We know that regarding the implementation of unmanned driving, it is mainly concentrated on two major technical routes.
One is the "radar" route used by most car companies, and the other is the "camera" route that Tesla must adhere to.
The reason for this persistence is that Musk's reason is quite the classical idea of what Chinese medicine eats and what to supplement, that "since people can distinguish everything by their eyes, they can also install eyes on the car."
The eye musk is talking about is the "camera", and the various radars are the "ears" that he believes are not advisable to draw snakes.
There seems to be no problem in logic, but in terms of technical difficulty, Musk's camera route is actually relatively difficult. Because the development of human science is still not very clear, how does the human eye recognize all things?
For an example that we find perfectly common, how do you recognize the person you are waiting for at a glance in a crowded pedestrian street?
You will say: nonsense, because I know him, I don't know others, my brain will definitely mobilize my familiar portraits.
Not really. We don't really know anyone.
A data generation technology
Because the principle of optics tells us that everyone's last second and the next second are different, even the photographer of the bull fork cannot take two identical photos.
"Because the brain has seen, we recognize", which is called "fixed rule algorithm" in the technical route, such as we want to identify a watermelon, fixed rule is to find the exact definition of the watermelon, and then write a program, so that the computer can identify any watermelon according to the fixed rule. This is more intuitive, because our brains can recognize someone or something by remembering certain rules and definitions.
But after nearly half a century of practice and research, scientists find themselves wrong.
Through in-depth study of the brain, they found that our sensory systems have been adapting themselves to the statistical laws of nature, which refer to the most important features in visual input for us. At the same time, the sensory system exerts great plasticity on these laws through hundreds of millions of neural connections.
Plasticity, on the other hand, refers to the brain's ability to reorganize connections between nerves. Plasticity not only allows the brain to repair from damage, but also allows the brain to allocate resources to the most important tasks of the moment.
Simply put, some cells are sensitive to the aspect ratio of a face, others are sensitive to the distance between the eyes, and some are more interested in the position parameters of the eyes. If we look at each parameter individually, none of them tells these cells very bluntly that "the face exists", but if these parameters are combined in some way, the cell can make a judgment as to whether the object in front of us is a face.
What is this algorithm called in the technical field?
You probably already have the answer – this is the hottest "machine learning algorithm" of all these years, known as "artificial intelligence." That is to say, the machine learning algorithm is more in line with the cognitive model of human brain vision, and its core concept is to build complex neural networks by imitating the working principle of the human brain, so as to gradually have human intelligence.
From the late 20th century to the present, with the development of data resources, that is, the sharp increase in sample size, and the powerful computing power of large-scale dedicated computers, the use of software to quickly identify faces is no longer a difficult task. Various applications of artificial intelligence, including speech recognition, intelligent translation, smart home, intelligent medical care, etc. have appeared in all aspects of our lives.
Machine learning algorithms do not need to be as artificially defined as fixed rules, also take watermelon as an example, the practice of machine learning algorithms is to provide a large number of various watermelon pictures, or provide some negative cases, such as strawberries, watermelon pictures, etc., so that the computer itself can find the law through learning.
This also tells us that if we want to achieve unmanned driving with only a few cameras instead of lidar, we must feed our own artificial intelligence with massive amounts of data to learn.
Well, that's where Tesla really doesn't get started.
The dispute over data is a dispute over sovereignty
Differences in technical routes are essentially differences in resource endowments.
We know that as the only two Internet powers in the world, there are still huge differences between Chinese enterprises and American enterprises. The population limit served by our Internet technology companies is basically about 1.4 billion, while the population limit served by Internet technology companies in the United States is about 4.5 billion (the number of people on the Internet in the world).
This gives Musk the confidence and feasibility of "bragging". We don't know exactly how much data will be eaten by artificial intelligence that achieves driverlessness through "vision", but it is clear that the more the merrier, the more the merrier, and 4.5 billion is definitely a priority number.
Volkswagen's CEO has expressed his "panic": there are now 500,000 Teslas around the world collecting receipts every day, which is something that other car companies simply cannot do.
Duoyunjun has said more than once that the reason why our country's electric vehicles have emerged is due to the huge accumulation of the Internet and smart phones in the past decade or so, which has prompted us to have far better resource endowments in this new era than Japan and Germany, which are traditional car powers.
Similarly, the United States also has such a strong resource endowment, and it is not worse than ours (in fact, except for the United States car companies can choose the "visual" route, car companies in other countries do not have such a resource endowment). Therefore, Multi-Cloud has also expressed that the future development of electric and intelligent vehicles only needs to focus on China and the United States.
Therefore, in the context of the times, in the past year, we can actually feel another subtle change, that is, China attaches more and more importance to "data sovereignty", from the rectification of Internet finance to the Didi incident, data has increasingly become a must for international games. The relevant departments have directly issued that "the overseas listing of companies with more than 1 million users must be subject to review", and at the same time require that "data do not leave the country", and the data collected by any company in China must be stored in China.
Whoever has more data has more initiative, and any company can soar with considerable data, and the same is true of today's national competition. The reason why Musk has publicly promoted the "visual" route on various occasions is essentially to persuade countries to open up data collection to themselves, and at the same time to instill in consumers the value that "data should be shared by the whole world".
Therefore, Tesla's "visual route" comes from its own data confidence, and our car companies mainly rely on radar routes, which is pragmatic and realistic about the current situation of data.
But the country's data protection will also determine that the assertion that "independent brands will become the mainstream of the Chinese market in the next three decades" will be true, and no car company will know more about Chinese data than independent car companies.
The change of the times in the automotive industry is a national war, competing for energy, wisdom, courage, timing, rare metals, and more on everyone's data.
-end-