Advanced intelligent driving is becoming insular, FeiFan R7 is uncompromising.

Author: Zheng Senhong

On September 27th, FF Automobiles launched its flagship model, the FF R7.

The post-subsidy price and user purchase rights of the FF R7 are as follows:

Before its launch, FF R7 had already made a lot of impressions on the outside world, such as the domestic first 4D imaging radar, the eye-catching 43-inch three-screen, the Huawei Vision Enhancement AR-HUD that can be projected and the RISING PILOT full-integration advanced intelligent driving system.

The FF R7 is positioned as a mid-to-large SUV, which is almost a sub-market that domestic luxury brands must fight for.

Especially this year, new forces such as NIO, Idean, and XPeng have successively gathered in the mid-to-large SUV market, further raising the product threshold of this sub-market in terms of quality, technology, demand and function.

This also means that, for newcomers who want to establish themselves in this sub-market, basic skills such as quality, technology, demand and function are a must, and unique and novel product highlights are bonus points.

If I had to describe the FF R7 in one sentence, I would say that this is a car that combines multiple extremely intelligent achievements in one body.

An epitome of intelligent high-end

From the design perspective, FF R7 has achieved bold and precise product features.

For example, the pressed front and rear tail not only breaks the traditional homogenized design concept of car models, but also improves its own aerodynamic performance. The wind resistance coefficient (Cd) is 0.238.

Compared to the 0.272Cd of the Idean L9, the 0.282Cd of the Audi Q5 e-tron, and the 0.25Cd of the BMW iX, the FF R7 undoubtedly has a noticeable advantage.

In other words, under similar body sizes, the FF R7 can save 10% of energy consumption compared to the Audi Q5 e-tron and BMW iX.

According to the information released at this press conference, the FF R7 is equipped with two types of battery packs of different capacities (CLTC comprehensive working condition):

  • 77 kWh battery pack: The single-motor version has a range of 551 kilometers.

  • 90 kWh battery pack: The single-motor version has a range of 642 kilometers, and the dual-motor version has a range of 606 kilometers.

In the current pure electric SUV market, models with a range of over 600 kilometers are almost the top players in the sub-field.Obviously, this requirement has been regarded as a threshold for high-end pure electric SUVs.

Secondly, there is a battle for the intelligence of the interior cockpit.

The FEIFAN R7 adopts a three-screen design, and the performance of the 15.05-inch, 2.5K high-resolution central control screen is not inferior to that of high-end digital products.

In addition, the FEIFAN R7’s co-pilot screen and instrument screen also use the same MiniLED screen as the 2021 iPad Pro and MacBook, providing ample space for the interaction and visual design of the vehicle system.

For example, the light sensing of the FEIFAN R7’s car system interface can simulate the sunrise and sunset of Kepler planet with the passage of time, and it can also simulate translucent color materials. When playing music and touching the screen, the local light sensing of the screen will also change accordingly.

“7nm process manufacturing, eight cores, 8TOPS computing power, integrated CPU, GPU, NPU, AI engine, including ISP for various cameras, support for multi-display DPU, integrated audio processing and other functions”, this is the hardware ability of Qualcomm 8155.

For example, the GEEK001 only used one Qualcomm 8155 chip to solve the problem of car system carton and delayed lighting exposed by 820A, which is the “magic power” of Qualcomm 8155 chip.

Also thanks to the assistance of Qualcomm 8155, the FEIFAN R7’s car system is smooth and silky, with simple interface logic. Even for first-time users, it is easy to operate.

For example, in the experience process, users can complete the “identification + understanding + feedback” action in about 1 second through voice assistant.

Compared to its competitors of the same level:Flyyin R7 is much larger than the Tesla Model Y with a similar crossover coupe design, and its luxurious interior far surpasses that of the Model Y. While both models have swappable batteries, Flyyin R7 leads in intelligence over the NIO ES6.

It’s fair to say that Flyyin R7 represents one of SAIC’s intelligent technology flagship products, and it wouldn’t be an overstatement to say it’s a full-length vehicle.

According to data from China Passenger Car Association, the share of SUVs priced above 250,000 yuan has risen from 7.25% in 2020 to 9.4% in the first half of this year. In the pure electric SUV market, the latest market penetration has risen to 22 percent.

This to some extent reflects the fierce competition in the large electric SUV market, and it also indicates that Flyyin R7 is currently in an expanding competition.

The upper limit of hardware capabilities determines the final software experience.

Looking simply at product strength, Flyyin R7 is a car with many “features”. It is so named because it makes full use of all kinds of perception hardware, with up to 33 perception modules on board, which is almost the ceiling of current smart electric vehicle hardware.

The official positioning of Flyyin R7 is “extremely smart high-end pure electric SUV”, what does that mean?

Sensors are the foundation and starting point for defining a product before the ultimate goal of a software-defined car.

Therefore, improving sensor quality and perfecting sensor fusion has become a key focus of intelligent driving players.

Tesla chooses a pure vision route, using deep learning algorithms to transform two-dimensional images captured by multiple high-definition cameras into bird’s-eye three-dimensional maps.

However, this approach still relies heavily on data training, and from the previous testing versions of FSD released by the US and Canada, cases of not recognizing roads and hitting barriers occur frequently.

Compared to Tesla’s unconventional approach, Chinese automakers tend to use multiple sensors, including cameras, millimeter-wave radar, and lidar.

Flyyin R7 is equipped with 12 high-definition cameras, including four 8-megapixel cameras and eight 3-megapixel cameras.

With the help of high-pixel cameras, the automatic driving system’s resolution for spatial and temporal dimensions can be improved, helping the vehicle see “clearer” day and night.Compared with the current mainstream 2 million-pixel camera, the 8 million-pixel camera adds perceived data for objects such as cones, distant pedestrians, etc. and more pixels can help the system perceive earlier and make more accurate judgments.

For example, an 8 million-pixel camera can ensure that when a vehicle is traveling at high speed, the detection distance can be increased to 250 meters; when driving on non-highways, it can also detect traffic signals 100 meters away earlier.

The 360° camera angle covering the entire vehicle guarantees that the Feeyo R7 can recognize and perceive dynamic traffic participants, static lane markings, ground markings, traffic lights, and speed limit signs.

Traditional mm-wave radar point cloud (left) and 4D mm-wave radar point cloud (right)

To solve the problem of low-quality information of traditional mm-wave radar, the Feeyo R7 is equipped with Continental’s 4D imaging radar, which is the first domestically-produced sensor for consumer-grade autonomous driving vehicles.

The 4D mm-wave radar includes four dimensions: distance, speed, direction angle, and elevation.

Traditional mm-wave radar, lacking vertical resolution, has to filter out recognition data for stationary objects. This was previously the “culprit” of many assisted driving failures, leading to collisions with high-speed barriers or construction vehicles.

Compared with conventional mm-wave radar, 4D imaging radar added a data dimension of vertical resolution, which adds the ability to measure height that conventional mm-wave radar does not possess.

This kind of mm-wave radar can solve Corner Cases in certain special scenarios, such as the 2015 Tesla Autopilot collision with a turning truck.

In terms of performance, 4D imaging radar has a resolution 16 times higher than traditional automotive radar and can perform full-scene object detection within a range of 350 meters.

At the same time, by capturing the spatial coordinates and speed information of targets around the vehicle, 4D mm-wave radar can also provide more realistic path planning and passable space detection.

Of course, the most critical factor is the product’s price-performance ratio.

The cost of 4D imaging radar is only about 10%-20% of that of LiDAR, which is why it is seen as consumer-grade sensing hardware in the industry.

It can be said that 4D imaging radar not only solves the shortcomings of traditional mm-wave radar in terms of performance but also satisfies the logic of lower-cost products.

Finally, there is “computing power.”

To some extent, intelligent vehicles are engaged in an “arms race” competition in terms of computing power. The Feeyo R7 is equipped with two NVIDIA Orin X chips, with a single chip computing power of up to 254 TOPS.After the last-generation chip Xavier was proven to be the best-performing SoC on the market, NVIDIA did not choose to upgrade gradually but directly boosted the computing power of this generation Orin chip to over 200 TOPS.

From the perspective of perception, the maximum sensing hardware capability is closely related to the technical route.

There is no accurate value for how much computing power advanced driver assistance systems (ADAS) require, and ADAS features need continuous iterative updates. To avoid the embarrassing situation of “being outdated once in mass production,” automakers generally adopt the strategy of “hardware embedding while software keeping up.”

Overall, many of the technological configurations of this car are roughly equivalent to NIO’s ET7, which is currently priced between 458,000 to 536,000 yuan.

At the Tech Brand Day event hosted by XPENG last year in March, XPENG expected that subsequent models, including R7, could achieve “hardware replaceable, software purchasable, and battery chargeable, replaceable and upgradable.”

Does this slogan sound familiar? NIO also mentioned similar statements regarding intelligence and energy systems.

This to some extent indicates that the early exploration of some automakers in intelligence is gradually becoming a mainstream view.

Alternatively, it can be said that the trend of the development of intelligence by industry leaders is “great minds think alike.”

Software Algorithm Becomes the Focus of Competition

To a certain extent, more sensor hardware is more advantageous for perception and accumulating more redundant environmental information, but how to integrate data from different sensors with different functions and output them into driving decisions is a problem.

This also means that XPENG needs to deliver a highly efficient and demanding sensor fusion algorithm for multiple sensors.

First, let us explain what “multiple sensor fusion” is.

In the current mainstream program, taking commercial vehicles equipped with lidar and cameras as an example:

Due to the different imaging principles, the former collects point cloud data, and the latter collects image data, and the time recording also has differences, which requires data from different sensors to be “fused.”

Nowadays, there are generally two approaches to fusion: front-end fusion and back-end fusion.

  • Front-end fusion mainly involves front-end fusion before the point cloud and image data are independently calibrated, and recognition is then done at the algorithm level.
  • Back-end fusion first lets the hardware level feedback different data and then does the fusion at the software algorithm level.

These two different technological directions have their pros and cons.

For instance, in back-end fusion, “single sensor” may have limited capability, which may lead to misjudgments and omissions in specific conditions. For example, cameras are not good at judging distance and location, while lidar is not good at judging color and texture. The system needs to verify their information against each other to achieve higher credibility.Before fusion, the system has extremely high requirements for the timeliness of data and hardware computing power, and requires higher security redundancy. If the algorithm produces a misjudgment, the final control layer will also make incorrect instructions.

For example, when a vehicle is driving on a vacant road, the center screen prompts that someone has passed by and displays a “ghost image”, which is a misjudgment caused by a standard front fusion algorithm.

The Fevan R7 adopts the industry’s first fully fusion algorithm, which is a triple fusion of post-fusion + front-fusion + mixed fusion algorithm, avoiding the risks of front fusion and post fusion in algorithms and perception.

In other words, this will make the perception judgment of intelligent driving more reliable.

Simply put, the workflow of full fusion is to first use the front fusion approach to obtain one solution, then use the back fusion approach to obtain another solution, and finally the system will Fusion, compare and check the two solutions, and come up with a perception and efficiency win-win result.

The most intuitive change this brings is that the success rate and stability of Fevan R7 in complex scenarios such as ramps, lane changes, and static obstacle recognition far exceed most competitors on the market.

Taking the three-way area scene of the multi-forked ramp as an example: at present, most intelligent cars cannot recognize in advance due to system failure or recognition failure, and directly miss the ramp and exit automatically, while the Fevan R7 can use comprehensive perception from a further distance in advance. Recognize the triangular area of ​​the ramp, and then have more sufficient time to change lanes and merge into the ramp in a safe direction.

Taking the accident scene where frequent collisions are caused by the inability to recognize static obstacles: Fevan R7 can detect and recognize static obstacles up to 350 meters away through the sensing ability of 4D imaging radar, and then quickly avoid them through the full fusion algorithm.

With the sensing hardware fully configured and the introduction of the first full fusion algorithm, Fevan R7 is also regarded as the “champion” in the intelligent driving industry.

In addition to its leading advantages in technology, Fevan R7 also has an advantage: it can be used upon delivery.

Under the trend of intelligence, the “arms race” of car companies is constantly upgrading, comparing hardware capabilities and delivery time, and the scent of gunpowder is getting stronger.

Behind these lively scenes, many car companies sell “futures” openly.

Even if the car company promises to upgrade through software OTA later and can indeed push it as scheduled, this approach also has certain drawbacks:

In the early stage, sales personnel cannot intuitively present the key selling points of intelligent driving to consumers.The delay in activating the “pre-installed hardware” for which users paid may be unfair to them. In some extreme cases, activation may not be possible even after a quarter or a year, such as in the case of Tesla FSD, the “fully autonomous driving capability” software package priced at 64,000 RMB in China that has yet to be launched.

In contrast, the Feeyo R7 has been able to use high-speed scenario point-to-point assisted driving function since its self-delivery. This is because RISING PILOT’s intelligent driving has undergone more than 170,000 kilometers of testing and more than 400,000 kilometers of overall testing data, providing enough data to support its delivery and use.

The launch of R7 is a breakthrough for Feeyo and a transformation that relies on self-renewal to emphasize its technological advantages. However, whether R7 is able to perform well in the market remains to be seen.

Feeyo believes that there is a significant differentiation in pure electric intelligent cars, and ultimately, the middle class will choose the price range of 300,000 RMB, a market that has accumulated 4 million vehicles, with strong demand.

According to the China Passenger Car Association, as of August this year, sales of new energy vehicles priced between 300,000-400,000 RMB reached nearly 490,000 vehicles, with Tesla Model 3, Model Y accounting for 42% and 23%, respectively, followed by Ideal ONE, accounting for 16%.

Clearly, Tesla and Ideal practically monopolize this segment of the market, but both have recently experienced rare declines in product competitiveness and sales, suggesting that the submarket of 300,000-400,000 RMB will enter a new interchanging period of products, and Feeyo R7 undoubtedly hopes to become an “opportunist” in this regard.

Backed by the strong resources of SAIC, Feeyo is creating a new path in the middle-to-high-end pure electric vehicle market by abandoning traditional car manufacturing concepts and exploring new models of data-driven, industry co-creation.

If the current automotive industry has entered the era of the Cambrian explosion of intelligent life, then the Feeyo R7 is a work of intelligent revolution that appears in this historical context. In simple terms, as hardware configuration and parameters gradually become homogenized, but functions and experiences remain in the “futures” concept, the delivery and immediate use of Feeyo R7 gives a sense of confidence in addressing the intelligent transformation of automobiles.This also means that FF doesn’t need to resort to new stories to lure in capital, nor does it need to make its users foot the bill for their “futures”.

On the path to advanced driver assistance systems (ADAS), the leading players in the industry unanimously expect continuous functional iteration, increasingly refined scenarios, and genuine user needs being met.

Admittedly, FF’s R7 was “late to the game”. But didn’t the “tortoise and hare race” also lead to a dramatic reversal of fortunes?

This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.