Under the premise of ensuring safety, autonomous driving requires more room for trial and error and time for exploration.
| Toretto
When machines become more and more human-like, will the person acting as a "template" be curious, surprised or panicked?
From having a good feeling of disgust, fear and then regaining good feelings, this is the "uncanny valley" theory proposed by Japanese roboticist Masahiro Mori, the second stage of which is that "if the similarity between robots and humans reaches a certain level, human reactions to them will suddenly become extremely negative and disgusting, even if the robot and humans are only a little different, it will appear very conspicuous and dazzling, so that the whole robot has a very stiff and terrifying feeling." ”
The self-driving industry, which was affected by the NIO NOP accident, seems to be heading towards the "uncanny valley".
With the fermentation of this accident, the discussion and debate on automatic driving technology has been expanded from "assisted driving" to include higher-level L4 and above unmanned driving technology, the entire automatic driving industry has been affected, and even Dong Yang, vice chairman of the China Electric Vehicle Hundred Association, deliberately emphasized in his writing, "The title of this article uses intelligent driving technology, and try to avoid the use of automatic driving technology."
Safety is the first priority of the transportation and automotive industry, the death of the owner of the Weilai car did sound the alarm for the entire autonomous driving industry, whether it is the government, enterprises or the public need to scientifically look at and use automatic driving technology, whether it is assisted driving or higher-level driverless.
The government in encouraging innovation while strengthening supervision, enterprises should not only actively develop and explore, cautiously test operations at the same time, to objective publicity, strengthen the guidance of users, do not treat users as guinea pigs; and the general public need to be cautious, can not relax their vigilance, but also dare to innovate while taking into account the safety of enterprises to give some tolerance, to avoid a stick to kill everyone.
Unfortunately, a small number of non-automotive technology media confuse L2, L3, L4 and other different levels of technology, and do not distinguish the difference between mass production models and test vehicles, concept cars and other cutting-edge technology products, in such a situation eager to rub hot traffic, blindly criticize the automatic driving industry and enterprises, easy to mislead the public, and is not conducive to the development of the entire industry.
Of course, mass production of assisted driving on the car does have some publicity and even technical problems, but can not be choked up, many companies and the entire commercial focus of the future period of time is still to force assisted driving, the top priority is to find problems, iterate and progress on technology, make up for loopholes, help the public out of the "uncanny valley".
The L4 and other higher-level unmanned driving is still in the stage of research and development testing and trial operation, should not be confused with assisted driving, under the premise of ensuring safety, such cutting-edge technologies require more trial and error space and exploration time.
The road is long and long, and the warriors who are searching up and down are already on the road.
<h1 class="pgc-h-center-line" data-track="26" > the origin of the "chaos" of automatic driving</h1>
Whether it is ordinary consumers or some media practitioners, the key to their difficulty in distinguishing between assisted driving and automatic driving is that the "authoritative" concept of automatic driving definition and classification has not been simplified and popularized.
Many definitions are full of professional and complex expressions and many dazzling and relatively vague limited words, which are not friendly to the general public, and invisibly raise the threshold for ordinary people to understand and understand autonomous driving technology and the industry. Not to mention that different countries and companies are actually controversial about many definitions, which increases the cost of learning.
At present, the relatively common grading method is the National Highway Traffic Safety Administration (NHTSA), the American Society of Automotive Engineers (SAE) for automatic driving classification standards, there are level 0 to Level 5 six levels, most companies use abbreviations L2, L3, L4 and other words to label, such expressions are actually not friendly to domestic and foreign consumers, seemingly concise but can not clearly convey information to them.
April 2021 Edition of grading standards Source: SAE international
According to the above figure, look at its corresponding professional interpretation, taking L2 as an example, this technical level is to provide steering and braking/acceleration support for the driver, can this expression be understood by ordinary consumers?
In terms of L3 and L4, seemingly professional and rigorous statements such as "limited conditions" and "all conditions are met" are actually vague, and the right of interpretation is attributed to experts or enterprises, and it is difficult for the public to truly understand the relevant meaning.
As Philip Koopman, chief technology officer at Edge Case Research and associate professor at Carnegie Mellon University, puts it, the SAE's classification criteria are "engineer-designated engineering standards for engineers to serve engineers" that are difficult for ordinary people to understand and master, so in early 2021 he suggested publishing a framework classification for consumers.
It's just that his standards don't jump out of the SAE framework, such as its second level "Supervised" (about equal to L2 and L3), which explains: "The vehicle is responsible for driving, but the human driver is responsible for ensuring safety", which is still vague and still not friendly to the general public.
China also has its own self-driving rating. The "Guidelines for the Construction of the National Vehicle Networking Industry Standard System (Intelligent Connected Vehicles)" issued by the Ministry of Industry and Information Technology in December 2017 pointed out that in terms of decision-making control, according to the roles and responsibilities of vehicles and drivers in vehicle control, they are divided into "auxiliary control" and "automatic control". The former can be divided into driver assistance (DA) and partial automatic driving (PA); the latter includes three sub-categories: conditional automatic driving (CA), highly automated driving (HA) and fully automated driving (FA).
In fact, similar to the problem of SAE, the two categories plus the following 2-3 sub-categories, there are clear definitions, but it is difficult for ordinary consumers to have the energy to understand the standard first, let alone to understand which level of the "automatic driving" function of a certain model is compared to this standard.
The result is that consumers and a small number of media in the field of non-automotive technology can only obtain information through the publicity of car companies and the interpretation of relatively professional media, and it is difficult to form their own judgment.
When the owner of Weilai died in the accident, the vehicle was in the NOP state, that is, assisted driving, but with the passage of time, there began to be a general public "questioning the higher level of automatic driving technology", such as the L4 level Robotaxi, that is, driverless taxis, and some media contributed to the attack on companies committed to the development and testing of such cutting-edge technologies, and attacked the entire autonomous driving industry.
In fact, it is desirable to reflect on the excessive publicity or confusion of "assisted driving" for "automatic driving" by some car companies, and call on consumers to use assisted driving technology scientifically and rationally, but it is undoubtedly unprofessional or even wrong to confuse mass-produced assisted driving with unmanned driving technology that is in the stage of research and development testing and trial operation, and requires the participation of all parties to get the public's cognition of the autonomous driving industry back on track.
<h1 class="pgc-h-center-line" data-track="163" > how assisted driving gets out of the "uncanny valley"</h1>
"Please note that the title of this article uses intelligent driving technology, and the statement of avoiding the use of automatic driving technology as much as possible is to avoid misleading users that the technology can be applied unconditionally." This is written by Dong Yang, vice chairman of the China Electric Vehicle 100 Association, in the article he published.
Even the titans of China's auto industry are beginning to avoid affecting readers, and autonomous driving may indeed be in the "uncanny valley".
The "uncanny valley" theory was proposed by Japanese roboticist Masahiro Mori in 1970:
Because robots and humans are similar in appearance and movement, when a robot's appearance becomes more human-like, the observer's emotional response to the robot becomes more positive and empathetic;
When the similarity between robots and humans reaches a certain degree, human reactions to them will suddenly become extremely negative and disgusting, even if the robot and humans are only a little different, it will appear very conspicuous and dazzling, so that the whole robot has a very stiff and terrifying feeling;
When the similarity between robots and humans continues to rise, equivalent to the similarity between ordinary people, the emotional response of humans to them will return to the positive, resulting in empathy between humans and humans.
The second stage can be regarded as the bottom, almost all of the "humanoid" intelligent products in some aspects have a history of uncanny valley before the technology matures and is generally accepted by the public.
"The self-driving car is close to L2+, and when it's close to L3, its driving experience is very good, people will believe it, but when it will have an accident we don't know, so it fell into the uncanny valley." Zhu Xichan, a professor at the School of Automotive Of Tongji University, said bluntly that whether it is Weilai or Tesla FSD and other self-driving technologies falling into the "uncanny valley", it just shows that this technology that tolerates driver distraction is maturing, because at least some people already believe in starting it.
There is always a difference in theory, and the theoretical difference between Mori Masahiro is that the assisted driving technology that is now mass-produced on the car is caught in the "uncanny valley" because some consumers do not know 100% about it, they have gained their trust, developed inertia, and used it in scenarios beyond its technical level.
In summary, the "uncanny valley" of assisted driving is caused by many reasons such as cognition, inertia and technical level, and it is possible to start from these aspects when it comes out of the bottom.
Car companies and technology companies should gradually abandon the use of overly professional but vague expressions to "fool" consumers, such as L2.5 or L2.9 and other specious concepts, and instead can cooperate with the academic community or several companies to launch a hierarchical expression of autonomous driving that users can understand.
Ideal Auto founder Li Xiang called for a unified self-driving Chinese name in the circle of friends, and the chairman of 360, who invested in Nezha Automobile, said that "L2 to L5 (terminology) are industry black words, entertaining themselves." ”
The thinking of Li Xiang and others is quite good, but changing to Chinese name, the interpretation and interpretation are still very complicated, and it is still difficult for ordinary consumers to understand its meaning. In fact, there have been attempts to stop defining the grade in the SAE standard, and the definition is more concise.
A Chinese executive from an international automaker that leads the world in active safety told HD Auto that the industry can still be called the autonomous driving industry, but they divide autonomous driving technology into two parts: driver-led "assisted driving" and "driverless" that does not require human participation, which will clearly cut the two.
This is quite similar to the "auxiliary control class" and "automatic control class" of the Ministry of Industry and Information Technology of China, but the subsequent subdivision sub-categories are omitted to make it easier for consumers to understand.
Shen Hui, chairman of WM Motor, also publicly stated, "Using the L2 level assisted driving function, the driver is the main body of functional operation and the main body of responsibility; Using the L4 or above level of automatic driving function, in the unmanned state of the car, the operating body is the vehicle itself, and the responsible entity belongs to the main engine factory. ”
Under the two categories of assisted driving and driverless driving, it is no longer subdivided, but added to the two items for consumers to choose through the way of "function", each function emphasizes the use of methods and specific scenarios, reduces the learning cost and difficulty of consumers, and it is easier to clarify the responsibility between people and vehicles.
Secondly, car companies are to a certain extent "trying to challenge the 'evil' in human nature - inertia".
When assisted driving is introduced, even if the publicity is misleading, it will remind consumers that this is not "automatic driving" and need to pay attention to road conditions at any time.
Many people say that drivers need to be "educated" to learn and even repeatedly instill "assisted driving is not equal to automatic driving", "can not relax when turning on assisted driving", etc., but this itself is anti-human, that is, "inertia" - many situations before the vehicle can be solved, why should I concentrate?
Engineers in the former Tesla Autopilot business unit once told HD Auto that they had done statistics and surveys on the accident rate of Autopilot and found that after the accident rate was reduced to a certain extent, there was a significant recovery, and even could not be lowered.
But can't assisted driving systems avoid the driver's inattention? In fact, there are also "patches", which may not be perfect.
Whether it is to hold the steering wheel or grip force induction, in the face of "laziness", consumers can always find various ways to hang a brick or put a bag of oranges, in order to make it easier, human creativity is endless.
In fact, DMS is the driver monitoring system can better help solve this problem, the camera for the line of sight to grasp, if a certain time to close the eyes or look in other directions, the system will alarm, if there is no operation, then slow down, pull over. The infrared camera can solve the problem of sunglasses, once it cannot be detected, the assisted driving system cannot be turned on.
This is a technological patch for assisted driving systems to overcome "inertia", and the more time-and-for-all approach is to develop and use a more advanced technology - driverless.
<h1 class="pgc-h-center-line" data-track="164" > driverless: tough but promising</h1>
From assisted driving to driverless, whether the two technologies can evolve gradually or leap forward, there is still controversy in the autonomous driving industry, but there is no doubt that assisted driving is (L2/L3) can not be confused with driverless (L4 and above).
Most car companies agree with the leap change, in their view, assisted driving is a human-led technology, unmanned driving is machine-led, the two are not a gradual development process, perhaps some data and technology are universal and reused, but the difference between the dominant players makes the two more parallel development and evolution.
In the case of GM, for example, in March 2016, GM spent $1 billion to acquire cruise Automation, a driverless startup, to keep the startup team running independently, hoping to "walk on two legs" with the team that developed assisted driving.
Qian Huikang, then executive vice president of GM and president of GM China, told HD Auto that the autonomous driving technology led by the Super Cruise team and Cruise Automation belonged to two completely different technology paths, each developed by two completely independent teams. The former is progressive, the latter is leapfrogging.
Of course, driverless technology is still in the early stages of development, and has not yet developed to the stage where large-scale mass production can replace assisted driving.
"In the next few decades, self-driving technology will not be able to be ubiquitous, and driverless cars will always have limitations." This is Waymo CEO John Krafcik pouring cold water on a once-hot driverless in November 2018.
With the gradual commercialization of autonomous driving, the future prospects are very bright. As the most commercially valuable scenario for AI, if autonomous driving can take root, according to Roland Berger's prediction, by 2030, the global revenue of driverless taxis will reach 1.5 trillion euros (more than 1.75 trillion US dollars).
It can be said that from road testing to commercial trial operation to the future of a new fully unmanned cockpit design, these can not be produced out of thin air, and need to be constantly tried and explored.
More importantly, the continuous advancement of unmanned driving technology can accelerate assisted driving to break through the "uncanny valley" to a certain extent.
Although there are differences in the technical logic of assisted driving and unmanned driving, to a certain extent, the technology of unmanned driving continues to mature, and some of them can be lowered to help improve and perfect the L2 level assisted driving technology.
The technology of the autonomous driving industry is nothing more than perception, decision-making, execution, etc., whether it is open roads or closed parks, passenger cars or commercial vehicles, assisted driving or driverless, universal software and algorithms still exist.
"For us, the cost of technology migration is not high, and software and algorithms such as perception, fusion, computing, and decision-making can be largely shared." The management of a domestic autonomous driving startup bluntly said that they have research and development of both technologies, assisted driving is profitable, driverless is relatively expensive, but the latter's technology can feed back assisted driving to a certain extent.
"In the future, whoever masters the real autonomous driving platform technology will be the king." Autonomous driving companies with this powerful technical capability will be able to implement dimensionality reduction strikes. This is the public statement of Han Xu, CEO of Wenyuan Zhixing, this year.
Startups are more willing to try platforms and hope to achieve dimensionality reduction, of course, technology companies are also taking this road.
Whether it is assisted driving or unmanned driving, its original intention is to solve human traffic accidents and improve travel efficiency. Although neither of them may be 100% accident-free, assisted driving can make human driving safer, and future unmanned driving must be safer than human driving.
McKinsey's data shows that self-driving technology can reduce accidents by more than 90 percent. The government's data is more rational, and an independent U.S. government investigation report shows that automatic driver assistance can reduce collision rates by 40 percent.
Waymo, a leader in driverless driving, released data from its January 2019 to September 2020 road tests last year, with 6.1 million miles tested in 2019 and 65,000 miles driven by its fully autonomous vehicles in the first three quarters of 2020.
In the equivalent of "500 years of driving time of the average licensed driver in the United States", Waymo admitted that the test vehicle was involved in 18 collision accidents, but the accident vehicles were all "minor injuries" and the responsibility was almost all on the other side. They hope to build public trust in self-driving technology by disclosing enough compelling data.
Even in the 2019 California Vehicle Authority's MPI (Average Takeover Mileage) ranking, Baidu surpassed Waymo to rank first. In China, Baidu Apollo's L4 level automatic driving safety test mileage has exceeded 14 million kilometers, which also strongly verifies the safety of autonomous driving.
Only in the past 5-10 years, assisted driving will be the mainstream of industry development and enterprise commercialization, and driverless technology is in the accumulation stage of accumulation. The autonomous driving industry has coexisted for a long time, which is an industry where multiple parties work together, cooperate with each other, understand and tolerate to continue to move forward.
All parties are really trying. On August 24, the Ministry of Industry and Information Technology issued an official letter replying to Baidu Chairman Robin Li, as a member of the National Committee of the Chinese People's Political Consultative Conference, "Accelerating the Popularization of Autonomous Driving Commercial and Intelligent Transportation, Making People Travel Greener and More Convenient, and Achieving the Goal of Carbon Peaking" submitted by Baidu Chairman Li Yanhong in March this year, stating that "we will continue to promote large-scale experiments and commercial deployment of the Internet of Vehicles in typical areas and typical scenarios, innovate and explore operating models and business models, and gradually form a replicable and generalizable construction and development model".
At the moment of talking about automatic driving discoloration, the Ministry of Industry and Information Technology replied to Robin Li's proposal at this time, in a way, giving driverless cars a clear support.
In terms of enterprises, Baidu Apollo has long begun to carry out commercial demonstration operations of autonomous driving in Beijing, Guangzhou, Changsha, Cangzhou and other cities.
On August 18, Robin Li released a new self-driving travel service platform - "Radish Run", which is an upgraded version of its self-driving ride-hailing software "Apollo GO" at this year's Baidu World Congress.
On the basis of 14 million safety test miles and 400,000 cumulative service trips, Baidu hopes to accelerate the commercial deployment of autonomous driving travel by launching "Radish Run", which is fully open to the public.
Only when more and more Robotaxi companies begin to commercialize trial operations and invest more vehicles, can they continuously improve the technology at the same time, so that more ordinary people can also contact driverless technology, slowly eliminate the fear of "no one", and let it gradually enter the lives of the people.
As Dong Yang said: "The emergence of high-level intelligent driving technology is a huge progress in science and technology, and this technological progress cannot be completed in a short period of time." At present, high-level intelligent driving technology is in the stage of further improvement in use, and only by scientifically looking at and applying can it truly benefit mankind and avoid unnecessary harm. ”
-END-