On July 3, Musk replied to netizens on Twitter that he had given up "fantasy" and publicly admitted that it was difficult to achieve autonomous driving.
A few days later, at the World Artificial Intelligence Conference (July 8), Su Jing, president of Huawei Intelligent Driving, commented more sharply: "The L5 fully automatic driving is a lighthouse, but I can't see it in my life."
Obviously, when the whole industry and the capital market are passionate or even desperate about the future of autonomous driving, the above remarks are like pouring cold water.
Recently, at the Autopilot Salon organized by the Automobile Industry and Commerce Organization, Zhu Xichan, a professor at the Automobile College of Tongji University, commented that the current autopilot market is indeed "overheated".
And what is the real situation of L4 and L5 autonomous driving in domestic and foreign automobile markets?
It may not be as optimistic as the implementation timetable announced by the company.
(note:? According to the automatic driving classification of SAE (International Association of Automata Engineers), in the L4 and L5 stages, vehicles do not need any human operation, which are collectively called "unmanned driving", or fully automatic driving or generalized automatic driving. Last year, the automatic driving classification standard issued by the Ministry of Industry and Information Technology was similar. )
On July 7, AutoReport Salon: Zhu Xichan (second from right), a professor at Tongji University Automotive College; Zhang Ning, Vice President of Ma Xiao Zhixing and Head of Beijing R&D Center (second from left); Sun Xiaohan (first from right), editor-in-chief of Car Things and host of salon; Huang Chi (first from left), deputy editor-in-chief of Automobile Production and Marketing and host of salon.
Play autonomous driving,
Is it a process of "hitting the face"?
The Guardian 20 15 predicts that by 2020, you will become a "permanent backseat driver".
A title of Business Insider 20 16 is "2020, 1000 Driverless Cars on the Road".
General Motors, Google's Waymo, Toyota, Nissan, Honda and Baidu all announced that they will produce driverless cars in 2020.
Musk bull blew earlier, and once said that it would realize automatic driving at 20 18. If 20 18 is not successful, it will eventually be realized in 2020.
The result ... as you can see.
In addition, Waymo, an unmanned vehicle company owned by Google's parent company Alphabet that adheres to the robotaxi route, has also experienced a similar process.
In 20 17, Waymo began to test unmanned driving on public roads in Phoenix, and took the lead in opening the commercialization road of L4 autonomous driving.
Objectively speaking, this is a landmark event of unmanned driving.
But now that four years have passed, Waymo's self-driving exploration is still struggling-this year, because the commercialization path of robataxi is still unclear, it has begun to expand into business B.
"Self-driving cars are really grinding." John Crafts, CEO of Waymo, said in an interview last year. "If you ask me, this guy is harder than putting a rocket into orbit."
Last month, waymo's fifth-generation driverless car just expanded its commercial operation from Phoenix to San Francisco (the combination of 29 cameras and 6 lidar sensors can be said to be one of the most perceptive driverless cars at present), and a traffic accident occurred within a few days of its opening.
According to the official clarification of waymo, the vehicle was leaving the automatic mode and driving in the manual mode when the vehicle turned left at the intersection. After turning, the vehicle came into contact with passers-by while still in manual mode, and no serious injuries were caused.
According to industry insiders, this small accident alone may affect the mood of investors and the public.
Uber is the worst on the road of autonomous driving.
Uber's autonomous driving department was established on 20 15. In the same year, a Uber car used for autonomous driving test was involved in a car accident in Tempe, Arizona, USA. Although no casualties were caused, Uber also suspended the road test project for self-driving cars.
Then on 20 18, Uber's self-driving car caused another death, which hindered its further development.
Until June 5438+February last year, affected by the COVID-19 epidemic, Uber's loss-making self-driving business was sold to Aurora for $4 billion.
Regarding the development status of global autonomous driving, Mark gottfried, a partner of Bain, said: "The reality has awakened fanatical hype."
Domestic autonomous driving began to be "fanatical"
"But this is just a starting point."
On the other hand, in China, in a year or so, Baidu, Didi, Ma Xiao Zhixing, AutoX, Wenyuan Zhixing and many other technology companies have joined robotaxi circuit.
In May of this year, Baidu Apollo launched Apollo GO, a self-driving travel platform in Beijing, to provide "shared unmanned vehicles" travel services;
Ma Xiao Zhixing unmanned vehicle joined Guangzhou Robotaxi team;
Auto X tried a driverless taxi in Pingshan, Shenzhen; ……
On the other hand, major domestic car companies have also announced their own L4 level implementation time:
Great Wall plans to realize the first L3-level automatic driving with full vehicle redundancy in China this year, and the first intelligent driving with L4-level capability in China in 2022.
Changan and Beiqi New Energy will set the L4-class autonomous driving mass production time in 2025;
The L4 self-driving model jointly built by Weimar and Baidu is planned to be released this year;
Li expects that in 2024, OTA will enable its models to have L4-class automatic driving capability;
……
Why has domestic autonomous driving become so "hot"?
Zhu Xichan believes that because the window of autonomous driving is about to close, everyone will not hesitate to exchange money for time. This year, capital went crazy.
However, compared with international car companies and technology companies that entered the autonomous driving track earlier, the domestic market is as hot as the United States a few years ago.
Tesla began to design his FSD when autopilot was launched on 20 16. 20 19 the first version of FSD beta has been released. It was not until 202 1 that Musk began to admit that autonomous driving was more difficult than he thought.
Zhang Ning, vice president of Ma Xiao Zhixing and head of Beijing R&D Center, said that the whole "mental journey" of Tesla's research and development of FSD took five years, including a "learning curve".
"Today, everyone's planes and cannons are armed to the teeth and started to do this, but you haven't walked many pits yet. In fact, you don't know how deep the water is in this place. "
The same principle can also be applied to the operation of robotaxi.
Also at the World Artificial Intelligence Conference on July 8, Li Yanhong, founder, chairman and CEO of Baidu, said that in the next 2-3 years, Baidu plans to fully open the previously public Apollo moon landing shared unmanned vehicle service to 30 cities across the country.
Zhou, the founder of 360, sarcastically said from a distance: "Blowing artificial intelligence to the extreme is harmful to the industry. It is better to tell consumers honestly that I can achieve automatic driving in some scenarios, but this car cannot be separated from people. "
And Zhou's "bitter friend" remarks reflect the gap between the ideal and reality of autonomous driving.
After all, even Waymo, which started the commercial operation of robotaxi from 20 17, complained that this road was extremely difficult last year.
Moreover, although Ma Xiao Zhixing has joined robotaxi circuit, Zhang Ning clearly pointed out that everyone is talking about the second half of autonomous driving, but this is only a starting point.
"I think we should work hard for at least five years. Even if it is truly commercialized like Waymo, it will take several years. "
Zhu Xichan revealed that most domestic companies operating robotaxi refused to upload the detachment report.
Autopilot deviation report is an important reference to judge the running situation of autonomous vehicles. Without the resignation report, it is impossible to know the true level of the company's autonomous driving.
How high is the public's expectation for driverless driving?
In the process of developing FSD, Musk once asked: "Does the safety of L5 autonomous driving need to reach twice the requirements? Triple? Five times? Or ten times? What is the acceptable level for human beings? 」。
Zhu Xichan mentioned that the United States has also conducted market research on autonomous driving and found that society has high requirements for the safety of autonomous driving. ?
In the questionnaire, is it acceptable that self-driving cars are as safe as human drivers? The result was almost totally unacceptable.
Reduce traffic accidents by 50%. Accept it or not? More than half of the people don't accept it.
Until the questionnaire, self-driving cars can reduce the traffic accident rate to the existing 10%, which was accepted by most people.
Zhu Xichan said that even so, it still cannot truly reflect the acceptance of ordinary people.
When automatic driving above L4 is applied, users may not tolerate accidents. Zhu Xichan said, "People are much more tolerant of their own mistakes than those of machines."
Su Jing once said that ordinary users tend to distrust new technology products at first, but once they try and have a good initial experience, they will trust this technology very much, which is the unexpected beginning.
Su Jing only talks about L2 and L3 assisted driving. So how can technology achieve higher expectations of autonomous driving? How can people avoid accidents after accepting new technologies?
Perhaps as Musk said, this is extremely difficult.
Moreover, at the salon, both Zhu Xichan and Zhang Ning admitted that from the design logic of self-driving cars, it is impossible and should not take all small probability events into account.
"For example, if this person suddenly jumps in and takes all these irrational things into account, you will find that your car is impossible to start."
Zhu Xichan said that it is unrealistic and impossible to build an unmanned vehicle without accidents at all.
Written at the end, even when most advocates or promoters of autonomous driving AI technology think that autonomous driving will be safer than ordinary drivers, some insiders pointed out:
Due to fatigue, carelessness, alcoholism and other factors, human beings do make many mistakes in driving, but the current sample number and data distribution can not accurately compare the accident frequency between drivers and AI.
In addition, compared with the AI algorithm behind self-driving cars, human errors may occur frequently, but there are few strange phenomena: for example, rational human drivers will not hit overturned trucks.
Therefore, around the future of autonomous driving, there are more unknown problems, even wrong premises, besides technical route selection, cost, policy and morality?
So, when will the real autonomous driving come?