The development of the industry requires the driver to be out of his hands while ensuring safety.
Author | Cheng Xi
Edit | Wen Liang
Recently, the 4th "Global Intelligent Driving Summit" hosted by Leifeng Network & New Intelligent Driving was officially held in Shenzhen.
At the summit, Professor Zhu Xichan of the School of Automotive Of Tongji University delivered a wonderful speech entitled "The Boundary between Autonomous Driving and Assisted Driving and the Technical Challenges of Autonomous Vehicle Safety".
This year, the smart car track is very hot, in addition to Wei Xiaoli, the three new car-making forces, this year there are many new players to join the car-making.
Zhu Xichan believes that this is largely related to the flow of capital - Tesla and Wei Xiaoli may have less than 0.1% market share, but the market value is almost equal to the market value of traditional cars combined. There is a split between the market capitalization of traditional car companies and the new car-making forces.
However, there is still room for improvement in the understanding of autonomous driving and assisted driving by ordinary users, especially for the definition of L2+. Zhu Xichan said that this state cannot exist for a long time, and this year governments have begun to recognize L3 certification.
In June 2020, the ECE Regulation came out with a draft ALKS. ALKS creatively solves the critical testing problem of hazardous scenarios:
Based on the collision avoidance ability of skilled and cautious human drivers, "unreasonable dangers" and "reasonable risks" are distinguished from the regulations; through the establishment of critical working conditions in dangerous scenarios, the "human rights" of "electronic drivers" of autonomous vehicles have been won, so that autonomous vehicles will not be too conservative, which is conducive to improving user satisfaction, improving traffic efficiency and reasonably controlling accident risks.
In order to ensure sufficient safety, many car companies have begun to add more accurate cameras, higher performance chips, and more lidar. However, Zhu Xichan stressed that to ensure the reliability of the automatic driving system, it is necessary to develop the V-shaped development process based on the scene library + the agile development of the closed loop of user data.
The following is the full text of Zhu Xichan's speech, Leifeng Network New Wisdom Driving has made a collation and editing that does not change the original meaning:
Hello everyone! I heard a lot of reports this morning, and now when it comes to smart cars, any one of them is inspiring for everyone.
This afternoon, I want to talk to you about autonomous driving and assisted driving. Many car companies have launched navigation Pilots in the past two years, including NOP (Weilai), ANP (Baidu), NGP (Xiaopeng), and also expressed their expectations for automatic driving of cars - when cruising aircraft, pilots basically do not have to manually operate the machine.
So, are the cars we currently have on the market autonomous driving or assisted driving? This boundary is very blurred, and even some people can't say that the thing is automatic driving, and the accident is assisted driving, and now L2+ is in this state, but this state cannot exist for a long time.
Therefore, you will find that starting this year, governments have increased the intensity of L3 license certification and driverless license certification, and the corresponding laws will also be changed to a certain extent, which we will talk about in detail later.
Over the past 100 years, the big changes that have taken place in the automotive industry have mainly revolved around power. At present, it is the industry's once-in-a-century transformation. I formed it into three battles:
Electrification campaign (2015-2020), the "carbon emissions" international convention pushes for the "electrification" of vehicles.
In the intelligent campaign (2021-2025), market demand drives the "intelligence" and "networking" of automobiles.
In the ride-sharing campaign (2026-2030), L4 will revolutionize the properties of cars.
It can be seen that in the past five years, traditional cars are slowly being replaced by electric vehicles, but this battle is not over, and internal combustion engines can extend their "life" through hybrid, extended range, plug-in and so on.
The two general directions of new energy vehicles in the future are pure electric and fuel cells, but at present, Tesla, Wei Xiaoli and other new car manufacturers basically rely entirely on power batteries, hydrogen as an important branch of new energy, has not been very perfect development, and even related companies are facing the risk of loss and bankruptcy.
The intelligent campaign, many companies that started with AI technology have also joined, and have begun to make good progress in the intelligent cockpit and ADAS, but the focus of this campaign will be L3/L4. Five years later, with the gradual landing of L4 autonomous driving, the attributes of the car will be completely changed.
From the perspective of market share, pure electric passenger cars in the A0 class and B class two market segments sales are relatively high, A class is relatively small, dumbbell type; of which B class cars are no longer pure means of transport, but gradually become the third living space.
The sales distribution of fuel vehicles is spindle-type, A0 class and B class type are relatively few, and A class cars still occupy a very large market share.
From the perspective of market capitalization, the industry has also undergone a major split - traditional automobile companies still account for 99.9% of automobile sales, but the market value of Tesla and Wei Xiaoli is almost equal to the market value of traditional automobile companies.
This also greatly affects the flow of funds in the investment and financing industry, one of which is the intelligentization of automobiles, which mainly includes three aspects:
Smart cockpit, if consumers find that there is no intelligent screen in the car when buying a car, many people basically will not consider it. Smart cockpits are a very necessary factor to attract consumers to buy a car.
Intelligent driving, which is also one of the most important functions of a smart car. For safety reasons, most of the smart cars on the market are now equipped with ADAS systems based on monocular cameras and millimeter-wave radar, and the number of vehicles is growing faster and faster. It is no exaggeration to say that this is the highlight of the ADAS system. ADAS systems can "save people", but there are also times when they are "messed up". Everyone realized that relying on millimeter-wave radar and cameras for a few hundred dollars was not enough to meet the needs of intelligent driving. So this year, everyone began to pile up materials, equipped with higher-definition cameras, chips with higher computing power, more powerful domain controllers, and even lidar and high-precision maps.
A domain controller will be a new flashpoint. Because the distributed electronic and electrical architecture cannot meet the needs of automatic driving, the new electronic architecture of autonomous vehicles has been industrialized, and the domain controller is an important part, and then it is the integration of the whole vehicle and the joint control of the vehicle and the cloud. At the same time, the development of domain controllers is inseparable from the support of high-power vehicle specification chips.
Over the years, with the improvement of the computing power of the car chip, the pixels of the car camera have been upgraded from the original hundreds of thousands to more than one million now, and even go straight to 8 million pixels.
But there is a very real problem is that the application of high-power chips will greatly increase the cost of the whole vehicle, although its cost is declining, but the current consumers can not accept. Even if consumers can accept it, the current development of intelligent driving technology will not allow them to truly take off their hands, feet, eyes, let alone leisurely drink coffee in the driver's seat and look at their mobile phones.
In June, I shared the safety of autonomous driving at an event that mentioned the concept of "uncanny valley", where users expect more than the system can. Soon after, a car company had a fatal accident suspected of being caused by intelligent driving, and then almost all car companies were revising their promotional copy, changing "automatic driving" to "assisted driving".
In addition, car companies have also made many improvements to prevent improper operation, such as once the driver's hands leave the steering wheel, the system will issue a warning reminder - even if the driver's hands can leave to some extent.
However, the development of the industry requires drivers to take their hands off their hands while ensuring safety, so we are seeing the advent of L3-related accredited certifications. In June 2020, ECE issued a draft regulation on ALKS (Automatic Lane Keeping System), which was the first to be interpreted as an international regulation on L3 autonomous driving.
In fact, various countries are now formulating laws and regulations, and the Society of Automotive Engineers of China is also establishing relevant standards. But what exactly are these laws and regulations/standards binding on? If accidents in the automatic driving system are not allowed, then in reality, students who have obtained a driver's license through the assessment cannot completely avoid accidents.
The same goes for cars, we're not going to develop a car that's never going to happen, a car like that can't be developed at all. What we have to do is to define the bottom line, which will require self-driving cars to be safer than skilled and cautious human drivers.
So, how to define "security"?
At present, ISO PAS 21448 is expected to be functionally safe, and the scenarios faced by automatic driving are divided into four quadrants related to safety, namely known safety scenarios, known unsafe scenarios, unknown unsafe scenarios, and unknown security scenarios. Among them, especially unsafe scenarios, we need to pay special attention.
In a known unsafe scenario, if an autonomous driving system is less capable of handling risks than human drivers, then we consider this to be an unreasonable risk, that is, a defect.
ALKS creatively solves the critical testing problem of hazardous scenarios:
Based on the collision avoidance ability of skilled and prudent human drivers, "unreasonable risks" and "reasonable risks" are distinguished from the regulations;
Through the establishment of critical working conditions in dangerous scenarios, the "human rights" of the "electronic driver" of the self-driving car have been won, so that the self-driving car will not be too conservative, which is conducive to improving user satisfaction, improving traffic efficiency and reasonably controlling the risk of accidents.
The hardest thing to deal with is the unknown unsafe scenario because we can't predict the future. At present, the national testing center is also building a scene library, but the mileage used to collect road data is only about 30-50 million kilometers. In order to cover the long-tail scene as much as possible, everyone is now building a data closed loop through the shadow mode, and constantly iterating on the automatic driving algorithm.
So far, the development of cars by traditional car companies has basically ended at the SOP stage, but for smart cars, the reliability of the autonomous driving system is inseparable from the agile development based on the V-shaped development process of the scene library and the closed loop of user data. There are six main stages to go through:
Development, test evaluation scenario library, simulation platform, HiL, ViL test platform, test field test, road empirical test, continuous improvement in user use.
Finally, let's talk about driverless driving. From the perspective of mass production landing, driverless companies may not see the dawn until five years later, and even many people begin to transform to low-speed scenarios. In fact, low-speed scenarios have their own challenges and challenges.
Technically, we feel the magic of self-driving cars, and we also believe that its safety will be higher than that of humans, but just like human drivers need to take driver's licenses, the landing of unmanned driving also needs to wait for policies. In particular, Uber's self-driving car has had fatal accidents before, which has had a great impact on the industry.
In the major demonstration areas in China, there have never been reports of autonomous driving accidents. What does this mean? I guess it's too conservative. I have been in the automotive safety industry for more than twenty years, and I know that as long as the vehicle moves, there is a risk of accidents, even if the speed is zero, it does not mean absolute safety.
Fortunately, this year, the Beijing Yizhuang Demonstration Zone first shouted out the slogan of "Policy Pilot Zone" and increased support for the automatic driving industry.
On the other hand, if the security of bicycle intelligence is not high enough, then we will take the route of network connection.
According to DMV's disengagement report, the best players in 2019 also achieved an average distance of 18,000 miles; if humans don't intervene, it is equivalent to 18,000 miles of an accident, which is unacceptable to us.
China is now accelerating the layout of vehicle-road collaboration to support the landing of intelligent and connected vehicles through 5G communication technology. In the previous 4G era, it may take 500 milliseconds for the information of the road-end camera to be transmitted to the car, which is not enough to meet the needs of auxiliary autonomous vehicles; today, the delay of the 5G network can be compressed to tens of milliseconds, which can better support automatic driving at the roadside.
Moreover, most of the current self-driving cars still have safety personnel, and it is impossible to achieve truly driverless commercial operations. Based on the support of the 5G network, security officers have been able to remotely assess the operational capabilities of the planned routes of the operating vehicles to determine whether the operating vehicles can operate safely.
This year's domestic release of GB/T 40429-2021 "Automobile Driving Automation Classification" mentions remote drivers and dispatchers. Previously, Germany also amended the content of the Road Traffic Law on automatic driving, including that autonomous vehicles must be equipped with (remote) technical supervisors, etc., and the amendment has now entered into force.
There are many industrial institutions in this regard that have issued research reports, and I will not repeat them.
Cars replaced horse-drawn carriages 130 years ago, and perhaps for the foreseeable future, self-driving cars will also replace humans driving cars. Thank you!
Previous articles on the 4th Global Intelligent Driving Summit
The University of Macau must be loyal: "Conservative" can not promote the progress of the intelligent driving industry | The 4th Global Intelligent Driving Summit
2021-12-15
Meituan Xia Huaxia: "Unmanned delivery", the technical difficulty is not low| The 4th Global Intelligent Driving Summit
2021-12-17
AutoX Xiao Jianxiong: Unmanned, different routes but different destinations | The 4th Global Intelligent Driving Summit
Su Shuping, General Manager of Innoviz China: LiDAR is a tool for intelligent driving vehicles, not to "show off wealth"
2021-12-16
Guo Zhenyu of Ali Damo Academy: Technological progress does not equal commercial success | The 4th Global Intelligent Driving Summit
2021-12-20
Zhijia Technology Liu Li: The commercialization of unmanned driving will not be achieved overnight| The 4th Global Intelligent Driving Summit
2021-12-23
Moshi Intelligent Gao Wenbo: "OEM + supplier" system will coexist for a long time | The 4th Global Intelligent Driving Summit
2021-12-21
Sun Zuolei, partner of the unmanned driving business of Xijing Technology: The user cultivation period of commercial vehicle driverless has been basically completed | The 4th Global Intelligent Driving Summit
Gao Xinxing Wu Dongsheng: Vehicle-road coordination, how to improve the safety of automatic driving | The 4th Global Intelligent Driving Summit
2021-12-24
SenseTime: Efficiently solving the long-tail scene is the premise of improving the coverage of the intelligent driving scene | The 4th Global Intelligent Driving Summit
2021-12-28
3 questions, 7 CXOs, brainstorming around "autopilot" | The 4th Global Intelligent Driving Summit
2021-12-27
![Is L2+ Autonomous Driving Just a Gimmick? How to eliminate "Uncanny Valley"[fig]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)