 questions. Does the card right now need to connect to the Internet when they do the self-driving? Does the card need to connect to the Internet? Yeah, we try not to get connected because then there's the security issue. People can try to hack into our car and try to do some funny things. Yeah, so eventually, well actually that is not entirely correct. So there is some form of connection. For example, what we used to say to tell the card where to go when we do our testing is actually through our, you know, a smartphone app. So we're actually developing the whole, so we made the decision of not, so the one thing that we do not work on is the car. So many other startups in this domain are actually developing their own new concept vehicle, you know, the design for special car owners. We decided we don't want to do that. So yeah, it's great, but it takes a lot of effort. And honestly, I don't know the first thing about designing a car. I know a thing or two about running a software for it. On the other hand, there are many people with automotive manufacturers who actually know how to make a car. So we buy all of that. But aside from that, and we buy the sensors and we buy computers and, you know, components, we decided we are essentially developing the whole end-to-end system. We are working on the in-car technology, you know, just to make it that way itself. We're working on the backend technology in the sense of, you know, how to manage the fleet of, say, potentially hundreds of thousands of vehicles. And we're also working on the mobile app for the booking. So to get back to your question, we do the booking. For that, you need, you know, a form of wireless connection. But at this point, it's the only point where the car communicates with the outside. The main reason is really that of security, cybersecurity. We don't want people from the outside to hijack our car. Sorry, yeah. So I'd like to know, like, for your new autonomy, right? Is it, like, would I have to buy a new autonomy car? Or is it, like, a package where I can buy and install on my own kind of thing, like the strategy going forward? Needed one. So, as I mentioned before, I think that, you know, before, so we don't see this as a product that you sell to the consumer. It's not a consumer-level product, okay? You know, what I, what I, you know, try to convey in my presentation before is that I think that that is very, you know, will take a very long time for us to get to the point for many reasons. You know, if you buy something as a product that, you know, if I tell you, okay, so give me $10,000 and I will give you a kit that you can put on your car and your car will become autonomous. Oh, it's great. It is the $10,000 that you go home. You install the thing and you want to push a button and be able to drive anywhere, right? Doesn't matter if it's raining or, you know, you're driving, you know, some place, you know, some weird place where you have no maps or you know, things like that, right? Also, typically when you buy something, not everybody is able to do the, or they require troubleshooting, maintenance, calibrating the sensors, things like that. Also, and then the last thing, and you know, this is a big problem for me with the OEM path. You know, the OEM, they want to sell you the car, right? But when you buy this package, would you buy it if I told you that it costs $100,000? Well, probably there may be a few people who still buy it, right? But it's not really a mass, it divides the product for the mass market, right? So, since the OEM are thinking mass market, right? So, they try to get away a lot of, you know, try to do it in the cheapest possible way. And unfortunately, the cheapest possible way is also a very dangerous way. Okay, so that's why, you know, these kind of commercial autopilots that you can access today are kind of limited in their capabilities. On the other hand, for us, since I'm not trying to sell you our car, but what I will do is I will rent you our car for the time, you know, for the time necessary for you to go from A to B, okay? Then the financial proposition is not that of selling you a product, right? But it's that of selling your service. And then what you have to compare to is what would it cost you or what it cost me as the service provider to provide the same service with the current technology that is with the carbon-based lifeform behind the wheel, right? If you think of it, if I want to provide 24-7 service with a car with a human behind the wheel, then I need to say hire at least 3% How much does a taxi driver make here in Singapore? I don't know, but let's say about $40,000, $50,000 a year, multiplied by 3%, that's $150,000 a year. Now, even if I told you that my system costs $100,000, that's actually a bargain, right? So financial, safety, liability, all of these take a completely different meaning. If you switch from thinking of business as a product, consumer-level product for a service, service makes a lot more sense and it's much easier to watch. On the other hand, let me conclude, we are actually installing, converting vehicles, many different vehicles of different brands. So in a sense, from our point of view, we can essentially get a vehicle and then instrument the vehicle. So we are not really that peculiar or, you know, they're choosing about a particular vehicle. Seems like every regional manufacturer is coming up with their own autonomous car right now. So how are we going to be different from theirs? And why not just go buy Mercedes or anything else? It depends on how much you like, ending up under the trailer. They only have intelligent cruise control that can follow the car for 100 miles. So again, if you go back today, and again, you're misled in your appreciation of what the self-driving cars, autonomous cars are. So you are conflating everything into, oh, everybody's doing it. What you have to understand is that, what all the OEMs are doing, they're putting all these autonomous driving capabilities into their production vehicles, it's going up that path that they drew on their life. So these are vehicles that are not really autonomous. They treat you into believing that they are autonomous, but they require you to constantly pay attention. That's not what I would call an autonomous vehicle. The one thing that really is game-changing in this technology is when the car is able to drive itself with nobody on board. The rest is attention. And none of the OEMs are developed. It's not that they're not commercializing, they're not even developing the car that is able to drive itself without anybody on board. So what are these articles you read about? Are the OEMs having fabulous cars that are being lapsed around Nuremberg or all this sort of thing? There are articles out there. What are they about? About cars? I mean, they're having different production, but they actually are able to drive. I mean, unless the press has been slid off. Yeah, but, again, no. So a lot of these people have very big PR operations, like marketing and things, that they make you believe that they're doing something right. For example, when you have how they made a big thing, or at some point they went to some Formula One race track and they had a car going, making one lap around the track. I could do that in my lap 20 years ago. Maybe not with an Audi going at 150 km an hour, but my claim is that driving a race car on a race track. You know how wide a race track is? 18.2 meters. So if you actually look at the video inside the vehicle, even though the car is going at 150 km an hour, so there is not really challenging of that. And Mercedes did this down where they drove this path between two points in Germany. But all these things tend to be kind of like a corridor. They claim that they've been the first completely autonomous drive from coast to coast in the United States. It was not completely autonomous. I mean, they drove the car, they put it on the highway, put it in the lane, okay, point this way, and then they pushed the button and the car was driving. Up to the next point where they needed to exit the highway and then somebody took over and drove the highway. So this way of thinking, from the way I see it, the only other companies who are pursuing this really level 4 economy completely autonomous with nobody requiring more, people like Google, or companies like Google, open. So they're not really the same. I think that we have a question here. Where do you... Okay, so under the liberation of the internet, do you plan for vehicle to vehicle communication? Absolutely. However, what you have to be careful about is the following. Essentially, we do not want to rely on V2V, V2I, V2X. For the simple reason that it's a chicken and egg. Because typically if you rely on V2V, vehicle to vehicle communication, then anything that you do relies on the fact that other vehicles have the same technology. So then anything that you do depends on the rate of adoption of how much of a big share of the market you're capturing. And if you do something like that, again, it's a chicken and egg, so your system will never work unless a lot of people adopt it. But then people will not adopt it if their system is not working. So what we say is that we do not rely on V2V, we will use V2V to the extent it is available. So for example, another thing that happened, it's a California intelligent highway project, actually 20 years ago, was very successful. And they were able to demonstrate autonomous platooning on the highways in California. That's great. So then the governor and all the politicians went to see the demo. The demo was great. Let's do it. How much will it cost? Because the thing was that this system was actually relying on some magnetic sensors that had to put in the pavement, the pavement on the highway. So good luck doing that across the whole California. So again, that's why I don't want to rely on V2I because I don't want V2I to be equal to infrastructure because I don't want to rely on the government installing the devices that went in at the roadside. Because if I go to the end, yeah, I had this offensive technology for townspeople. But you need to install this device at every single intersection that would never work. So I think that the benefit of the conflict of development is one that doesn't require anything. The only infrastructure that we require is the existence of roads. But roads are a repeat. And you mentioned that you don't design cars. So are you selling the software to car manufacturers? So this is part of our development business strategy. So at this point, I would rather buy the cars from the companies and keep control. So essentially, I think that as a company, you want to be as high as you can on the roadside. So we are trying to maintain control of everything. We are considering licensing deals, and it is not clear. So we are developing before the conflict to be able to operate the service. It's unclear at this point if we will be ourselves offering the service, or maybe we will be licensing the whole thing to somebody else. But this is still a thing that I'm trying to figure out. Can you follow up my question? Hi. So my question is, from now to 2018, there's two years time. So my question is, why is it two years? Because from what I see on the demo, everything, it seems that you have all the necessary technology. So do you need any other major breakthroughs in the R&D side or will the next two years be spent on testing and finding the existing technology? Right. So I wouldn't say it's a certain major breakthrough, but there are a lot of corner cases. And you must be, it's fairly easy to be able to handle 80%, right? But you want to handle 90, 99. So essentially we are just covering that tail of issues that we cover all corner cases. So I think that was a question here. Okay. My question is still surrounding the connectivity and what we call between stuff is that, I'll notice that, is there any other sensing other than visual on the car? Because what I've noticed just now on the demo is that the car can see against the visual. Our main sensor is actually a laser sensor. So the main sensor, and going back to your question, please, what many, for example, of the autopilots or automation features that you find in production vehicles typically do not have LiDAR because it's kind of expensive. But we think it's necessary in order to be able to drive safely, okay? So our main sensor is LiDAR. We also have video cameras. We have radars. We have GPS and accelerometers in actual sensors and so forth. But the main sensor is LiDAR with video cameras. Thanks for sharing. I have two questions. The first one is about technical. I just wonder how different is your automotive car from automated driving vehicles that are already in this main course or in the factory model of this model? I want to say a mobility as a risk concept. Is it your ultimate goal is to develop light transportation and more operating systems? That. Regarding what I call the image units, you know, typically they operate for industrial automation. They operate in somewhat controlled, close environments. They typically operate on fixed tax. So if you go to the port, they actually have... Actually, they told us because what they're using now... I don't remember exactly, but they're using some tax or some fiduciary marker or something that is actually embedded in the pavement. You know, going by the same thing. It's similar to what they're doing in California. And actually, you know, because they have these heavy vehicles that moving containers were back and forth, they actually had to resurface the pavement every year or two. And when they do that, they had to reinstall all these markers. So this is a hugely expensive operation for them. So they were telling us and others to have alternate solutions. So the main difference with all these industrial vehicles and industrial automation vehicles is that these vehicles are just going back and forth on fixed tax. Whereas we don't do that. So we are working in an environment that is not as controlled. There are many more unexpected things that can happen. And by the way, I cannot lay down markers for Singapore. So it's a completely different picture. You were asking about the business model. I don't really see that, you know, certainly you can imagine that there is a central operator, a big brother that manages central all cars in Singapore. I don't really see that happening, honestly. Probably what will happen is that will be and you see it already with taxi companies, right? So maybe you do have several taxi companies that compete. Maybe there's one that is bigger than others. Then you have other entrances. You have Uber now. You have Grab that is doing other things. So in that, you know, my expectation is that this would be something that would be more left to in a situation to see not what we have today where the government is regulating but then private providers should provide the service. Okay, from a consumer point of view, where the occasion where the human needs to take intervention. For example, you know, as a breakdown we find that I need to take a different route, a different route. Or something I don't know, something that, you know, requires the human to make the transition because the car is inside of whatever. Yeah, we haven't decided exactly how that kind of things will work. You see, I mean, the person has a good driver, a good life. Right, yeah. So, yes, that would be, you know, for me, for example, it's a very different thing to say. What I say, you know, what I say, the danger zone, the red things are. When you require a human to take over in order for the car to be safe. On the other hand, I don't think that it's anything wrong in allowing the human operator to take over control. What you don't want to do, if you allow that, what you don't want to do is the reverse. Right, so, I don't want the human operator to just drive around, then hit a button and say, automation, we are in charge. Right, so, because now we are in the opposite situations, which also expect, you know, please do not expect. Right, so, you have to be a little bit careful in the switching over. So, for example, for me, once you take over control, then you are in charge. Until you stop, for example, and the car tells you I'm ready to take over. So, these are all human interface issues that we are still looking at. Now something that we are working on is, you can imagine that today, LTA requires us to have a safety driver on board. Eventually, we will want to remove the safety driver and maybe have the safety driver, instead of having safety driver on board, we may want to have a safety driver off board. Somebody that is monitoring, you know, the car from some operation center or something, control center. Eventually, you know, initially this would be one remote safety driver per car. Eventually, this would be maybe one remote safety driver monitoring five cars or 12 cars. So, these are all things that are yet to come and still look at possible ways that we can handle. There was a reason that basically the others, some of them are using the cameras and the camera sensor because the laser sensor is too expensive for it to be commercially viable. So, going forward, do you see that the price of the laser sensor will drop to a level? So, for me, this commercial variability, the cost of the sensor, for us, when we design our car, we really don't care about the cost of the sensor. Because in any case, it's negligible with respect to the cost of a person performing the same function. So, keep that in mind. That said, of course, nobody likes to pay more than they have to. So, what I see happening is that nowadays we are using LiDAR, still essential. We are working hard on trying to use more vision, computer vision. My belief is that computer vision is not ready yet. So, there are two opposite trends. On one hand, LiDARs are becoming cheaper and cheaper and cheaper. On the other hand, computer vision algorithms are becoming better and better and better. So, if you ask me, ten years from now, what will be the configuration on your car? I don't know, right? Because either, maybe now we're getting much better computer vision, or LiDARs are getting so cheap that we don't just buy 20 LiDARs and it will be done. So, I don't know. So, this is something that is evolving. I don't think we'll completely get rid of the LiDAR because I think a lot of this is complementary. LiDAR, the camera usually doesn't work well in many situations. When there's a sunlight opposing, then it's completely blind. But in that case, LiDAR is still working very well. In contrast, LiDAR doesn't see certain objects like black cars sometimes just disappear from LiDAR. In that case, vision can be used to complement this function. So, I think of them as complementing functionality. We have to pick one or the other. A second question is, one challenge is obviously to think about what are the unexpected situations that may happen. But I guess you can design for what you know that might happen. What if there are situations that you don't infuse and then that it just happened. And because as human brain you react to the unexpected things. But for cars, if it's on program in their system, do you use any machine learning or how do you tackle this? We are doing some of that. For me, machine learning, deep learning, or whatever, it's not a... by magic, a solution to everything. So, we're trying to figure out what are the good ways of using these technologies. I think there are a lot of misconception. Misconception in the research and in the community about these things. First of all, the easy answer to what you're asking is that our car tends to be to try to behave in a very defensive way. One of the reporters that was at the event, at the demo that we did in May, she wrote in an article, I took a ride in there, Clumsy but safe. So, our car, by design, is programmed to be very defensive. If there is a situation that it's not clear how to handle, it tends to slow down and stop. It can be unnecessarily so. But I rather stop than hit somebody. So, in a sense, our objective in the years to come is to maintain safety and make the car less clumsy. But I wouldn't make the car very efficient and smooth and at the cost of compromising safety. So that's the thing. Then there is something that is very important to me is some of my colleagues and other people who work in this area they claim that well, there are too many rules of the road, so it's impossible to code all of them up. So the only thing that we can do is learn from students. That's below me. Actually, the rules of the road, there are not too many. In fact, I don't know the driving age here in Singapore, but in the US, anybody at 16-year-old, they can go to the Department of Motor Vehicle, get a booklet with a bunch of rules. A couple of weeks later, take the driving test, the written test. So actually, the number of rules of the road is very small. Actually, not all of them are in our cars. So it's actually the number of rules of the road is very small. Now, what is very large is that these rules actually interact with one another and with the possible realization of the environment. Okay? Now, if I had to learn what is the right reaction to all possible combinations of rules and position of other cars, pedestrian lane markings and things like that, I mean, I need to look at gazillions of miles, right? Whereas if we have an efficient way of combining the rules that are actually not that many, I think it's a much better approach and that is actually the approach that we are favoring. So that's one thing that I learned from the urban challenge. Back then, I think what many teams did in terms of guaranteeing that the traffic rule is satisfied is we caught up this humongous finite state machine that, at some point, nobody understands anyone in that finite state machine anymore. But then we started doing more and more research on how to automatically construct this finite state machine so that by construction, they are correct, right? So there's a way to just convert the traffic rules that is written mathematically into something that can govern the behaviors of the car. And I think that's something that we think is important because if you just follow purely machine learning, then essentially you don't take the advantage of having this clearly written rules, right? And I think that this is why you could claim that humans are pretty good at learning, right? That's how we figure out a lot of things, right? So you could imagine that okay, in order to get the driver's license, just drive around with your dad or your mom or somebody supervising you for some time and then you know how to drive, you know, just take the driving test. That's not how we do it. The way we do it is we make sure that you know, driving students, they first have to pass the written test. We have to make sure that they know their piece. There are not many, but you need to know them. And then you can demonstrate your proficiency driving, right? So I think that there is no point in trying to learn from experience things that are actually written down specifically, right? And this literature, I mean deep learning is great. Machine learning is fantastic, right? I think that people are very often are abusive, right? And people doing research or we're using deep learning to discover the laws of nature and then they turn the clock for days and days and I'll have my machine learning algorithm figure out that 2 plus 2 is equal to 4. Really? Right? I saw some papers on machine learning to figure out traffic patterns. Oh, we found out that in the morning all the roads leading to the business district are jammed and the afternoon they're jammed in the other way. Really? So I think that there are other things that we know and they're written specifically as roads and by the way, these roads are written by us, right? So we know that, because we know that. So it's better to just cover them up, not that many have intention to do that. Okay. Do you think on that, right? And also the question just now, I think driving I think what part of the real thing there in Singapore is not the use of other drivers on the road especially like expressways where you merge in between the signalling and motorcycle lanes and all that. So how do you intend to do it? Because it's a very different environment from one road where it's slow and there's a lot of traffic. Absolutely. That's a good point. So what we try to do is as part of our system we also try to see other people and other cars and pedestrians we also try to figure out what is in the devil next that those guys could do next, right? And that is a way why this is one of the reasons why our car tend to be clumsy. Because if I see you that you're approaching maybe the curve I will stop for you, right? Because I don't know if you're going to jump into the road or not, right? Before I proceed I want to make sure that actually not doing anything stupid like that, right? Or nothing unexpected. And in a sense the better and better we try to develop our models of what other people are doing the less classy we will get. Something else that I think is very important for us is and this is the reason why we did the public prior in the garden. And he said that I think it's important for us to push the technology to the public to educate the public and the authorities on what this technology can do what it cannot do and what kind of benefit they can get. Because what I found is that the population, general population can be divided into two groups. One group is the naysayers. So people say this thing is never going to work. I'm never going to touch this. Okay fine, I can work on convincing that so the burden is on me. But then I will claim that the other part of the population this is the believers this is actually more dangerous. These people think oh this is a computer control car. So this car cannot have accidents. No matter what the laws of physics are saying. And actually have otherwise intelligent people do incredibly stupid things. For example, many years ago with the cars we did and we had a TV crew do you remember that? TV crew coming and filming of the car driving at 35,000 hours something and this guy just jumped on the road in front of the car. What are you doing? The car is going to stop actually the van was in the car almost hitting. He had to step back not to avoid being hit. But I thought the car would avoid me, would not hit me. What happens is that the car hit the brakes hard as soon as he jumped into the road. The inertia is such that the car was not physically able to stop by in order to avoid him. Another case and this was actually with a golf cart on the university town of England. We are driving this golf cart around. We had this delegation of very important people and was this nice lady very elegantly dressed wearing this fancy sandal. And she just stuck her middle foot in front of the approaching vehicle. What are you talking about? What are you talking about? How? The car will see my little tiny foot wearing sandals. No! Would you do anything like that with a human driven car? So I think that people have this kind of belief that this technology is magic and that magic will avoid any kind of accidents, violating the rules of physics. So for us it is important to show people what this car can do and what it cannot do. Other experiences we are talking about the steering wheel. I don't know if you noticed, but in the experiment we did at the garden we actually removed the steering wheel and we just put like a giant eyeball with a typewriter instead of the steering wheel. Why is that? Because what we found doing some early pre-trial trials where we had normal people from the public driving in the vehicle. Usually the first reaction is that oh my goodness, I'm scared to death and this thing is going to kill me. And then people just hold on to everything and they're just terrified. Within 30 seconds this thing is not killing me. It's when they start relaxing. What happens, what you realize is that when you're driving this thing is actually pretty boring. So because nothing happens people start getting bored. And then some point they say, what happens? So people are constantly trying to break the thing. Or you know, still in their hands it's like, yeah, you know what, we don't need this. But we left the brake pedal so that people were able to in case they thought that the car was going too fast or they were scared and they would slow down. So for normal cars something goes wrong, it goes too early for more. So for this need, something goes wrong. Or it's like the brakes feel shut down more. But what does it do? Well, I mean this is also something else that is, you know, for now we drive with safety barriers. So in case something goes wrong there's a person who is able to take off. But if you think of it as something okay, so what to do if nobody is able to cover then it's really not much. So if you're able to lean back to some safe place just go to the side of the road and stop, maybe not in the middle intersection. So if you can go to some safe place or worse comes to worse you just slow down try to get out of the way if you can. If you cannot, just stop. But this is pretty much what people in your car would do. So if you have a flat well it depends on how much you can run but you try to get to the side of the road and call. Do you need to quote certain quotes or epics to be appropriate? Passenger safety versus public safety? No. Everything is very focused on the technology of things. I used to talk that the OEM didn't want to go into this because of potential liability. If I say it's a GM car getting an accident with some kind of this. Again, if this is something that like an OEM selling and then it becomes unclear honestly I think it's a very headache in the sense that it is already very well established the fact that if you get into an accident and the accident is due to a manufacturing defect of the car then the manufacturer burst the responsibility. I mean this happened with the other like the gas pedals and all kind of things like that. So it is well established that when the responsibility of the accident clearly resides with the person making the car it's their fault. Now in the case of autonomous vehicles of course then you have the Tesla approach that they say but we wrote in the contract that if you want to use the new pilot you are committing to an artificial driver which actually worked both ways and for me this is kind of disheartening how can people get away with this because ok so you are at fault if you are not paying attention. There was another case where the driver was paying attention so the autopilot was kind of just about to hit the car so she was paying attention so she hit the brake still was not able to avoid collision but then Tesla said oh you hit the brake so you disengage the autopilot so the collision happened when you were driving ok people get away with this kind of stuff from the point of view for example the kind of service that we are envisioning then we are the ones writing the software we are one operating the system so we will clearly accept the liability coming from any potential accident so when you get in a taxi it's already that way so if the taxi gets in an accident it's not your fault it's whatever the company is so for us that is something that of course we are simply liability coming from that the code of ethics for me is the same as that applies to human drivers I mean it's not really code of ethics it's a set of rules so that is the same code and then you read a lot about nonsensical questions like for example so you are driving and you end up in a situation where you need to hit somebody would you hit Hitler or Mother Teresa what kind of question is that it's not all ok so of course it's a reasonably educated person I know who is the reason who Mother Teresa is and I know what is the value to the world, to humanity to society so if you ask that question from the point of view of an autonomous vehicle I mean it's not as I'm driving around I'm downloading criminal records and the CPO oh well that guy is a PhD another guy this kind of doesn't make a lot of sense so and also I think there is a danger of and again this is part of the belief to believing that this technology is magic now there is this attempt to anthropomorphize given an ethical status to this technology ok so the fact of the matter is that the car the computer is not really able to make those ethical decisions because it just doesn't have enough information about everything you can clearly make there are some clear value propositions that you can make for example if you know that you're going to hit something I would rather hit something that is part of the fixed infrastructure otherwise I would hit if I had to hit something that moves I would rather hit something that is being potentially and hopefully it possibly has more than millions or more if I cannot then I will hit something that is at least solid and the last resort I will hit something that is squishy like a human and the last result I will hit something that is squishy and small so you can you can say that for the theory but in a sense that is to think of it we ask this kind of questions and it's really important that I think it's not up to the developers necessarily to come up with this priority I think that this is part of the responsibility of the authorities in a sense it's not the responsibility let's say it's not the right of the developers to make these choices I think that these are choices that are political in the best sense of the world in the sense that these are choices that are made by society for itself whom in our form of government we usually give our power to make those decisions to our elected representatives who come up with the government bodies who are responsible for these things so for me it's the role of the authorities to decide in a sense what are these priorities and what are the rules and then we as developers will adopt those decisions so I don't think it's my prerogative to decide so you have no idea about this right and you know something that's important that people don't realize is that if you look at the rules of the world they're actually not necessarily complete, they're not necessarily sound and clearly they're not rigorous so for example there may be situations in which it's not clear what should happen according to the rules of the road so the rules of the road do not cover every possible case in some cases the rules of the road they're actually being complete with one another and in many cases the rules of the road are actually not rigorous for example we had a very lively internal discussion on what it means to walk right away means okay so as humans we kind of understand so if you had to give right away to somebody else means that you had to let them go first right but what if the person that potentially has the right away is one kilometer away do you wait for them to go first or you kind of go anyway these are decisions that we make every day several times a day but it is no rigor in the rules so for us something that would happen out of all this evolution the technology is also clarification of the rules as they apply to humans as well going back to your examples there is no rule for humans that says that you should save yourself versus another person right so if you are involved in an accident then for some dramatic sequence of events you end up hitting somebody and you did all you could to reasonably good to save them but nobody is putting you for all you didn't drive down the cliff to save that person right so that is kind of unfortunate the kind of things that are difficult decisions that are part of it that is why you think that this is not something that I should decide so if it comes where it comes to the point where we really have to assign then again society as a whole should come up with that but also you shouldn't much credit to what the computer can actually understand so which is the same thing as in humans when you are involved in a traffic accident you have a fraction of a second to come up with a solution maybe you do something and you hit the child not because you decided to hit the child because you didn't see the child so I think that there is a whole set of things that you have to take into account boundary rationality timeliness of the decision and really a whole lot of things that the car doesn't necessarily know because I think it's a very heavy burden for the developer for me the burden of the developer is ok in the absence of regulation we can come up with our reasonable interpretation of things right but it's really for me it's more of a burden for the developer now what we try to do as I mentioned we try to be transient safe we always are in a situation where we are exposed to that kind of time so we never have to make a decision to save A or B we just save everybody by slowing down but this is another reason why we don't want to talk about the behavior from the computer to the human we don't want that to happen so we want the computer to always be in control because if the computer is always in control the computer can always make sure that it stays within the boundaries of an operational envelope that it knows how to handle so my car will never get into a situation where it doesn't know what to do so I'm sorry about it so what's, what's about this this model right I'm just trying to understand cause I came in a bit late so I'm just trying to understand are you trying to sell this kind of like the car itself or you're selling the the service the service so it's like when you're saying you're selling the service are you going to rent this thing out you're selling it well rent these out in some taxi life services services your target is trying to like for example you're going to align with a reasonable or some grab taxi or something align maybe a strong word but let's take over it in the same in the same space I think that's all I think that's where our target and the second question is actually I'm trying to understand if you are comparing yourself with Tesla right so what's your advantage and disadvantage in Singapore as well as in the US right so Tesla is not a self driving car but we understand Tesla is a technology that requires you to constantly pay attention which I think is a fundamental flaw and keep in mind that in many cases Tesla autopilot is mostly meant for highway driving that's not our target we are mostly interested in urban driving and in full automation that does not require essentially you know for me the one thing that changes everything is the beauty of the car to drive itself when nobody is on board Tesla cannot do that so Taki is actually the Singaporean well the issue Taki is the Singaporean and then there is the whole of the world business at least not that person okay I have a function so I presume that you have like a range of circumstances that your package or service can't afford to understand and some circumstances I think are your expectations so making a car not react automatically so in that case when a car say there is an accident who would be responsible for that again you are developing the software in the service so we will take responsibility for things that depend on the software if the car breaks down then of course it will be a manifestation when there are things that we do not understand that we do not expect and things like that we tend to slow down and eventually stop in case something goes wrong it doesn't go really wrong due to the time constraints can we have the last few questions so no questions thank you no solution for the past can it be adapted to a much smaller way those that are not used for transport passengers like some small delivery items actually we are working primarily on the local mobility and personal mobility market because we think that is the way that we can get a bigger impact on the other hand you have to if you remember the picture that I showed about the waiting times what you will find is that the waiting times are significant substantial during the rush hour during any other time of day the waiting times are essentially zero what it means is that we have a whole lot of vehicles that are not being used so one of the things that we are considering is to just drop the seats and convert the car into a delivery and then use the car for passenger delivery for the perception part I do do an online localization for the car because I see from the video you are matching with the existing one we drive around we collect math data it should be a huge amount of data for the car I think that is another misconception so what do you mean by huge amount of data you worry about the space or memory space or you are worried about the connection of the data okay so actually the way that we do our localization is actually very parsimonious we estimated that we could store the math data that we need for the whole United States actually the whole United States in about 5 terabytes I will go to Best Buy in the US I don't know what is the what is the electronic store so you go to any electronic store neighborhood electronic store and buy a 5 terabyte hard drive for a few hundred bucks so memory space is not really that much of an issue second I think that you may have heard about this mapping division of Nokia that was acquired last year for about 3 billion dollars by a German construction honestly it was always too money because my expectation is that within a few years maps will be $100,000 in the sense that as soon as you have now what is the cost of getting maps or the Google Street View and things like that you need to put cameras and sensors of cars and drive them around the city and the whole world so now it's expensive because you have a handful of cars doing that when we transition to the point where we have 100,000 cars in Singapore with these sensors as they drive around we can cost around the map of Singapore every 20 minutes so even the data collection and updating and things as this technology takes place and you know the fleets scale up doesn't really become much of an issue so for example if you call Uber the Uber fleet they cover 90% of Manhattan every hour okay so now imagine mapping one of the sensors of every car that Uber has and mapping is no more than a problem now that you've talked about the funding box the funding box is stand for the obstacles funding box so what's good is the question and the planning I don't understand where the question is because with your model the obstacles are using one color model or the funding box model to save time when we get the funding box we can do both we have both versions I guess you mean like using the OGM kind of thing versus abstraction into this funding box after both are quite equivalent we can do both okay okay he's getting late