"Tesla won't need a driver next year."

Musk recently made a high-profile announcement that he is confident that Tesla can reach a new height next year. Of course, this is not the first time he has made such high-profile propaganda. As early as 2015, Musk predicted that "within 2 years, Tesla will be able to achieve fully autonomous driving", but his last prediction has obviously failed.

Although Musk has always emphasized that Tesla's autopilot technology has reached the L4 level, the outside world generally believes that the actual level is also the L2 level - the assisted driving level. After a year, Tesla can really achieve the "triple jump". Coming to L5 level?

｜What is L5 level?

SAE (Society of Automotive Engineers) defines different levels of autonomous driving L1-L5 for the definition of driving standards.

L0 level: manual driving. According to the definition of SAE, this level is completely operated by the driver, including steering, braking, accelerator, etc., all of which are judged by the driver. The vehicle is only responsible for the execution of commands, so this level does not count Autopilot.

L1 level: It can assist the driver to complete certain driving tasks, such as the adaptive cruise (ACC) function installed in many models, and the radar controls the vehicle distance and vehicle acceleration and deceleration in real time. It is used in many domestic models.

Level 2: Partial automation, capable of automatically performing acceleration, deceleration and steering operations at the same time, which means that the adaptive cruise system and the lane keeping system can work at the same time.

L3 level: This level controls the vehicle through a more logical trip computer. The driver does not need to be on standby. The vehicle can complete the operation and driving independently in a specific environment, but the driver cannot sleep or rest. When artificial intelligence cannot accurately judge , still need manual operation.

L4 level: highly automated, which can realize the whole process of driving without the need for a driver, but there are also restrictions, such as limiting the speed of the vehicle to a certain value, and the driving area is relatively fixed. And after the realization of this level of automatic driving, the installation of brake and accelerator pedals can be eliminated.

Level 5: Fully automated, capable of fully adaptive driving, adapting to any driving environment. However, it involves legal, high-tech breakthroughs and other restrictions, which need to be further developed.

Level 5 autonomous driving technology, also known as "full autonomous driving technology". After the car reaches the L5 level, the driver does not need to stay in the cab at all, and the car can automatically complete a series of actions such as steering, lane change, overtaking, and U-turn.

This technology is built on a very important foundation - safety. If safety aside, all self-driving cars can now complete the above technologies. L5-level autopilot can reduce the probability of accidents to less than 1%, allowing the driver to release the steering wheel with confidence. How does Tesla do this? ? If Musk dares to say this, he may continue to be slapped in the face.

｜How far is "L3" to "L5"?

Undoubtedly, the popularization of autonomous driving technology will greatly improve traffic safety, traffic efficiency, reduce energy consumption and emissions, and profoundly change our future traffic and social models. However, as an emerging technology, autonomous driving requires a process from technology research and development to full commercial application. At present, L3 level and below driver assistance systems have been mass-produced, and some applications of L4 level in specific scenarios have also been gradually developed. However, the development idea for L5 level autonomous vehicles has not been clear. So how far is it from the L5 level, the ultimate stage of autonomous driving that we are most looking forward to?

Although many automakers have successfully developed L3-level autonomous driving technology, most of the autonomous driving technologies in practical application are still at the L2-level. So, to truly achieve L5 -level fully autonomous driving, what else is there? Technical problems need to be broken through?

One is the "deep perception" of complex traffic scenes

A self-driving car must detect road characteristics in all conditions, and it must respond safely in situations where weather and road conditions are constantly changing. In other words, autonomous driving must be an artificial intelligence system that cannot make mistakes. Because if it makes a mistake, it will cause a traffic accident, and the self-driving car must be able to perceive its surroundings carefully and reliably.

The second is to understand the "pre-action"

In addition, there is a difficult problem, which is to make the self-driving car accurately judge the expected behavior of surrounding objects or surrounding vehicles. The so-called pre-behavior is that before the surrounding objects or vehicles do not produce this behavior, the self-driving car can determine what changes will occur to surrounding vehicles or moving objects at the next moment. Human drivers communicate their driving intentions based on the pre-action of the vehicle ahead or the surrounding moving objects. For example, we are driving on the road and there is a car in front of us. Through the driving status of this car, we can judge whether the driver is a novice or a veteran, so that we can know what kind of driving should we do in the next moment. Action, there is a pre-judgment. But with current autonomous driving technology, it is difficult to explain subtle pre-actions.

The third is the response to "accidental encounters"

Human drivers can understand traffic scenarios based on body language and other contextual cues. For example, when we drive to a certain area, the traffic control is suddenly stopped. The human driver uses the traffic police’s gestures to make decisions about the next driving. However, it is difficult for the current automatic driving technology to understand such abnormal situations. In addition, there is such a situation, for example, there is a child on the side of the road, he is holding a toy in his hand, and suddenly the toy falls to the side of the road, he has to pick up the toy, in automatic driving, we have no way Code such an unexpected encounter in advance and write it into the algorithm of autonomous driving.

The fourth is the natural human-vehicle interaction

There is also how to create an environment where people interact naturally with self-driving cars, that is, when a passenger sits in a self-driving car, he needs to be able to interact with the self-driving car in his own language. Talk to the self-driving car on the way: "Where am I going?" "How much time do I have to reach the destination?" This natural communication can not only provide passengers with a comfortable experience, but also a safety certification. The self-driving car understands where the passenger is going and can answer it in words, so the passenger can feel at ease. Another point is that in the future, cars driven by human drivers and self-driving cars will definitely share the road. At this time, the interaction between cars and cars is also very important.

Fifth, network security

Self-driving cars need to update their maps through the cloud to obtain relevant data, and autonomous driving that obtains and updates maps through the cloud will face greater risks. It is conceivable that such a malicious behavior will occur in the future. For example, if a hacker invades a self-driving car, he can disable the self-driving car, and may also cause traffic accidents while driving.

｜L5 autonomous driving can’t be rushed

As mentioned just now, the SAE classification method does not have a clear application scenario, but most car companies transition from L2 to L4 in actual research and development to delineate autonomous driving functions according to scenarios.

As an innovation fighter, Musk has never followed the rules and regulations of those international institutions. Naturally, he does not take SAE's grading rules into consideration. Musk has taken a different approach, not mentioning the concept of grading, and directly promotes the scene. Musk may know that Tesla can't achieve the L3/L4 classification level, so he simply created another statement, creating Tesla's FSD is equal to the perception of automatic driving. Everyone from the industry to the people who eat melons mistakenly think that Tesla's FSD is a masterpiece.

Everyone knows that autonomous driving must be the trend of future development, but it is obvious that it is not realistic to realize this technology in the short term.

According to the investment logic, the expectations of each new technology will be infinitely exaggerated when it first appears, especially the expectations of autonomous driving. This is a technology that can completely change human beings. Everyone has very high expectations for autonomous driving. The logic of transformation is different. Electrification, especially power battery technology, is relatively mature. Enterprises solve problems from 1 to 100, while autonomous driving companies need to solve problems from 0 to 1.

An industry developing from 0 to 1 often has two bottlenecks, one is technology and the other is money. Autonomous driving is an expensive entrepreneurial competition, and the attitude of some financial investment institutions towards autonomous driving is quietly changing: At present, financial investment institutions are at the point of transition from full support to attitude differentiation.