How many companies have come up with the slogan of zero emissions, zero accidents, and zero casualties? Jaguar Land Rover, General Motors, Gaohe, Mercedes-Benz, and even Bosch, ZF and so on, I am afraid that I can't count on both hands.
However, Professor Zhu Xichan of the School of Automotive Of Tongji University has put forward a new point of view: we cannot pursue zero accidents.
To understand this point of view, we need to start with L2+ automatic driving.
At present, many car companies are saying that their products are equipped with L2+ automatic driving systems when they are spreading, mainly at level 2.5 and 2.9. Some people think that this is just a word game for car companies, after all, there is no such level as L2+ in the SAE J3016 driving automation rating released in May 2021.
As the industry's latest general guideline, SAE J3016 deliberately defines the difference between L2 and L3 levels in order to better define autonomous driving technologies that have been mass-produced and are being developed. The rating system names the L0 to L2 system "Driver Assistance System", while the L3 to L5 level is considered an "autonomous driving system".
So, is the L2+ autonomous driving currently prevalent in the industry really just a gimmick?
Professor Zhu Xizhan believes that L2+ automatic driving is actually L3 level automatic driving in the form of man-machine co-driving, which is completely different from L2 level automatic driving or ADAS driving assistance, and L2 level cannot grow into L3 through software iteration, and L2+ will eventually grow into L3.
He said: "L2+ hardware development is done in accordance with L3, due to the need for software iteration, the government's recognition and certification process is a bit lagging behind, and enterprises out of legal responsibility for automatic driving considerations, the current NP products can only be promoted according to L2+." ”
Professor Zhu Xichan pointed out that in the process of iteration from L2 to L3, although the company emphasizes the auxiliary driving system in the user manual, many drivers will leave the driving for a long time and use it as an automatic driving system. Such products have certain safety hazards, and the liability for accidents is not clear.
He called this safety hazard the "uncanny valley" because it doesn't have a high probability of happening, but it is likely to be fatal if it happens. The reason for the existence of "Uncanny Valley" is mainly because the user's expectations exceed the capabilities of the system.
How to eliminate "Uncanny Valley"?
Professor Zhu Xichan said: "At present, most manufacturers are constantly reminding users that they cannot leave driving for a long time, but this way will lead to a decline in user experience and satisfaction. So a better way is to quickly upgrade the ability of autonomous driving to meet the requirements of L3 level. ”
Professor Zhu Xichan pointed out that from the portrait of the L3 level highly autonomous driving car, there are several characteristics: a domain controller with high computing power, a taiwan network in the car, a high-precision map (supporting automatic driving functions), and an environmental perception sensor from millimeter wave radar to a high-definition camera.
He said that since the release of the SAE J3016 driving automation rating, L3 autonomous driving is the most controversial level.
For example, Volvo, which has always been safety-oriented, was the first to stand up and say that it would not do L3 level automatic driving, and they believed that in the case of human-machine co-driving, the responsibility could not be determined.
But this debate stopped after the ECE R157 came out last year, mercedes-Benz passed the approval of the German Federal Transport Administration, becoming the first company in the world to legally use L3 level autonomous driving in Germany, and its L3 level autonomous driving technology DRIVE PILOT will be the first to be equipped with S level and EQS.
Professor Zhu Xichan believes that ECE R157 not only shows that L3 level is certifiable, but its biggest contribution is to allow a self-driving car with a certain risk of residual accidents to be recognized and certified, but the probability of this residual accident risk is very low, safer than a skilled and cautious driver.
"ECE R157 defines the unreasonable risk of a self-driving car, as long as the level of risk is not higher than that of a skilled and prudent driver, then there is no unreasonable risk in the car." He said so.
Professor Zhu Xichan pointed out that ECE R157 has formulated critical working conditions for three typical dangerous scenarios on the highway, such as cutting in and out, encountering dangerous vehicles in front and emergency braking of vehicles in front.
Through critical working conditions, for these known dangerous scenarios, according to the obstacle avoidance ability of human drivers, the limit of the risk of self-driving cars participating in accidents is set. Then, by combining it with the black box, the problem of delineating the liability for accidents in self-driving cars can be effectively solved.
Some of the things that used to be debated have been settled with the advent of the ECE R157.
In Professor Zhu Xichan's view, the method of determining the critical working conditions of dangerous scenes is a topic that must be studied.
Because he hopes that when the critical working conditions are determined, the government will be able to recognize and certify a vehicle with a small risk of residual accidents, as long as it is safer than this skilled and cautious driver.
Professor Zhu Xichan stressed: "We can't pursue zero accidents of self-driving cars, and if we pursue zero accidents, this car will become too conservative, which will seriously affect the user experience and affect traffic efficiency." ”
He believes that L2-level driver assistance systems should pay equal attention to functional evaluation and safety assessment, but at present, it does not matter
Both NCAP and C-NCAP are too focused on safety assessment, and there is no evaluation of functions based on user satisfaction with autonomous driving. NoP, a type of product that positions L2+, is a product that attaches equal importance to driving experience and safety.
Of course, although it is impossible to pursue zero accidents, it is still necessary to pursue zero deaths by studying the zero-death safety strategy of the residual accident risk of automatic driving, Professor Zhu Xichan added.
In the past, when we talked about intelligence and automatic driving, most of them were aimed at zero stories, and car companies were also based on safety as the only premise when they carried out technology research and development, so there would be a so-called arms race, and even lidar could not be less than 4.
We all overlook one thing, security is indeed the most important, but too much, can not just think about security issues, but do not pay attention to the user experience. Therefore, Professor Zhu Xichan called on the industry to introduce a new evaluation system that attaches equal importance to functional evaluation and safety evaluation as soon as possible.