I’m Danny. I talk about cars and tech. At the end of the last episode of DannyPal’s video, Emma asked David an interesting and sensitive question: David has analyzed the various development paths of Google, Tesla, Didi and others, and said that the traffic situation in China is more complicated compared to the U.S. So David, as a front-line practitioner in the autonomous driving field, how are you going to develop the autonomous driving tech in China? As the host of the show, I’m looking forward to seeing how David will answer Emma’s question. Let’s go watch.
So what’s your plan as a practitioner?
That’s good question。我们实际上还是要针对中国这样的模式，把城市里面的大部分问题解决掉，同时再把这样的技术出口或者说转化给新兴城市。就是说如果中国有某个城市，从交通灯到人行线，都跟你自动驾驶，为了你而改变，那就比较好。但是我们不会奔着这个方向去做方案，因为这样就相当于直接把方案阉割了一部分。我们还是会奔着在没有任何外力的情况下，这个单车智能是怎么样的，奔着这个方向去的。
In fact, we still need to solve most of the problems in the cities of China. Then at the same time, we can export or transform this technology to the emerging cities. That is to say, if there is a city in China, let’s say, from traffic lights to sidewalks can all be changed for the autonomous driving. We’re not going to run to this approach because it would cut off a part of the solution. We’re still going to work on the one-car intelligence without any external force. We’re heading to this direction.
I understand that. I want to ask why this is the trend all over the world. For example, there will be many self-driving companies in Silicon Valley in the United States. It becomes a storm center for autonomous driving. Then in China, Guangdong is that type of center. Why are these two places standing out?
People in Beijing are not happy anymore. Beijing says there are still many companies.
I understand the question you are trying to ask; you mean, why there are industrial clusters? This industrial cluster can be in any industry. Let’s take the United States as an example of why all financial hedge funds are on Wall Street or somewhere in Connecticut. Why filming in LA? Why do the offshore oil gather in Houston?
Why many awesome people are at DannyData, right?
Why there are the industrial clusters? That has something to do with the people in those cities. However, in China, these companies focus on autonomous driving from those leading companies like Baidu. It depends on where these companies’ employees stay; then, we will know the location of those companies. Not quite the same as the United States. In the United States, those companies did different things are in different cities. The companies which are building planes are in Seattle, right? The companies are working on offshore oil in Houston. China is not like that; most of China famous companies are concentrated in these cities. And these cities are made up of these original big state-owned enterprises or private enterprises like BAT. Those people who come out from those companies, then they try to do something new. That industry will take location in the place where are these people stay.
When we talked in private, you mentioned a particularly interesting point. The United States makes a place a representative place of an industry. For example, when we talk about movies, we think of Hollywood. Wall Street is linked to finance. Speaking of technology, we will think of Silicon Valley. This is indeed very interesting. China actually has this similar model because of the United States. For example, we have Zhongguancun etc..
I think the Huaqiangbei business district in Shenzhen is representative. When it comes to starting with electronic components, it is easy for everyone to think of Huaqiangbei. Although the place of Huaqiangbei may have stagnated in Shenzhen now, just like Zhongguancun Hailong and Dinghao Building are not selling products now, the spillover effect of the cluster effect is strong. Putian is another matter. Putian is a group of people who came out of the village. However, we are concentrated into a village, Zhongguancun.
The breakdown of the automated driving module
David，刚才我们聊了很多自动驾驶，但是我看了你之前提供给我的资料，你其实对自动驾驶，无论是软件还是硬件方面，都是非常硬核的，包括我看你把比较头部的自动驾驶模块都拆解出来了，包括Waymo、 Uber ATG还有Cruise的硬件。David你能不能给我们大概介绍一下，这几家公司的特别之处？
Just now we talked a lot about automated driving, but according to the information you provided to me before, you actually have a deep understanding of the software and hardware of automated driving, including that you give us a breakdown of these head automated driving modules to us. David has analyzed Waymo, Uber ATG and these Cruise hardware. David, can you tell us something about the special features of these companies?
In fact, these companies are based on multi-sensor fusion. Multi-sensor fusion is, for example, the blind grandpa needs the deaf grandma to cross the road. Multi-sensor fusion is actually using the advantages of different sensors under different characteristics. For example, millimeter-wave radar has a relatively accurate judgment on speed, but its judgment on the identity of objects is weaker than that of vision. There are many defects in vision. For the specific coordinates in the three-dimensional space, the visual control of them is more fuzzy than the lidar. So we used lidar to get its specific location with a point cloud.
So every sensor of every company has its advantages and disadvantages in physical properties and identifiable range. Then you combine it together. Basically, each performance index is better in terms of its own. Then you combine it together, whether it is long distance, close distance, visibility or other circumstances, basically the combination is better than each alone on every performance index. Range from Waymo, Cruise, Uber ATG, Baidu, and these domestic companies basically use multi-sensor fusion.
Then in the multi-sensor fusion, including multiple lidars, a main lidar is responsible for many things. Each company has different technical differences in this regard. If you see Cruise’s car, you will find that there are lidars.
Cruise is the project of GM right?
Right, this company got GM’s investment. It is actually a splicing using multiple mechanical lidar and vision plus millimeter wave radar. Then one of Waymo’s main sensors corresponding to it is above the center axis. All these sensors are on the bottom bracket. There is a supplemental radar on the side as an aid. There is also a car made by neural AI, a start-up company in the Bay Area, which actually translated the entire L4 solution to his car. This car is not simple. Let’s take a look at his roof. We can understand this car in the same way as an unmanned taxi.
A lot of comments say that David‘s examples are very vivid.
The deaf grandpa met the blind grandma.
Two people holding hands across the road are like millimeter wave radar and lidar, and a few people are holding hands to drive automated driving car.
Do you speak the same way to investors about this? I’ve heard from my friends that investors in Silicon Valley actually know a lot about the industry, but maybe some investors in China don’t necessarily know so much about autonomous driving. Can they understand what you mean?
Well, the investors are my father, right? I can’t offend dad. Here’s the thing, what they prefer to hear is a story. Of course every investor likes to hear a story, and I’ll tell you what story they prefer to hear, which is to say how big is the unmanned taxi industry in China. Let’s say it’s 1 trillion, I can be very confident saying that I’ll take 1% of it, and that’s 10 billion. Once the investors hear this, 1% of 1 trillion is also 10 billion, it’s a 10 billion project, this can be invested. So, at this time you don’t need to tell him about any particularly in-depth technology and the time synchronizing space. They cannot listen to more than that, you know? They have fallen on deaf ears on this issue.
But if you’re speaking to an investor in this particularly hard-core technology, they’re going to ask you again how you’re going to realize this technology and make profits. And they’re going from the basis to ask you a question. So one is standing on the top with no feet on the ground, and the other is standing on the bottom asking you to tell them tomorrow how this turns into money. These two questions come from quite different places.
If I have to answer, I’d prefer to answer the second question, which talks about how does this technology monetize? So actually, for our company, we want to put the technology into a full stack package and give access to the auto-taxi companies. To give them a so-called big hat, audience can understand a hat which makes a car to achieve the automated driving. Our work is to figure out this solution, rather than the operation or anything else related to the vehicle. This is a little easier to realize. To put it bluntly, we’ll make the money from the future part of people’s imagination. To make money from the hardware first, and I won’t tell you any long-term stories. I won’t say anything like “I’m 1% of the 1 trillion market.” Let’s just talk about this current bunch of sensors plus my algorithm, kind of like Tier 1, but not quite the same as Tier 1. Because Tier 1 is dealing with the car companies and we’re dealing with these operating companies that have taxis.
So there’s a question from our audience, what will the relationship be like between automated driving companies and car OEMs in the future?
It won’t be very close since it’s a business relationship. The car OEMs are still working as a computer assembly company according to my understanding. How can an assembly company study every single component? That’s unlikely.
那么自动驾驶公司实际上就是把方案拆分。比如说我现在手里有一个很成熟的方案，叫L4，我把它切成8段，从L2开始一点点喂给你。其实主机厂也很开心，这样他在做广告的时候不用一次all in。如果一次all in，这个客户可能买完这一次就疲劳了。而是一点一点喂给他们，就变成还是像传统这个模式。
Then the autonomous driving company is actually selling the solution piece by piece. For example, I have a very mature solution called L4 in my hand now, I’ll divide it into 8 segments and feed it to the OEMs bit by bit from L2. Actually those companies will be very happy. This way I don’t have to be a ll in when advertising. If it’s an all-at-once mode, the customer may get tried quickly and only purchase this one time. Feeding them little by little and this method turns out to be like the traditional one.
You guys have a lot of business tricks.
That’s true and necessary. Just like selling shampoo.
Someone asked, can David talk about the new domestic power plan? What we just talked about are mainly foreign, such as Cruise and Waymo. What do you think of China’s new car-making forces?
The plan of the new forces must not come from the new forces of car manufacturing, but from its technology suppliers. But it must be far worse than some companies that do autonomous driving solutions. He certainly can’t afford companies with autonomous driving solutions, but prefers to use vision. So this is actually not comparable. The first company that sells cars thinks about how to sell cars. As for the technology, it can be used by consumers first. He can draw a blueprint first and show what it will look like later. Companies such as autonomous driving, such as Baidu’s Apollo, and other companies differentiated from Baidu, his ultimate goal is actually to sell services, so his technology must be stable. I find it difficult for me to answer his question at once.
Yes, I’ll just give you a brief introduction, here. Now there are two main routes, and that main route is the use of lidar. Radar is also divided into three kinds, radar is the major one, including lidar, millimeter wave radar, ultrasonic radar. Google, for example, they use more lidar in the program. It’s a different scheme from what David said about the new domestic car companies. The new forces may highlight the second solution, that is, they may take the visual as the outstanding point, like Tesla and also NIO, Lixiang. They’re using this relatively cheaper solution than the lidar solution, right? I don’t know if I’m getting this intro right?
It has to be much, much cheaper to get it. In fact, we also have to do some math: what is the median spending of a car in China? And how much is the added value of autonomous driving in a car? For example, let’s say a certain car manufacturer spends a few thousand dollars to changes the color of a seatbelt, and this seatbelt can bring it maybe 5 to 10 times added value to the business. If you use this logic to think about an autopilot package that can bring 5 to 10 times added value to the car company and the price is within the consumer’s acceptable range. How cheap do you think this tech have to be at Tier 1?
I believe we should have an idea of what such a cheap technology can achieve, and it won’t be particularly advanced. Compared to an automated driving solution companies, I mean there is certainly no way to compare with them in a short period of time. Because like Waymo’s solution, I think if the BOM will cost more than a hundred thousand dollars. Based on this amount of cost, if the solution can help the car factory generating another 10 times the added value, which is over a million. That’s impossible, right?
The system will cost more than a car.
Yes. So it doesn’t aims at selling cars.
So in the short term, investors will certainly ask you how to make money. For example, if the whole system cost more than a car, and as for Didi and Uber, as we talked before, they have already been the monopoly in the industry, but they are not making profits.
Yeah, so that model is highly suspect. Like I said, company like Didi. Well, it’s not polite to say that directly, let’s just give an example. If this the unit cost of a car is high, it’s even less likely to be commercially sustainable in the long term in China where labor is relatively cheap.
Taking our products as an example, we hope to package the whole hardware and software of automatic driving, pricing around 200,000 to 300,000 yuan. This is equivalent to a full-time car driver’s annual salary. It may be possible that the taxi company will accept it, but if you tell him that it’s a million yuan, or two million yuan if the car is included. It’s unlikely that it will accept the package.
The relationship between autonomous driving and the society
The difficulties encountered in the application of autonomous driving in China and in the United States are actually quite different. For example, cost prices will be considered in China. After all, the price of manpower will be cheaper in China, right?
Yes, the unit price of passengers in the United States is higher, so in the United States, even if it is logistics for driverless vehicles and logistics in the community, its unit price is higher and it is easier to pay back.
In China and the United States, there are companies that make unmanned taxis and people who make unmanned trucks. In fact, we also talked about this. This is one you need to see whether it is an incremental market or an existing market. Travel in the US is actually an incremental market. Every household in the United States needs a car, because they live in the suburbs, and travel services are not enough. So there is a huge deficit in this market. But speaking of logistics, because the United States generally drives large and low-income groups, and these immigrants of Latino are doing. Therefore, the government’s willingness to promote it is not strong, because this is equivalent to robbing the middle and low-income people for jobs. Then if the target is taxi, you are equivalent to supplement the lack of travel capacity, this is ok.
How do we understand this problem in China? Which part is stronger, and which part needs to be supplemented, right?
The auto union is actually avery big hurdle in America, right?
So in fact, the technology in the US has its advanced nature, but there is also a warm side. That is, we can’t eat the whole fish in some places. Technology itself is set to benefit the mankind, but now, instead, you let these low-and-middle-income people fall into another dilemma. That’s not beneficial to all human beings.
Yes. Europe will protect this kind of things more extremely. For example, Uber is still not legal in France.
In fact, this is understandable. When the same technology takes root in different regions, its blooming flowers can be different.
This does have negative effects. You will see that China and the US are developing at a fast pace, but they will be limited.
Let’s turn to another question. I was constantly asked that why the QR-payment in Japan is not advanced? What’s wrong with Japan? I have been to Japan countless times, and I can catch the point of this. In Japan, other aspects are very developed. Convenient stores like 711 are everywhere. They don’t need any revolution in this. Is the food delivery service really necessary for them? Why must you use your so-called advanced method to change them? Their own advanced methods have enabled people there to live in a very dignified way. If you stripped off their traditional lifestyle, can you guarantee that your way would be better? The technology itself is cold, but the warm side is how people regulate the technology. Do you have to let these Japanese people shut down his physical store and then engage in online shopping? In other words, it is wrong to engage in online payment. The Japanese government will not do such a thing. So what do you say about Uber in France? Affirming that the French government is out of protecting the well-being of his people, there will be some examples of this. There is a bill.
I noticed there is another argument in this market, that is, where can autonomous driving develop faster, in China or the United States? Some people feels that the development will be faster in China, because Chinese people may be more open-minded for those innovative things. Like, if you want to build a special building, you cannot build on a semi-finished product, because then you cannot lay the foundation, which may be faster, I don’t know. What do you think?
I think that in a commercial society, we should consider from the commercial point first, which is affordability. We don’t even need to use unmanned taxi as the example. Software replacement is enough. According to netizens’ understanding, Chinese have higher acceptance, then, do we have a software company who has been making easy money? No. People are unwilling to pay, or that their income level is relatively low. Our national conditions are here. I don’t think this question is difficult to understand. Think about the average income in China, or as Premier Li said, 600 million people live with a monthly income of less than 1,000 yuan. How do you compare with the United States? Americans are more willing to pay for technology. So technology is easier to be mature there. If Netflix was born in China, and I don’t think it could succeed.
Fair point. Money is a huge problem here. If this technology can drive down the price, we will pay.
This kind of business model in China should be called: Wool comes out of the pig. America is different. If you used its technology or enjoyed its services, you would pay naturally.
Right. I saw a comment: Taking advantages hinders technological development. I agree.
Absolutely. Taking advantages definitely hinders the development of science and technology. Not only the development of science and technology, but also the development of culture and any businesses. How can excellent literary and art workers survive without money?
Unless it’s a live show.
How would there be good content creators without paying, right? Just kidding. Actually in China the business model is completely different. For example, I agree with David that “getting everything for free” are not just a barrier to the technological development but also to the arts. For example, like we all know, some singers actually lived a miserable life in the earliest days. Pirated records are everywhere, and if you work so hard to produce a record but the money will be earned by these pirates. I mean, if you can’t sell many of your tapes or CD albums, you can only join in some variety shows or some concerts, right? Or even make a movie to get money. It’s understandable. Even though there are a lot of stars shooting many bad movies on the market now, I think it’s a process, right? They may have no choice, at times.
Yeah, if they can make a good movie, or the conditions allow him to make a good movie and make profits on it, why would they run away, right? Why would they play dumb?
Yes. There’s been some improvement now, like QQ Music, Netease Cloud Music. For example, Jay Chou releases a song and everyone supports one yuan.
It depends. I still watch MVs, and MVs are free.
Well, that may hinder Jay Chou releasing new songs.
What is now called the smart car cannot even parking a car well. When will the complete autonomous driving realize? I mean the L5 level. When will that be possible? Is this a particularly difficult question to answer?
这个问题要把它拆一下。 你问一下这位网友是他停车的时候费劲，还是他在高速上开车的时候费劲？肯定是停车的时候费劲。所以整个无人驾驶的大链条当中，要把交通分成大交通和小交通。所谓小交通实际上就是 AVP和APA，自动泊车和停车，这个实际上是作为单独一个课题，在单独做的，是不一样的。我们说的广义上的自动驾驶，是指这个车从A点到B点的这么一件事情，和你在一个区域内找停车位关系不太大。而且在中国停车和在美国停车完全不是一个体验。在美国，傻傻地停在路边，或者停在一个空旷的地方，比如说沃尔玛前面你就直接插进去就行了，反正有的是位置。中国地下停车场绕两圈下去，还得找车位，完全不是一个体验。
This question can be broken down a bit. Do you ask this user if he struggles to park, or if he struggles to drive on the highway? It’s definitely a struggle to park. So the whole driverless big chain, to divide the traffic into big traffic and small traffic. The so-called small traffic is actually AVP and APA, that is, automatic parking and parking, this is actually as a separate subject. That’s different. We’re talking about autonomous driving in the broadest sense, which is such a thing as this car getting from point A to point B, and this is not really related to the process of looking for a parking space in an area. And parking in China is not at all the same experience as parking in the US. In the U.S., park stupidly on the side of the road, or park in an open space, like in front of a Walmart where you can just stick it in. There’s plenty of room anyway. It’s not an experience at all when you have to go around the Chinese underground parking lot twice and still have to find a parking space to go around.
Yes,there are audience who says that the Chinese parking problem is still parking space problem, and it is, indeed, complicated.
Right, so what we’re talking about now is solving the problem of large traffic. For companies like us at autonomous driving solutions or taxi operating companies, we don’t pay much attention to small traffic. When a car that’s coming back to the starting point from the heavy traffic, and you get a guy driving it to a fueling station or gas station to fill it up and maintain it. This has saved over 90% of the manpower. Maybe this is a problem that needs a company like a component supplier to figure out how to do an AVP in a parking lot and APA. The autonomous driving company may not think much of it.
Then I have one last question, which I particularly want to ask. Isn’t SpaceX’s Starlink project very popular? Then some people said that is a big move for Tesla’s future unmanned layout. Can David explain it to us? Does this have anything to do with autonomous driving?
In fact, it doesn’t matter much, because the automated driving we talk about depends more on one-car intelligence than on external ones. For example, whether it is 5G or Starlink, the relationship is not very big, because in fact, our so-called communication, bandwidth and speed can only outsource some of the computing work to a certain place. But in the scenario of autonomous driving, we are not really allowed to outsource these computing work because your implementation is beyond control. No matter how good the connectivity is, there will be occasional breaks. So it is only possible to outsource such things as high-definition maps, you download maps, you go to a certain area, and load the map of the next area in advance. This is possible, but the smart outsourcing of the car is not too realistic.So in fact, I think that the Starlink may be tied together in the future, and what flowers can be bloomed, but it must not be the result of the car’s intelligence, but the ecology of the car that may be based on the ecology of the car.For example, what service do I want to buy and enjoy with higher bandwidth? Watch YouTube in the car, right?
A friend asked how about NVIDIA’s autonomous driving? He has invested in its stock.
In fact, the company NVIDIA is quite powerful. It provides a structure. Among the three major open source platforms, the NVIDIA family has contributed an architecture, and so far most of the calculations we have called automated driving solutions companies, or computing units still use graphics cards. This is because the graphics card handles these deep learning models, and NVIDIA is basically a monopoly in this field, unless there are other companies that say they have made a chip acceleration called FPGA.They have made a FPGA, add their own algorithm, otherwise all are using NVIDIA graphics cards. This company is very powerful.
Investment has risks and should be cautious, we do not provide investment advice, ok?
Okay.The last question. Will autonomous driving affect the development of vehicle insurance?
I’ve just actually mentioned this a little bit before, but first we need to find it a model pair. Right? Just like we average person buys insurance, a car insurance policy will have a reference model. And then the autonomous driving with the what can be referred to as a model? There’s not enough mileage or collision at the moment. It’s just crashing all sorts of oddities case that are not enough to support this model. So actually the insurance company at this time they are not very good for this matter, that is to say they are taking one directly from the traditional model. According to my understanding, they can only be treated as a machine that had a work injury.
It need more data.
Yes, and it takes a long time.
Would David introduce you Whale Offshore AI again?
我们惠尔智能目前就在做一个L4的方案。实际上我刚才有说到，我们希望把它做成一个软硬件一体的package，大家可以理解为一个“大帽子”。这个“大帽子”套在车上，我们可以让这个车实现自动扫地，当然终极目标是无人出租车了，实现这些我们定义的功能。所以我们这个公司更多的是跟实际的运营公司去合作，比如说实际做出租车运营的，实际做扫地运营的，当然也会跟车企有一些绑定的关系，他们需要把车的底层线控给我们，我们再在上面去做，是这样的一个 business model，而不是像特斯拉或者其他全栈公司那个模式，不是的。
As for Whale Offshore AI, we are currently working on an L4 solution. In fact, I just said that we hope to make it into a package of software and hardware, and the home can be listed as a big hat. As for cars, our cars have achieved automatic sweeping function, and the ultimate goal is the unmanned taxi and to achieve these functions we defined. So our company is more cooperating with the actual operating company, for example, the actual taxi operation, the actual sweeping operation, and of course there will be some binding relationship with the car company. That is, it will need to give us the underlying wire control of the car, and then we will build the tech on top. It is such a business model, not the full-stack company model like Tesla or other full stack, no.
Cool, it’s a hard-core share! I think it’s a combination of Google and Waymo’s perceptual methodology.
Ok, thanks David.
I hope you will continue to watch on Danny’s show, and I hope that I can bring more tech friends to you here. Wait for my friend after the launch of his rover and also the Boring Company.
The Boring Company is Musk’s ground hole company, the hole digging company.
And then there’s Red, my favorite camera company.
Yeah, the Red, the one used by another channel, Yingshijufeng.
I can invite some more awesome friends from these companies to have a chat with you on Danny’s platform, sounds pretty good.
Right, Thanks David. if there is any big names in the future, we should remember that it was David who helped us to invite them. Let’s all be more supportive of David’s projects in the future.
Thanks David. I learn a lot from him for free.
Do you guys notice that David is smiling when talking, even the topic is autonomous driving.
But experts know that the autonomous driving that he’s working on is super difficult. The famous formulas are hard enough to understand, but they’re basic requirements for the autonomous driving. Even the scientific concepts like light speed are affecting the tech of lidar, and also limiting the development of the autonomous driving. Anyway, it’s as complex as launching a rocket. It’s mental and money consuming. And in the short term, it’s hard to be profitable.
David inspires me with his knowledge and his attitude. Though he’s anxious, he can still face it with a positive heart in daily work and find out some fun and unlimited things in the work.
Conversation with David reminds me of The Sun Also Rise by Ernest Hemingway. The protagonist once asked Mike:”How did you break up?”
麦克回答：Two ways. Gradually, then suddenly.
Mike said: Two ways. Gradually, then suddenly.
自动驾驶的进程，大概也会这样。慢慢地开始，突然地爆发。Tech is magic, 我们下回再见。
The development of autonomous driving might be the same:gradual development and sudden burst.
Tech is Magic, See You Next Time.