenge really began. Wei Jianjun also emphatically introduced the road conditions in Baoding.

Compared with super-first-tier cities like Beishangguang, it is characterized by many irregular and asymmetrical intersections:

There are many sudden violations of traffic regulations:

And, in the case of narrow roads, it is necessary to complete the diversion of turning lanes:

Renovation of the old city, the construction on the road occupies many roads and is irregular:

And this set of map-free NOA also successfully challenged Baoding’s unique "six-way mouth" with hell-level difficulty:

The whole intelligent driving system has zero mistakes, no emergency exit or demotion, and has perfectly completed various games, detours and avoidance actions.

Wei Jianjun said,Autopilot of the Great Wall has been working for 10 years., but everythingTake "safety" as the key link, never show or brag if you are not sure.This live broadcast in person has enough confidence in the progress of its own products.

Actually, there is a doorway in the car circle, and the "gold content" can be divided into three grades.

The third file is the "zero takeover" Demo that is often seen in press conferences and official numbers. You know … you can’t say that you can’t show any strength, but it is definitely the effect of the official "polishing", and the degree of reference is limited.

The second file is provided by the government for everyone to experience, such as the media test drive meeting and the 4S shop’s intelligent driving experience line for ordinary users. The route is selected, "tempered", and it has been run in the background for countless times, which belongs to the open-book exam. We can see some basic skills, but in the face of complicated and changeable open roads, universality and stability are not guaranteed.

The first gear is to throw it directly into the rolling traffic without any special adjustment and preset, which can best test the true level of intelligent driving. It is common for users and fans to test spontaneously.

So we can see two key factors of intelligent driving test, one isHave you optimized the debugging for a specific route?; Second, yes.Is there a high-precision map in the background?.

Wei Jianjun live measured NOA of the Great Wall, which file does it belong to?

Great Wall Motor revealed to the smart car reference that the NOA just measured in Wei Jianjun,There is definitely no high-precision map.. Completely real-time perception, and OCC, which is hot in the industry, also got on the bus.

Whether it is a special tune or not, the Great Wall has not deliberately concealed it. becauseThe live broadcast was suddenly decided by Wei Jianjun last Monday.There is no notice and plan.

Therefore, in the past seven days, the Great Wall Intelligent Driving Team has been optimizing the software algorithm, but this optimization is aimed at the algorithm itself at the vehicle end, not at the test route.

In fact, the road conditions in Baoding are too complicated and there are too many emergencies, and the road renovation construction is changing day by day.It is impossible to have an "open-book exam".

That’s why in the live barrage,Some users bluntly said that Wei Jianjun is a "real person.".

Because of his first live broadcast, in addition to the demand and desire of the Great Wall for traffic (who doesn’t ~), the core purpose is to show the technical strength and product progress of the Great Wall in Zhijia.

Besides, the Great Wall is not just talk. Today, NOA, the city tested by Lao Wei, is still an engineering version, but Great Wall revealed that the production version will get on the bus in the middle of this year, and the first model is the new Wei brand Blue Mountain.

In fact, the interior of the new Blue Mountain has been "leaked" in the live broadcast. I wonder if you found it?

Great Wall Motor is the first of the old car companies.Demonstrate an accurate understanding of intelligent driving technology and AI systematization abilityPlayers.

It is also the only traditional car camp player who really understands AI and enters AI, instead of trying to solve the problem once and for all through the "turnkey" scheme of purchasing suppliers.

In fact, this will not solve the problem at all. The "first brother", an independent new energy source, has suffered greatly. So far, he can’t get rid of the hat of "unintelligence", so he can only blur the concept of electrification and package it into intelligence to show people.

So, what kind of understanding and technical system does the Great Wall have?

The NOA function actually measured in Wei Jianjun today is the latest of the Great Wall. SEE integrated large model.

The point is,integration. The core connotation is the link integration that realizes the integration of perception and decision.

It is not a strict end-to-end big model, because this "big black box" lacks interpretability.

There are still sub-models of perception, planning, decision-making and so on, but they are no longer independent modules. Firstly, the perception model based on Transformer architecture extracts features from the input data of each sensor.be harmlessInput to the subsequent planning and perception model.

Integration is embodied in the fact that the Great Wall uses data-driven instead of rule-driven.Instead of defining what is good driving behavior by rules, the model can learn "how to drive" directly like a human driver, which can greatly improve the complexity of the road conditions and make decision-making driving behavior closer to human beings..

The "lossless" is reflected in OCC occupying the network to get on the bus, subverting the "white list" mechanism of previous perception and recognition. The system does not need to know what obstacle is in the way, only needs to know where it is, and the double redundancy of lidar is enough to output highly accurate perception results.

The planning, decision-making and other modules behind perception all use data-driven models, and such a large model of massive human excellent driving case data is itself large enough.

Going back to the core of the Great Wall’s most emphasized intelligent driving safety, in engineering practice, the algorithm will inevitably have false detection and missed detection, or some relatively extreme driving behaviors that are inconsistent with human habits, some of which are security risks, and some are not conducive to building user confidence.

Therefore, the Great Wall specially added a layer of artificially written rules "shell" outside the data-driven algorithm as a bottom-up measure in extreme scenes.

Therefore, the meaning of SEE is actually like this: multi-sensor redundancy in hardware, OCC in software, "safety" security represented by pre-fusion algorithm, and "Efficiency" efficiency driven by data under the framework of large model.

Finally, it is an "Experience" experience that you can drive anywhere (with navigation) and stop anywhere (all kinds of difficult parking spaces, memory parking, etc.).

If you still question the intelligent strength of the Great Wall, there is probably one last angle:

What about computing power?Without infrastructure, it’s no use talking about it.

The Great Wall also announced the latest situation:nine administrative divisions of China in remote antiquity —— finally, the force scale reaches 1.64 EFLOPS. Huawei, which was "far ahead" two days ago, announced the figure of 3.3 EEFLOPS.

One EFLOPS is equivalent to 10 trillion floating-point operations per second.

In addition, the Kyushu Supercomputer of the Great Wall can also have 5T/s high-performance storage and a high-performance network with a communication bandwidth of 3.7TB/s. It can realize real-time capture of hardware anomalies, minute-level anomaly recovery, continuous training of thousands of cards for several months, and the training cost of hundreds of billions of parameter models is reduced to the level of 100 cards per week.

From the first live broadcast of intelligent driving by Wei Jianjun, the head of Great Wall Motor, we can see the level of "leading" intelligent driving in the industry at present.

First of all, the function experience no longer depends on how many L2 functions come standard, because it’s useless to have more standard items, a bunch of function ODDs don’t overlap, and the experience is sparse. In 2024, such a product car can’t get on the table.

Must be standard for high-speed NOA entry.Full coverage of intelligent driving from toll station to toll station, including macro-following navigation in and out of ramp, as well as micro-game lane change, overtaking detour and so on.

Secondly, subject to the end of 2024, the city NOA must also be mass-produced and get on the bus, and it does not rely on the experience of high-precision maps-it can be opened if there is a road.

From the technical system construction displayed by the Great Wall, we can also see what knowledge and strength the head players need.

The first is the construction of computing infrastructure. In China, Huawei is taken as an example, and NOA without map should be based on at least several EFLOPS.

Secondly, the framework system, there are two key points, one is the BEV perception model based on the Transformer architecture, and the other is the data-driven AI model that permeates all aspects of intelligent driving.

Data-driven is very important, which is directly related to whether smart driving products dare to use or not.

Finally, the Demo test is turned into the engineering ability of mass production, which includes the verification and testing of technical solutions, sensor calibration, safety certification of vehicle regulations, software and hardware adaptation and so on.

With such product progress and the systematic ability behind it, domestic autonomyWisdom driving the "five permanent members"Situation, is gradually taking shape:

Representatives of large technology companiesHuawei, DJI, representatives of new forces building carsXPENG Motors, representative of autonomous driving startup company MomentaAnd the only representative of the transformation of established car companies,Great Wall Motor.

And doing smart driving, the Great Wall has other advantages that players who have landed in the north, Guangzhou and Shenzhen do not have- Baoding.

Momenta is also one of the partners of Great Wall Intelligent Driving. CEO Cao Xudong said to Great Wall not long ago:

"The effective data of intelligent driving collected in Baoding in one week is more than that collected in Shanghai in one quarter".