What is the first nature of autonomous driving? Can we achieve the “mission must be fulfilled”?

Autonomous driving is one of the most talked-about topics in the global technology industry today. Regarding the irreversibility of this emerging technology of autonomous driving, in the previous series of reports, we have demonstrated from multiple dimensions such as industrial laws, physics, economics, and intelligence principles.

From the perspective of international comparison of industrial development, this competition for the commanding heights of intelligent technology has been fully rolled out in the fields of China, the United States and Europe in the field of automobile and technology camps. Chinese autonomous driving manufacturers and intelligent driving solution manufacturers represented by Baidu and Huawei are building one of the strongest clusters to promote the evolution and popularization of this cutting-edge technology.

At that time, from the downtown area of ​​Yizhuang in Beijing, to the Tianjin Port Terminal, and then to the Luoyang Luoyang Molybdenum Mine mining area, autonomous driving has initially demonstrated the effectiveness of its exponentially advanced productivity. At the National “Thirteenth Five-Year Plan” Science and Technology Innovation Achievement Exhibition that just ended on October 27, many autonomous driving R&D and implementation achievements were unveiled, giving more people a clearer understanding and confidence in autonomous driving.

The future is bright, but just as every new thing will encounter various doubts when it comes, for autonomous driving, its safety has always been a topic of greatest concern to all sectors of society. Especially in the recent period, some safety accidents caused by the implementation of automatic driving in the name of assisted driving have made this topic even more hot, and it has also caused the automatic driving to bear undue misunderstandings and prejudices.

There is no safety guarantee, not enough to say automatic driving. As for the safety of autonomous driving, it is actually time to put it on the table and talk about it.

01 The first nature of autonomous driving: safety

Yes, the first priority of autonomous driving is by no means commercial interests, but safety.

Different from the logic that cars are to carriages, high-speed rails are to Pulie, planes are to ground transportation equipment, and space shuttles are to civilian aircraft, the technology of autonomous driving for the automotive industry is not about breaking through physical space constraints, making People efficiently explore the unknown space, but it lies in maximizing (optimizing) the safety of driving, operation and transportation system per unit mileage, followed by efficiency, which leads to new forms of life and business.

Autonomous driving is a natural extension of AI technology. It is based on the engineering coupling of software and hardware, and based on the high rationality of computing power and algorithms relative to human functions. It realizes the analog replacement of human perception, operation and other systems, with exponential reliability. Push the safety factor of vehicle driving to 2.0, 4.0, 8.0, 16.0 … era.

Specifically, according to statistics, on average, human drivers will have one non-fatal collision every 300,000 kilometers. That is to say, as long as the self-driving technology is twice as safe as human drivers, it is equivalent to an exponential improvement, which means that the human driving safety factor has reached 2.0, a milestone evolution.

So can autonomous driving achieve “mission must be achieved”? From the realization of the situation, we can be optimistic:

Taking Tesla’s intelligent assisted driving system Autopilot, which is equivalent to L2.5-L3 level automatic driving capability, as an example, according to its “Vehicle Safety Report” released in Q2 of 2021, it shows that when Autopilot is turned on, it occurs once every 7.1 million kilometers on average. Accident. When Autopilot is not turned on, an accident occurs on an average of 1.93 million kilometers (Note: Tesla vehicles currently account for a relatively small proportion of the car ownership in the United States, so the accident probability data is relatively low).

According to this comparison, the intelligent driving assistance system that can achieve at most L3 level of automatic driving capability is relatively 30 times more secure than manpower. Of course, it must be pointed out that the limitation of the number of Tesla vehicles, This data is greatly distorted.

But in any case, the ability of autonomous driving to simulate the replacement of manpower has already shown its advantages in the first dimension of safety.

The data from the leader of China’s autonomous driving force also supports this point: Baidu is the only autonomous driving company in China with a real road test mileage of tens of millions of kilometers. It reaches 99.99%, and the overall safety is higher than that of human drivers.

In addition, from an engineering point of view, this expectation is also logical: the motion patterns of a car’s four wheels, braking system and steering system are basically all predictable and simulated, controlled by the autonomous driving software system. The number of action combinations is actually a relatively small value. Another reason driving can be automated is that driving is a relatively rigid and repetitive reactive activity—as evidenced by the fact that a driver’s license can be obtained regardless of intelligence.

In fact, from the perspective of ordinary people, it is not difficult to understand this logic:

According to the World Health Organization, car accidents are the highest cause of premature death among young people aged 15-29 globally, and the second most important cause of death in all age groups of car owners. Most of these crashes are not caused by a malfunctioning part of the car, but by human factors that could have been avoided, or the “4Ds” — drunkenness, drug use, drowsiness, and distraction. As long as the steering wheel remains in human hands, fatalities from car crashes are likely to continue to rise.

In addition, there is another risk factor for human driving cars – distracted driving continues to grow. Look at the drivers around you, phones, communication software, text reading and short videos are invading their driving space.

In contrast, in a sense, it can be said that the cost of questioning and delaying the development of autonomous driving technology is almost calculated in the number of human lives.

Of course, autonomous driving definitely does not mean zero accidents. Any imagination with such rigid expectations is essentially ignoring the laws of physics—there is no perpetual motion machine in this world, which means that there is no perfect energy system.

From the perspective of practical obstacles, like human driving, autonomous driving is confronted with extreme road conditions (corner cases). This randomness corresponds to the uncertainty principle of quantum theory. In the face of extreme road conditions, the responsiveness and operational rationality of autonomous driving are theoretically exponentially better than those of manpower; from an engineering point of view, no AI technology, including autonomous driving, has 100% robustness, but From the expectations of the entire industry, autonomous driving will eventually achieve a safety limit of 99.99999%. And such scientific and technological progress is already civilization-level scientific and technological progress.

02The Chinese model of autonomous driving is more worth looking forward to

Autonomous driving cannot be achieved overnight. So far, its intelligence is still relatively immature compared to the unconscious ability of human beings to survive. Therefore, it is the consensus of all current autonomous driving manufacturers to face up to its disadvantages, rationally use and optimize its advantages.

The growth path of some overseas manufacturers who value bicycle intelligence and verify and improve autonomous driving technology at any cost. The inherent oriental culture of the Chinese actually makes the Chinese competitors of the same factory have higher requirements for the fault tolerance rate of autonomous driving safety. high.

For example, Tesla adopts a single-vehicle intelligent automatic driving mode, which mainly relies on the vehicle’s own vision, millimeter-wave radar, lidar and other sensors, computing units, and wire-controlled systems for environmental perception, computing decision-making and control execution; Baidu chose The vehicle-road collaborative autonomous driving mode, based on the intelligent autonomous driving of a single vehicle, organically links the “people-vehicle-road-cloud” traffic participation elements through the Internet of Vehicles, helping the autonomous driving vehicle to perceive the environment and calculate the decision-making. and control the execution and other aspects of the ability to upgrade.

Specifically, the Apollo Moon, the latest generation of self-driving mass-produced cars from Baidu Carrot Run, has two lidars installed in front of it. It should be noted that only one is running under normal driving conditions, and the other is a redundant configuration. Step in when the former fails or goes wrong. In addition to lidar, including cameras, millimeter waves and even brakes, the idea of ​​Baidu Apollo Robotaxi is a 1-fold redundant configuration. On the one hand, it is more concise, and on the other hand, it is necessary to eliminate all unnecessary security as much as possible in the early stage of commercialization. doubt.

In addition to all the software and hardware redundancy, the safety guarantee system for Apollo Robotaxi is “5G cloud driving”, that is, through intelligent cloud monitoring, once there is a vehicle problem, after receiving help, the manual service system in the operation headquarters It will start immediately to fill the position of the driverless system until it gets out of trouble. The addition of this business strategy has basically solved the “last mile” problem of the L4-level Robotaxi commercialization.

The Baidu model is actually a microcosm of the Chinese model of autonomous driving: a highly secure and redundant configuration. To sum up, it includes not only the sensor redundancy and safety redundancy at the vehicle end, but also the redundancy in the operation process (5G cloud driving), and also includes the road-end infrastructure that China is advancing at a high speed – the improvement brought by vehicle-road collaboration. High dimensional redundancy.

For the comprehensive promotion of such a Chinese model, the entire Chinese policy and the industry and economics are actually a game of chess: In February 2020, the “Intelligent Vehicle Innovation and Development Strategy” issued by the National Development and Reform Commission and 11 departments clearly pointed out that before 2025, China’s standard intelligent vehicle The technological innovation, infrastructure, regulations and standards, product supervision and network security systems should be basically formed, and the L3-level autonomous driving smart cars should be mass-produced to realize the application of L3-level autonomous vehicles in relevant scenarios.

If so, the Chinese model is more worth looking forward to in dealing with the issue of the first priority of autonomous driving—safety.

Fear often comes from “ignorance”, and even more from “ignorance and fearlessness”. To deal with fear, only scientific rigor and expected reality can guide us step by step towards a higher technological civilization. The future is there, and the Chinese autonomous driving force still needs to go a long way. As the public, we must be realistic and optimistic.

The Links:   C070VW02-V0 MG75J1ZS50

Related Posts