Meng Xing of Shunwei Capital: Self-Driving Car Technology Forms Great Opportunity for Tech Entrepreneurship, with Highest Ceiling Being Image Sensing|An interview with AI Drive Capital
Author: Wu Dexin 2017–03–10 15:02
Lead: The building of an entire self-driving car system involves a myriad of parts and technologies from the sensor to sensing technology, to planning, and then to mapping and controls, etc. In my opinion, the ceiling is probably with sensing technology.

Notes from leiphone: Meng Xing is currently vice-president and Entrepreneur-in-Residence at Shunwei Capital. Prior to joining Shunwei Capital, Meng was the founder of one company and then CEO of another in the field of artificial intelligence. Orbeus, a company that he had founded in Silicon Valley, makes use of deep learning to provide developers with image and facial recognition capabilities. On the other hand, Cogtu Tech seeks to enhance internet advertising with the aid of image recognition technologies. Cogtu has been acquired subsequently by Sina Weibo.
After joining Shunwei Capital, Meng has been mainly responsible for the surveying of tech-driven companies. Artificial intelligence and self-driving car technologies have been the foci in this regard. To date, Shunwei Capital has already entered the fields of mapping and new-energy vehicles with its investments in Careland and NEXTEV. Meng has disclosed to AI Drive at Leifeng.com that in the last year, Shunwei Capital has also made investments in a self-driving car tech company and a company working on laser radar technologies.
Let us take a look at what Meng Xing, who has transferred from entrepreneur in the field of AI to the field of venture capital, has to say about self-driving car technologies.
#AI Drive Capital Interview# See what the foremost sector experts in capital say about self-driving car technologies. AI Drive at Leifeng.com (Official account: “leiphone-sz”) welcomes discussions with you on this topic at our official account “AI-Drive”. Understand future cars, and the future of cars with AI Drive at Leifeng.com.
AI Drive: What are Shunwei Capital’s investments in the area of self-driving car technologies like today? What are the projects that are involved?
Meng Xing: Currently, we have invested in companies on the mapping level, the automobile assembly level, and on the components level.
I have been with Shunwei Capital for a year or so, and here I have mainly been looking at tech-driven companies. Artificial intelligence is a key area of focus. I feel that the automobile and self-driving car industry is a bit special, and also quite interesting. The field applies technologies like machine learning or computer vision to what is a universalist robotic application scenario.
The investment cycle for self-driving technology all the way to the complete automobile takes a very long time. However, if the project succeeds, the market would be massive. And this is a very clear direction.
If the start-up were to choose to integrate machine vision with everyday applications, for instance in finance or security, they would need to surmount severe commercial challenges. And for the majority of start-up teams, they would find it very difficult to create products that are ten times better than the existing products.
As such, it is not very possible for them to bypass sales or promotional channels and obtain victory with “steamroller” products.
With automobiles, once fully-automatic self-driving has been achieved, the driving experience would be ten times better than what it is now. The upstream and downstream industry chain for the automobile is relatively mature, and self-driving car start-ups can place technology and the product at the core. Even when the start-up’s business capabilities are just so-so it would still be possible for an automaker to get in touch with it.
Hence, in my opinion taking this approach would be an advantage for tech start-ups. For a relatively long time, the start-ups would be able to focus on its technologies without having to consider many other things.
AI Drive: What does Shunwei Capital think about investment opportunities in the area of self-driving technologies?
Meng Xing: The entire market for self-driving cars will be a very, very large one. The automaker will evolve in the course of its development. It would not hope to evolve from OEM manufacturer to “pipeline company”. The operations company will also grow, and eventually the two may converge.
This way, automobile use may increase and the rate at which the market purchases new vehicles may slow dramatically.
Let us make a bold hypothesis here: we can regard the experience provided by Didi Chuxing — whether it is with a driver or without — as a kind of “unmanned driving” (Chinese term for “self-driving”).
I feel that such “unmanned driving” experiences are not too bad, but a little pricey. If the cost of each trip within Beijing, regardless of distance traveled, were to be capped at CNY5 or CNY10 (let’s say this really happens), what kind of impact would it have on the entire driving market?
In my opinion, if we can hit the CNY5 level, that may take 50% of all cars off the roads. At this juncture, the pleasure that an automobile can give to a car enthusiast or its collectible value may exceed its actual use value.
Is there the possibility that a company working on self-driving car technologies would end up completely dominating the market like Mobileye has?
From a purely technical point of view, in terms of ADAS, Mobileye has essentially won 80–90% of the high-end automobile market. Can a similar company emerge in the self-driving car market?
If the possibility is there, then the risk of getting the wrong pick is very high for VC players. Personally, I feel that it would not be possible for this sort of company to emerge again.
First of all, automakers have very strong motivations for preventing the emergence of a company like this. The value of a self-driving car solutions is not the same as that of ADAS. Today, ADAS is a supplementary product. Assuming that a car is worth CNY100 today, CNY85 of this amount would be accounted for by the engine, the transmission, and the interior decorations, etc. Even if the car were to be equipped with ADAS, it would only account for the remaining 15%.
If fully-automatic self-driving technology were to appear, then 80% of the car’s value would be expressed in terms of whether it is capable of automatically entering and exiting the expressway and refuel without manual intervention, and how long it is capable of staying in cruise control or standby mode.
In the future, the way we use cars may be similar to the case of Mobike or ofo: when it becomes common and cheap enough, we may not pay too much attention to things like whether the bicycle comes with a fixed gear or with multiple gears.
As such, automakers will not allow a self-driving car tech company to dominate this sort of change, or at least they would put up very strong obstacles.
In my opinion, the development of self-driving car technologies will eventually lead to a one-to-one relationship with automakers. That is, each automaker will align itself with a technology supplier, and then align itself with emerging companies that can provide it with technologies and algorithms in the corresponding areas.
Conversely speaking, from the companies that have first achieved self-driving in the United States we see that companies that seek to become common platforms may lead in terms of technology but face multiple challenges in going commercial. An example would be Google. If self-driving car solutions are to become the Android of the automobile world, this would be way too sensitive for automakers.
In summary: there will be many independent self-driving car tech companies catering to their respective clients. However, how many (first tier) automakers there are, there will be as many (more or less) self-driving car tech companies. Hence, from the investment perspective, you don’t have to invest in the best company. If you invest in one of the top ten companies globally, generally speaking, it will be a very good company as it is.
Now, self-driving car tech companies both in the U.S. and in China can raise around USD10 million in the first round and they can be valued at USD50 million to USD100 million.
More shockingly, when the company shows its demo, typically it would raise a massive sum of money in the second round, sometimes between USD30 million and USD50 million or even more. I feel that if a number of companies reach this round, the capital market may not really have so much funds.
Hence, companies that have been able to obtain so much funding early on have it good. They will have an advantage compared to companies that have managed to obtain funding at a later time window.
AI Drive: Which self-driving car projects does Shunwei Capital think will do well?
Meng Xing: We are beginning to see differentiation among companies that work on self-driving car systems:
1. There are those that make it in one step to L4 with a demo at every intermediate stage. Here, commercialization is not a great concern over the course of the process;
2. Some teams have already established L4 as a target, but they introduce a product at every stage such as digital training platforms, digital calibration platforms, and map operation tools. With each product, there is the possibility of becoming the upstream counterpart for other self-driving car companies or for other automakers. Some of these intermediate products may form a very large industry chain, and in the short term may generate more solutions than self-driving.
3. There are also teams that have, for the short term, sought an entryway through low-speed cars. Personally, I feel that working on low-speed cars does not really help the work on high-speed cars, for the technical pathways involved are completely different. However, low-speed cars do have their value. The present market of existing stock is actually massive. Together, the demand for low-speed cars by various industry parks and logistics facilities may amount to billions of yuan in value.
Of the aforementioned few types of companies, those that have chosen to get to L4 in one single step bear the greatest risk. This may require one or two hundred million US dollars and five years of work. Assume that the company has met all of its market milestones for this year. However, it is difficult to predict how next year’s market will go. This sort of risk stacks with each year.
Companies that have “landed” in the intermediate stages have pioneered a very interesting market. Their actions may be even more aligned with the rhythm of the general environment, that is, at a speed that is acceptable to their clients and the government. The customer may not require L4 self-driving systems today. Perhaps the company would first build a system, and look at how to apply the data collected by data vehicles into high-precision maps. Of course, this is not an easy task.
On the flip side, take for instance Google’s unmanned car project that was established in 2008 or 2009. Two years later, Google began to run a demo on city roads with barely any human intervention. In the subsequent few years, Google’s unmanned car project resolved another few core issues and extreme situations.
For instance, there is an inordinate number of willow trees in a particular area, and when the catkins land on a particular part, a new reaction is created. And also previously I have seen how in Google’s sensing solution the sensor is used in certain ways to solve certain problems. Thus, the car may not work if switching to another hardware calibration scheme. After that demo, Google went on to resolve more than a thousand similar issues. Today, the project is close to the state of refined operations.
To take a few steps back: in the area of self-driving technologies, first, there are the early movers. Those that are near the second round of funding have an advantage. At the same time, I feel that it is easier for companies to “land” along the way, whether it is with low-speed cars or with data training platforms.
AI Drive: For instance, in the areas of ADAS and self-driving technologies today, many companies have begun with imaging and computer vision. You have also had experience in this area. In your opinion, what are the strengths and weaknesses of such companies?
Meng Xing: The building of an entire self-driving car system, involving parts and technologies from the sensor to sensing technology, to planning, and then to mapping and controls, etc. In my opinion, the ceiling is with sensing technology, where continuous innovation is most possible. Those who have come from vision teams are strongest in this area, and have the greatest room for continual enhancement.
As for weaknesses, I feel that teams that have come from computer vision may not have enough respect for the automobile. They may be more inclined to use vision to solve all issues, or to build a common solution for all problems.
AI Drive: Today, Shunwei Capital has already invested in companies that work on laser radar and self-driving technologies. What other kinds of projects are you looking at?
Meng Xing: Previously, I had paid more attention to companies that focused on overall self-driving solutions. The cycle for self-driving automobiles is very long. Out of several possible pathways, there may be some that would be proven false in the next few years. However, the overall market has always been there and growing.
At the same time, I feel that on the level of services, there may be new opportunities after self-driving has been realized, for example, with app products, including full-screen-display apps together with technologies like image recognition and AR. Will these mature at the same time as self-driving tech? When self-driving tech is realized, there will be little difference from one car to another. Then, will these services become an important point of differentiation for everyone?
In addition, in my opinion, I feel that certain new radar or sensing technologies may emerge now and create new changes to the situation.
There are already many companies that are working on data mining and SLAM algorithms using computer vision. However, there are few companies that perform deep data mining using laser radar point clouds. Basically, this is only done by the radar companies. However, algorithms are not the strongest suit of these companies. Or, some teams are working on integrating computer vision with point cloud algorithms, but are not realizing the full potential of such an approach.
Thus, the question is: is there a possibility of teams that can truly create radar algorithms emerging?
When laser radar technologies become better and better, and are no longer limited to self-driving applications but in multiple industries, would there be a possibility of making use of improvements in the area of laser radar algorithms to render the sensor as a standard aid apart from the usual RGB camera? This will reversely lead to the prosperity of laser radar technologies in the area of self-driving.
