These days, there is no money to start a business without talking about AI.

Big companies engage in artificial intelligence, big company executives switch jobs and start businesses to engage in artificial intelligence, and investors invest in artificial intelligence, making artificial intelligence companies undoubtedly become the gold sink of current Internet companies. Let alone reliable companies, as long as they are reliable employees, they can get a large amount of venture capital, which is enough to spend a period of time in the tide of artificial intelligence.

But to be practical, the current research directions of these technology companies' artificial intelligence are actually three: deep learning, natural speech processing, and machine vision. Deep learning is a required course of artificial intelligence. Natural language processing can directly apply technology to virtual intelligent assistants, while machine vision can be directly transformed into automatic driving technology.

Smart speakers equipped with virtual voice assistants have been around for a long time. However, from the feedback of consumers, they do not think that smart voice assistants are intelligent things, and even smart speakers look like chicken ribs at present. Unlike smart speakers, which are hot in the market but cold in the consumer circle, automatic driving seems to be a promising project for both investors and ordinary consumers. Many people expect that automatic driving can improve the future traffic conditions and solve the current congested and inefficient traffic problems. The capital market also particularly favors automatic driving companies and automatic driving talents.

This morning, Momenta, an autonomous start-up, announced that it had received a round B investment of 46 million dollars. This round of financing was led by Weilai Capital, followed by Daimler Group (parent company of Mercedes Benz), Shunwei Capital, Innovation Works and Jiuhe Venture Capital. In 2016, Momenta obtained round A financing from Blue Lake Capital, Innovation Works and Zhenge Fund. At the beginning of 2017, we obtained the A1 round of financing led by Shunwei Capital.

Within one year of its establishment, the company has raised several rounds of funds, and 46 million yuan in round B is indeed a considerable investment, especially at present, the company has no external output.

Let's dig deep into this company.

Momenta, an autonomous driving company in Beijing, was established in September 2016. Cao Xudong, the founder, previously worked at Microsoft Research Asia and Shangtang Technology. The goal of this company is to build a brain for autonomous driving. Its core technologies are environment perception based on deep learning, high-precision maps, and driving decision-making algorithms. The products include different levels of automatic driving schemes and derived big data services.

In other words, Momenta is an autonomous driving solution company.

Compared with other companies, the technological advantage of Momenta lies in visual recognition. According to Cao Xudong, Momenta has the world's top depth learning experts, the author of Faster R-CNN and ResNet, the most advanced frameworks in the field of image recognition, and many champions in ImageNet 2015 and MS COCO Challenge 2015. The team comes from Tsinghua University, Massachusetts Institute of Technology, Microsoft Asia Research Institute, etc., and has profound technical accumulation and strong technical originality. Ren Shaoqing, the R&D director of the company, is the inventor of Faster RCNN, the most widely used object detection framework in the world.

Such advantages of visual recognition technology enable this company to obtain a relatively large advantage in the most basic perception part of the automatic driving scheme. For now, visual recognition will become the main solution of the automatic driving perception part. Compared with the laser radar widely used in the early days of the automatic driving car, the cost of the camera sensor is lower, and the data generated will be smaller than the data generated by the laser radar mapping the surrounding environment. It is a practical solution to achieve automatic driving or driving assistance.

However, the difficulty of visual recognition lies in the high-precision visual recognition algorithm. Only if the algorithm is optimized well, the on-board computer can recognize and process the images collected by the camera and divide the driving area, pedestrians, street lights and other road traffic elements.

And algorithm is exactly what Momenta is good at.

But to achieve automatic driving, it is impossible to have good perception alone. Decision making and control are the most critical technical modules of automatic driving. If we want to adjust the decision-making algorithm of automatic driving, we need a lot of actual road test data to optimize the algorithm, so that we can achieve the final intelligent and safe automatic driving control.

However, the accumulation of automatic driving data is a money burning job, which is difficult for small companies to do, so it requires huge financial and human support.

So we can look back and see where the $46 million financing will be used. Cao Xudong said: "The financing will be used in three aspects: 1. strengthening the core capabilities of AI, including big data, big computing and excellent AI talents; 2. productization of visual based environmental awareness and high-precision mapping technology; 3. developing L4 unmanned driving technology for high frequency and rigid demand scenarios."

At the current stage when all companies are accumulating autopilot technology, talents are the most expensive investment. Because all companies are attracting autopilot technicians, the income of talents in this field has risen. Many large companies are even willing to invest hundreds of millions of dollars in order to acquire small autopilot teams established by several technicians.

On the other hand, for self driving startups, capital will not allow long-term input and no output consumption. Everyone thinks that self driving can earn money in the future, but how much money it can earn and how to earn it when self driving is not popular are still difficult problems. Therefore, startups still need to gradually commercialize their technology to give investors enough confidence and company prospects, Can continue to invest in the development of their own technology. For an autopilot company like Momenta that is good at machine vision, The preliminary productization plan is likely to provide vehicle manufacturers with technical support for driving assistance to make profits (It can also be seen from the investors that Mercedes has also participated in the investment, which is also to make up for its own technical weakness in visual recognition of driving aids. It should be known that Mercedes Benz is not later than Tesla in the research and development of driving aids, but it is because visual recognition lags behind Tesla that Mercedes Benz's high-level driving aids seem to have been tepid in the consumer market.).

Of course, the ultimate goal of each company is to achieve L4 level fully automatic driving. The goal given by Momenta is limited to L4 driverless technology in high frequency and just needed scenes. In response to the traffic conditions in China, it is likely to be low-speed automatic driving in traffic jams.

At present, most startups are targeting L4 level full-automatic driving, which is particularly attractive to investors. However, automatic driving is not something that can be solved in a while. The year 2020 or 2021 predicted by major car manufacturers is definitely not a year in which automatic driving can be fully profitable. How long will the wind energy of artificial intelligence and automatic driving last? It depends on how patient the investment circle is.