With the advent of Xiaomi cars, the last big man in the new energy vehicle industry has officially entered the market. This is both the end of an era and the beginning of an era. It's the end, because the characters who should appear are basically there. It is said to be the beginning, because the new knockout competition is about to kick off, and the focus of this knockout competition will be intelligent driving.
In the past two days, the "mutual war" between Zhiji Automobile and Xiaomi Automobile has been very lively, and with the intensification of competition, similar mutual confrontation or business war dramas in the market are believed to be staged one after another.
Zhiji and Xiaomi broke out in a "mutual hatred" war
Although each product has its own merits, if you look at the path selection of intelligent driving, domestic manufacturers can basically be divided into the same faction, whether you are Huashan faction, Hengshan faction or Songshan faction, they all belong to the Wuyue Sword faction.
If we open our horizons, we will find that there is still a kind of "route dispute" in intelligent driving, and the "route dispute" is mainly between China and the United States, which is behind the comprehensive competition of policy, infrastructure, industrial chain, technological innovation, cultural habits and even political beliefs.
The battle for routes has intensified
Will the future be beautiful or will it be the same?
In the field of intelligent driving, China and the United States are showing a situation of one super war and many strengths, Tesla represents the fundamentals of autonomous driving in the United States in a sense, and China is a new force competing for the lead.
Tesla's self-driving system Full-Self Driving (referred to as "FSD"), the Chinese name "full self-driving", adopts a pure vision solution. Pure visual perception scheme, completely abandon lidar, millimeter wave radar and other sensors, only use cameras for perception, Tesla cars use a total of eight cameras distributed around the car body, the overall realization of 360 degrees of global surround vision, the camera is like the "eye of the car", providing a visual perception system similar to a human driver for the car.
Musk is extremely repulsive to radar, which has both commercial considerations, when Tesla first went public, the cost of radar was relatively high, some as high as tens of thousands of dollars. Another reason is his belief in "first principles" – since humans drive with their eyes, and road structures and signs are designed for the human eye, cameras as the "eyes of the car" could theoretically achieve the same effect.
At the same time, the technical shortcomings of millimeter-wave radar are also an important concern for him. Because millimeter-wave radar cannot distinguish high-dimensional information, the confidence level of stationary object recognition will be reduced in use, and if there is no other reliable sensing, it may lead to false braking or missed braking, resulting in "ghost braking".
These factors, combined with Musk's confident and stubborn personality, have made Tesla firmly embark on the road of "going to the radar".
In April 2021, the V9.0 beta version of Tesla's self-driving system FSD announced the adoption of a vision-only solution.
A month later, in May '21, new deliveries of Model 3 and Model Y models in North America were no longer equipped with millimeter-wave radar. However, after the launch of the 8-camera version of Tesla, the number of complaints about "ghost brakes" surged. Tesla also recalled 11,000 vehicles in November 2021 due to some issues with FSD.
It can be said that the initial pure visual solution did not achieve the expected effect, but became one of Tesla's main slots. But Musk is Musk after all, just as the Starship rocket launch failed so many times, and it never affected his confidence. The temporary unsatisfactory did not affect his admiration and preference for pure visual solutions, but continued to increase innovation and improvement.
And the ability to continue to evolve is also the scary point of Tesla's FSD.
In July 2021, Tesla officially launched the BEV (Bird's Eye View) + Transformer technical solution, which further improves the camera's range, speed measurement, and ability to adapt to some extreme scenes.
In 2022, Tesla proposed the Occupancy Network, which introduced the concept of voxels - the basic unit of three-dimensional space, and further improved the recognition ability of object volume information through the occupancy network. At the same time, Tesla also continues to feed videos of various scenes, so that the AI algorithm can understand as much of a wider range of driving situations as possible.
On March 1, 2024, Tesla's FSD v12 was updated, and some industry insiders said that FSD v12 is the most important update in Tesla's history, bringing unprecedented self-driving capabilities to Tesla vehicles. "FSD Beta v12 upgrades the city street driving stack to an end-to-end neural network trained on millions of videos and replaces more than 300,000 lines of C++ code. In the update notes, Tesla said. FSD v12 is regarded as a game-changing technology, and the end-to-end algorithm in the entire system is almost entirely built with neural networks, which has the advantage of being very flexible, adaptable, and high-ceilinged, and can be driven in a highly anthropomorphic state, greatly improving the driving experience.
Here are some real-life footage from bloggers' reviews.
Wait for a pedestrian to cross the street:
Pull over the side of the road:
Avoid sudden braking of the vehicle in front:
Overall, the results are very good.
However, there are some shortcomings to the purely visual solution, and the main challenge comes from bad weather. Bloggers who tested the first batch of tests claimed that testing the FSD V12 Beta in heavy rain had several takeovers.
Musk responded to the feedback of netizens on the X platform, saying that the FSD V12 performed well in California, but it still needs to train more in areas with a lot of rain.
But overall, it can be said that FSD v12 has shown the ability to approximate human drivers, and Tesla's pure vision solution has also seen the light of day.
Different from Tesla, domestic manufacturers generally adopt the LiDAR solution, which relies on hardware equipment such as LiDAR + camera + millimeter breaker and ultrasonic radar, and relies on real-time mapping and AI algorithm fusion perception to achieve intelligent driving. For example, Huawei Wenjie series, Li Auto's 2024 L series Pro version, ZEEKR 007, AVATR 12, Xpeng G9, Wenjie M5, etc.
LiDAR has huge advantages in measuring distances and identifying object information, especially in extreme environments such as fog and rain. Andrews, a contributor to Wired magazine, tested three Chinese cars (NIO, Li and Xpeng) equipped with semi-autonomous driving capabilities, and found that all of them were superior to similar autonomous driving systems in the United States. The reason seems to boil down to the fact that American passenger cars are not yet equipped with one feature: lidar. He said many companies are steadily and far more successfully moving toward a similar goal, but through different routes.
The main story line of intelligent driving of Chinese manufacturers is relatively simple, and the direction of the plot is basically like this: the loading capacity of radar is increasing, the thread of radar is getting higher and higher, and the cost is getting lower and lower. According to the data from the automotive intelligent network insight report released by the Passenger Car Association, 55.3% of new energy passenger vehicles will be equipped with L2 and above assisted driving functions in 2023.
The most representative of domestic manufacturers is the Huawei Wenjie series. At a new car launch conference, Yu Chengdong said that his intelligent driving technology "surpassed Tesla's FSD, surpassed all peers at home and abroad, and achieved NO.1". Still far ahead, leaving no room for politeness. Judging from the effect of the actual car on the road, it has also achieved convincing results.
And with the continuous progress of the radar route, in 2024, China's new forces will begin to prepare for a large-scale "entry into the city". ZEEKR 007, AVATR 12, Xpeng G9, Huawei Wenjie, etc. have all opened urban and high-speed intelligent assistance functions. Autonomous driving in urban scenarios has great practical significance and symbolic significance, and "entering the city" is an important symbol of moving towards a higher level of autonomous driving.
The intelligent route choice between China and the United States seems to be an active choice between manufacturers, but in fact, the invisible hand of the market is playing a role all the time. The intelligent driving industry chain is very complex, covering sensors, chips, high-precision maps, algorithms, Internet of vehicles and other technical links. China has certain advantages in some aspects, such as radar, sensors, etc., but in some areas such as on-board chips and algorithms, the United States is superior. This is the most primitive driving force for the formation of the radar route and the pure visual route, and the bifurcation of the route between the two countries further affects the development of the industrial chain. At the same time, the United States has ushered in leapfrog development of autonomous driving chips and algorithm technology, while China is slightly lagging behind in these areas.
In 2022, the merger of the American star company Velodyne and Ouster attracted market attention, with the highest valuation of Ouster and Velodyne reaching $2.6 billion and $8.7 billion, and now the total market value of the new company cooperated by the two sides is only $400 million.
"At present, either the main product is in the mechanical radar area, which mainly relies on autonomous driving companies, and autonomous driving companies are not having a good time at present, or the layout of the next generation of solid-state lidar products, which cannot be mass-produced and commercialized. A practitioner of a domestic lidar company analyzed this.
On the other hand, in China, due to the crazy stacking of technology and volume products by domestic manufacturers, Hesai Technology, Suteng Technology, Tutongda and other companies have been able to bind car companies, and the number of sales has steadily increased with car companies, gradually moving towards large-scale profitability.
Of course, the United States also has its own advantageous projects. The U.S. has unrivaled advantages in generative AI, AI chips, and top AI researchers, which are the most comfortable technical environments for U.S. car companies in the field of intelligent driving, which provides warm soil for the practice of Tesla's vision-only solutions.
So far, Tesla has created several unique weapons: the FSD chip of the autonomous driving system, the Hardware series of automatic assisted driving hardware (which has been iterated to version 4.0 last year), the HydraNet detection network structure, and the end-to-end deep neural network.
Zhang Yongwei, vice chairman and secretary general of the China Electric Vehicle 100 Association, said in 2022, "The self-sufficiency rate of domestic automotive chips is less than 10%. "Basically, more than 90% of the chips in every domestic car come from imported or foreign-funded local companies. Not only that, the U.S. government has also issued several bans to prevent some on-board chips and AI chips from being sold to China.
However, we also see another phenomenon quietly occurring, and each other's technical routes are slowly merging on the basis of their own focuses. Tesla is secretly developing 4D millimeter-wave radar radar technology, after all, radar can always provide some safety redundancy, and at a time when the hardware cost is already extremely low, Musk has no reason to turn it away. On the Chinese side, at the Xiaomi SU7 press conference, Lei Jun officially announced the two sets of Xiaomi SU7 intelligent driving systems. These two sets of intelligent driving systems are: Pro pure vision version, Max vision + lidar version. The M9, launched by Huawei last year, is equipped with a 192-line LiDAR newly designed by Huawei, 3 millimeter-wave radars, 11 lateral and forward-facing visual perception HD camera groups, and 12 ultrasonic radars. Represented by Huawei and Xiaomi, the technical solutions of pure vision and radar solutions are moving towards convergence.
Single-bike intelligence isn't enough
The era of vehicle-road-cloud collaboration is coming
The track of intelligent driving does not only belong to private passenger cars, but also shows a different "speed" in the field of taxis.
On August 10, 2023, General Motors' Cruise and former Google's Waymo operated robotaxis in San Francisco, setting a precedent for self-driving commercial operations in the United States. However, recently, due to two self-driving vehicle accidents in succession, the California Vehicle Authority (DMV) began to restrict them, and caused concern about the safety of autonomous driving in the United States.
According to follow-up reports, GM recalled 950 vehicles, and the collision detection subsystem of the in-car cruise automatic driving system (ADS) software may have failed, which could have caused the car to try to pull over instead of remaining stationary after a collision, which is most likely the culprit of the accident. The incident led to the suspension of Cruise's driverless car business in the United States, the resignation of the CEO, and GM used the remaining funds to buy back shares to appease investors (A-share shareholders expressed great envy). Cruise will cut spending by about $1 billion in 2024, but will remain committed to the business.
In China, however, there is a different landscape. In 2022, the Yizhuang High-level Autonomous Driving Demonstration Zone successively issued China's first batch of unmanned RoboTaxi road licenses and Beijing intelligent connected passenger bus test licenses, as well as unmanned retail vehicles that have been in operation for many years. At present, there are two companies that provide fully unmanned self-driving taxi services in Yizhuang, Baidu Radish Express and Pony.ai, and it is easy to see and distinguish driverless taxis in the Yizhuang area.
Driverless taxis on the streets of Yizhuang
After two years of "road running", Beijing is planning to build an autonomous driving demonstration area on a larger scale, according to the plan, the autonomous driving demonstration area will complete the 3.0 phase in June this year, expanding to 600 square kilometers including Tongzhou and Shunyi. At the same time, five railway stations and two airports frequented by ordinary people will also be opened for autonomous shuttles.
Judging from the results of the operation of intelligent driving taxis in China and the United States, there is no doubt that China's side is going more smoothly. There is a different philosophical thinking behind this, with the Chinese focusing on the overall interests and the Americans focusing on individual interests and emphasizing individual heroism. From the perspective of overall interests, the car is both an independent individual and a part of the overall transportation of the city. The car is not only an independent intelligent brain, but also a bridge with the outside world. Every car is driving on the road, and every moment is connecting between cars and cars, cars and people, cars and roads, and even cars and cities. We need to ensure the safety of bicycles, but also take into account the efficiency of traffic and urban operation. In order to achieve group intelligent driving, it is not enough to rely on single-vehicle intelligence, but also need vehicle-road collaboration, such as the support of infrastructure such as communication end, road end, and cloud.
In January 2024, the Ministry of Industry and Information Technology and other five departments issued the Notice on Carrying out the Pilot Work of the Application of "Vehicle-Road-Cloud Integration" for Intelligent Connected Vehicles to promote the implementation of vehicle-road-cloud integration, which also represents the gradual standardization, systematization and ecology of intelligent driving in mainland China.
The policy steering wheel is left and right
Intelligent driving covers a variety of cutting-edge technologies, which not only contain huge commercial benefits, but also become the focus of the game between countries. It can be said that in intelligent driving, the policy steering wheel has been playing the rudder left and right, trying to control the direction of operation.
"The technology used in autonomous driving, lidar, cameras, artificial intelligence, and other advanced sensors and semiconductors, can all be used to collect data about the American people and infrastructure that could be transmitted back to China. On December 7 last year, a coalition of U.S. groups representing the self-driving car industry wrote a letter to U.S. officials warning that the U.S. Department of Transportation needs to support the development of self-driving cars, or the nascent industry risks losing out to competitors such as China.
On March 1 this year, U.S. Secretary of Commerce Raimondo said in an interview, "Today's cars are like iPhones on wheels, imagine that there are 3 million Chinese cars on American roads, and Beijing can shut them down at the same time." ”
American businesses and politicians, in addition to their own development, but also always pay attention to the trends of their Chinese counterparts, behind this concern, is the sanctions stick that may fall at any time, which will bring major variables to the development of intelligent driving and new energy vehicles in China and the United States.
The United States is a country on wheels, and the American auto industry has begun to take off before the Qing Dynasty is dead. In the 1950s, shortly after the end of World War II, the U.S. auto industry accounted for more than 80% of the global auto market. The U.S. still has the highest car penetration rate in the world, with an average of more than 80 cars per 100 people, compared with more than 20 in China.
As Lei Jun said, "the next decade will be a decade of intelligence", which is enough to subvert the history of automobiles in the past 100 years, and will also reshape the destiny of the two countries.