Future Car Buying Guide Guangzhou Auto Show Intelligent Driving Survey

[Autohome?Technology]?Although times are progressing, it seems that our concept of buying cars has not changed much. In the past, we may have paid more attention to fuel consumption, space, and the sound of closing the door. Although today's consumers are more mature and rational, there are still some "suggestions" in sales words, such as autonomous driving, ACC, and lanes. Keep… Most users don’t understand the meaning behind these concepts. Of course, we also have to admit that with the upgrade of technology, many cars have begun to introduce these technological functions that you do not often use. The theme of this Guangzhou Auto Show is "Intelligent New Technology", which shows the importance of technology in the current market. This article will start from the perspective of smart driving and talk about some of our observations of this auto show.

◆The implementation of L2 driving assistance is accelerating. New car companies continue to lead the way, and traditional car companies are closely following. With the rise of autonomous driving, the market for driving assistance, which is the cornerstone of autonomous driving, is gradually heating up. , and the entry of Tesla and the success of the Autopilot driving assistance system have made more car companies begin to pay attention to the added value brought by the driving assistance technology. It can not only effectively improve driving safety, but also shape the technological side of car companies in some aspects. For the current consumer market, technological capabilities are an important label to measure the strength of a company, which can also virtually enhance the brand content of a car company.

1) Don’t buy Tesla as a second-hand car? From a technical perspective, there is indeed a certain threshold for driving assistance functions from hardware to software. Let’s take Tesla as an example. The Tesla Autopilot driving assistance system first used a monocular camera as the main sensor of the system. It was equipped with the Mobileye EyeQ3 platform and was also equipped with an NVIDIA Tegra 3. In addition, a single camera was added to the front of the car. A millimeter-wave radar with a detection range of 160m. Later, there was an accident involving a collision with a large truck. Since then, it parted ways with Mobileye, a top visual solution provider, and switched to Nvidia. The upgraded second-generation Tesla began to use the front-facing trinocular camera we see now, and the autonomous driving platform was upgraded to NVIDIA's DRIVE?PX2. The upgraded multi-camera and higher computing power autonomous driving platform are the second The biggest change in the modern autonomous driving platform. Later, it was upgraded again on NVIDIA DRIVE? PX2, which was considered a half-generation upgrade. At that time, the number of processing chips was increased (from one processing chip to two) to improve the performance of the camera because it needs to perceive color and brightness. computing power. The most groundbreaking thing is the release of the third generation product, HW3. This was in 2019. Tesla officially launched its self-developed FSD chip. The new platform was equipped with two self-developed FSD chips. FPS made rapid progress - 20 times higher than the previous generation (2300FPS), and the computing power increased by 11 times reached 144TOPS. However, HW3 cost is reduced to 80% of HW2.5. Do you know what this means? If you buy a second-hand Tesla from a few years ago, even if it is equipped with the Autopilot system, it is still far from the current products. Similar to our computers, the hardware has already been determined before buying. Later, as the game software is upgraded, it will become more and more stuck. Only better hardware can meet a higher degree of "picture smoothness", and those so-called algorithm upgrades It may improve the experience to a certain extent, but it will always hit the ceiling of these hardware. Tesla Model 3 currently uses this autonomous driving platform. In our previous driving assistance tests and active emergency braking tests, the new Tesla Model 3 achieved excellent results in all aspects. The new autonomous driving platform Played a very important role. It took Tesla about 6 years to complete the upgrade and transformation of the driving assistance system, which does not include previous research and development, but the final results were outstanding. Tesla's current driving assistance systems can be considered among the best in the entire market. From relying on external suppliers to self-developed chips and algorithms, Tesla is forming a closed-loop ecosystem similar to Apple. Its autonomous driving development has little impact on upstream suppliers, but it poses a great threat to other car companies.

2) China’s new power car manufacturers continue to increase their focus on “autonomous driving” (driving assistance). Tesla, as a foreign new power car manufacturing company, has a great influence on China. Domestic new power car manufacturing companies also use "Autonomous driving" (currently assisted driving) is the focus of publicity. At this auto show, Xpeng, which plays at home, released its next-generation autonomous driving architecture. Xpeng Motors stated that it will upgrade the driving assistance hardware and software algorithms on mass-produced models in 2021.

『Xpeng Motors Regional Marketing Director Yu Ding』 However, the Audi A8 equipped with lidar at that time was obviously more showy than practical. The detection range of this low-cost lidar was only about 80-100m. Since this car There is also the existence of long-distance millimeter wave radar, so lidar is more likely to play a role in accurate judgment of objects, similar to the multiple redundancy of the system. At this stage, Xpeng's XPILOT autonomous driving system has been upgraded to version 3.0. This system is equipped with 13 cameras, 12 ultrasonic radars, 5 millimeter wave radars, dual-frequency high-precision GPS, and supports high-precision maps and V2X. technology.

The computing platform comes from NVIDIA DRIVE? Far from it. However, this system can now fully meet the L2 level driving assistance functions. At the same time, after completing the OTA upgrade early next year, it can complete high-speed autonomous navigation and driving functions. In addition to Xpeng, NIO and Li Auto, two other domestic new car manufacturers, are also making frequent moves in smart driving. NIO's driving assistance functions on mass-produced cars are similar to those of Xpeng, but the driving assistance platform is based on Mobileye EyeQ4, the data processing unit system uses the EyeQ4 chip, and the camera uses a trinocular camera like Xpeng , at the same time, 5 millimeter wave radars, 4 surround-view cameras and 12 ultrasonic radars are also arranged around the car body. This set of NIO hardware can complete hardware support for L3 level autonomous driving, and for the current L2 level driving assistance functions That's more than enough. Note: TOPS is a performance indicator that evaluates the computing power of a processor. 1TOPS means that the processor can perform one trillion operations per second. Li Auto currently has a major gap with Xpeng and NIO in terms of hardware. First of all, the sensor type is significantly smaller than that of Xpeng and NIO. This means that although it can complete the same functions, it requires manual input and cannot complete full automation. Operations, such as the automatic lane changing function, the Ideal ONE requires the driver to judge the rear environment. In addition, due to the lack of angle radar, the ability to judge vehicles coming from oblique obstacles should not be weakened. Therefore, the current ideal mainly relies on algorithm iteration to upgrade its intelligent driving functions.

3) Traditional car brands followed closely. At this Guangzhou Auto Show, BMW finally opened the automatic assisted lane change function to the country. Coincidentally, Mercedes-Benz also tested the E-class before the auto show. At the meeting, the automatic lane change assist function of the new E-Class was demonstrated to everyone. It can be said that the domestic smart driving functions of foreign high-end brands have covered most of the L2-level assisted driving functions, and some high-end models can already support more advanced automatic driving functions from the hardware. The BMW iX3 is equipped with BMW’s autonomous driving assistance system Pro. This system is the latest driving assistance system currently promoted by BMW in China. This system has been fully upgraded in terms of hardware, including 12 ultrasonic sensors, mainly used for parking assistance systems; 1 front-view trinocular camera, It is mainly used to capture images for image recognition; in addition, there is millimeter wave radar, which is mainly used to detect vehicles and obstacles in the area in front of the vehicle.

"BMW's automatic lane change assist needs to be activated at 70km/h or above" Regarding this brand-new system, the BMW technical team started redevelopment and verification based on Chinese road conditions two years ago. BMW's technicians admitted that this system has been specially optimized for traffic signs, China's unique traffic signs, high-speed fishbone markings and obstacles on the road. China's road conditions have also improved BMW's own technical level. Mercedes-Benz has also demonstrated similar functions on the new E-Class. At the same time, the Mercedes-Benz technical team also said that it has made new optimization and localization upgrades for China's road conditions. Aion? LX's ADiGO? 3.0 system is equipped with 12 ultrasonic radars, 4 short-range lateral millimeter wave radars, 1 forward mid-range millimeter wave radar, and 5 cameras (one of which is mainly used for forward direction) on the body. Object recognition, the remaining 4 are mainly used to achieve 360° panoramic image function). These sensors can be used to detect objects around the vehicle. The driving assistance functions displayed by other Chinese brands such as Geely at this auto show are quite satisfactory. Most models have a standard monocular camera + millimeter wave radar configuration, which can meet common L2 driving assistance functions in terms of hardware. . However, according to the performance of UNI-T, UNI-K, as a higher level, will definitely not lower its requirements in terms of driving assistance functions. Then this car will also cover most of the L2 level driving assistance functions. At the same time, high precision may also be introduced in the future. Map, then it should also be able to complete highway lane change assistance. Of course, this is all speculation at the moment. Summary: Next, let’s summarize some of the situations of car companies from various countries at this auto show: Chinese traditional brands: In terms of traditional car companies, all brands, regardless of the price of the model, are promoting L2 driving assistance functions. The type and number of sensors are relatively important in positioning. There is a significant increase in high-end models, while the number of sensors in mid- to low-end models is smaller. In terms of functions, the main ones are common driving assistance functions (adaptive cruise and lane control functions). It can be seen that the Chinese brand first solves the problem of driving assistance for each model. From a technical perspective, except for some high-end models, most models have a certain gap with new car manufacturers in terms of hardware composition and intelligent driving platforms. Regarding new domestic car manufacturers, "driving assistance functions" are still one of the core competitiveness of these brands, so car companies are more willing to make in-depth investments. In terms of hardware, Xpeng and Weilai have certain redundancies. They belong to the first group and their ideals are slightly behind. There is still a big gap between the computing power platform and Tesla. However, whether the driving assistance function requires L4 level computing power at this stage is still a question mark.

Generally speaking, the above three Chinese new forces are at a relatively high level among mass-produced vehicles in terms of intelligence from computing power to hardware level, but there is still some gap between them and Tesla. In terms of foreign brands, Tesla leads the way, both in terms of technical route and functional coverage. Although traditional high-end brands are relatively conservative in terms of intelligent driving, it can be seen that they all have strong technical reserves. At the same time, foreign car companies are constantly absorbing Chinese road condition information to feed back their own driving assistance functions. In addition, BMW, Mercedes-Benz, etc. have completed the hardware configuration to support more advanced autonomous driving, and may launch more comprehensive driving assistance functions in the future.

◆The driving assistance technology route of mass-produced models is changing again, and cost may be the biggest driving force behind it. At this Guangzhou Auto Show, in addition to driving assistance functions becoming more popular, we found that the driving assistance technology route has also become more popular. Something changed again.

1) Pure monocular cameras may become the first choice for driving assistance in low-cost models. We know that the current mainstream driving assistance hardware technology solutions are vision solutions and vision + radar solutions. Pure vision only uses visual sensors, and hybrid uses both vision and radar. The previous consensus in the industry was that purely visual routes will gradually decrease, and millimeter-wave radar + monocular camera routes will become mainstream (for relevant knowledge on technical routes, please refer to: Who can see clearly at night? Three-vehicle night automatic brake test). But judging from the observations at this auto show, it is obvious that the purely visual route has not disappeared, but has become more and more popular with the breakthrough of monocular vision technology. Although it may be affected by bad weather conditions, the functional implementation is enough to achieve the goals of car companies, and the pure monocular camera route is cheaper. Suppliers including Bosch, Continental and others have completed related products. Among the suppliers The link is not the problem. It is foreseeable that in the future, it may be possible for mid- to low-end models to be equipped with pure monocular cameras to realize driving assistance functions.

2) High-end models still develop multi-sensor fusion, and lidar begins to advance. If low-cost models will use monocular cameras to popularize driving assistance functions, then high-end models still consider multi-sensor fusion. In terms of vision, it has begun to evolve from a single camera to a multi-camera. For example, Tesla, BMW, etc. mentioned above have all evolved from a monocular camera to a trinocular camera to improve visual perception. Manufacturers like Subaru, which started with binocular cameras, have begun to integrate millimeter wave radar into the system to improve the stability of the system in bad weather. Lidar has been mostly used in L4-level self-driving test vehicles in the past, but at this year’s auto show, we also began to see top luxury cars beginning to be equipped with lidar, such as the Maybach S-Class. Xpeng Motors also announced at the auto show that it will soon be equipped with lidar. In fact, some time ago, Volvo and Great Wall both had news of lidar being installed in cars. Lidar is gradually being accepted by car companies due to its advantages of high precision and high detection range. , and with the rise of Chinese lidar suppliers, the cost of lidar is also constantly decreasing. It can be said that autonomous driving is accelerating. ? Editor’s comment: Looking back at the Guangzhou Auto Show, we found that car companies are paying more and more attention to the concept of driving assistance. Although the exhibitors are all high-end or top-end models, from a price point of view, top-end models above 100,000 level The models have comprehensive coverage in terms of driving assistance, and the advancement of L2 driving assistance functions is faster than we imagined. Purely visual monocular cameras are likely to occupy the low-end driving assistance market in the future due to technological upgrades; mid- to high-end segments will continue to develop radar + camera fusion technology, and the intelligent driving platform will still be dominated by foreign suppliers (such as Mobileye) In addition, lidar is already eager to try, but due to cost constraints, I think the time for mass production and installation will not come soon.

(Photo: Yang Peng, Shu Ning? Text/Autohome? Leng Xiaoyang)