Yesterday, China Automobile Center, Tongji University, and Baidu jointly released the "White Paper on Traffic Safety of Autonomous Vehicles", which summarized that there are actually two main sentences: human driving is not safe; automatic driving is more adaptable to the traffic environment than people.
On the matter of intelligent driving than manual driving, technology companies and car companies have spent a lot of words, Tesla's quarterly financial reports, will be attached to the "Millions of miles driven between collisions" data, to prove that the use of AP will be more than 8 times more secure than the overall data of the whole people.
The White Paper on Traffic Safety of Autonomous Vehicles (hereinafter referred to as the White Paper) screens 6967 accidents caused by passenger cars from 2011 to 2021 through the road traffic database CIDAS, and finds that 81.5% of the passenger car accidents are caused by human factors of the driver.
Of these, about 79.9% are caused by subjective errors by drivers, and about 20.1% are due to limited driving ability.
There are many kinds of subjective errors, of which 43.4% are not given way according to regulations, 9.7% are too fast, in addition to traffic lights violations, drunk driving, fatigue driving, etc.
The accident caused by the driver's limited ability refers to the accident caused by the driver's own insufficient ability or limited by environmental conditions, and the behavior of other traffic participants is not fully observed, including encountering limited vision scenarios such as high beams, rain and fog weather, and blind spots of vision, including the untimely response to sudden obstacles.
L4 automatic driving data data, only because China's open test area is very scattered, dozens of cities across the country have cars running, the lack of a unified statistics, so there is no public data can be used. However, refer to the data of the DmV in California, North America for a general understanding.
In dmv California's self-driving accident collision column, there have been 89 autonomous vehicle traffic accidents so far in 2021. The number of Robotaxi fleets in California is comparable to the national total, so speculation is that the domestic fleet is probably in the range of dozens of cases per year.
Of course, because the current volume of L4 automatic driving is not large, and it is all within the limited road, there is little data and data, and the comparison significance is not large.
The white paper is mainly from the three perspectives of perception, decision-making and control, demonstrating that the machine has great superiority over people, and is more competent than people in complex traffic environments to achieve safe driving.
First of all, in terms of perception, the white paper synthesizes data from the CIDAS database to pull out the distribution of pedestrian centers of gravity in the detection range of 30°, 45° and 60° for human drivers and autonomous vehicles in the 1.0 seconds before the collision. The conclusion is that the obstacles detected by the autonomous driving sensors are richer and more comprehensive.
This is only bicycle intelligence, if you take into account V2X, intelligent driving car to obtain the amount of information for decision-making, in fact, is much greater than human, the driving intention of other traffic participants, can reduce traffic conflicts. The white paper estimates that when the penetration rate of self-driving cars reaches 25%, 50%, 75%, and 100%, traffic conflicts will be reduced by 12%-47%, 50%-80%, 82%-92%, and 90%-94%, respectively.
Richer perceptual data means that the path planning of the machine is more reasonable than the path planning of the human, and the correct decisions can be made. In addition, people's personalities are very different, and even if they are the same person, the driving decision-making style is different at different times and in different emotions, such as driving a bucket car. Relatively speaking, the machine without emotion can more stably do the role of cautious "old driver".
In terms of execution, the human eye sees the brain processing, and then to the hands and feet to operate the steering wheel accelerator pedal, the entire link is relatively long, there is a certain delay. Relatively speaking, intelligent driving is the transmission of electrical signals in the system, which is faster and easier to control the vehicle to make operations such as sharp braking in milliseconds to avoid accidents.
The white paper basically and reasonably explains the advantages of machine driving over all-round crushing of human driving. But even so, can consumers really trust autonomous driving?
At present, a considerable proportion of people cannot use self-driving cars with confidence, and even if it is a very basic L2 level assisted driving function, many people have never used it after buying a car. On the one hand, the technology is indeed not mature at this stage, and most assisted drivers are not clear enough in visualization. On the other hand, the experience of assisted driving function is not very good and cannot be reassured, which will also affect the judgment of L4 autonomous vehicles.
It has become a consensus that autonomous driving will be implemented according to scenarios, and in the next 3 to 5 years, people will be more and more exposed to the experience and services of intelligent driving. While expanding the scene of functional progress, the cultural level allows consumers to truly eliminate the fear of automatic driving, which is also the direction that needs to be worked on.