We go riding with George Hotz and his $1,000 autonomous carComma.ai has most of the technology down, but needs the help of the masses.
Hi, my name is George Hotz. I play a character named geohot on the internet. I used to hack cellphones, and now I build self-driving cars. You may have seen me in Bloomberg last December. I had a self-driving car. I still do, but now I have a company as well. And now we have a self-driving car. And we have an incredible self-driving car. And by the end of the year for under $1000, you too will be able to have an incredible self-driving car. And this isn't like you gotta bring it to some guy and he's gonna do it. We wanna put it on Amazon Prime. We wanna make it a go to Amazon Prime, search self-driving car, comma AI self-driving car kit, add to cart. It'll be here in two days. And then, it's as easy to set up as a piece of IKEA furniture, if you think that's easy. And then, you have a self-driving car. And that's our dream. We're coming AI. Was it good? [LAUGH] In three, two, one, we're [INAUDIBLE] Okay. [INAUDIBLE] Alright. My hands are off everything. [CROSSTALK] See those solid green [INAUDIBLE] there? The car knows what it's doing. So I will have to make some lane changes up ahead. Notice how we did a merge right there. Whoa, that guy hit the brakes hard. Yes he did. Yeah, and we hit the brakes hard too. Now you programmed it for two car lengths, three car lengths behind? We programmed it for 1.7 seconds. Okay. But we drive very conservatively. Because I need time to get the car [UNKNOWN]. So this is all the car right now. We don't want to get off this exit So I'm just gonna nudge it over a little bit here without disengaging the autonomy. Okay, so just a tiny little input. Tiny little nudge, right. And now I can just let it go and it'll recover and pull us right back into the lane. Okay, now we're coming up to a turn on the road here. Coming up to a turn on the road here. Okay. Let's see what it's gonna do. Well, it likes to drive. You set it for 65. Yeah. This is a little aggressive. It is, honestly it's a little surprising that it's quite this aggressive. Do you find that it has a personality? I thought you'd like it a little more aggressive. No, when you say it's being aggressive tonight, you find it. It adds a little bit of personality. Yeah? Yeah. But this is a accelerating into this turn Woo, [NOISE] let me take control. [LAUGH] But yeah, it would take you around that turn, it's up to 65. Now- And we can do, let's see if we can stop behind this stationary car. [NOISE] So this is something that's fairly tricky. [NOISE] Yeah, no. It did see it, now it sees it, but it's a little bit late. Hit on that. So what are you going to do to make sure that that's- That's a bug. That's a bug. Yeah but stopping behind a stopped car should be like the first thing it should do. By the way. It would have stopped. You would not have enjoyed it. Okay. It did see the car. It just saw it late. Okay. So, like I said, everything this car does should be conservative. If it's aggressive I take back all that. Like right now it's got a solid [INAUDIBLE] Now, here again, coming up on a stationary vehicle probably gonna late brake, hit the brakes at all. [LAUGH] Solus, don't say that to me dude. Yeah, five, four, three- No, there it goes- No, not that one What? Still good, still going. Look at that. I wouldn't do it until we got to one, he still has plenty of room. That was a smooth break-in. That was pretty smooth, that was pretty smooth. Three is about when I start to hit the breaks, yeah. So you would heard that ding if I had stepped on anything. Like, I swear to god that's true, right? There's no trickery here. No, that's true. Yeah, but did you still tape at any time? I mean, a little. [LAUGH] But does the car do anything sketchy is what I'm saying? No, no, it was more aggressive than I thought that it was going to be. System is re-engaged. Okay. It was aggressive, but it wasn't tailgating. No, no, no, it wasn't tailgating. It sped up because I have the thing set at 65, right. Some people like to go fast, we like to go- So here are these little, these green lines are what it thinks are- The lanes. The lanes. But my lanes are obviously not all wiggy-waggy like that, they're straight. No, no, that's just because we moved a little bit in the lane. Okay. So actually, this here is our car, and that's our position. pushing the lanes>> OK>> so we're a little bit to the right now a but now were centering out again OK so yeah I'm going to let this guy pass us then I'm going to resume manual control in three two one It feels so much safer Should you Yes having this twenty six year old Kids driving me around. Driving me around. [CROSSTALK] Do you really trust me more than the computer? I don't know yet. All right. I don't know. I think this kind of stuff, people are gonna have to get used to it in stages, right? Yeah. And that's the thing, right? You saw how easy it was to disengage the system. Right. Put this system in your car. And then You know, grow with it right? Trust it for a second, trust it for two, trust it for ten, get in the back seat, that's that, you know? [MUSIC] Is this even legal? Without a doubt. So, what we're building is a lot more following the presidents of driver assistant systems. I have a little bit about this in the presentation, but because we, the use is still liable and in control of the vehicle at all times, it's a lot more like cruise control but it helps you out a whole lot more than cruise control does So even though we did get a cease and desist from the California DMV, they don't understand what their own laws say. Under their own laws, we've consulted with many legal experts, this is absolutely legal. So, what do their own laws say? Their laws prohibit an autonomous vehicle, where an autonomous vehicle is defined as a vehicle without human monitoring. If you're in the driver's seat, and you're watching It's technically not an autonomous vehicle. And then the law goes on to a specifically exempt lane keeping assist system and adapter cruise control systems. Which is, larger than what our system is.>>Right.>>We keep you in the lane. We make sure you don't hit anything in front of you. That's highway driving. That's traffic driving. And that's even quite a bit of city driving.>>So how its that any different from For example, what's already in a lot of Volvo cars, where you've got adapted cruise control and it has lane-keeping assist. Yup. And you have lane-keeping assist, and you turn it on, and it'll start taking the wheel and it'll be like, I don't wanna cross that lane line, I don't wanna cross that lane line, I don't wanna cross that. And literally it bounces back and forth. You'll get pulled over for driving drunk. What we are a lot more like is Tesla auto pilot which is actually a system that feels I see the future of self driving cars not this hyper conservative auto maker if we make it too good they might use it we cant have them using it right that's literally that's literally I got this guy V Every shipping car today except for the Tesla, these systems are just atrocious. The fact that we built what we built so quickly, compared to, these manufacturers have been working on it for years, and they don't even have what we have. Even Tesla, they didn't build that in-house, they licensed [UNKNOWN]. We built an in-house solution that works In some cases better than Tesla in like I nine months. And the reason we were able to it so quickly, huge advances machine learning. Huge advances in what? Machine learning. Okay. We do not code in the rules of the driving problem. We learn from hours and hours and hours of data. So when you're say you learn from hours of data you go out, you drive the car and then you take that data that happen Put it into your computer and then a computer learns that way. Yup. Okay. So, that means that the car we're going to drive today, will that potentially be you driving? Jake's done some driving. Ricardo's done some driving. Okay. Mostly the three of us. But, we have a product announcement today. And we're hoping to to a whole lot more drivers I'm listening. How do we succeed? Big Data. How do we get big data? From the world. Let's have the world teach us to drive. So announcing the first product of Comma.ai. It's Dropcam plus Fitbit. For your car. And you can earn incredible comma points by just driving. Mount our app, our app is called chauffeur well we were gonna call it chauffeur but we realized nobody knows how to spell chauffeur so we got rid of some of the letters. But it's a tiny chauffeur, it's not a big chauffeur, it's tiny. It's just a little chffr, chffr. First product of comma ai. Well you said app so I didn't think phone. It is a phone. I did say app, I'm not supposed to say app. But yeah. So that's chffr right there, go ahead and open it up. [BLANK AUDIO] [BLANK_AUDIO] We want you to mount that on the windshield of your car and point that out. And point that out in such a way that you would feel comfortable driving from the feet. OK. So you mount that up on your windshield. It automatically starts talking [INAUDIBLE] Yeah. It records all that data, it's a dash cam, incredibly useful if you get into an accident or some of those something sketchy. You have that login to Cloud. It's also fit it, you can see all of your out You can see where you drive, you can see how long your commute's taking everyday for people who are really into that stuff. So is this app, will it be free? Yeah, definitely. Yeah, yeah, yeah. Yeah, we're honestly, our thousand dollar price point does not reflect a huge profit for us either. We just want to get these things into as many cars as possible. We want to get this thing into users hands. To me that's worth so much more than nickel and diming. Why? Why? Because I think its cool So just because its cool I mean part of it too is like I'm really interested in artificial intelligence right. And building huge data sects that can teach an AI about the world. So. Kids do nothing for like the first two years. They kind of just sit there, you have to feed them and they're looking around, right? They're learning about the world. When you look at the data rate going into that baby's brain, it's insanely high, there's so much data. Bigger than any, probably bigger than any at least human-annotated data set that's out there. Not bigger than like the amount that you do videos, that's larger. We need data sets of that magnitude if we want to train human level AIs. So I care a lot more about that than I do about making a profit. There are so many other monetization routes here. And we're not gonna lose money on the hardware, we don't wanna sell these things at a loss. But it's not like we're building these things for $50 and selling them for $5000. If we can build them for $50, we'll sell them for $60. So, yeah, the app is definitely free. So, yeah, I'm an idiot. I come to Las Vegas, right? And I thought it would be cool if I went to Vegas. Sure. I mean, Vegas is awesome. But have you actually ever looked at a road in Las Vegas? Yeah they suck. Yeah so this is a road right they don't just suck they don't have lane markings Well I mean they do they just not painted They're box dots so mobile had to spend engineering time to write a special detector for box dots. And it doesn't work very well. You can read in the tester user forms about people trying to use it around the Vegas area. And you don't want to have two detectors because then you have a statement to switch between them. As we were saying, comma FI statements kill people Our car has literally never seen [UNKNOWN] like that before. So two nights before, I'm crashing at my friend's place in LA, working on street view. I'm like, they're everywhere, you can't get away from them. I had a night about [UNKNOWN], seriously. And We get here and it doesn't work. It doesn't work at all. But then we looked at what time of day it was. It's dusk. Our car has dynamic range issues at dusk. As soon as the sun fell, it started working on [UNKNOWN]. Woke up this morning, worked flawlessly. On the box dots despite never having seen them before. Mobilize spend engineering time and can't build something that works. Our system is so robust and reliable that it drives on something it's never seen before. Because a human would obviously think they're lanes. Right? And there's a lot of other intuition in that highway image that I'm showing you. Look, look at where like the tire tracks Right. Right. It can look at all of that and understand, this is how I should drive. And that's what makes us so different from every other approach out there. Cuz that's what it is. The hard part of this is not building the thing with the motor that can turn the steering wheel, and the gas, and the brake. It's writing the software that can take sensors, figure out what to do, and actually do it. Well we didn't get to spend too much time with George Hudson, his self driving Acura but he's based in San Francisco, we're based in San Francisco, maybe we can meet up again soon. We don't know yet if this product is gonna be successful but I for one can't wait 'til the end of the year to find out. [BLANK_AUDIO]