Latest / Autonomy Insiders / How WeRide Built a Global Robotaxi Business in 9 Years | Tony Han, CEO WeRide

Transcript

How WeRide Built a Global Robotaxi Business in 9 Years | Tony Han, CEO WeRide · Autonomy Insiders

• Daniel Abreu Marques: Welcome to Autonomy Insiders, the show where global industry leaders in autonomous driving unpack their real-world insights. Tony Han: European market is one of the most important markets for weride We want bring the prosperity economics to the local community. So one one principle weride is we want to do local hire if possible we want to as much as possible to collaborate with the local collaborators. no companies I mentioned Have passed the other reverse qualification process. the competitors you mentioned, where are their robotaxi? Where are their driverless robotaxi? Can I hail them? No. hallucination in digital AI may be entertaining, but it is fatal in physical AI. Daniel Abreu Marques: Welcome back to Autonomy Insiders. I'm your host, Daniel, and most conversations about robotaxis are still conversations about one city, maybe one country. But today's guest runs a company that has tested or operated in more than 60 cities across 13 countries and holds autonomous driving permits in eight of them. So my guest today is Tony Han co-founder and CEO of WeRide. He was a tenured professor in the US, then chief scientist for autonomous driving at Baidu, and in 2017 he left both to start WeRide. And today WeRide is a really well known company in the AV industry. It's a publicly traded company, dual listed on the Nasdaq and in Hong Kong. And today we will cover the business model, the Middle East, Europe, the numbers, and where the whole industry is going. So Tony, I'm so happy to have you on the show today. Tony Han: Hello Daniel, all the friends. I'm very happy to attend this interview and I feel like it's really a great pleasure to me to talk about autonomous driving and with Daniel and with all the friends. Daniel Abreu Marques: So Tony, maybe let's let's start with with you personally, because your CV has also a turn in that most people in this industry don't have. So you were already a tenured professor when many people would consider their careers complete. So why did you decide to leave academia and start WeRide? Tony Han: Okay, at that time, you know, I was promoted to a tenured professor and you know, according to the academia tradition in the US, you know, we have a luxury that is we have sabbatical leave. So basically you got paid, but you can visit wherever institute you want and do some collaboration. So I shoot out two emails. One is to Google X. At that time there was not Waymo So I shoot more emails out. One is to Google X, the other one is to Andrew Ng because Andrew just started the Baidu AI lab. You know, I send them emails basically meaning saying you know this is my background, I'm expert, in computer vision, I've been doing research for this for many years. And now I want to visit you guys seeking collaborative opportunities. And but my main interests are in autonomous driving and do you have such a position. and google x response is like you know they just say okay we are a secretive lab and we don't host visiting scholars and andrew replied by email to me saying okay we don't have autonomous driving projects so far but we have a very exciting deep learning project you want to have a visit? so i visit him and seems like you know at that time if i remember correctly that ai lab only had 10 people and then you know So we actually discussed about all the possibilities. Andrew is a great scholar and all the exciting projects. So I turned out, I said, okay, I would just want to enter the Baidu AI lab. But the exciting thing is like at that time there are so many great researchers attending the same time, like Dario Amodei the founder of Anthropic. We entered the Baidu AI lab on the exact same day. You know, HR people help us to register together and he turned out to be my desk mate. we worked together for all kinds of deep learning projects, quite fun. And I remember there was Saurabh who now started another company doing AI for mathematics. And you know, there's David Seetapun you know, who is famous for that David Seetapun conjection and also Bryan Catanzaro now the SVP of Nvidia doing computing and Adam Coates used to work at Apple cars And so so many, so many very good good researchers. Andrew is a great researcher. So we work on deep speech. We built up the first end-to-end the deep deep learning-based speech recognition. At that time, it beat the traditional hidden mark model-based speech recognition engine and beat human performance in a single sentence context. And our publication was published in ICM 2022. 2016. So it is still very well cited paper. And so we worked there, you know, we do we did research and worked there for two and a half years. well actually one and a half year, and then Baidu started autonomous driving project. So I basically urged Andrew and all the authorities saying, I got to go to work on this project. So I become the chief scientist of Baidu autonomous driving unit and then later, you know, at that time it was just like If you have ever worked for DeepMind or Google brain and Open AI, and then the investors will rush to you saying, okay, do you want to start your own company? So we start at the beginning, you know, we don't have business plan at all, but we raised like 57 million dollars. You know, at that time it was quite a significant amount of funding with just our resumes and as a seed money. So that's that's how we started. So of course you know if you want to start such exciting projects you have to focus all your time and resource on it. So I quit my tenure professorship to start this company. So far after nine years, yeah after nine years we have been deploying in I don't know how many cities around 40 to 50 cities and close to nine countries. So that's how we started. Daniel Abreu Marques: Yeah, super fascinating story. and staying maybe also with how you think for a moment. So because there's something else in your background that I think I almost never have seen in a CEO bio in this industry. So you have a long standing interest also in philosophy. So has that influenced the way you think about AI and autonomous driving? Tony Han: absolutely. but Daniel, thank you for your homework. You know I I like philosophy a lot, you know. I I just like philosophy really a lot. because you know in China, Asian China, we have Taoism, you know, start from Lao Tzu and we have Confucianism of course start from Confucius. And there are a lots of ideas you if you read in the literature is about Taoism it's really about the universe and there's lots of things that I feel like it may have some ideas shared with quantum mechanics so it really help you to boost your understanding and recognition to the next level and then you know My old read I admire Bertrand Russell a lot. You know, if you look at Bertrand Russell's writing and the history, right? The famous Russell paradox. And it's so simple but it's so beautiful, just with one sentence, it's made it displaying the third crisis of mathematics. So all of this just like you know when we read philosophy, when we read this literature, all of these things. I'm so proud to be a human, okay. It's it's humane understanding nature of our wisdom. Like you can experience so many things. and then I like mathematics. So all this just make me feel like I just simply don't have enough time. If I have enough time, I would rather sit, you know. I I want to sit in my office, sit in my room and read it all the time. like there's a famous book I'm not sure whether you read it or not. it's called The Grand Design by Stephen Hawking. When I read this kind of book it's just like so enjoyable. So lesson background, you know. And also why to perform driving the same thing, you know, why we philosophy and do technology simply help you to understand the problem deeper and maybe help you to solve the problem. Daniel Abreu Marques: Super interesting. I like that. So maybe let's now take that thinking and make it very concrete. Because on paper, WeRide doesn't look like one company, it sometimes looks like five because WeRide has rapidly expanded also its autonomous vehicle business across multiple segments and regions. So you have a portfolio of robotaxis, robobuses, robovans ADAS, robosweepers. And can you walk us through WeRide's core business models, particularly also how they differ across the different segments in terms of revenue generation and partnerships. Tony Han: So I think the core of WeRide is basically autonomous driving technology, okay, supported by the all AI infrastructure and state of art AI algorithms. but for different what we call the product lines, we have like different business models. For example, Robotaxi, right? We team up with Uber international market and we have our own car hailing app called WeRide Go app and also in China you can also hail a WeRide robotaxi through WeChat mini program. And so that is quite popular it's in a super app in China okay it's it's just like you have Facebook plus Twitter plus everything you know it's called WeChat in China and so for Robotaxis simple basically we are building robotic drivers and you know we can team up with car hailing platform so that, you know, we can reduce human labor by a big extent at the same time make the experience a much more pleasant experience. robotaxi simply drives smoother and safer and also you have your own privacy as a taxi passenger. So that part we just want to have the recurrening revenue for the taxi fare. We basically speed taxi fare with a car hailing platform or if we own fleet and then we can take the whole taxi fare. And so that's the business model for robotaxi. for our RoboBus, you know, currently to my best knowledge, we used to have several competitors, but now we think like we are the only company that can deliver full autonomous RoboBus all over the world. And today if in the public road you find a full driverless robobus you even don't need to think about which company it's from. It's for sure from WeRide because I don't see any other company have that driverless robobus running on public road. and our robobus has been deployed in so many cities. so many European cities And I think it's pretty suitable for European market, because the public transportation costs lots of money and the bus drivers including their welfare is quite expensive. Many small size city, towns, they even feel very under huge pressure to support effective or efficient public transportation system. We believe that is where AI can help, autonomous driving can help and I'm so proud WeRide is the only company that currently produce robobus that can run full driverlessly on public roads. And then comes to our robosweeper, you know, we have this kind of autonomous driving robosweeper and which can clean the street and do the sanitation business for the city all the time. And in our headquarter where it's located, the whole island actually. It's like a future island, you know, every street, every road was cleaned by our robotic sweeper. So we also have robovan okay. This is just doing logistics and robovan has been adopted by many logistics giants, including Jingdong. and we also have ADAS system, you know, we actually we won six consecutive championships in China urban intelligence driving challenge. We beat many, many renowned brand, all trend brands except Tesla, because Tesla haven't been active in China fully activated in China. But by the fourth quarter I heard the news may come that they would Deliver a full-fledged FSD in China, and we look forward to do head-to-head comparison with Tesla with FSD. But ADAS in the business model is also quite simple. We just charge license fee to the automobile company. And when they sell the car in China, currently when they sell your car with the ADAS functionality, they just charge you once. Basically, it's included in the whole car price, and you can use the ADAS all. time it's not like in US you have to pay $99 for your FSD subscription mode. In China it's not that way. But we look forward if one day like every competitor is trying to follow the Tesla's approach and adopt the subscription mode and we can also start to charge fees to our customer. So that's a major business model for our five different product lines. may sound complicated but actually here I want to quote lao tzu okay Da Dao Zhi Jian in Chinese that means like if I try to translate to English Da Dao Zhi Jian means the grand principle in the universe is simple. Okay. So to us, you know, this looks complicated, but then the core is so simple. Autonomous driving technology. And then it's this autonomous driving technology as a chunk and it's branch out different products, but they all share the same computational platform, same AI infrastructure, same data labeling, same technology, because if you have your autonomous driving car running on a public roads you have to be able to detect signal light and the info from the Counting numbers and you have to take care of cyclists. Sometimes you have to take care of carriage jaywalkers, road range, accidents caused by other cars. All of these things you have to be able to handle no matter you are running a RoboSweeper or a RoboBus or a robotaxi. So by doing this way, we boost our efficiency. And that's the core idea of WeRide. And I'm very confident to tell you in this world there's only one company adopts this kind of strategy that is WeRide Tesla is doing ADAS and they haven't touched RoboTaxi yet, you know, they haven't got like a full driverless robotaxi operation. Waymo never done ADAS and but WeRide have done RoboBus, RoboSweeper, RoboTaxi, ADAS, all of this together. So that part I'm proud of it. Daniel Abreu Marques: And and what prompted you to diversify into those multiple product lines and maybe also how do you decide which use cases or vehicle categories to prioritize in a given market? Tony Han: I think you know to us like we always think one question before we branch out a business unit. That is that based on our core technology. For example, we didn't do Robotruck, you know, that 40-ton big one, although Waymo and also our other competitors they do Robotruck, but currently we we don't do it because Robotruck they only drive on the highway and their control and their planning method is kind quite different from the the Sedan and SUVs we are used to. So if this one cannot share the technology and like we don't do the robotic mining truck because it's basically running the minefield and it's in unstructured road they're different so we currently we don't do this kind of business because we want to build up one set of algorithms and apply to all applications that's when we violate our basic principles so we don't want to do it but once you know if we can use just the same idea and can do a very small change small adaptation and then we can use it. Of course we will do it because the variety of business Help you to build up your robustness. You know, you know why we want to have a diversified society? So the diversity actually boosts the robustness of the whole human society. Suppose we only have one race, only one belief, only one thing, you may be correct for a certain shorter period. But what if the environment changes? Previous advantages become disadvantages. And then the whole human society will be in a big danger. But if you have diversified society, we have tolerance and then you know if as long as human beings there is some nature inside is we tend to help each other. So if the human society is tolerant society and help to each other Today you think some people are weaker and you are taken care of. When the environment changes, they may be become a strong people and they are coming back to help you. So it's the same thing applies to our business. So we become a profitable company. At the same time, we have very robust business because there are so many companies used to be giant, but when technology evolved and because their business is so simple and they die off extinction. I can go with the names like a Kodak, you know I visit Kodak's headquarter Rochester, it looks like a history. Okay. So we don't want to be that kind of company. We want to be a company like continuing using our technology to branch out and continue evolution like Apple and Amazon. So that's the philosophy. Daniel Abreu Marques: Makes sense and maybe let's let's also talk about what the the fleet is actually made of. So you reported roughly 3400 level four vehicles at the end of July, and more than 1800 of those are robotaxis. So robotaxis a little more than than half of the fleet. The rest is then the robobuses, robo sweepers, and also robovans. But what I find interesting is also the movement. So robotaxis grew by about five hundred vehicles since your latest April release. And the total fleet grew by roughly 600, which means the large majority of every new vehicle you put on the road last quarter was a robotaxi. So how do you actually make that call when you have one more vehicle to deploy? What makes it a robotaxi rather than a robobus or a sweeper? And also where do you want that mix to be in three years? So does the robotaxi keep taking share or will the other lines catch up? Tony Han: I think it's mainly depends on orders, okay. for us, you know, I don't want to deliberately change like say we plan like the ratio is like sixty percent robotaxi, twenty percent robobus, fifteen percent robovan five percent robot sweeper. We don't plan like this, okay. we are listening to listening to the demand from the market. And therefore, like what is the percentage and the numbers, basically I regard them as the result of function, you know, the parameters, the inputs, the variables are actually from market. And what we don't do is what we want to respond to the needs of the market quickly. And so it's a dependent variable. so to us like it's we just want to find out all the opportunities. Like for European market, right? European market like our customers just love RoboBus. They they just say, Okay, of course we do need robotaxi. They said don't get us wrong. We do need robotaxi, but we want to foster public transportation. And if you can make Robo bus available and in many streets, it actually helps us To release the burden of the municipal government. And we want to foster this first. And some some cities just say, okay, like for example, in the Mid East market, they say, okay, you know, probably I don't want to wait for a bus in the bus station at a temperature of like 45 Celsius degree you guys are in Europe you know the formidable summer feels like right I was in Abu Dhabi and Dubai During the summer, stand outside without air conditioner. Just five minutes can dry you out. So with this kind of environment, of course I would want a RoboTaxi because I know while robotaxi arrives, I can jump into a RoboTaxi with AC. I don't want to wait in a bus station. So these have different needs. What we want to do is using our advanced technology to cope with their needs. For us, we have one philosophy and our believe is technology should make people live better, live happier instead of making people jobless. That's our goal. So that's how we develop our technology, how to develop our how to design our products and services. Daniel Abreu Marques: And when you look at that fleet, who who owns and and operates the vehicle in your deployments, are there also differences in the different segments, also in regards to maybe China versus international business? Tony Han: Yeah, it depends on the markets and depends on the cities actually depends on the business. I want to say this is a global market and we are all human beings are booming in this global economy. we don't want to build up companies that just suck the profit and run away and leave the local community not developed. We want bring the prosperity economics to the local community. So one one principle weride is we want to do local hire if possible we want to as much as possible to collaborate with the local collaborators. work need to be done by the local workers. And we want to have just local workers local jobs available. but in China, like you know, sometimes you know we can't find very efficient local collaborations and we are Chinese company and sometimes we own a fleet. but for the international market we use the asset-light approach. And also depends on the business, right? robobus some municipal cities just say okay we will maintain a bus and we want to run a bus because it's public transportation, so we own a bus. That's very natural. So it all depends on the the business and all depends on the market and the geography position. Daniel Abreu Marques: Absolutely makes sense. And when we look at one of those regions, I think one is standing out. So like the Middle East, they're definitely doubled in a single quarter from roughly 200 to 400 vehicles. So let's get a bit more deep here because I think it's also one of the most important proof points of your international business model. So maybe let's start with your Abu Dhabi operations. Can you walk us through the three way structure? So weride provides technology vehicles and Uber provides the platform. Tawasul transport handles fleet operations, and then who decides how many vehicles run on a given day, who controls pricing, and who also bears the cost if a vehicle is offline, for example, for maintenance. Tony Han: You know, the number of vehicles, the deployment plan is actually co-decided by Uber and the WeRide Management Team. We have a plan decided we discuss quite frequently. We have management level meeting quite often. And so far our goal is trying to as soon as possible to deploy driverless robotaxi at hundreds level because Dubai and Abu Dhabi we recently actually you know we have different zones. We start from independent zones in the city and then we directly just like grow our zones and finally the zones start to connect to each other and cover the whole city. That's the plan. And we gradually add at depends on the needs of different zones and we gradually add at number of cars. And the nice thing about Uber is like Uber is very popular in Abu Dhabi and Dubai market, and it's especially For international travelers. When I got to Dubai, so suppose you visit Dubai, when you get out of the airport. And of course you can hail a local taxi, but it's kind of quite expensive, and sometimes you probably need to speak Arab in order to hail a taxi. But for Uber it's universal, right? It's you it's an app in your phone and you just input address to get Uber. so depends on the needs, like there are some there's a limited amount of Uber driver. available in certain areas and we can quickly fill the gap with the robot taxi as we gradually grow. So that number and deployment plan is actually decided together by WeRide and Uber Daniel Abreu Marques: let's flip the side and look at also the riders. I think in September 2025, Uber launched also a dedicated autonomous category in the Abu Dhabi app. so that was a global first, as far as I know. And after that launch, weekly orders also reportedly rose about 20%. So what surprised you about the rider behavior once people could specifically choose a robotaxi rather than being randomly matched to one. Tony Han: I think you know, we of course want to foster the adoption of autonomous driving. But as ordinary user or for the first time they want try it. They may want to wait for twenty minutes to try robotaxi. But after you have tried robotaxi and next time if you are you know, we all use Uber App, right? my way is like as long as the price is acceptable, whichever category comes first, the the list of ETA estimated time of arrival my waiting time the shortest waiting time category is the one I picked up right so basically we just distribute our robotaxi there and when there's order there a Uber will just dispatch the orders and find out the nearest car and they assign that to the customer. So that's the way to do it. For me, I think the user experience and the safety. Actually safety is most important and then the user experience. We want to put our priority towards safety and the and user experience. And then we start to think about how can we booster or foster the adoption of autonomous driving robotaxi Daniel Abreu Marques: but I think it's also quite an cool perk because when I read kind of experience reports from autonomous trips, everyone says, Okay, I tried a a robotaxi and I'm totally in love and I just want a robotaxi. So they don't want to have this randomly assigned because they love just the experience, is smooth, is it's like private experience. So it's a good way to to give that value to the customer. Tony Han: Yeah, yeah. Because it's you know, we use Farizon GXR for our Mid East market, it's a quite spacious van you know. You basically have a business class experience in your taxi. So it's very spacious. You can you can halfly lie down. Yeah, of course you cannot lie down flat because it's not safe. Your seatbelt won't work if you lie down completely flat. But you can halfly lie down on like 140 degrees and it's really relaxing. So you can just enjoy and because there's no driver in front of you. One of the other things if you want if if you are a tourist, you want to see around. So basically you can look 360 degrees all the time and there's no no drivers blocking your view because we all always know the front side view is the best. And previously, you know, you either sit in the passenger side, or if you see the back row and the and your 50% of view was blocked by the driver. But now all the views available and you can lie down. It's just like great experience. So you know, we want to add more robotaxis in Abu Dhabi and Dubai so that more and more people can experience this kind of great feeling and and hopefully we can help them to reduce the waiting time because you know I'm a busy guy and to me waiting time is essential. If if some service can save me five minutes, I would rather you know pay more to to get that car. Daniel Abreu Marques: Well, I can imagine that. So maybe let's move to my home continent. because Europe is also where a lot of my listeners sit. So and it's also where you have been very busy this year. So you have announced a number of Tony Han: Mm-hmm. Mm-hmm. Daniel Abreu Marques: achievements in the European market. So in the first half of this year, I think it it includes launches in Spain, in Slovakia, in and Denmark. So can you give us more detail on these recent European projects? Tony Han: European market is one of the most important markets for weride and we team up with Uber and Uber is quite popular for the European market. so to us we have a plan of different cities, you know. we will let you know once we are ready to announce. And we are actually working on build up the the technology for different countries. We have to prepare for all of this kind of technology to make sure they are kind of consistent with the local environment. that costs us lots of research resource. At the same time, you know, we have to solve all the local hire issues. so we have to cope with all of This difference. You know, we like a diversified society, but we have to admit the difference and we need to cope with that. So there are lots of issues we have to take care of. in some markets people need to use public transportation. And then we have to make sure be very careful where we pick up the drop off our passengers and in some cities you are allowed to pick up and drop off passengers in the bus station. Okay that's shareable. But some cities absolutely no even the bus station is empty. You are simply not allowed to to drop off and pick up passengers in a bus station. All of this subtle difference you have to take care of Daniel Abreu Marques: Yeah. And you you already mentioned that these are all different countries, all different environments. which makes me want to understand also maybe the selection logic behind it. So what made the launch markets your targets in the European region? Tony Han: Very good question. there are several factors. so first of all I think the population and the number of the taxi available in that city is one important factor. And also the local regulatory authorities openness to autonomous driving technology is a consideration. For example, you know, based on our previous conversation currently for the European market, only Netherlands allow the trial version of FSD, currently it was a big surprise when I actually I heard this thing in Stuttgart when I do the test drive with my Bosch colleague. I said, why don't we try Tesla? He said, Tony, you probably didn't notice that FSD is not allowed to activate in Germany. I said, what? Not allowed? it's it's really hard for me to imagine because I feel like look, Germany, France, Spain, they are very similar to US. I think whatever you can do in US, you can do in European countries. That's my feeling. But that's just like FSD cannot activate so depends on the the local local regulatory authorities acceptance to autonomous driving technology and also depending on the needs, you know, as I mentioned, you know, we want to fill the gap between the society needs and the labor supply. If the city population is relatively young and the people need that job as a taxi driver, we will halt our deployment because to me I have this Again, technology is to make human live better and happier. It's not used to destroy working class life, take their job away. No, that's that's not at least that's not my belief. I want technology to kill cancers. I don't want technology to take away nurse jobs. So that's our factors we are considering. Daniel Abreu Marques: Makes sense. And maybe that brings me to a selfish question. And I know maybe you cannot answer all of it, but someone has to try it at least. So of these fifteen cities, you want to launch with the Uber Weride expansion plan. So how many are in Europe? And for me as a german would be also really interesting to know if Germany is also on the table and maybe even Stuttgart, where I'm located. Tony Han: I personally like Germany a lot and as you mentioned, you know, I'm I'm a lover of philosophy and I'm a fan of Ludwig Wittgenstein. Actually in our recent large language model, we name it WITT you know because in for Chinese it's so hard even for me to pronounce this long German name Wittgenstein but you know I want to show our solution to Ludwig Wittgenstein I like Germany a lot. I like the music, I like the architect. So always if possible we want to deploy in Frankfurt Stuttgart so but all depends on the plan. For me, you know, I enjoy being a tourist in Germany. I would never get bothered if I can spend two months in Germany. So many great cities, great places, great towns to visit. I like beer, I somehow like German food a lot, like like the Daniel Abreu Marques: Mm-hmm. Tony Han: beer. So so if possible we want to deploy in Germany as fast as possible. Daniel Abreu Marques: Sounds great. I keep watching. Maybe let's move from the map now to the P&L for a moment, because there's also a number that I think maybe also gets misread constantly. So WeRide announced that they're unit economics break even. And can you give us some context on what that means and in which regions you are also break even or close to break even? Tony Han: first of all you know, according to our calculation and the numbers, in Mid East we are getting close to regional break even. That is, you count all the operational cost and all the local management cost and the rentings you can break even. You need break even close to even in China because in China taxi fare is relatively low. we so far, you know, each taxi can get average order like ranging from nineteen to twenty-one, they can afford an amortization and electricity and the parking fees for the car. But for R and D forget about it. We still need several years to make sure, maybe one or two years to make sure R & D fees we distribute to each car and then we can break even But for Europe, I think from day one we can become profitable is because of the fare is relatively high. And you know our business model is like we have a share the fare with Uber and with car hailing platform. so that part is really heavy depends on the fare, the taxi fare of the fleet of that specific city. So of course we if possible we would like to deploy all or robot taxi in some cities with relatively higher taxi fare so that's the explanation so but most of the time think about it the cost of our cars is like ranging from let's be trying to convert to to the to the dollar, you know. in my brain I always count it in RMB It's like roughly You know, everything hardware costs kind of fifty to sixty thousand US dollar. But today, you know, if you want to get a Tesla in US, including tax, you know, I got a model Y it costs me more than fifty-five thousand US dollars. Maybe more expensive in Europe. So I think the price is relatively low Daniel Abreu Marques: And maybe now let's zoom out a bit from WeRide for the last part of this conversation and talk about the industry in general because a lot of my listeners are also investor strategy people trying to understand where this is going. and many people see autonomous driving as one of the most demanding real-world applications of AI. So What does it take for AI to move from generating content to making decisions safely in the physical world? And also how do the requirements for physical AI differ from those for digital AI, so to speak? Tony Han: Yeah, that's a great question. I think one day I wrote a sentence, I I'm quite proud I wrote that sentence. I said hallucination in digital AI may be entertaining, but it is fatal in physical AI. So that means like for physical AI, in robotaxi humanoid robot. Reliability is the key issue, the safety. Okay, you simply cannot make any mistakes. If I ask ten thousand questions to Chat GPT, they give me nine thousand nine hundred and ninety-nine questions, correct answer, and with one answer, you know, some irrelevant answer, maybe entertaining, but made a mistake. You just say, okay, let me just change my rephrase my questions and ask again. But think about for RoboTaxi, if I give you nine thousand nine hundred and ninety-nine trips safely, but there's one trip the the car hits the pole and what do you think they said forget about it I will drive myself I will never risk this kind of my life over convenience so that's the thing so that's why we try to adopt what we can learn from the from the aviation industry like the redundancy and the design structure all of this. That's why you know there are so many countries can build cars. But for the large airplanes, it's only Germany and France they can build the the Airbus. And then the other one is Boeing from US and now China can can build some and it's relatively mid size airplane So that's the difference. So we have to make sure we push to the limit about safety, about we use redundancy as a measure to to to add another layer of safety. it's not easy. but we have we will we have to do it. So physical AI is my feeling is much more challenging than what we can achieve in digital AI because simply tolerance for error is minimal. So we have to be very, very careful. Daniel Abreu Marques: so most of your competitors or a lot of your competitors are climbing, so they sell like level two and L2++ into mass-produced cars today, so for the revenue. Tony Han: Mm. Daniel Abreu Marques: and the plan is then to climb from there to level four and WeRide went More less the other way around. So you started at level four, driver out, and then took that stack down into ADAS. And it's working. As you mentioned, your ADAS delivered about 30,000 Tony Han: yeah. Daniel Abreu Marques: units last quarter, I think. and you won six straight championship at the China urban driving competition. So why do you Tony Han: Yeah. Yeah. Daniel Abreu Marques: believe that your direction is the right one? And more fundamentally, maybe also do you see level two and level four as two points on the same path or as two generally different problems? Tony Han: So you know, frankly speaking, I think there are so basically the two paths from L4 graduate to L2++ is also with lots of challenge, but the the two directions, the challenges are different. From for a L4 company go to L2++, there's a lot of engineering details, verification details and how do you do the optimization. We are facing different difficulties like It's just like I will call it from L4 to L2 is called projection. Basically, you are suppose you are a four-dimensional creature, you want to understand what two-dimensional world looks like. So basically it's a projection from high dimension to low dimension. From low dimension to high dimension, it's called another inverse procedure called lifting. You have to lift. Think about we are three three-dimensional creatures. We want to imagine four-dimensional things. You have to goes through another layer of difficulties. So but which one is more difficult? It depends on the company. Depends on how you solve the problem. So here that's why I call it, I propose a kind of thing I call the qualification process. What do I mean by qualification process? The qualification process means like for an L4 company like Weride if you want to claim you're capable of doing ADAS, you have to make sure your system not just simply for demo purpose You have to make sure your system can be put into mass production car and then the users, the customers buying your products. Now we have achieved that, right? GAC N60, Cherry Exceed and GAC I60. And we have so many car OEMs I will I let you know once we can announce it. Now we are cars already selling in hundreds of thousands level. Hundreds of thousands level, people using our ADAS and using them for for commute every day. So that means weride has passed the qualification process. We can claim our ourselves as L4 company and L2 because our RoboTaxi is available everywhere. People can hail our RoboTaxi, driverless robot taxi in in Beijing, in Guangdong, Abu Dhabi in Dubai, everywhere. Now for the L2++ or ADAS company, their qualification process is like this. You should have at least 100 cars. make it available to public so that everybody can test your service, can run your service, and at least make it available for one hundred and eighty days, half year without any severe accidents or halted by the legal authority. I proposed this qualification process probably one year ago and weride has passed this qualification process for a long time, but nobody, no companies I mentioned Have passed the other reverse qualification process. the competitors you mentioned, where are their robotaxi? Where are their driverless robotaxi? Can I hail them? No. It's just like they show you some so-called robotaxi service with a safety driver sitting behind the steering wheel, which makes the whole business model doesn't make sense at all. You know, if I have a safety officer sit there, I can drive the car. Why do I need you to do the robotaxi? So unless they have robotaxi service available to public for half a year without any problem at a magnitude at a fleet level of one hundred. I wouldn't qualify them as a robotaxi company. Daniel Abreu Marques: Makes sense. And and maybe now let's also so widen that field out to beyond the level two players. So competition and autonomous driving is intensifying. So with both tech companies, traditional automakers making bold moves and claims. how do you see the competitive landscape evolving and also what will separate the long term winners from the rest? Tony Han: I think you know no that's the beauty of the market economy, right? I think the winner and survivor of this very intensive competition is like whoever can finally make their company profitable and get enough market share will be the winner. Just like the bloody competition in the electric car industry, right? Previously, like back to five years ago, nobody thinks BYD is a great company. one year ago, nobody thinks Leap Motor become a strong player. But now with their market share and with their profit, they become very, very important. Weride ten years ago or nine years ago, we are just one of thirty startups doing autonomous driving. But today WeRide to become one of three major players for robotaxi globally. And for ADAS. Two years ago nobody thinks we are a player, but now we got six consecutive consecutive championships and our our ADAS system are deployed at hundreds of thousands level, then that's tells everything. So I I do believe results manifest themselves. Daniel Abreu Marques: Tony, this this was a really packed conversation and we covered a lot of ground today. And I have to say it is Tony Han: Mm-hmm. Daniel Abreu Marques: generally an honor to have you on the show. you have been in this industry since before most people believed it was a real business, and you built one of the very few companies that can actually point at driverless vehicles operating on public roads. So I want to thank you so much for taking the time today and also thank you for being this open on the details. That's not something every CEO does. So You have a standing invite to come back on the show, especially when the European expansion lists gets longer, and I keep asking for Stuttgart and yeah. Tony, thanks again. Tony Han: Maybe we can get maybe we can get together to have coffee or brezel together at Stuttgart One thing actually during our interview, one fun part is Even today, we cannot prevent internet or disconnect the hiccup from happening. Now you see the great, great challenge or the formidable challenge we are facing for physical AI. Think about in your robotaxi if we have this kind of hiccup in our robotaxi what will happen? So the reliability is a great, great challenge. But that but we are so proud we can handle this kind of formidable challenge and we are so proud about our safety record. And also it's a great pleasure for me to attend your podcast. Daniel Abreu Marques: Thank you, Tony. Tony Han: Mm. Thank you. My great pleasure.