news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

NIO is the first to move from "software-defined cars" to "AI-defined cars" in the industry. Whether it is chips, vehicle operating systems, or new-generation electronic and electrical architectures, they have laid a solid foundation for generative AI to dominate a car's performance, control, and digital interaction between people and smart cars. This foundation can support a hundred-story building.

Text丨Wang Xin and Li Muyu from Zhijia.com
Editor | Langlang Mountain and Mingzhi Mountain

At 2:00 p.m. on July 27, NIO IN, Shanghai World Expo Center.

During the nearly three-hour NIO Innovation Technology Day, NIO showcased a series of its latest technologies and products, including the world's first 5nm intelligent driving chip "NIO Shenji NX9031", the vehicle-wide operating system "SkyOS Tianshu", end-to-end intelligent driving based on the world big model, and the new NIO Phone.

This is actually more like an AI DAY for car companies, but unlike in the past, no one is saying that NIO is not doing its job.

Last year, NIO's R&D investment reached 13.43 billion yuan, but its monthly sales have been slow to achieve the target of 20,000 units set at the beginning of the year. In November of that year, NIO launched a round of layoffs. Under outside scrutiny, President Qin Lihong responded in public: NIO will not and cannot go bankrupt.

From the outside, it seemed like NIO was in a precarious situation at that time.

But in 2024, NIO's sales exceeded 20,000 units for several consecutive months. The "squat and jump" that it had been shouting for two years finally began to reap positive feedback in the terminal market. In May this year, its second brand, Ledao, was launched and received rave reviews. The third brand, Firefly (internal tentative code name), will be released in the second half of this year. At the same time, the battery swap model that has been questioned for many years has established the broadest alliance in the industry, and has carried out all-round strategic cooperation on battery swap with seven automakers including Changan Automobile, Geely Holding Group, JAC Group, Chery Automobile, Lotus Automobile, GAC Group, and China FAW, as well as many companies in the fields of transportation and energy.

With a full-stack layout in the smart electric vehicle industry chain, NIO has launched a high, medium and low-end combination of NIO, Ledao and Firefly in terms of vehicle categories, while establishing a complete energy Internet architecture based on the battery swap model. NIO, which is "looking at the present from the future", has gradually shown the big chess game it has laid out to the market and opponents relatively clearly (see "" for details).

The continued upward growth in sales has greatly enhanced NIO's confidence in further increasing its research and development. On this Technology Day, Zhijia.com could clearly feel that Li Bin and NIO's confidence has returned, and they are no longer worried about the outside world's doubts about their expansion.

01.
5 nm intelligent driving chip taped out successfully

Regarding NIO's full-stack self-developed technology puzzle, Li Bin has shown the following picture on more than one occasion.

A closer look reveals that NIO has single-handedly built a NIO-style smart electric vehicle ecosystem.

Compared with the 2023 Nio In Technology Day and the eT9 launch conference, this year can be said to be a day for displaying the results of the big plan that Li Bin has drawn.

What Zhijia.com is most concerned about is the operating system and intelligent driving chips.

Last year, Li Bin demonstrated two chips. One was "Yang Jian", the first domestically developed and mass-produced laser radar master control chip, and the other was the Shenji NX9031, an intelligent driving chip.

The biggest progress this time is the official tape-out of the Shenji NX9031.

Li Bin, unable to hide his excitement, suddenly took out the smart driving chip from his pocket and announced loudly: "Tapeout was successful."

Li Bin said that it is the industry's first intelligent driving chip manufactured using 5nm automotive-grade technology, with more than 50 billion transistors. Based on the CPU computing group with super-strong task concurrent processing capabilities and ample memory access bandwidth, fast and stable processing delay is achieved.

Here it is actually benchmarked against Nvidia Orin-X and Horizon Robotics 6 with 7nm process.

For chips, a higher process speed does not naturally mean higher performance, but it is definitely a reflection of design capabilities.

However, NX9031 did not show more parameter details, such as the NPU computing power that is almost equivalent to intelligent driving capabilities. Li Bin only focused on demonstrating the high-performance ISP effect on the big screen.

In order to verify the high performance, Li Bin released some comparative test clips on site, especially in harsh environments such as dim light at night and rainy days.

The comparison results with the industry's flagship intelligent driving chip show that the self-developed ISP of Shenji NX9031 has clearer imaging in conditions such as rainy days, foggy days and nights.

I will not elaborate on its innovations here, but simply show Li Bin's description of its performance: Shenji chip's BEV algorithm performance improvement4.3 times, Transformer algorithm performance improvement6.5 times, Lidar point cloud processing algorithm performance improvement4 times。

"It's not just an ordinary lead," Li Bin emphasized.

However, Zhijia.com has actually been paying attention to the sales of the laser radar main control chip "Yang Jian" released last year, but this time Li Bin did not introduce relevant information about this chip.

The reason why Zhijia.com pays attention to this information is that many people in the industry have expressed to Zhijia.com their pessimism about car companies' self-developed chips.

The reason for this is very simple, that is, the investment in chip research and development is large and the payback period is long, and it is difficult to recover the cost in a short period of time. Chip research and development is a typical economy of scale.

For NIO alone, its demand for chips is difficult to cover the cost of chip design and R&D, which means that NIO must find the needs of non-NIO brand car companies as soon as possible.

"Even if NIO achieves 300,000 units of sales per year, it cannot guarantee a positive cash flow for its chip business," a chip supplier once told Zhijia.com. "At the same time, the chip iteration cycle is getting shorter and shorter, which means continuous investment is necessary. At the same time, more and more brands are starting to consider whether the ecosystem based on this chip is mature when choosing a chip."

Of course, the Wintel alliance between Intel and Microsoft has facilitated the global conquest of the PC ecosystem based on this combination.

Of course, Li Bin and NIO have always had a strong ability to form alliances. The battery swap model, which has been criticized by the industry, has established a powerful battery swap alliance across brands and models in China.

Whether the Shenji chip will gradually evolve into the Shenji chip ecological alliance depends on Li Bin’s integration capabilities.

In addition, NIO is not the only car company to develop its own intelligent driving chips. Tesla was the first to do so. After Tesla abandoned the Mobileye and NVIDIA platforms, it launched HW3.0 and HW4.0 one after another. The computing power was upgraded from 144 TOPS to 300-500 TOPS, which is not inferior to NVIDIA at all.

Li Bin not only develops his own chips, but also develops his own vehicle-wide operating system "SkyOS Tianshu" to create a truly efficient hardware and software integration for smart cars. It can be said that he is learning from Tesla while paying tribute to Apple.

However, judging from the founder's personality, Li Bin is closer to Musk, as both are gentle and sensitive Cancer men.

With the successful tape-out of the Shenji NX9031, its mass production and delivery is approaching. Li Bin said in the first half of the year that it would enter the mass production stage in October.

Next, let’s focus on analyzing its released vehicle-wide operating system “SkyOS · Tianshu”.

02.
SkyOS Full Release and Banyan 3

Regarding the R&D investment in SkyOS and chips, Li Bin told media including Zhijia.com after the press conference:

"This is not something most companies are willing to do. Most companies still think in terms of components, but smart car companies should think in terms of layered decoupling."

At present, NIO’s battery pack has been decoupled from its vehicle models. If the battery high-voltage platform is stabilized, the fourth and fifth generations of models can use the same battery pack.

"At first, everyone may think that they have invested too much money, but it actually gets easier as time goes on. The foundation is laid according to the number of floors, which is very different. We have to lay the foundation of a 100-story building, so the speed of getting out of the ground is slower, but after getting out of the ground, the speed will be faster."Li Bin said.

So is the foundation of this 100-story building almost completed?

Obviously, chips and operating systems are the core foundation.

Li Bin has always promoted NIO's SkyOS as the first smart electric vehicle full-vehicle operating system developed and released by a car company in the automotive industry.

Zhijia.com has analyzed the controversy in the industry regarding the vehicle-wide operating system in many articles (see " " for details).

However, as China's automobile industry takes the lead in intelligence and electrification, it does face the opportunity to build a China-led automotive operating system.

However, what kind of operating system can be called a vehicle-wide operating system?

Cao Bin, General Manager of Neusoft Rich, described such a system:

"As intelligence drives the new E/E architecture to upgrade to domain controllers, domain controllers that adopt SOA architecture ideas are gradually put into mass production. The challenges faced by central domain controllers and autonomous driving domain controllers will inevitably give rise to a new generalized automotive operating system.An operating system with a central computing unit and autonomous driving domain control as its core and with certain distributed characteristics: "one brain controls the whole body."This is something that the industry should pay attention to.”

The SkyOS released by NIO this time was built by NIO to connect all its systems. Not only the body control domain, but also the intelligent driving domain and the smart cockpit can all achieve the interconnection of the underlying logic.

From the architecture diagram, we can see that at the bottom are chips, devices, and other hardware in the car (intelligent driving perception, doors and windows, etc.), and above these, SkyOS is running.

SkyOS solves the problem of software, hardware and service integration by building a "1+4+N technology cluster", while also solving problems such as security and real-time performance in different "domains" of the car.

The bottom layer is SkyOS-H, which serves as a virtualization platform. It virtualizes the hardware resources of the entire vehicle, dispatches the heterogeneous large computing resources of the entire vehicle, and provides a secure and isolated operating environment for each system.

On the upper layer, there are 4 cores, which have different responsibilities:

SkyOS-L：The first real-time operating system to achieve domestic substitution of AutoSAR and large-scale commercial use.

In comparison with AUTOSAR, SkyOS-L can stably execute high-priority, medium-priority, and low-priority instructions. AUTOSAR is not as good as SkyOS-L in accurately delivering medium-priority and low-priority instructions.

SkyOS-M：It is a microkernel architecture that runs in the central brain and mainly controls the body, chassis, suspension, etc. The kernel is more stable than traditional Linux, and the service isolation is better. At the same time, on the basis of security isolation, there are four-layer monitoring and three-layer recovery security mechanisms.

SkyOS-C：Based on a deeply customized operating system of Android, it carries the functions of the smart cockpit, has a self-developed TOX protocol stack, more stable data transmission, and also has AI smart experiences including NOMI.

SkyOS-R：Improve the system's load capacity.

By connecting the underlying data, intelligence is no longer limited to the cockpit and driving assistance, but truly realizes the full-domain intelligence of the entire vehicle.

Judging from the above introduction, SkyOS meets the description of the vehicle operating system by Cao Bin, general manager of Neusoft RichAuto.

However, NIO's SkyOS has similarities with Huawei's HUAWEI IDVP smart car digital platform. The basic software platform of HUAWEI IDVP covers Huawei's three self-developed operating systems - HMOS (Hongmeng cockpit operating system), VOS (intelligent vehicle control operating system), and AOS (intelligent driving operating system).

The core of creating and leveraging the advantages of a vehicle-wide operating system is to achieve the decoupling of vehicle software and hardware. Such a system is, of course, an unavoidable technical direction for smart cars.

NIO explained the process of developing this system before the launch:

Starting with the NT1 platform, we first developed the middleware of the vehicle's global operating system, that is, the protocols, interfaces, and interactions between the operating system and application software. On the NT2 platform, we mass-produced the SkyOS-L lightweight real-time operating system, which replaced AUTOSAR and outperformed AUTOSAR in all aspects. On the NT3 platform, we will realize the full-featured mass production of "SkyOS · Tianshu".

This is a remarkable achievement.

Li Bin revealed that SkyOS will first be installed on NIO and Ledao's own models.

In a live broadcast after the release, Li Bin also answered netizens' questions about whether the system would be open to other car companies. He said: "It depends on external demand. If there is, we will also open it to the outside world."

The Banyan 3 intelligent system was released simultaneously with the release of this complete vehicle system.

The smart cockpit system Banyan will usher in a major version update at the end of August. The new Banyan 3 will become NIO’s first smart cockpit system based on AI underlying capabilities.

The background image of Banyan 3 below is generated based on difussion technology and NOMI GPT.

The first major upgrade of Banyan 3 is the AI ​​smart chassis, which connects the super-sensing system, body perception, chassis perception and actuators. Li Tianshu, vice president of product experience at NIO, said on the spot that the real soul of today's smart chassis is AI capability.

In Banyan 3, 4D comfort navigation has been upgraded to 4D global comfort navigation, which records various road events including potholes and speed bumps.

For example, the suspension will be actively adjusted during the next pass, and 4D prediction will not be affected by ambient speed and visible distance, which can greatly improve safety and comfort.

Moreover, as long as the car models have CDC dynamic damping control, it can be used, including not only ES8, ET7, EC7, EC7, but also ES6 and EC6 owners can experience 4D comfortable navigation. However, the two models of ET5 do not have this function for the time being.

However, for embodied intelligence, NIO has concretized its AI capabilities into NOMI.

NIO shared a set of data: in the past three months, users have interacted with NOMI GPT more than 15 million times.

NOMI under the new system will bring more new experiences in three areas: anthropomorphic interaction, capable assistant and emotional partner.

The author’s deepest feeling about NOMI is that NOMI has grown into an intelligent AI assistant with active and natural interaction.

Weilai wants to create an AI with emotions. There was a scene that touched me at that time - when NOMI was broadcasting information, the driver happened to receive a call. The driver raised his finger and made a "hush" gesture, and NOMI tactfully muted the call.。

NIO also introduced NOMI Agents multi-agents for NOMI, which reconstructed NOMI's cognitive and task capabilities based on a large language model. In summary, there are six major capabilities:Parking Assistant, Guard Summary, Service Butler, Discovery, Travel Memories and NOMI DJ。

Musk once said that making AI is actually like educating a child. In addition to teaching him knowledge, the most important thing is to instill values ​​in him, let him understand human nature, and make him a good AI with both IQ and EQ.

NOMI is already one of the most successful designs for smart cars. With the addition of generative AI, it is enhancing the unique emotional capabilities of the NIO brand.

03.
Weilai's world model is here

In terms of intelligent driving, NIO has launched two technologies: one is the world model and the other isEnd-to-end architecture.

The current person in charge of NIO Intelligent Driving is Ren Shaoqing, vice president of intelligent driving, a science and engineering man who likes white T-shirts and jeans.

Ren Shaoqing said: "Last year at NOI,My friend told me that the story is well told and the technology is good, but the speaker doesn’t talk much, so this year I will try to tell something that everyone can understand.。 ”

On the car side,Before the press conference, NIO announced that it would upgrade AEB to an end-to-end architecture. Let's take a look at the picture below, which shows the effect of NIO's end-to-end AEB in dealing with the performance of two-wheeled vehicles.

This time, NIO launched a world model in the cloud, which draws on the human brain's ability to understand space and time through imagination and deduction. By inputting the vehicle status and video, it can deduce multiple scenes in the "parallel world".。

At the same time, this is also the first time that SenseTime Jueying, Horizon Robotics, Ideal Auto and Nullmax have joined the ranks of the top 100 companies in the industry.NIO has also introduced brain-like thinking into its end-to-end intelligent driving architecture.。

It is reported that NIO used more than 10 million Clips of data for learning to make the reconstruction more realistic and to better understand the complexity of the real world.

Pay attention, the following video, which looks like the real world, is actually imagined by the NIO world model.

For example, when on a traffic jam, by providing NWM with video, vehicle status and other information, the vehicle can "imagine" thousands of different roads ahead and overtake the slow vehicle ahead at a certain point in time.

The so-called world model is similar to what Nvidia calls the Twin Earth, which uses digital simulation of the real world to establish a virtual testing ground for intelligent driving.

The NIO world model, end-to-end active safety model, plus all processing mechanisms and safety mechanisms, form NADArch2.0.

When introducing NOP+, Ren Shaoqing mentioned that NIO is supported by the new generation of static perception network NADLane 2.0, cloud large model NADWWM, layered value network NADHVN, and NAD architecture NADArch.Ability to continuously verify and operate new pilot routes through "crowdfunding"。

This concept can refer toCrowdfunding map means that after a car owner drives through an unfamiliar road, these data will help subsequent car owners to achieve intelligent driving when passing this road.

This is similar to the "smart driving for commuting" and "memory driving" proposed by many car companies.

At the end of September, after the NIO APP opened the wish list for pilot routes, every user can propose the route they need, and then gradually iterate it during everyone's daily driving to "develop" a mature route. At the same time, they can also share the routes that other users have already adjusted.

Although this "opening journey" route is not the first of its kind by NIO, it is also one of the technical routes that many car companies are currently trying.

However, the core of end-to-end is to enable the vehicle to understand the scene. Although "Kaicheng" can help the vehicle deepen its understanding of the road, it cannot directly improve the underlying capabilities of intelligent driving. I believe that the world model can help NIO improve its understanding ability through 24-hour uninterrupted training.

At present, NIO's full-area pilot mileage has been opened for 3.899 million kilometers, and all users have driven 1.112 billion kilometers. NIO's total computing power of the car cloud is 287,100 POPS, the actual road verification mileage is 420 million kilometers, and the sample analysis and mining volume reaches 220 million per month.

This is the basis for NIO's intelligent driving to continue to iterate, and big data is an irreplaceable resource for intelligent driving.

04.
Li Bin: The new NIO Phone cannot bear not to be the main phone

In addition to chips and battery replacement, the NIO Phone was once the most criticized product.

In less than a year, NIO Phone has entered a new generation. After Ren Shaoqing returned, Li Bin took out a new NIO Phone from his pocket and said, "I can't bear not to make it my main phone."

At last year's NIO IN, Li Bin earnestly answered the question of why NIO wanted to make mobile phones.After today's Technology Day, the question that NIO needs to answer is "How can NIO make a good mobile phone?"。

At the media communication meeting after the press conference, Li Bin said: "Objectively speaking, there has not been much innovation in the mobile phone industry in the past few years."

The new NIO Phone uses a 6.82-inch four-curved screen with equal depth, named "NIO Sky Dome Screen", with a bending depth of 724 microns, which is a great challenge for the manufacturing process. The greater bending depth brings a better grip.

However, at present, NIO Phone can only attract some NIO owners to buy it. For those users who have never purchased a NIO car, NIO Phone cannot bring a fresher experience.

For NIO owners, NIO Phone is not perfect either. After all, most NIO owners (NIO owners and Apple users have a high overlap) are still unwilling to give up the iOS ecosystem and switch to the Android camp.

Li Bin also clearly knows this, he said:

"NIO's logic for making mobile phones is not to compete directly with Huawei, Xiaomi and other mobile phone manufacturers, but more like NIO Life, it aims to meet the needs of car owners around NIO's automotive business."

The final price is basically in line with the rumors (6,499-7,499 yuan), but the rights of NIO car owners are that they can use 10,000 points to deduct payment. At the same time, if the mobile phone is regarded as an optional configuration of NIO cars, then this price is acceptable.

Li Bin said in July last year: "We are definitely not comparable to the companies that sell 100 million mobile phones now. It is not difficult to make mobile phones now, but it is still challenging to make good mobile phones. What NIO needs to do is to make a good mobile phone for NIO users, and develop one mobile phone every year, just like Apple."

If we start from the user experience of NIO car owners, and if we have to find a bug in NIO phones, it is that NIO phones currently have a very obvious defect that does not conform to the "rechargeable, replaceable and upgradeable" requirement - they do not support battery replacement.

So is smartphone a product line that must be adhered to?

As generative AI is becoming the foundation of a new generation of digital products, smartphones and smart cars can combine to create powerful scenario innovations under an operating system with unified underlying data.

At present, the biggest difficulty for Nio Phone is the inability to achieve large-scale cost reduction, but Qin Lihong once told Zhijia.com that his team is not large and NIO can afford it.

05.
Weilai is the first to embrace "AI-defined cars"

How do you evaluate the technologies and products released by NIO on this Technology Day?

Simply put, NIO is one step ahead in the industry in moving from "software-defined cars" to "AI-defined cars."

Whether it's chips, vehicle operating systems, or a new generation of electronic and electrical architecture, all have laid a solid foundation for generative AI to dominate a car's performance, control, and the digital interaction between people and smart cars, a foundation that can support a hundred stories.

So what is the difference between software-defined cars and AI-defined cars?

Before this year's Beijing Auto Show, Wu Xinzhou, head of autonomous driving at NVIDIA, compared and analyzed the two from the three stages of development, testing, and deployment of a car when communicating with media such as Zhijia.com.

Software-defined cars require a large number of engineers, and the coding and engineering design of all components is a huge workload. The recognition and integration of edge scenarios are slow, and a large number of road tests are required. A large number of corner cases need to be identified and tested one by one, and efficient infrastructure is needed to support the closed loop of test data. Once a new version is formed after a large number of tests, it needs to be updated through OTA. In the current Chinese market, each OTA must also complete the compliance requirements of relevant national laws, and be filed and reviewed, which also requires a large number of people to participate.

AI-defined cars can greatly simplify the process.

First of all, most models can be trained in the cloud (world model), which is completely data-driven to solve edge cases and use Internet-scale data in FM to enhance generalization capabilities.

That is to say, the self-iteration of the model is completed through automatic data-driven, and the verification and testing are completed through simulation in the cloud. This can not only reduce the large-scale equipment deployment and testing on the vehicle side, but also greatly reduce the cost investment. Each OTA upgrade is actually an update of the model, no longer a huge code update.

NIO’s technology demonstration this time basically followed this idea.

If NIO really achieves AI-defined cars in the development of new products, then it can be said that after experiencing huge and continuous investment of real money in the early stage, NIO will reap the dividends of AI-defined cars:Technology iteration is faster and costs are falling massively.

This will be a huge technological dividend.

06.
Controversy: Robotaxi and weekly sales rankings

After the press conference ended that day, some media asked Li Bin and Qin Lihong two major questions - how they view Robotaxi and the weekly sales chart.

Li Bin stated firmly that NIO would never make Robotaxi.

"We believe that 'freeing up energy and reducing accidents' is the real social value of smart driving. The value of smart driving is not to eliminate the jobs of today's hard-working private car drivers and taxi drivers. The real value is to reduce accidents, free up some of the energy spent on driving, make driving less tiring, and be safer."

This conclusion is of course controversial in the view of Zhijia.com.

This is obviously inspired by the conflict between Baidu's Carrot Run and local online car-hailing drivers in Wuhan.

Li Bin explained:The value of intelligent driving is not to eliminate the jobs of private car drivers and taxi drivers. Robotaxi has no real social value. "

He also believes that Robotaxi cannot solve the problem of traffic volume.

"The Robotaxi market has very limited capacity. This is a restricted business model. If so many Robotaxes fill up the city, the government will not agree. NIO will never do this. NIO's goal is to reduce accidents by 10 times."

In the view of Zhijia.com, intelligent driving is reaching the critical point of autonomous driving. In urban road conditions, fully autonomous driving has already seen the light of day. The large-scale implementation of this technology is bound to give birth to new business models.Whether Robotaxi can form a commercial closed loop by replacing taxis and online car-hailing services remains to be seen, but it will definitely provide a large number of travel services to the travel market again.supply.

However, it is too early to draw a conclusion about Robotaxi. Let time tell.

Another controversy is Li Bin and Qin Lihong's dissatisfaction with some companies releasing weekly sales charts.

Qin Lihong said that at the 2024 China Automobile Forum held not long ago, he advised industry leaders to avoid low-quality internal competition and said: Is it possible not to publish weekly rankings?

"We at NIO have never authorized any other organization to publish a weekly chart with our name on it. This number has never been approved by us. Only four or five people in our company know the weekly sales data. First, the data is not serious. Second, our brand is not authorized. If it is a national statistical agency, then there is nothing we can do. That is your job. If it is another company, why would they use our brand for promotion? Can they exclude us? The relevant departments can still take care of this matter because it is very cruel."

Li Bin said:If we become the first, we will never publish a weekly list.。”

These words were obviously directed to Li Xiang and Ideal Auto.

Is it possible not to release the weekly rankings?

Why not make the work of the spectators and the media easier?

[Follow smart cars and follow Zhijia.com video account]

For cooperation or news clues, please contact: [email protected]
To share information and contact the author, please add WeChat: xinwell0709