China EVs & More

China EVs & More MAX Episode #3 (ME) - Maxwell Zhou - DeepRoute CEO/Cofounder

February 11, 2022 Tu Le & Lei Xing
China EVs & More
China EVs & More MAX Episode #3 (ME) - Maxwell Zhou - DeepRoute CEO/Cofounder
Show Notes Transcript Chapter Markers

Tu and Lei speak with Dr. Maxwell Zhou, CEO of DeepRoute.ai, a 3-year-old 2nd wave autonomous vehicle (AV) startup.

They ask Maxwell the recent $10K AV system that DeepRoute launched using solid-state LiDAR, when robotaxis will be ubiquitous, and why Tesla doesn't use LiDAR?

A very candid, entertaining, and informative guest, Maxwell answers those questions and much, much more.

Tu Le:

Hi everyone, Tu Le here, one-half of the China EVs & More duo. Lei and I are always thinking about different ways to bring you, our audience, relevant compelling content about the China EV, AV and mobility sectors. Especially now that several companies that we’ve tracked over the last 50 or so China EVs & More episodes have become global phenomenon. China EVs & More MAX is where we bring you special content, in the form of conversations we have with special guests from those sectors. 

In this episode, we talk to Dr. Maxwell Zhou, founder & CEO of DeepRoute.ai, one of the so-called second wave of Chinese AV startups and perhaps the fastest growing. Founded just three years ago and headquartered in Shenzhen, DeepRoute.ai made headlines late last year when it announced the industry’s first $10,000 Level 4 self-driving system, this just a few months after receiving a $300 million round of financing led by Alibaba, officially making it a unicorn. 

Lei Xing:
This is your co-host Lei Xing. 

 It’s pretty amazing that for someone who’s only in his 30s, Maxwell already has two AV startups under his belt and calls himself an “old soldier” in the AV space, having worked at prominent companies including DJI, Texas Instruments and Baidu, before founding Roadstar.ai and DeepRoute.ai. Tu and I caught up with Maxwell right before the Chinese New Year and asked him about his professional journey, the $10,000 equation, AV competition, Tesla, LiDAR, Robotaxi commercialization timeline, and more. 

 Here is our conversation with Maxwell.

Tu Le:
Hi, Maxwell, thank you for sitting down with us to chat AVs and DeepRoute.ai. First of all, let's start out by you telling us a little bit about yourself, how you got into autonomous vehicles and how you started DeepRoute.ai.

Maxwell Zhou:
I started with robotics at a very early stage, like 10 years ago, there is no autonomous driving in the markets. When I was in primary school, I liked DIY and do model planes. I started learning programming and circuit design in junior school. In high school, I got a chance to join a Robot Olympics. This is my first time actually doing something related with robotics. Then I won the Gold Medal in the Robot Cup. So I realized I got talent in this area, then I enrolled in a talents program at Tsinghua University, they gave me more experience. This talents program is not on robotic track but focused on mathematics and physics. So they gave me very special experience to explore deeper in mathematics and physics. After that, I pursued my Ph.D. in artificial intelligence. 

So there maybe a few changes. Initially I started my Ph.D. in physics. After I joined a few conferences, I realized physics is different, this is not what I want to be. Then I switched my majors, so I go to the (United) States having an artificial intelligence Ph.D. During this period, drones were very popular, so I got a chance to join DJI’s development challenge, and I won the first place. During my Ph.D. I was doing Co-Ph.D. with Texas Instruments. So I worked on projects with QB Labs, and you see, especially there is agriculture robots. So that's my experience before I went to work. 

Then I finished my Ph.D. degree, I needed to face what to do in the next step, so there were a couple of choices. The first choice is go to DJI. At that time, DJI is popular, back in 2015, that's very popular. But I believed autonomous driving is the future. So especially I chose my offers between Waymo and Baidu, but I chose Baidu because they are the team starting from 0 to 1. Waymo they are already in 100 to 10,000, so being there, you will learn less. Then I decided to go to Baidu and then in their Bay Area office. So that's my first job. Then I left Baidu to start my several companies. The first company in Roadstar.ai, and DeepRoute.ai is my second company, founded in 2019. Right now DeepRoute.ai is already three years old.

Tu Le:
Still a baby.

Lei Xing:
To us, it's ONLY 3 years old.

Maxwell Zhou:
Yeah. But for autonomous driving, definitely I am like an old soldier for the AV industry, especially in the L4 domain. Before 2016, there was only Waymo in the world doing autonomous driving, and Baidu was the first team besides Waymo. So I have a very intensive experience in autonomous diving, and I know a lot of people. And for my universities, my programs in Bay Area, then I founded the group in Bay Area, then we relocated to Shenzhen.

Tu Le:
How large is DeepRoute.ai right now? How many employees in the U.S. versus Shenzhen?

Maxwell Zhou:
Right now we have 400 employees, most of them are in the technical fields. In the Bay Area, about 30 to 40.

Tu Le:
I lived in Silicon Valley for almost seven years, and I worked in Fremont for about five of them. So where exactly in Fremont is DeepRoute.ai located?

Maxwell Zhou:
In Fremont, I would say near, the…

Tu Le:
 The W Hotel?
 
 Maxwell Zhou:
Very close to Tesla, like one mile away. 

Tu Le:
Ok. So I drove by the Fremont factory almost every day to get to work. Very Cool. We're just going to get right to it. So can you help us? I think our listeners aren't sure how L4 is defined. Can you define what L4 means to DeepRoute.ai and what it means to you? Are we talking point to point autonomous driving? Or what are certain scenarios when you're saying L4,  how do you define that?

Maxwell Zhou:
In my opinion for L4, it is in a special domain. For example in Fremont, the vehicle is able to (go) from any point to any point without safety drivers and maybe you can have remote access, but there is no safety drivers. That's L4. Because by definition, L4/L5 and L1/2/3, the total difference is, if there's some accident, who will be responsible, for L4/5 the system will take this responsibility. That's the difference.

And by the technology side, L4/L5, their technical stack is different. For example, L4/L5 they are a centralized system. For example, normally they will have very powerful processors, maybe GPU or CPU based. But for L2/L3 they are distributed system. For example, they have AEB functions, the MCUs to doing the work. They have lane keeping functions, there are processors doing this. So they are that different, in systems and in hardware they are different. But in functions, by safety I already explained that there are different definitions. For L4/L5 right now, normally they will need HD map. These are still different.

Tu Le:
Unless you are Tesla.

Maxwell Zhou:
So Tesla is without HD map. Actually, maybe Tesla can eventually do L4/L5 without HD map, but it’s very hard. That technology side you have no prior information regarding the environment. That makes the technical challenge is tougher.

Lei Xing:
So Maxwell, a follow up question on L4. Mobileye CEO Professor Shashua said at CES this year, you might have heard this: the endgame for the auto industry is the consumer AV. What is DeepRoute.ai's endgame? Do you even believe the so-called consumer AV or personal AV that GM announced with L4 autonomy that you can buy, actually exist? Or do they even make sense? 

Maxwell Zhou:
Yes, for my views, the commercial AV, that exists. Maybe the final ending is Robotaxis. People don't need personal AVs. They don’t need personal vehicles, they just use it as a subway.

Lei Xing:
Public transport, almost.

Maxwell Zhou:
Yeah but in the middle, I believe a bunch of people, maybe a lot of people, they will buy AV cars. For example, like Teslas. Maybe Tesla is not L4 ready, maybe L2. But people they will buy it. Because we need a lot of road test mileage, like billions of mileages. So you cannot do this by only Robotaxis. You will deploy your systems on the commercial AVs by…, you need the data, you need billions of data, tens of billions of the mileage data. That's what we think is that's the end game. Recently we just announced our production ready L4 solution, the total price for the AV kit is under $10,000. And all this hardware is meeting auto grade, using solid-state LiDARs, using (NVIDIA) Orin chips, everything is auto grade. We announced this last year, in December. 

Tu Le:
I got to tell you Maxwell, when that announcement came out, a lot of AV companies crapped their pants. Because months before it was Baidu saying they had hit half a million RMB. But you went way below half a million RMB.

Maxwell Zhou:
Yeah, because all these people are using mechanical LiDARs, because mechanical LiDARs can give you very good quality point cloud. But for the solid-state LiDARs right now, they are very poor performers. But how you can use these sensors to achieve the same performance, that's by your algorithm. We have advanced algorithm, especially in the early sensor fusion. I was almost one of the first people, in this industry to propose this method. So we have bunch of experience of using low-price LiDARs and cameras that fuse together to achieve the best performance.

By the videos we posted on YouTube, you can see our performance. Our vehicle is definitely providing the L4 functions with no disengagement in the rush hour in Shenzhen. This is just regular testing, we don't choose from a bunch of tests. So a lot of people could go to Shenzhen or go to Shanghai and test our vehicles. They are all surprised by the price, how you can do this at this price. But this is the truth. There are some companies, they are following our ways, they make some announcements with PPTs, no videos.

Tu Le:
Yes, or they edit the videos.

Maxwell Zhou:
This is very, very tough to use such cheap sensors. So this sensor setup is designed for L2, but we can do L4.

Lei Xing:
So my next one question is you mentioned about LiDARs. You recently signed strategic cooperation with both RoboSense and Z Vision. I think they're both going with the MEMS route for LiDAR. What factors made you choose these two suppliers? And how will they be involved in the $10,000 equation?

Maxwell Zhou:
I can give you the rough price, because I cannot tell the exact price.

Tu Le:
No, that's fine.

Maxwell Zhou:
I can give you the price the OEMs purchase (on a volume basis). They are like $600 per LiDAR, maybe $700 per LiDAR.

Lei Xing:
It's definitely under $1,000.

Maxwell Zhou:
Yes. And you can see $1,000 as far as that they don’t make the equation that much (different). For the Orin chips, one chip, you roughly need $1,000. So usually you use two Orins, $2,000, and use five LiDARs, $5,000. The rest you have still got $3,000. Cameras maybe $50. You can calculate it, that's easy.

Lei Xing:
We can do the math.

Tu Le:
How much lower do you think these solid-state LiDARs can go? Like price wise?

Maxwell Zhou:
$300 or $400. And by the volume production, you need three of them. So they don't actually need five. We use five LiDARs, but we can cut this to three LiDARs, maybe they can only do L3 functions, not L4, but make this price more competitive.

Lei Xing:
So my question was on RoboSense and Z Vision, how do they complement what you are doing? Because you now have two different suppliers. How are they differentiated in what you are trying to do?

Maxwell Zhou:
They have different products. So one is for the long-distance LiDARs, the RoboSense, they are with narrow FOV, but they can see far, they can see 100 m or 150 m. Z Vision, they have wider FOV, this is for blind spots. That's the difference. We mount them in different places. That is why we chose them.

Lei Xing:
Yeah, there's no brainer on RoboSense because they've been all over the place, recently.

Maxwell Zhou:
But to do L4/L3 you need LiDARs for the blind spots.

Tu Le:
Can you tell us a little bit more about DeepRoute.ai where your pilots are in China? What you guys are doing in the U.S.? If you guys have any pilots, do you have a license with the DMV in California? Can you give us a little bit of background on some of the pilot programs you're running in China and the U.S. if you are.

Maxwell Zhou:
In China mainly are the Shenzhen pilots. Right now we have like 70 vehicles in Shenzhen and in Wuhan as the second largest pilot around 40. A lot in Hangzhou, around ten. So that's all in China for the Robotaxi only. In the Bay Area, we have the license from the DMV, we have three or four vehicles in Fremont, where we do the road test.

Tu Le:
Can you share with us any plans that you have for more rollout for 2022 in China? Because I know that Baidu and Pony.ai have licenses to run Robotaxis without the safety driver. Is that something that you're going to try to get this year as well?

Maxwell Zhou:
This year is the government's permit. So we have this technology doing this. For the cities we operate they don't have the regulations or policies for us. For Beijing, yes, they do. They have this policy, but you see only Baidu and Pony.ai doing there. If you want to operate in Beijing, they will charge you a lot. For example, they will charge you RMB1 million or RMB2 million for one license. So why I want to go there? They don't make sense. This license is much more expensive than your vehicles. They don't make any sense. Initially, they charge you RMB2 million for one license in Beijing. But in Shenzhen it’s just RMB100,000. It’s only 10%. For the DMV license, it’s free. 

Tu Le:
Yeah, I didn't know that actually that they were charging so much for the licenses.

Maxwell Zhou:
So because you need to you need to do the test in the private place. They have a private test place. The private test place charges a lot. They don't admit the rest of your test results. So that's horrible. So for example, for us, we can do the testing in Chongqing and in Shenzhen. This is cheap. In Shenzhen and Wuhan they accept this route, but in Beijing no, you can use only this one. This is very expensive.

Tu Le:
It's the guanxi thing right? So I think your investors will be very happy to hear that you're trying to manage their money well or manage your burn rate very well. Can we move on to the commercial side because I know that you're looking at launching some commercial pilots as well. Can you tell us, are you working with trucking companies? Or how are you entering the commercial tracking side?

Maxwell Zhou:
Yes, right now, we have the light-duty (note: Maxwell meant medium-duty) trucks. We have collaborated with some companies, for example, in Hema, that's the Alibaba-based company. They are doing the food deliveries. Tu, you are living in Beijing so you know Hema, they sell high-end seafood.

Tu Le:
I know Hema every day, so especially with COVID.

Maxwell Zhou:
We have the program with them. And then we have the relationship with the Deppon Express. So we do deliveries for these logistic companies. Then I cannot tell more, some is not on PR yet. This is what I can tell.

Tu Le:
You can hint us.

Maxwell Zhou:
Maybe more in Ali base.

Lei Xing:
Let me follow up with the trucking because you have two routes, the Robotaxi and also the trucking. What is that position? Is that like, you know there's the last mile, middle mile, long haul. I think your position is in the last mile, is that correctly? Last mile trucking, delivery, like Gatik, you know Gatik in the U.S. Is that what you're trying to do?

Maxwell Zhou:
It is like UPS deliveries. So for the final deliveries UPS they will have our ligh-duty (note: again, medium-duty here) truck, they deliver packages for you. It is not last mile. It is hub to store based, for example, the central hub in California, maybe in the Bay Area. The Bay Area has fleets that deliver to Fremont, to San Jose, Sunnyvale, this kind of hub to store scenarios.

Lei Xing:
Hub to store. Got you. 

Tu Le:
I think it is like Gatik.

Lei Xing:
I think Gatik, yeah, but maybe slightly different, maybe longer routes, I suppose?

Tu Le:
Yeah Gatik is pretty long route.

Lei Xing:
I think there was a question on your competitors. The WeRides, the Ponys, even the Baidus, where you worked before, and AutoX. How are you different than these guys? Whether it's from the hardware or software stack? You talked about algorithms, but what you've done, maybe besides the $10,000 kit, they've done as well. So where are you versus these guys?

Maxwell Zhou:
For the $10,000 solution, that's the outcome. DeepRoute.ai, we are founded much more recently, but we move fast. The technology we are doing, we would just go before the competitors. Other competitors are like our friends. We move faster than our friends. That's the outcome. So the internal reason for that is we have much (more) efficient teams and much more efficient burn rates. So recently we raised a lot of money, not like the years before, but we still calculate (down to) the pennies. We're using this to achieve the best performance. Maybe you see there are companies that raise billions of dollars, but still their technologies, they are the same, or even slower. So our key difference is we move much faster. You consider three years ago where we are and see where we are now. We move much faster.

Tu Le:
I look at DeepRoute.ai as the second wave, and I put Qcraft into that, where the pioneers, the Ponys, the WeRides and the Baidus have been in the market for quite some time. And then these younger AV startups like DeepRoute, like you said, they move faster, but also because you're able to take advantage of some of the price decreases in the hardware. And so is that part of your advantage then just being able to kind of piggy back off of some of the things that they've been able to accomplish?

Maxwell Zhou:
It is not that way. You can see there is a bunch of second waves, and only the DeepRoutes moved to a good place. There are some special chemistries. For technology speaking, will always need in this industry. For example, I was the first one to propose this early fusion methods within a bunch of the benchmark, like KITTI, we have papers accepted by CVPR every year with one or two papers. In the second waves, we are the first one having the machine learning engine teams, we are the first company having this technology. We know what's the future and how we can plan our R&D. This is our other advantages. People are following us. The first waves are following DeepRoute.ai. That's the truth.

Tu Le:
You've made a lot of headlines in a short period of time. So, very impressive.

Maxwell Zhou:
Yeah, in my previous one they followed the early fusions. Second they followed the influence engine. We are always leading this industry in technologies. That's also the reasons make us fast. So what's the end game? I can say I was for almost the first guy to realize the end game must to be cheap and better L4 solutions. So that's why we plan our technologies. Then eventually we have this $10,000 solution, and no other in the world have this. Maybe some companies have PPTs, but we are the only one that can run the vehicles.

Lei Xing:
So then I go back to the earlier question, Tu kind of eluded to it, which is you talk about this cost side of it, then the commercialization side of it, when do you see this Robotaxi being as ubiquitous as going out there to hail a DiDi or an Uber like we're used to today. And they're all on public roads everywhere, point to point. When do you see that happening?

Maxwell Zhou:
I will say you still need 4-6 years. I can tell you my calculations by DeepRoute.ai methods. Today we have low-cost solutions, less than $10,000, and it can achieve driverless capability in certain areas. And in the coming two years, this will be in-depth collaboration with the automakers. So normally you will spend two years having cars ready. By 2024, you can see they have some vehicles installed with this similar sensor setup by pre-installed or factory-installed on the roads. And then you need data. You need at least two years of data to convince the governments anyway, to develop your program, your technology, you need at least two years. And then by 2026, maybe this is the earliest date you will see bunch of the vehicles on the roads. I don't believe one million, maybe half a million hardware you can do mass production, there's no way. You cannot use money by this way, even Waymo’s price, is very expensive. Even Waymo has only 1,000 vehicles. It’s too expensive by using the old methods.

Tu Le:
I completely agree.

Maxwell Zhou:
Yeah, so 2026 that's the earliest stage, because mass production is not possible before 2024, and then you need at least two years of data to convince the governments, to have billions of mileage data. This needs time. And then you convince the governments, they will allow you to do this. Today the governments allow you in very limited areas. Even in Beijing, it’s just 4 km. By walking, it’s less than one hour, almost useless.

Lei Xing:
So then the follow up question is, looking back when you started maybe at Baidu or when you were at Roadstar.ai and now DeepRoute.ai, let's go back 5 or 6 years, to 2015 or 2016 was the fervor of the autonomous vehicles. What were your expectation then? Has it accelerated? You said 2026, or has it slowed?

Maxwell Zhou:
It has definitely slowed. This is much tougher than you expected, because initially all people are going leading by Waymo. Waymo said that they can go directly to Robotaxi. But you can see Waymo only has 1,000 vehicles. It is too expensive. But Tesla tells you that one million is just one year you can sell. The price is the ceiling. Anyway you need a lot of mileage, a lot of vehicles, then to make the cost down, you have to overcome this. We are the first company to realize that. Back to 2019, when we were founded, the first days, I realized you have to do the mass production, but not by L2, because L2 data is useless for L4. You need L4 solutions for mass production, then you can achieve the Robotaxi. There is no second way.

Tu Le:
So that brings up a great question, Maxwell. Do you see the automakers like an Xpeng, a NIO and a Tesla as real competitors to DeepRoute.ai, because they're going to be able to get thousands of cars on the road. Tesla already, just last year they sold almost 1 million cars. Just from a data standpoint, they're way ahead of everyone else. So do you see a NIO or an Xpeng becoming a viable competitor or their autonomous driving system, do you see that being competitive against a DeepRoute.ai system?

Maxwell Zhou:
Right now, they are a lot, I will say. I will say right now they are using the L2 solutions, the different architectures, different solutions. Maybe it's in the future they want to explore the L4 algorithm. But I will say there are a lot of, still gaps they need to catch up, because they want to sell their cars. Their performance must meet some certain levels. If your own algorithm is below that, you cannot sell your cars. So for some company, maybe there are some leading companies, Teslas, for example, Tesla has a very good performance vehicle, everyone wants to have these technologies. Everyone want to have its technology, but go to the market, you have idiot cars, Tesla is very smart cars. You are unable to sell your vehicles. You have to use third party ones. This definitely happening in the most of the companies, not all can do this.

Tu Le:
Do you ever see DeepRoute.ai getting into vehicle manufacturing in order to get more cars on the road?

Maxwell Zhou:
No, we're not doing like some company want to make the cars. So we are collaborating with them, like co-pilots. So you don't have this resource, maybe the capability to doing this. Why not co-pilot with them for the OEMs? So just like that, the name you say in the previous cases. If they cannot do actually good algorithms, why not co-pilot with some company like us?

Lei Xing:
So I was going to ask that you do work with partners like Dongfeng or CaoCao Mobility, who will be on the operation side of it and you don't want to be operating the fleets, right? You just want to provide the solutions. Is that correct? That's your way.

Maxwell Zhou:
Maybe we have more, but not PR yet, more powerful ones, but I cannot tell in today's interview. This is years ago. I can say by different ways. For example, this is commercial, so we don't do anything. But if this purpose is for Robotaxis, then we will do this together with them, for some OEMs they just want to sell you cars, they don't want to do the operation. It depends on the demands by them, our partners.

Tu Le:
Let's change gears a little bit. Maxwell, can you tell us from a difficulty standpoint, what the difference between a Robotaxi driving around in Fremont, California versus a Robotaxi driving around in downtown Shenzhen at 6 o'clock in the evening, how much more difficult is that? What is the vehicle looking at that it is not looking at or considering in California? 

Maxwell Zhou:
There is a lot of difference. For example, when I drive in Fremont, I will not see people on the road. Normally there are only vehicles, very few times you can see cyclists. But in China especially we operate in the CBD area in Shenzhen, there are a lot of scooters like the Kuaidi xiaoge (express delivery guys) or Meituan. There are a lot of them. They don't follow the traffic rules. They are just random things on the road. So that's the toughest thing compared to Fremont, that's totally different, so pedestrians and cyclists that's totally different. But for the vehicle side, is very similar, maybe in the U.S., the driving average speed is about 1 km/h, but in China, because of the traffic, normally you cannot drive more than 60. They have in different ways for when we test in Fremont, we drive in our vehicles at 60 or 70 mph, always they will be used. They are allowed to drive fast.

Tu Le:
So would you say because I’ve had questions where people have asked me, could there be a global autonomous vehicle? So a DeepRoute.ai L4 Robotaxi, could we ship it to Europe or the U.S. as is and it would work effectively and safely? Or would you need to do specific things to make it regional friendly or specific? Could there ever be a global autonomous vehicle? Or would there always need to be certain considerations made because of the region that it's being used in?

Maxwell Zhou:
There are two different things. The first is technology side. Definitely your technology developed in China is easier to develop in U.S. by algorithm side. Because it’s much tougher, there are much tough data. But still there are differences by the traffic rules. In Europe, there are some countries driving on the left side, but China and the U.S. drive on the right side. You need to change your traffic rules, but this is like a small modification. They don't add to your difficulty on the technology side, algorithm side, that maybe you add more constraints. But it's easier for us, easier to deploy our technology to the U.S. For example, our Bay Area office, initially we are unable to drive the vehicles. After we set up everything in China, we just download the program to the U.S., and it works with no disengagement. Maybe you need to stop at the stop sign. There are some certain scenarios you need to change that. Because in China they don't have stop signs, But back to my first company, Roadstar.ai, we first ran the vehicles in Bay Area, but then back to Shenzhen, no, you cannot run your vehicle, algorithm is definitely no ways, no way. They're very painful.

Lei Xing:
This is very interesting.

Maxwell Zhou:
Because actually in my experience, we have things happen. In Roadstar.ai, it’s from the Bay Area to Shenzhen. No, it's not working. And in DeepRoute.ai, China to Fremont, it works in a few days.

Lei Xing:
I have a question, what remains to be the two or three or four difficult scenarios or cases in your mind to be solved, whether it's DeepRoute.ai or Robotaxi in general.

Maxwell Zhou:
There are general ones, like double parking. These are buses parking on the street, and your car maybe parking, maybe you go through this bus. Because there are blind spots, right? And unprotected left hand turns, this is the second one. And any unprotected behaviors that's tough. But the most tough one is e-riders, scooters, e-scooters, that’s the toughest one. Believe me. In the States, you only consider double parking and unprotected left turns. But for China, this is the toughest. Even protected turns, they have these people. Even in late nights, you have these people. So this is always unprotected, you need to consider all road is unprotected, anywhere you go out. 

Tu Le:
So double park and unprotected left.

Maxwell Zhou:
Turns, unprotected turns. And sometimes there are very narrow space, for example, in the place right now, the road is very narrow and sometimes there are cars parking on it. So if you are driving this road, you need to drive carefully, because they say safety distance is very close. You need to slow down your speed.

Lei Xing:
What about extreme weather conditions and special, in the morning sun glare, in the evening, those special times of the day plus weather conditions. What about those?

Maxwell Zhou:
For the extreme conditions we don't have snow, because our office is in Shenzhen, in Wuhan, in Bay Area, we don't have snow. Forget about snow, and for raining normally it is fine. But for the very, very heavy rains we will stop the operations. But for normal rain, there's no problem.

Tu Le:
Let me offer you this Maxwell. If you ever want to open an office in Detroit where there's four seasons, I’m happy to help you do that. If you want to test in Detroit, let me know.

Maxwell Zhou:
I will say that’s tough because the current technology use HD map. For heavy snow, they change the terrains. For current solutions, if the terrain changes, you don't know where you are. That’s tough. For the vision based, you cannot see the lane markers. For the localization, there is no landmarks to do the comparison, that's tough. I will say for the heavy snow, definitely, this will be a problem. Maybe you need data for the snow scenarios.

Tu Le:
So Maxwell, why doesn't Tesla use LiDAR, especially at a sub-$1,000 price point

Lei Xing:
He is probably tired of this question.

Maxwell Zhou:
I was saying for today, you can believe before, because LiDAR is expensive and there are no way to install them in vehicles. But for today I will say for some days I believe they are going to change, because by our opinion, if by pure vision Tesla can do this performance, maybe there are a lot of L4s. How about Tesla using LiDARs? That's horrible. I believe Tesla has very good engineers. If Tesla has LiDARs, maybe you can see the rest of the OEMs, their R&D is not ok. But for Tesla, no, they have the best talents. So that's very horrible things if Teslas are using LiDARs. Yeah, normally in the OEMs, their R&D is very...

Tu Le:
Mama Huhu (so so).

Maxwell Zhou:
Yes! But Tesla is a technology company, it’s not an OEM company. They are a different company.

Lei Xing:
Software company.

Maxwell Zhou:
Tesla has the same level engineers like Waymo and Cruise. And they have a lot of vehicles. If they install LiDARs that will be horrible.

Lei Xing:
Soundbite!

Maxwell Zhou:
Sometimes when the LiDAR is released. Right now, we purchase this in low price, but I cannot guarantee it can work in more than one year.

Tu Le:
You got to work with your partners, you got to make it reliable, right? So you got to work with your partners, demand more.

Maxwell Zhou:
Yeah, that's the reason Tesla why they don't use it, because there's no proven record that say the second year they still work. But when the products, when they are developed in the next generation, next generation, when this is reliable, it is auto-grade, why not? Just a few hundred bucks. They have the best engineers, that’s horrible competitors. But for the normal OEMs, we don't believe that because their R&D I know in this area, even they have thousands of people, that's different.

Tu Le:
As someone who grew up in Detroit in the automotive sector, and then worked in Silicon Valley, and now in China, I tend to agree that.

Lei Xing:
You’ve seen it all.

Tu Le:
The level of engineering and understanding of software is a little bit. It's completely different in Detroit versus Silicon Valley.

Maxwell Zhou:
Completely different. Silicon Valley they have only one OEM: Tesla. And they are in Silicon Valley. They are the OEM. Maybe Apple, the second one. But they believe they are horrible competitors, Apple and Tesla, not the traditional ones. 

Tu Le:
So here's my take, Maxwell. I still think a Volkswagen, and a General Motors, and a Toyota have something to say, right? Because maybe their autonomous vehicle systems, hardware, software, stack, software, dev team isn't as talented. But if I am an App developer and Volkswagen approaches me and says I can flash you into 10 million cars tomorrow. If you give me an 18- month exclusive on your App, on your service, I think that's still pretty compelling, right? I'm not saying on autonomous vehicle stuff, there's a place for the OEMs to still, there'll be a couple, two or three in the next 4-5 years that’ll still be very, very competitive, but the rest of them really need to focus their efforts on probably two or three different things that they want to be really good at. So that's just my opinion. Lei, do you have any final questions for Maxwell?

Lei Xing:
No I was just going to say that you, DeepRoute.ai or whoever, you still need to depend on these automakers after all, to have these Robotaxis.

Maxwell Zhou:
Yeah I will say for the automakers that's collaboration is for us. We have no capabilities to make cars, but for automakers collaborating with the startups, that's the option. Because our talents team, we have very good talents teams in software side and they can make cars. This joint marriage where we are good. Otherwise, all people lose, only like Tesla and Apple, they win.

Tu Le:
I must say Maxwell, it's companies like DeepRoute.ai that are making things, you're going even faster than the rest of the sector, which is, so the automakers thought they had 10 years, 7 years and then companies like DeepRoute.ai come online and now they go from 5, 7, 10 years to like 4 or 5 like you said. So this is concerning and this is why they have to partner, because they can't bring up a software development team and then train that team to compete with your team in a short period of time. So that's why they're going to be forced to partner, most of them are anyways, so.

Maxwell, thank you for your time and you've got friends in Beijing and in Massachusetts now Maxwell, so we're definitely going to keep track of what's going on. And once you have these super-secret announcements in 2022, we'll have a follow up pod with you.

Maxwell Zhou:
Thank you guys. 

Lei Xing:
Well that was a fun episode, wasn’t it. Lots of laughs. And just to clarify, when he said “horrible” about Tesla what he meant was actually serious competition for DeepRoute and the rest of the AV industry, and what he meant by the RMB1 million or 2Million is actually the testing cost that's expensive and not the license, which is free.

I must say, Maxwell is THE most interesting and entertaining guest we’ve had on the MAX series so far. Jokes aside, I sense that there is both confidence and cautious optimism from Maxwell on his own company as well as the state of the AV industry. He was actually very specific when he said 2026 would be the year that we would really see Robotaxis out on the streets in large scale providing paid service to the general public and gave compelling reasons why he thought so. One thing is certain: the cost of AVs will continue to fall as hardware and software that make up the system continue to “democratize.” 

Tu Le:
Lei and I will be sharing more of conversations with the men & women around the world moving the EV/AV mobility sectors forward as part of this China EVs & More MAX series. Some folks will be instantly recognizable, but some will just be people that are doing amazing in the space that we think deserve to be highlighted. 

Don’t worry though, Lei and I will continue to host our live weekly China EVs & More Twitter Spaces room that summarizes that week’s most important news coming out of the China EV, AV and mobility space. For those that can’t catch the live show, you can find the China EVs & More pod on all major platforms or wherever you normally get your podcasts. As EV adoption reaches its global tipping point, it will be even more important to stay updated on everything that’s happening here.

Lei and I are confident that China EVs & More is the best resource to do that. Until next time, as always, thanks for listening!

 

Professional Journey
What is L4
AV end game and Consumer AV
LiDAR & $10,000 equation
Plans for 2022, rollouts
Trucking
AV compeition
Robotaxi commercialization timeline
Automakers collaboration/competition
AV commercialization hurdles
Tesla