writing / Li Gen [email protected]
qubits Produce | official account QbitAI
Autopilot ＝ No accident ？
Weilai's first case NIO Pilot After the fatal car accident , Discussion on topics related to automatic driving safety , intensified , A trust crisis is affecting all players .
It even formed bipolar representative views ：
One pole thinks ,“ Self driving accident ” In front of , All autopilots should be revalued , Because the technical capability has such obvious defects , a case involving human life is to be treated with the utmost care . R & D should be stopped , Prohibit the promotion of , Fundamentally eliminate similar accidents .
The other emphasizes , Driving aid , Don't let the autopilot back ,“ True autopilot ” There will be absolute safety , Autopilot ＝ No accident .
One party limits , The other side defends .
But I'm sorry , There are cognitive biases , May hinder the healthy development of a new technology that benefits the country and the people and benefits mankind .
You can't stop eating because of choking , It is not appropriate to Automatic driving and zero accident In the equal sign .
But it's time , reform from the bottom , Clarify the various historical problems left over from the barbaric growth period of Technology .
Problems left over from the history of automatic driving ？
The most important thing is the classification of automatic driving technology , And the aggravation of Translation Terms “ misunderstand ”.
In the hot discussion of Weilai accident , The founder of the ideal car Li Xiang 、 Invest in the company to build cars 360 Chairman of the board of directors Zhou Hongyi , And the founder of Weima automobile Shen Hui , It's all about Automatic driving skill level Discussion on related issues .
Some people say that they are friends in “ References to ”, But if you have a deep understanding of what they denounce “ Professional slang ” And the uniform terminology initiative , Will really feel —— This is their spontaneous self-help under the sense of crisis .
Because the historical problems related to the technical level of automatic driving are unknown , Brutal growth of the industry 、 The current situation that bad money drives out good money will remain unchanged .
So-called Automatic driving skill level , The core is this picture ：
Among them level（ level \ Grade ）, Namely L Several “ Industry technical slang ”.
L2, Namely level 2, That is, the second level .
L3, Namely level 3, That is, the third level .
This set of “ Grade ” standard , With a family named Society of Automotive Engineers Related to the organization of , abbreviation SAE, Chinese call American Society of Engineers .
stay 2014 Global automatic driving R & D , At the beginning of the first boom ,SAE Took the lead in formulating a set of technical grade standards for automatic driving .
The goal is to set different levels , It can guide the targeted development of the industry .
The formulation of the whole grade standard , Referring to many influencing factors .
But the core can be summarized as ：
AI Participation of the system in vehicle driving .
The whole standard , Divided into 6 A level .
L0,AI The lowest level of participation , It's basically to assist in an emergency .
L5,AI The highest level of participation , So high that the whole driving behavior does not need humans at all , It's also called Completely driverless .
For example, what we often hear now RoboTaxi Driverless taxi , Or a smart car that no longer has a brake, accelerator, pedal and steering wheel , Is towards L5 Efforts in direction .
Here 6 Of the three levels , There is a key watershed ——L4.
L4 following , Human drivers are ultimately responsible for driving behavior .
L4 And above ,AI The system is the ultimate responsible person .
L4 It is the watershed of power and responsibility .
In other words ,L4 following , Whether it's L2 still L3, It's all man-machine driving mode ,AI The system provides auxiliary Ability .
The difference lies in ,L2 What do you operate ,AI Just do what , For example, keep the lane centered 、 Adaptive cruise control , And your command to change lanes and AI Wait for the chance to change lanes …… All are L2.L2 Under the circumstances ,AI Not at all “ reflection ”, And don't take the initiative to make decisions .
and L3, The big difference is this AI Starting to help drivers make decisions . The simplest example is , When to change lanes , Human drivers don't have to give instructions .
While driving ,AI Will make decisions according to your itinerary and road conditions , Even in non open road conditions such as high-speed and ring roads , Help complete most driving behaviors .
Just this stage , Humans are still the core decision makers of driving behavior , So special circumstances 、 Emergency decision making , Human drivers are still needed to perform .
So at this stage , Different products have different names .
NOA、NOP、NGP…… But it basically means at high speed 、 Scenes with relatively controllable road conditions such as the ring road AI Driving AIDS .
quoted SAE Formulated terms and Translation Terms , They can really be called “L2 Autopilot ”、“L3 Autopilot ”, And even “L2.5 Autopilot ”……
But at the end of the day , It's all assisted driving , All are AI Provide assistance , The final decision to drive , It's still up to the owner .
It also means that , In case of an accident , Provide this “AI auxiliary ” Provider of capabilities , Yes “ The final driving decision belongs to the owner ” The space available .
The problem is , Autonomous driving itself is a new thing , The technical grade standard is “ Industry slang ”, Ordinary car owners and users , Lack of complete cognition …… In addition, interested technology providers , Maybe just talk about... Before sales “ Autopilot ”, Use cognitive bias to mislead .
That's why it happened “ Automatic driving before buying , Accident assisted driving ”, as well as “ The owner failed to use the correctly ” Such God ridicule 、 God replied .
So now there's an accident , This set of SAE Technical grade standards and translation issues , It's really to blame .
That's why the industry needs Uniform language When —— User trust and regulatory confidence , Can't stand the challenge of similar accidents .
But objectively speaking , The pot is not all in SAE.
Because of the 2014 When the grade standard was formulated in , They saw it RoboTaxi The ultimate goal , But unpredictable , Around the development of automatic driving technology , Will come out of different opinions The two route .
Two major routes for automatic driving ？
Tesla route and Waymo Route .
first SAE When setting standards , These two routes , Just different ideas of promoting automatic driving .
It's not time to blossom and bear fruit , The stage of practical impact .
And these two routes , Nature is With L4 As a watershed To practice .
I just said ,L4 It's a —— Whether humans play a decisive role in driving behavior —— The dividing line of .
among , Represented by Tesla , Hope from L4 The following iteration goes to L4 above , Later called Mass production automatic driving .
In the Google Group Waymo As a representative , It is assumed that only directly from L4 Start , Is in line with “ Safety first heart ” How to do it , Later, as a further distinction, it was called Completely driverless .
Between the two propositions and lines , And often you attack me , The war of words has not stopped .
And the founders and earliest practitioners of these two lines , All of them are Google .
But later , Google decided that it could not go, which was later called “ Tesla route ” The plan .
because Tesla route , It is considered that automatic driving can be in the state of man-machine driving 、 Under the driving data iteration , Keep upgrading , Finally from L2 All the way up to L5.
What Tesla calls “ Shadow mode ”, It is in this state of man-machine driving , Give Way AI Learn human driving behavior .
For example, in the AI When driving , Suddenly encountered difficulties , The human takeover completed the challenge , Will be marked by the system , Followed by AI Model training and learning are similar problems , Then get the model promotion .
actually , Tesla route also continues to show its possibility and feasibility .
In recent years , Tesla's autonomous driving ability , Progress is indeed obvious to all .
And the advantages of Tesla route , It is mainly reflected in cost and large-scale data iteration .
Where the car is sold , It turns on autopilot “ Road test ” The place of . The core requirement is to sell more cars , also “ Autopilot ” The software and hardware cost of the scheme , Within the user's tolerance .
in addition , Tesla route also has obvious paradox .
One side , Mass production automatic driving can replace human car owners driving in some scenarios , But on the other hand, the driver is required to take over the car in case of emergency .
This requires a person to relax for a long time , You must always focus on driving .
Another experience that more people have had , Distraction is allowed in class , But as long as the teacher asks questions by roll call , You can give the answer right away .
The blood and tears of many car owners show that , A relaxed driver , Can watch videos 、 Will fall asleep , And over trust Tesla's visual perception system .
So when “ The teacher asked questions by name ” coming , They are often caught off guard .
However , Drive on the road , It's about life safety , The cost is sometimes extremely heavy .
Musk said before , The owner can't go back when he uses the autopilot .
I make complaints about it , Some car owners do “ Can't go back ”.
And under this paradox , Tesla is not an example .
As long as it is man-machine driving , You can't ask all drivers , One side, relax , The other side can take over at any time .
people , Has always been the most uncertain factor in driving behavior .
It is out of humanity and “ paradox ” Insight , Google （Waymo） Take the initiative to change course , Decide to challenge the more difficult completely driverless route .
As the pioneer of this wave of self driving commercial wave in the world , Google originally wanted to build a man-machine driving system , Step by step RoboTaxi Of L5 The ultimate goal .
But later in the development and testing phase , I found that as long as there are the most uncontrollable factors in driving decision-making —— human beings —— There is , There is no real security .
At that time, with the development of the system more and more perfect , The test vehicle is taken over less and less after it is on the road , Some engineers simply ran to the back seat ……
So Google began to rethink the promotion route of automatic driving .
Because from Beginner's mind Speaking of , Google decided to spend a lot of money on autopilot , Just to use stable and skilled AI Drive , Improve the overall human driving level , Effectively reduce a large number of traffic accidents .
Safety is the starting point , It's also the ultimate destination .
But if you can't 100% Control the whole situation , For example, even relatively mature AI The driving system gives a less responsible human driver , Finally, due to the driver's over trust , Cannot effectively take over when encountering extreme scenes , It will cause safety accidents .
Of course , There is also a saying ,“ accident ” It's also AI System debug The process of .
But in this way to promote autopilot , One will make a thousand bones wither , Is it too ruthless ？
So Google is Google , Not for no reason , They decided to take a more difficult path ：
Directly put the human representative of uncertainty , Get out of the driver's seat . Final product form , Namely RoboTaxi.
So this one goes straight L4 The route of the , It's called Waymo Route .
This is a road test , After a mature system with enough security , Then open the route of mass production and commercial .
The advantages of this route are obvious , Safety first , Cost priority is not high , The most redundant sensor scheme can be used to ensure safety —— After all, the car doesn't face C End sales .
But the disadvantages are obvious , It doesn't push so fast , The speed of commercialization and scale , Also strictly limited .
stay Waymo On the route , All research and development 、 Landing costs , All need to bear their own , Large scale profit , It's far away .
So this 2 Over the years , Tesla and Waymo Encountered bipolar treatment in the capital market .
Tesla's market value and share price are rising ,Waymo Our valuation has been lowered again and again .
And it's also between this and that , There is also a third route ：
Cruise Route .
To put it bluntly “ Fusion route ”, hold Waymo Of “L4 Class a autonomous driving technology ”, After taking into account the cost , Use it on a mass production vehicle .
All in all , On the one hand, I saw the Tesla route “ Data iteration ” Power of , On the other hand, I hope to pass Waymo Route to enhance security .
But even so , As long as you don't get rid of “ Man machine driving ” Pattern , The paradox of human nature and Ownership of rights and responsibilities The problem of , It has never been eliminated .
Safe hidden trouble , There will be no cure .
So the problem comes , Since automatic driving at this stage can not cure the safety problem , Why develop ？
Still burning so much money 、 Give so many favorable policies to support development ？
Autopilot ≠ No accident , But only automatic driving can fundamentally reduce accidents
Yes , There is indeed an idealized view that ：
Develop automatic driving , Let the world be free of traffic accidents .
Also on the , But it is also not .
Yes , Because if the ideal situation is really reached , There are autonomous vehicles on the road , Maybe there won't be “ Kasai ”、“ Uncivilized driving ”、“ Violation of rules and regulations ” Behavior , Cars are civilized , The cars are on the road, safe and counselled , Traffic order is as orderly as automated assembly lines .
But the reason is wrong , Because of this idealized view , Not in accordance with the laws of mathematics and Technology .
In essence , Autopilot , Machine driving , It's a computer problem , It's a mathematical problem , In this question , There are always some subtle variables , Make the accident probability impossible to be completely 0.
What's more? , The complex long tail scenarios and conditions faced by autopilot , Accident probability , It's impossible to do it for you 0.
So autopilot = No accident , It doesn't work , And no one can be sure .
But even so , Autonomous driving still has higher safety , It is also the closest approach to the essence of human traffic accidents .
Traffic safety based on human driving , What is the status quo ？
Quote last year 12 month “ The ninth global traffic safety day ” The data of , When the number of motor vehicles exceeds 3.6 Billion 、 The motor vehicle driver reaches 4.5 Billion of China , Tens of thousands of people die in traffic accidents every year .
More intuitive conversion , The average 8 Minutes 1 People died in a car accident .
Worldwide , according to WHO stay 2018 Data disclosed in , The global death toll from road traffic is as high as 135 ten thousand people , Mean every 25 One person was killed in traffic in seconds .
among , The main causes of traffic accidents , Is the biggest uncertainty factor —— Human drivers .
Both at home and abroad , According to the statistics , The cause of the accident is prominent 10 Major traffic violations are ： Failing to yield as required 、 Speeding 、 Driving without a license 、 Drunk driving 、 Not keeping a safe distance from the car in front 、 retrograde 、 Violation of traffic signals 、 drive after taking a drink 、 Illegal overtaking 、 Illegal meeting .
meanwhile , And drunk driving 、“ Three super one fatigue ”、 Traffic violations such as running red lights .
But if you replace it with AI The driver , These hidden behaviors and factors , Will be fundamentally eliminated .
AI Drivers not only won't have the fatigue of human drivers 、 Drunk driving, drunk driving 、 emotional , And all kinds of dangerous driving and uncivilized driving behavior .
And it has something that humans can't “ Scale replication ” The advantages of , One AI Old driver cultivation , That means thousands of “AI old hand ”ready—— The corresponding is , Human beings have developed from novices to old drivers , Can't get around the linear time and the sharpening on the road .
So the road is straightforward , Although automatic driving can't guarantee that there will be no more accidents , But there is autopilot , Instead of the biggest uncertainty in traffic , It will make the traffic safer .
Besides , Another set of data can also be compared .
Mass production automatic driving route , Tesla with the highest accident exposure , Where there is an accident, there is almost nowhere to hide .
So there are good people , Tesla and AutoPilot Accidents and casualties caused .
Statistical display , From the tesla 2013 The first accident exposed in has been , The global death toll caused by Tesla accident is 201 people , among 9 people And AutoPilot relevant .
Waymo On the route , The only fatal car accident in the world , from Uber The test results in .
And by contrast , Every year around the world 135 Ten thousand people were killed in road traffic .
Although in recent years , Autonomous driving is still a new business , be just unfolding , It is far from reaching quantitative popularity , But the above two sets of data and frequency , You can still see .
So even automatic driving ≠ No accident , But only automatic driving can fundamentally reduce accidents .
What's more? , Now there are more and more redundancy To strengthen security .
For example, the sensor redundancy at the vehicle end 、 Security redundancy , Redundancy during operation , And China's high-speed road end infrastructure —— Higher dimensional redundancy brought by vehicle road coordination .
Under these multiple redundancy guarantees , If you really come to “ All autopilot ” state , Uncertainty in accidents , It will further approach 0.
The assumption is , When these infrastructures are completed , It can ensure that all vehicles on the road are driven automatically , No more man-machine driving , There are no more human driven vehicles and autonomous vehicles .
Human drivers , The biggest uncertainty in road traffic , It's gone .
Even as cars replace carriages , Human driving behavior is just like riding behavior , It has become an entertainment , Even if it's not a chartered road , Human driving also violates the law .
Autonomous driving will completely take over the urban travel network .
In this network , Electric cars （ New energy ）、 Carpooling 、 Car networking , And intelligence have realized series connection .
Call as you go , You don't have to stop , When there is no electricity, it automatically plans to charge .
Really achieve high efficiency 、 green 、 Safety, low carbon, environmental protection and sustainability , The car really became “ Use carrier ”, instead of “ Have the subject ”, Private cars have since withdrawn from the stage of history .
Under this network , Autonomous vehicles can be operated integrally 、 Dispatch , Just like the current communication operation mode .
So that's why , There is another hidden view in the industry , Think behind autopilot , Also have 5G Same “ standard ” battle .
Because autopilot is not a single technology , Is a from nothing 、 From individual cases to large-scale landing System ecology .
And this is not a region 、 A system that a country needs , It's a system needed all over the world .
For this system , Who is the first to have complete solutions and technical capabilities , Can become a standard setter , Can lead the development of the whole landing Ecology .
Exactly so , In the U.S. “ some goods in short supply , hoarded or cornered for making excess or enormous profit ” Our advanced technology restricts exports , Automatic driving has a high status .
But the technical development and iteration of automatic driving , It is inseparable from large-scale data and rich scenes .
This is also China's landing on autopilot , The key reason for showing more and more advantages .
It has been on the same starting line with the world , There are also rich and most challenging road data 、 scene .
So that's why the development of automatic driving , We can't stop eating because of choking .
give up eating for fear of choking Not only can it not fundamentally solve the problem , But also to the detriment of the country and the people , Bad luck will last forever .
Basically , Automatic driving is not only an advanced technology , It's also about smart transportation 、 Smart city and other future oriented global standards and voice , Even economic development and GDP One of the new power engines .
however , At this stage, the development of automatic driving , It is not that there is no perfection in the safe path .
Safer development of autonomous driving , What else can be done ？
Now that the problem has arisen , It's time to push for consensus in the hot debate .
actually , At the regulatory and policy levels , China may already be the most cautious country in the world .
Especially before automatic driving and RoboTaxi The qualification of the test and the issuance of the license plate are in progress , From program to process , Than California DMV, Too strict .
According to the current situation in China The only official Automatic driving test report ：《 Road test report of self driving vehicles in Beijing 》 Disclosure , The assessment criteria are quite strict .
Regardless of the length of the open road 、 In terms of standard setting , They are all meticulous .
It's also specially set up 64827 Km unmanned test verification , The feasibility of the test technology is confirmed 、 And the reliability of test methods and parameters .
And before that , Automatic driving road test with safety officer , It also makes clear requirements from multiple dimensions such as license issuance and operation .
Also under this kind of Supervision , Baidu, which has the most mileage in road test in China Apollo, Complete multiple cities RoboTaxi Landing operation , Yes 1400 After the actual road survey of 10000 kilometers , Be able to achieve zero accidents .
Abroad ,Waymo Start the industry again , Released last year “ accident report ”.
Disclosed from 2019 Early in the year 2020 year 9 month ,RoboTaxi Accidents during landing .
Displayed in the Waymo Of 610 Thousands of miles （982 Thousands of kilometers ） On the way , The test mileage including completely driverless without safety officers is also 65000 miles （10 Thousands of kilometers ）, A total of 18 Time accident , And then there is 29 Time Potential accidents avoided by the intervention of safety officers .
But these accidents have two common characteristics ：
First of all , None of them is serious enough to endanger life .
second , No active accident , Are passive accidents caused by other human driving vehicles .
So for RoboTaxi Autopilot in direction , From supervision to testing , Always have the impression of security and credibility supported by data .
But mass production autopilot —— Or more precisely , Auxiliary driving that requires man-machine driving , There was a gray area to go , In Savage growth .
Once there were only moral standards , There are no strict regulatory regulations .
But these days are also becoming a thing of the past .
At home ,8 month 12 Japan , The Ministry of industry and information technology issued 《 Opinions on strengthening the management of intelligent networked automobile production enterprises and product access 》, For smart cars, in terms of data and security , Made clear provisions .
One of the most striking , It is aimed at product management involving automatic driving function , Different levels of requirements are made before and after use , Strengthen the responsibility of vehicle enterprises in safety .
And clearly stipulates , Autopilot related functions OTA It needs to be approved first .
Abroad , tesla AutoPilot Related accidents , It also began to be investigated by American regulatory agencies .
So even if it's on the edge of the ball path “ Mass production automatic driving ”, The savage growth period is over .
Just before Wei Lai's accident ,“ Completely driverless ” and “ Mass production automatic driving ” Between the two routes , There are arguments and quarrels , Both sides thought they could fight each other .
But who can think of , In the crisis of distrust after the accident , Public opinion is not divided “ Route ”, Everyone is regarded by the same boat .
Prosperity and prosperity , Suffer all losses . No one can be alone .
Whether it's L Several new technologies , Whether it's mass production autopilot routes , Or a fully autonomous route .
Whether it's a car factory 、 supplier , Or system technology players .
So it's time .
It's time to learn from this accident , Face what has always been cognition and trust The crisis .
And only all players Unite , Reach a convention , To make development safer 、 sustainable .
At least it's time to change , And there are Three things You can start with me ：
First of all , Standardize the use of industry terms , Unified expression , Clarify the ownership of rights and responsibilities .
But where man and machine drive together , Need a driver 、 The owner took over in an emergency , It's all called Auxiliary driving .
Otherwise there will be an accident , Whether or not “ The owner failed to take over in time ”, It is the responsibility of the capability provider .
No driver required 、 The owner took over in an emergency , Or the responsibility is clear, that is, the technical service provider , To promote automatic driving .
second , Assisted driving on the car , There should also be multiple safety redundancy cooperation .
Not only before use , Clearly show the advantages and disadvantages of auxiliary driving system to the owner , Strengthen the access assessment and supervision of function users .
We also need to prevent unreliable car owners , You can easily bypass “ Auxiliary driving ” The necessary rules for opening , And get your hands off the steering wheel 、 Dangerous driving behaviors such as inattention .
Besides ,DMS Wait for driving behavior monitoring , It should be standard for assisted driving , Not just to protect the owner's safety , It is also to ensure the safety of other road traffic participants .
Without permission “ Cheating device ” Such behavior , Should be treated like dangerous driving . It is also necessary to take the initiative to call for a crackdown on sales “ Cheating device ” act , Call for legislative regulation , Hit by major retail platforms , We can't let such complicity in wealth and life , It's very popular .
Third , Regularly and proactively disclose security mechanisms and data .
Security mechanisms can enhance industry trust ,“ Security \ accident report ” Can make everything more open and transparent , Only by getting rid of “ Talk about accidents ”、“ Carefully conceal mentioning ” State of mind , Only in this way can the industry develop healthily .
This is also a constant popular science , Help the public better understand the ability of automatic driving 、 Necessary means of state and stage .
All in all , The accident is currently , The crisis of trust is now , The alarm bell rings for all autopilot players .
If here and now , We can't do it in a more conventional way , Develop and promote autopilot safety and trust , Wait until a bigger crisis comes , No one can stand out .
If you still want to “ Auxiliary driving ”、“ Autopilot ” And so on , If you let every time “ Self driving accident ” Become the focus of public opinion , If industry players have to “ Exaggerated propaganda ” In order not to look backward ……
The avalanche is coming , No snowflake is innocent .
It's time .
（ notes ： The author is qubit editor in chief , The column represents only personal views ）
Reference resources ：
Tesla accident statistics ：
Waymo Autopilot report ： Average 21 An accident in 10000 kilometers , Serious accidents are the pot of human drivers
WHO Road traffic fatalities report
Read the original