vtt/episode_028_small.vtt

WEBVTT

00:00.000 --> 00:03.120
 The following is a conversation with Chris Ermsen.

00:03.120 --> 00:06.040
 He was the CTO of the Google self driving car team,

00:06.040 --> 00:08.880
 a key engineer and leader behind the Carnegie Mellon

00:08.880 --> 00:11.240
 University, autonomous vehicle entries

00:11.240 --> 00:14.120
 in the DARPA Grand Challenges and the winner

00:14.120 --> 00:16.160
 of the DARPA Urban Challenge.

00:16.160 --> 00:19.480
 Today, he's the CEO of Aurora Innovation,

00:19.480 --> 00:21.360
 an autonomous vehicle software company.

00:21.360 --> 00:23.600
 He started with Sterling Anderson,

00:23.600 --> 00:26.000
 who was the former director of Tesla Autopilot

00:26.000 --> 00:30.160
 and drew back now Uber's former autonomy and perception lead.

00:30.160 --> 00:33.160
 Chris is one of the top roboticist and autonomous vehicle

00:33.160 --> 00:37.440
 experts in the world and a long time voice of reason

00:37.440 --> 00:41.320
 in a space that is shrouded in both mystery and hype.

00:41.320 --> 00:43.600
 He both acknowledges the incredible challenges

00:43.600 --> 00:46.560
 involved in solving the problem of autonomous driving

00:46.560 --> 00:49.680
 and is working hard to solve it.

00:49.680 --> 00:52.440
 This is the Artificial Intelligence Podcast.

00:52.440 --> 00:54.720
 If you enjoy it, subscribe on YouTube,

00:54.720 --> 00:57.920
 give it five stars on iTunes, support it on Patreon,

00:57.920 --> 00:59.760
 or simply connect with me on Twitter

00:59.760 --> 01:03.280
 at Lex Freedman spelled FRID MAN.

01:03.280 --> 01:09.160
 And now, here's my conversation with Chris Ermsen.

01:09.160 --> 01:11.960
 You were part of both the DARPA Grand Challenge

01:11.960 --> 01:17.040
 and the DARPA Urban Challenge teams at CMU with Red Whitaker.

01:17.040 --> 01:19.720
 What technical or philosophical things

01:19.720 --> 01:22.280
 have you learned from these races?

01:22.280 --> 01:26.640
 I think the high order bit was that it could be done.

01:26.640 --> 01:32.880
 I think that was the thing that was incredible about the first

01:32.880 --> 01:36.440
 of the Grand Challenges, that I remember I was a grad

01:36.440 --> 01:41.440
 student at Carnegie Mellon, and there we

01:41.440 --> 01:46.320
 was kind of this dichotomy of it seemed really hard,

01:46.320 --> 01:48.800
 so that would be cool and interesting.

01:48.800 --> 01:51.720
 But at the time, we were the only robotics

01:51.720 --> 01:54.960
 institute around, and so if we went into it and fell

01:54.960 --> 01:58.320
 in our faces, that would be embarrassing.

01:58.320 --> 02:01.160
 So I think just having the will to go do it,

02:01.160 --> 02:03.360
 to try to do this thing that at the time was marked

02:03.360 --> 02:07.120
 as darn near impossible, and then after a couple of tries,

02:07.120 --> 02:11.360
 be able to actually make it happen, I think that was really

02:11.360 --> 02:12.360
 exciting.

02:12.360 --> 02:15.120
 But at which point did you believe it was possible?

02:15.120 --> 02:17.000
 Did you, from the very beginning,

02:17.000 --> 02:18.360
 did you personally, because you're

02:18.360 --> 02:20.320
 one of the lead engineers, you actually

02:20.320 --> 02:21.800
 had to do a lot of the work?

02:21.800 --> 02:23.840
 Yeah, I was the technical director there,

02:23.840 --> 02:26.120
 and did a lot of the work, along with a bunch

02:26.120 --> 02:28.440
 of other really good people.

02:28.440 --> 02:29.760
 Did I believe it could be done?

02:29.760 --> 02:31.120
 Yeah, of course.

02:31.120 --> 02:33.400
 Why would you go do something you thought was impossible,

02:33.400 --> 02:34.880
 completely impossible?

02:34.880 --> 02:36.280
 We thought it was going to be hard.

02:36.280 --> 02:38.080
 We didn't know how we're going to be able to do it.

02:38.080 --> 02:42.880
 We didn't know if we'd be able to do it the first time.

02:42.880 --> 02:46.000
 Turns out we couldn't.

02:46.000 --> 02:48.400
 That, yeah, I guess you have to.

02:48.400 --> 02:52.920
 I think there's a certain benefit to naivete,

02:52.920 --> 02:55.400
 that if you don't know how hard something really is,

02:55.400 --> 02:59.560
 you try different things, and it gives you an opportunity

02:59.560 --> 03:04.080
 that others who are wiser maybe don't have.

03:04.080 --> 03:05.680
 What were the biggest pain points?

03:05.680 --> 03:09.360
 Mechanical, sensors, hardware, software, algorithms

03:09.360 --> 03:12.760
 for mapping, localization, just general perception,

03:12.760 --> 03:15.440
 control, like hardware, software, first of all.

03:15.440 --> 03:17.840
 I think that's the joy of this field,

03:17.840 --> 03:20.040
 is that it's all hard.

03:20.040 --> 03:25.160
 And that you have to be good at each part of it.

03:25.160 --> 03:32.280
 So for the urban challenges, if I look back at it from today,

03:32.280 --> 03:36.200
 it should be easy today.

03:36.200 --> 03:38.880
 That it was a static world.

03:38.880 --> 03:40.720
 There weren't other actors moving through it.

03:40.720 --> 03:42.440
 That is what that means.

03:42.440 --> 03:47.080
 It was out in the desert, so you get really good GPS.

03:47.080 --> 03:51.320
 So that went, and we could map it roughly.

03:51.320 --> 03:55.160
 And so in retrospect now, it's within the realm of things

03:55.160 --> 03:57.800
 we could do back then.

03:57.800 --> 03:59.200
 Just actually getting the vehicle,

03:59.200 --> 04:00.680
 and there's a bunch of engineering work

04:00.680 --> 04:04.200
 to get the vehicle so that we could control and drive it.

04:04.200 --> 04:09.520
 That's still a pain today, but it was even more so back then.

04:09.520 --> 04:12.920
 And then the uncertainty of exactly what they wanted us

04:12.920 --> 04:17.080
 to do was part of the challenge as well.

04:17.080 --> 04:19.360
 Right, you didn't actually know the track hiding it.

04:19.360 --> 04:21.520
 You knew approximately, but you didn't actually

04:21.520 --> 04:23.560
 know the route that's going to be taken.

04:23.560 --> 04:26.560
 That's right, we didn't even really,

04:26.560 --> 04:28.640
 the way the rules had been described,

04:28.640 --> 04:29.840
 you had to kind of guess.

04:29.840 --> 04:33.440
 So if you think back to that challenge,

04:33.440 --> 04:37.000
 the idea was that the government would give us,

04:37.000 --> 04:40.360
 the DARPA would give us a set of waypoints

04:40.360 --> 04:44.240
 and kind of the width that you had to stay within between the line

04:44.240 --> 04:46.840
 that went between each of those waypoints.

04:46.840 --> 04:49.280
 And so the most devious thing they could have done

04:49.280 --> 04:53.720
 is set a kilometer wide corridor across a field of scrub

04:53.720 --> 04:58.520
 brush and rocks and said, go figure it out.

04:58.520 --> 05:02.200
 Fortunately, it turned into basically driving along

05:02.200 --> 05:06.800
 a set of trails, which is much more relevant to the application

05:06.800 --> 05:08.760
 they were looking for.

05:08.760 --> 05:12.080
 But no, it was a hell of a thing back in the day.

05:12.080 --> 05:16.640
 So the legend, Red, was kind of leading that effort

05:16.640 --> 05:19.120
 in terms of just broadly speaking.

05:19.120 --> 05:22.040
 So you're a leader now.

05:22.040 --> 05:25.000
 What have you learned from Red about leadership?

05:25.000 --> 05:26.360
 I think there's a couple of things.

05:26.360 --> 05:30.880
 One is go and try those really hard things.

05:30.880 --> 05:34.760
 That's where there is an incredible opportunity.

05:34.760 --> 05:36.560
 I think the other big one, though,

05:36.560 --> 05:41.720
 is to see people for who they can be, not who they are.

05:41.720 --> 05:46.080
 It's one of the deepest lessons I learned from Red,

05:46.080 --> 05:51.000
 was that he would look at undergraduates or graduate

05:51.000 --> 05:56.120
 students and empower them to be leaders,

05:56.120 --> 06:01.400
 to have responsibility, to do great things,

06:01.400 --> 06:04.760
 that I think another person might look at them and think,

06:04.760 --> 06:06.600
 oh, well, that's just an undergraduate student.

06:06.600 --> 06:08.720
 What could they know?

06:08.720 --> 06:13.520
 And so I think that trust, but verify, have confidence

06:13.520 --> 06:14.880
 in what people can become, I think,

06:14.880 --> 06:16.680
 is a really powerful thing.

06:16.680 --> 06:20.480
 So through that, let's just fast forward through the history.

06:20.480 --> 06:24.200
 Can you maybe talk through the technical evolution

06:24.200 --> 06:27.480
 of autonomous vehicle systems from the first two

06:27.480 --> 06:30.920
 Grand Challenges to the Urban Challenge to today?

06:30.920 --> 06:33.600
 Are there major shifts in your mind,

06:33.600 --> 06:37.240
 or is it the same kind of technology just made more robust?

06:37.240 --> 06:40.880
 I think there's been some big, big steps.

06:40.880 --> 06:46.600
 So for the Grand Challenge, the real technology

06:46.600 --> 06:51.400
 that unlocked that was HD mapping.

06:51.400 --> 06:55.200
 Prior to that, a lot of the off road robotics work

06:55.200 --> 06:58.920
 had been done without any real prior model of what

06:58.920 --> 07:01.400
 the vehicle was going to encounter.

07:01.400 --> 07:03.960
 And so that innovation, that the fact

07:03.960 --> 07:11.320
 that we could get decimeter resolution models,

07:11.320 --> 07:13.560
 was really a big deal.

07:13.560 --> 07:17.480
 And that allowed us to kind of bound

07:17.480 --> 07:19.680
 the complexity of the driving problem the vehicle had

07:19.680 --> 07:21.040
 and allowed it to operate at speed,

07:21.040 --> 07:23.800
 because we could assume things about the environment

07:23.800 --> 07:26.400
 that it was going to encounter.

07:26.400 --> 07:31.320
 So that was one of the big step there.

07:31.320 --> 07:38.520
 For the Urban Challenge, one of the big technological

07:38.520 --> 07:41.960
 innovations there was the multi beam LiDAR.

07:41.960 --> 07:45.720
 And be able to generate high resolution,

07:45.720 --> 07:48.680
 mid to long range 3D models the world,

07:48.680 --> 07:54.120
 and use that for understanding the world around the vehicle.

07:54.120 --> 07:59.120
 And that was really kind of a game changing technology.

07:59.120 --> 08:02.880
 And parallel with that, we saw a bunch

08:02.880 --> 08:06.640
 of other technologies that had been kind of converging

08:06.640 --> 08:08.960
 half their day in the sun.

08:08.960 --> 08:16.800
 So Bayesian estimation had been, SLAM had been a big field

08:16.800 --> 08:18.600
 in robotics.

08:18.600 --> 08:20.800
 You would go to a conference a couple of years

08:20.800 --> 08:23.800
 before that, and every paper would effectively

08:23.800 --> 08:25.640
 have SLAM somewhere in it.

08:25.640 --> 08:31.560
 And so seeing that Bayesian estimation techniques

08:31.560 --> 08:34.040
 play out on a very visible stage,

08:34.040 --> 08:38.680
 I thought that was pretty exciting to see.

08:38.680 --> 08:41.760
 And mostly SLAM was done based on LiDAR at that time?

08:41.760 --> 08:42.400
 Well, yeah.

08:42.400 --> 08:46.720
 And in fact, we weren't really doing SLAM per se in real time,

08:46.720 --> 08:48.120
 because we had a model ahead of time.

08:48.120 --> 08:51.560
 We had a roadmap, but we were doing localization.

08:51.560 --> 08:54.080
 And we were using the LiDAR or the cameras,

08:54.080 --> 08:55.920
 depending on who exactly was doing it,

08:55.920 --> 08:58.080
 to localize to a model of the world.

08:58.080 --> 09:00.720
 And I thought that was a big step

09:00.720 --> 09:07.160
 from kind of naively trusting GPS INS before that.

09:07.160 --> 09:10.400
 And again, lots of work had been going on in this field.

09:10.400 --> 09:14.080
 Certainly, this was not doing anything particularly

09:14.080 --> 09:17.400
 innovative in SLAM or in localization,

09:17.400 --> 09:20.160
 but it was seeing that technology necessary

09:20.160 --> 09:21.800
 in a real application on a big stage.

09:21.800 --> 09:23.080
 I thought it was very cool.

09:23.080 --> 09:25.600
 So for the Urban Challenge, those already maps

09:25.600 --> 09:28.120
 constructed offline in general?

09:28.120 --> 09:28.600
 OK.

09:28.600 --> 09:30.920
 And did people do that individually?

09:30.920 --> 09:33.600
 Did individual teams do it individually?

09:33.600 --> 09:36.440
 So they had their own different approaches there?

09:36.440 --> 09:41.720
 Or did everybody kind of share that information,

09:41.720 --> 09:42.880
 at least intuitively?

09:42.880 --> 09:49.560
 So DARPA gave all the teams a model of the world, a map.

09:49.560 --> 09:53.720
 And then one of the things that we had to figure out back then

09:53.720 --> 09:56.720
 was, and it's still one of these things that trips people up

09:56.720 --> 10:00.240
 today, is actually the coordinate system.

10:00.240 --> 10:03.000
 So you get a latitude, longitude.

10:03.000 --> 10:05.120
 And to so many decimal places, you

10:05.120 --> 10:07.800
 don't really care about kind of the ellipsoid of the Earth

10:07.800 --> 10:09.520
 that's being used.

10:09.520 --> 10:12.720
 But when you want to get to 10 centimeter or centimeter

10:12.720 --> 10:18.480
 resolution, you care whether the coordinate system is NADS 83

10:18.480 --> 10:22.720
 or WGS 84, or these are different ways

10:22.720 --> 10:26.720
 to describe both the kind of nonsphericalness of the Earth,

10:26.720 --> 10:31.560
 but also kind of the actually, and I think when I can't remember

10:31.560 --> 10:33.560
 which one, the tectonic shifts that are happening

10:33.560 --> 10:36.920
 and how to transform the global datum as a function of that.

10:36.920 --> 10:40.400
 So getting a map and then actually matching it

10:40.400 --> 10:41.880
 to reality to centimeter resolution,

10:41.880 --> 10:44.000
 that was kind of interesting and fun back then.

10:44.000 --> 10:46.800
 So how much work was the perception doing there?

10:46.800 --> 10:52.440
 So how much were you relying on localization based on maps

10:52.440 --> 10:55.720
 without using perception to register to the maps?

10:55.720 --> 10:57.960
 And I guess the question is how advanced

10:57.960 --> 10:59.720
 was perception at that point?

10:59.720 --> 11:01.920
 It's certainly behind where we are today.

11:01.920 --> 11:05.800
 We're more than a decade since the urban challenge.

11:05.800 --> 11:13.080
 But the core of it was there, that we were tracking vehicles.

11:13.080 --> 11:15.600
 We had to do that at 100 plus meter range

11:15.600 --> 11:18.280
 because we had to merge with other traffic.

11:18.280 --> 11:21.200
 We were using, again, Bayesian estimates

11:21.200 --> 11:23.800
 for state of these vehicles.

11:23.800 --> 11:25.560
 We had to deal with a bunch of the problems

11:25.560 --> 11:28.240
 that you think of today of predicting

11:28.240 --> 11:31.040
 where that vehicle is going to be a few seconds into the future.

11:31.040 --> 11:33.680
 We had to deal with the fact that there

11:33.680 --> 11:36.000
 were multiple hypotheses for that because a vehicle

11:36.000 --> 11:37.640
 at an intersection might be going right

11:37.640 --> 11:41.440
 or it might be going straight or it might be making a left turn.

11:41.440 --> 11:44.080
 And we had to deal with the challenge of the fact

11:44.080 --> 11:47.520
 that our behavior was going to impact the behavior

11:47.520 --> 11:48.880
 of that other operator.

11:48.880 --> 11:53.400
 And we did a lot of that in relatively naive ways.

11:53.400 --> 11:54.720
 But it kind of worked.

11:54.720 --> 11:57.000
 Still had to have some kind of assumption.

11:57.000 --> 12:00.640
 And so where does that 10 years later, where does that take us

12:00.640 --> 12:04.200
 today from that artificial city construction

12:04.200 --> 12:06.920
 to real cities to the urban environment?

12:06.920 --> 12:13.600
 Yeah, I think the biggest thing is that the actors are truly

12:13.600 --> 12:18.680
 unpredictable, that most of the time, the drivers on the road,

12:18.680 --> 12:24.000
 the other road users are out there behaving well.

12:24.000 --> 12:27.040
 But every once in a while, they're not.

12:27.040 --> 12:33.320
 The variety of other vehicles is, you have all of them.

12:33.320 --> 12:35.760
 In terms of behavior, or terms of perception, or both?

12:35.760 --> 12:38.320
 Both.

12:38.320 --> 12:40.480
 Back then, we didn't have to deal with cyclists.

12:40.480 --> 12:42.800
 We didn't have to deal with pedestrians.

12:42.800 --> 12:46.240
 Didn't have to deal with traffic lights.

12:46.240 --> 12:49.360
 The scale over which that you have to operate is now

12:49.360 --> 12:52.240
 as much larger than the airbase that we were thinking about back

12:52.240 --> 12:52.720
 then.

12:52.720 --> 12:56.280
 So what easy question?

12:56.280 --> 12:59.720
 What do you think is the hardest part about driving?

12:59.720 --> 13:00.480
 Easy question.

13:00.480 --> 13:01.320
 Yeah.

13:01.320 --> 13:02.600
 No, I'm joking.

13:02.600 --> 13:07.440
 I'm sure nothing really jumps out at you as one thing.

13:07.440 --> 13:12.920
 But in the jump from the urban challenge to the real world,

13:12.920 --> 13:16.200
 is there something that's a particular euphorcy

13:16.200 --> 13:18.480
 as a very serious, difficult challenge?

13:18.480 --> 13:21.120
 I think the most fundamental difference

13:21.120 --> 13:28.960
 is that we're doing it for real, that in that environment,

13:28.960 --> 13:31.840
 it was both a limited complexity environment,

13:31.840 --> 13:33.240
 because certain actors weren't there,

13:33.240 --> 13:35.360
 because the roads were maintained.

13:35.360 --> 13:38.720
 There were barriers keeping people separate from robots

13:38.720 --> 13:40.880
 at the time.

13:40.880 --> 13:44.480
 And it only had to work for 60 miles, which looking at it

13:44.480 --> 13:48.960
 from 2006, it had to work for 60 miles.

13:48.960 --> 13:52.720
 Looking at it from now, we want things

13:52.720 --> 13:57.200
 that will go and drive for half a million miles.

13:57.200 --> 14:00.960
 And it's just a different game.

14:00.960 --> 14:06.080
 So how important, you said Lyder came into the game early on,

14:06.080 --> 14:08.880
 and it's really the primary driver of autonomous vehicles

14:08.880 --> 14:10.240
 today as a sensor.

14:10.240 --> 14:12.880
 So how important is the role of Lyder in the sensor suite

14:12.880 --> 14:14.760
 in the near term?

14:14.760 --> 14:18.680
 So I think it's essential.

14:18.680 --> 14:20.520
 But I also believe that cameras are essential,

14:20.520 --> 14:22.160
 and I believe the radar is essential.

14:22.160 --> 14:27.400
 I think that you really need to use the composition of data

14:27.400 --> 14:28.920
 from these different sensors if you

14:28.920 --> 14:32.600
 want the thing to really be robust.

14:32.600 --> 14:35.440
 The question I want to ask, let's see if we can untangle it,

14:35.440 --> 14:40.240
 is what are your thoughts on the Elon Musk provocative statement

14:40.240 --> 14:45.840
 that Lyder is a crutch, that is a kind of, I guess,

14:45.840 --> 14:49.600
 growing pains, and that much of the perception

14:49.600 --> 14:52.160
 task can be done with cameras?

14:52.160 --> 14:56.920
 So I think it is undeniable that people walk around

14:56.920 --> 14:59.680
 without lasers in their foreheads,

14:59.680 --> 15:01.840
 and they can get into vehicles and drive them.

15:01.840 --> 15:05.560
 And so there's an existence proof

15:05.560 --> 15:10.840
 that you can drive using passive vision.

15:10.840 --> 15:12.680
 No doubt, can't argue with that.

15:12.680 --> 15:14.320
 In terms of sensors, yeah.

15:14.320 --> 15:14.800
 So there's proof.

15:14.800 --> 15:15.960
 Yes, in terms of sensors, right?

15:15.960 --> 15:18.720
 So there's an example that we all

15:18.720 --> 15:23.280
 go do it at many of us every day.

15:23.280 --> 15:28.200
 In terms of Lyder being a crutch, sure.

15:28.200 --> 15:33.080
 But in the same way that the combustion engine

15:33.080 --> 15:35.240
 was a crutch on the path to an electric vehicle,

15:35.240 --> 15:40.840
 in the same way that any technology ultimately gets

15:40.840 --> 15:44.640
 replaced by some superior technology in the future.

15:44.640 --> 15:47.720
 And really, the way that I look at this

15:47.720 --> 15:51.720
 is that the way we get around on the ground, the way

15:51.720 --> 15:55.280
 that we use transportation is broken.

15:55.280 --> 15:59.720
 And that we have this, I think the number I saw this morning,

15:59.720 --> 16:04.040
 37,000 Americans killed last year on our roads.

16:04.040 --> 16:05.360
 And that's just not acceptable.

16:05.360 --> 16:09.440
 And so any technology that we can bring to bear

16:09.440 --> 16:12.840
 that accelerates this technology, self driving technology,

16:12.840 --> 16:15.720
 coming to market and saving lives,

16:15.720 --> 16:18.280
 is technology we should be using.

16:18.280 --> 16:24.040
 And it feels just arbitrary to say, well, I'm not

16:24.040 --> 16:27.800
 OK with using lasers, because that's whatever.

16:27.800 --> 16:30.760
 But I am OK with using an 8 megapixel camera

16:30.760 --> 16:32.880
 or a 16 megapixel camera.

16:32.880 --> 16:34.640
 These are just bits of technology,

16:34.640 --> 16:36.880
 and we should be taking the best technology from the tool

16:36.880 --> 16:41.600
 bin that allows us to go and solve a problem.

16:41.600 --> 16:45.160
 The question I often talk to, well, obviously you do as well,

16:45.160 --> 16:48.320
 to automotive companies.

16:48.320 --> 16:51.880
 And if there's one word that comes up more often than anything,

16:51.880 --> 16:55.320
 it's cost and drive costs down.

16:55.320 --> 17:01.440
 So while it's true that it's a tragic number, the 37,000,

17:01.440 --> 17:04.880
 the question is, and I'm not the one asking this question,

17:04.880 --> 17:07.160
 because I hate this question, but we

17:07.160 --> 17:11.680
 want to find the cheapest sensor suite that

17:11.680 --> 17:13.400
 creates a safe vehicle.

17:13.400 --> 17:18.240
 So in that uncomfortable trade off,

17:18.240 --> 17:23.680
 do you foresee lidar coming down in cost in the future?

17:23.680 --> 17:28.000
 Or do you see a day where level 4 autonomy is possible

17:28.000 --> 17:29.880
 without lidar?

17:29.880 --> 17:32.880
 I see both of those, but it's really a matter of time.

17:32.880 --> 17:35.080
 And I think, really, maybe I would

17:35.080 --> 17:38.760
 talk to the question you asked about the cheapest sensor.

17:38.760 --> 17:40.440
 I don't think that's actually what you want.

17:40.440 --> 17:45.720
 What you want is a sensor suite that is economically viable.

17:45.720 --> 17:49.480
 And then after that, everything is about margin

17:49.480 --> 17:52.320
 and driving cost out of the system.

17:52.320 --> 17:55.400
 What you also want is a sensor suite that works.

17:55.400 --> 18:01.280
 And so it's great to tell a story about how it would be better

18:01.280 --> 18:04.560
 to have a self driving system with a $50 sensor instead

18:04.560 --> 18:08.720
 of a $500 sensor.

18:08.720 --> 18:11.560
 But if the $500 sensor makes it work and the $50 sensor

18:11.560 --> 18:15.680
 doesn't work, who cares?

18:15.680 --> 18:21.680
 So long as you can actually have an economic opportunity there.

18:21.680 --> 18:23.760
 And the economic opportunity is important,

18:23.760 --> 18:27.800
 because that's how you actually have a sustainable business.

18:27.800 --> 18:30.440
 And that's how you can actually see this come to scale

18:30.440 --> 18:32.520
 and be out in the world.

18:32.520 --> 18:36.400
 And so when I look at lidar, I see

18:36.400 --> 18:41.200
 a technology that has no underlying fundamentally expense

18:41.200 --> 18:43.240
 to it, fundamental expense to it.

18:43.240 --> 18:46.120
 It's going to be more expensive than an imager,

18:46.120 --> 18:51.400
 because CMOS processes or FAP processes

18:51.400 --> 18:56.200
 are dramatically more scalable than mechanical processes.

18:56.200 --> 18:58.160
 But we still should be able to drive cost

18:58.160 --> 19:00.440
 out substantially on that side.

19:00.440 --> 19:05.880
 And then I also do think that with the right business model,

19:05.880 --> 19:08.440
 you can absorb more, certainly more cost

19:08.440 --> 19:09.480
 on the below materials.

19:09.480 --> 19:12.600
 Yeah, if the sensor suite works, extra value is provided.

19:12.600 --> 19:15.480
 Thereby, you don't need to drive cost down to zero.

19:15.480 --> 19:17.120
 It's a basic economics.

19:17.120 --> 19:18.840
 You've talked about your intuition

19:18.840 --> 19:22.720
 at level two autonomy is problematic because

19:22.720 --> 19:27.280
 of the human factor of vigilance, decrement, complacency,

19:27.280 --> 19:29.600
 overtrust, and so on, just us being human.

19:29.600 --> 19:33.000
 With the overtrust system, we start doing even more

19:33.000 --> 19:36.480
 so partaking in the secondary activities like smartphone

19:36.480 --> 19:38.720
 and so on.

19:38.720 --> 19:42.960
 Have your views evolved on this point in either direction?

19:42.960 --> 19:44.760
 Can you speak to it?

19:44.760 --> 19:48.240
 So I want to be really careful, because sometimes this

19:48.240 --> 19:53.000
 gets twisted in a way that I certainly didn't intend.

19:53.000 --> 19:59.360
 So active safety systems are a really important technology

19:59.360 --> 20:03.400
 that we should be pursuing and integrating into vehicles.

20:03.400 --> 20:05.680
 And there's an opportunity in the near term

20:05.680 --> 20:09.400
 to reduce accidents, reduce fatalities, and that's

20:09.400 --> 20:13.400
 and we should be pushing on that.

20:13.400 --> 20:17.280
 Level two systems are systems where

20:17.280 --> 20:19.480
 the vehicle is controlling two axes,

20:19.480 --> 20:24.800
 so breaking and thrall slash steering.

20:24.800 --> 20:27.200
 And I think there are variants of level two systems that

20:27.200 --> 20:30.200
 are supporting the driver that absolutely we

20:30.200 --> 20:32.560
 should encourage to be out there.

20:32.560 --> 20:37.920
 Where I think there's a real challenge is in the human factors

20:37.920 --> 20:40.800
 part around this and the misconception

20:40.800 --> 20:44.920
 from the public around the capability set that that enables

20:44.920 --> 20:48.000
 and the trust that they should have in it.

20:48.000 --> 20:53.880
 And that is where I'm actually incrementally more

20:53.880 --> 20:55.800
 concerned around level three systems

20:55.800 --> 20:59.960
 and how exactly a level two system is marketed and delivered

20:59.960 --> 21:03.240
 and how much effort people have put into those human factors.

21:03.240 --> 21:07.000
 So I still believe several things around this.

21:07.000 --> 21:10.760
 One is people will over trust the technology.

21:10.760 --> 21:12.720
 We've seen over the last few weeks

21:12.720 --> 21:16.280
 a spate of people sleeping in their Tesla.

21:16.280 --> 21:23.240
 I watched an episode last night of Trevor Noah talking

21:23.240 --> 21:27.160
 about this, and this is a smart guy

21:27.160 --> 21:31.040
 who has a lot of resources at his disposal describing

21:31.040 --> 21:32.880
 a Tesla as a self driving car.

21:32.880 --> 21:35.640
 And that why shouldn't people be sleeping in their Tesla?

21:35.640 --> 21:38.800
 It's like, well, because it's not a self driving car

21:38.800 --> 21:41.120
 and it is not intended to be.

21:41.120 --> 21:48.400
 And these people will almost certainly die at some point

21:48.400 --> 21:50.400
 or hurt other people.

21:50.400 --> 21:52.640
 And so we need to really be thoughtful about how

21:52.640 --> 21:56.280
 that technology is described and brought to market.

21:56.280 --> 22:00.760
 I also think that because of the economic issue,

22:00.760 --> 22:03.320
 economic challenges we were just talking about,

22:03.320 --> 22:06.960
 that technology path will, these level two driver system

22:06.960 --> 22:08.400
 systems, that technology path will

22:08.400 --> 22:11.560
 diverge from the technology path that we

22:11.560 --> 22:15.800
 need to be on to actually deliver truly self driving

22:15.800 --> 22:19.120
 vehicles, ones where you can get in it and sleep

22:19.120 --> 22:21.480
 and have the equivalent or better safety

22:21.480 --> 22:24.600
 than a human driver behind the wheel.

22:24.600 --> 22:28.440
 Because, again, the economics are very different

22:28.440 --> 22:29.800
 in those two worlds.

22:29.800 --> 22:32.720
 And so that leads to divergent technology.

22:32.720 --> 22:36.920
 So you just don't see the economics of gradually

22:36.920 --> 22:41.520
 increasing from level two and doing so quickly enough

22:41.520 --> 22:44.400
 to where it doesn't cost safety, critical safety concerns.

22:44.400 --> 22:48.600
 You believe that it needs to diverge at this point

22:48.600 --> 22:50.600
 into different, basically different routes.

22:50.600 --> 22:53.760
 And really that comes back to what

22:53.760 --> 22:56.840
 are those L2 and L1 systems doing?

22:56.840 --> 22:59.800
 And they are driver assistance functions

22:59.800 --> 23:04.360
 where the people that are marketing that responsibly

23:04.360 --> 23:07.960
 are being very clear and putting human factors in place

23:07.960 --> 23:12.400
 such that the driver is actually responsible for the vehicle

23:12.400 --> 23:15.200
 and that the technology is there to support the driver.

23:15.200 --> 23:19.880
 And the safety cases that are built around those

23:19.880 --> 23:24.320
 are dependent on that driver attention and attentiveness.

23:24.320 --> 23:30.360
 And at that point, you can kind of give up, to some degree,

23:30.360 --> 23:34.280
 for economic reasons, you can give up on, say, false negatives.

23:34.280 --> 23:36.200
 And so the way to think about this

23:36.200 --> 23:40.760
 is for a four collision mitigation braking system,

23:40.760 --> 23:45.080
 if half the times the driver missed a vehicle in front of it,

23:45.080 --> 23:47.640
 it hit the brakes and brought the vehicle to a stop,

23:47.640 --> 23:51.200
 that would be an incredible, incredible advance

23:51.200 --> 23:52.960
 in safety on our roads, right?

23:52.960 --> 23:55.080
 That would be equivalent to seatbelts.

23:55.080 --> 23:57.560
 But it would mean that if that vehicle wasn't being monitored,

23:57.560 --> 24:00.560
 it would hit one out of two cars.

24:00.560 --> 24:05.080
 And so economically, that's a perfectly good solution

24:05.080 --> 24:06.200
 for a driver assistance system.

24:06.200 --> 24:07.360
 What you should do at that point,

24:07.360 --> 24:09.200
 if you can get it to work 50% of the time,

24:09.200 --> 24:11.040
 is drive the cost out of that so you can get it

24:11.040 --> 24:13.320
 on as many vehicles as possible.

24:13.320 --> 24:16.920
 But driving the cost out of it doesn't drive up performance

24:16.920 --> 24:18.840
 on the false negative case.

24:18.840 --> 24:21.480
 And so you'll continue to not have a technology

24:21.480 --> 24:25.720
 that could really be available for a self driven vehicle.

24:25.720 --> 24:28.480
 So clearly the communication,

24:28.480 --> 24:31.640
 and this probably applies to all four vehicles as well,

24:31.640 --> 24:34.440
 the marketing and the communication

24:34.440 --> 24:37.080
 of what the technology is actually capable of,

24:37.080 --> 24:38.440
 how hard it is, how easy it is,

24:38.440 --> 24:41.040
 all that kind of stuff is highly problematic.

24:41.040 --> 24:45.680
 So say everybody in the world was perfectly communicated

24:45.680 --> 24:48.400
 and were made to be completely aware

24:48.400 --> 24:50.040
 of every single technology out there,

24:50.040 --> 24:52.880
 what it's able to do.

24:52.880 --> 24:54.160
 What's your intuition?

24:54.160 --> 24:56.920
 And now we're maybe getting into philosophical ground.

24:56.920 --> 25:00.040
 Is it possible to have a level two vehicle

25:00.040 --> 25:03.280
 where we don't overtrust it?

25:04.720 --> 25:05.840
 I don't think so.

25:05.840 --> 25:10.840
 If people truly understood the risks and internalized it,

25:11.200 --> 25:14.320
 then sure you could do that safely,

25:14.320 --> 25:16.200
 but that's a world that doesn't exist.

25:16.200 --> 25:17.560
 The people are going to,

25:19.440 --> 25:20.800
 if the facts are put in front of them,

25:20.800 --> 25:24.480
 they're gonna then combine that with their experience.

25:24.480 --> 25:28.400
 And let's say they're using an L2 system

25:28.400 --> 25:31.040
 and they go up and down the one on one every day

25:31.040 --> 25:32.800
 and they do that for a month

25:32.800 --> 25:35.200
 and it just worked every day for a month.

25:36.320 --> 25:37.400
 Like that's pretty compelling.

25:37.400 --> 25:41.880
 At that point, just even if you know the statistics,

25:41.880 --> 25:43.520
 you're like, well, I don't know,

25:43.520 --> 25:44.840
 maybe there's something a little funny about those.

25:44.840 --> 25:47.000
 Maybe they're driving in difficult places.

25:47.000 --> 25:49.960
 Like I've seen it with my own eyes, it works.

25:49.960 --> 25:52.480
 And the problem is that that sample size that they have,

25:52.480 --> 25:54.000
 so it's 30 miles up and down,

25:54.000 --> 25:58.800
 so 60 miles times 30 days, so 60, 180, 1,800 miles.

26:01.720 --> 26:05.240
 That's a drop in the bucket compared to the one,

26:05.240 --> 26:07.640
 what 85 million miles between fatalities.

26:07.640 --> 26:11.400
 And so they don't really have a true estimate

26:11.400 --> 26:14.440
 based on their personal experience of the real risks,

26:14.440 --> 26:15.640
 but they're gonna trust it anyway,

26:15.640 --> 26:17.720
 because it's hard not to, it worked for a month.

26:17.720 --> 26:18.640
 What's gonna change?

26:18.640 --> 26:21.600
 So even if you start a perfect understanding of the system,

26:21.600 --> 26:24.160
 your own experience will make it drift.

26:24.160 --> 26:25.920
 I mean, that's a big concern.

26:25.920 --> 26:29.480
 Over a year, over two years even, it doesn't have to be months.

26:29.480 --> 26:33.720
 And I think that as this technology moves from,

26:35.440 --> 26:37.800
 what I would say is kind of the more technology savvy

26:37.800 --> 26:41.480
 ownership group to the mass market,

26:41.480 --> 26:44.640
 you may be able to have some of those folks

26:44.640 --> 26:46.320
 who are really familiar with technology,

26:46.320 --> 26:48.880
 they may be able to internalize it better.

26:48.880 --> 26:50.840
 And you're kind of immunization

26:50.840 --> 26:53.400
 against this kind of false risk assessment

26:53.400 --> 26:56.960
 might last longer, but as folks who aren't as savvy

26:56.960 --> 27:00.200
 about that read the material

27:00.200 --> 27:02.200
 and they compare that to their personal experience,

27:02.200 --> 27:08.200
 I think there that it's gonna move more quickly.

27:08.200 --> 27:11.320
 So your work, the program that you've created at Google

27:11.320 --> 27:16.320
 and now at Aurora is focused more on the second path

27:16.640 --> 27:18.520
 of creating full autonomy.

27:18.520 --> 27:20.920
 So it's such a fascinating,

27:21.800 --> 27:24.600
 I think it's one of the most interesting AI problems

27:24.600 --> 27:25.640
 of the century, right?

27:25.640 --> 27:28.320
 It's a, I just talked to a lot of people,

27:28.320 --> 27:30.400
 just regular people, I don't know, my mom

27:30.400 --> 27:33.840
 about autonomous vehicles and you begin to grapple

27:33.840 --> 27:38.080
 with ideas of giving your life control over to a machine.

27:38.080 --> 27:40.040
 It's philosophically interesting,

27:40.040 --> 27:41.760
 it's practically interesting.

27:41.760 --> 27:43.720
 So let's talk about safety.

27:43.720 --> 27:46.240
 How do you think, we demonstrate,

27:46.240 --> 27:47.880
 you've spoken about metrics in the past,

27:47.880 --> 27:51.880
 how do you think we demonstrate to the world

27:51.880 --> 27:56.160
 that an autonomous vehicle, an Aurora system is safe?

27:56.160 --> 27:57.320
 This is one where it's difficult

27:57.320 --> 27:59.280
 because there isn't a sound bite answer.

27:59.280 --> 28:04.280
 That we have to show a combination of work

28:05.960 --> 28:08.360
 that was done diligently and thoughtfully.

28:08.360 --> 28:10.840
 And this is where something like a functional safety process

28:10.840 --> 28:14.360
 as part of that is like, here's the way we did the work.

28:15.320 --> 28:17.200
 That means that we were very thorough.

28:17.200 --> 28:20.560
 So, if you believe that we, what we said about,

28:20.560 --> 28:21.480
 this is the way we did it,

28:21.480 --> 28:23.440
 then you can have some confidence that we were thorough

28:23.440 --> 28:27.000
 in the engineering work we put into the system.

28:27.000 --> 28:30.160
 And then on top of that, to kind of demonstrate

28:30.160 --> 28:32.000
 that we weren't just thorough,

28:32.000 --> 28:34.000
 we were actually good at what we did.

28:35.320 --> 28:38.240
 There'll be a kind of a collection of evidence

28:38.240 --> 28:40.480
 in terms of demonstrating that the capabilities

28:40.480 --> 28:43.960
 work the way we thought they did, statistically

28:43.960 --> 28:47.200
 and to whatever degree we can demonstrate that

28:48.200 --> 28:50.320
 both in some combination of simulation,

28:50.320 --> 28:54.720
 some combination of unit testing and decomposition testing,

28:54.720 --> 28:57.000
 and then some part of it will be on road data.

28:58.200 --> 29:03.200
 And I think the way we'll ultimately convey this

29:03.320 --> 29:06.800
 to the public is there'll be clearly some conversation

29:06.800 --> 29:08.240
 with the public about it,

29:08.240 --> 29:12.080
 but we'll kind of invoke the kind of the trusted nodes

29:12.080 --> 29:14.360
 and that we'll spend more time being able to go

29:14.360 --> 29:17.280
 into more depth with folks like NHTSA

29:17.280 --> 29:19.760
 and other federal and state regulatory bodies

29:19.760 --> 29:22.600
 and kind of given that they are operating

29:22.600 --> 29:25.120
 in the public interest and they're trusted

29:26.240 --> 29:28.680
 that if we can show enough work to them

29:28.680 --> 29:30.040
 that they're convinced,

29:30.040 --> 29:33.840
 then I think we're in a pretty good place.

29:33.840 --> 29:35.040
 That means that you work with people

29:35.040 --> 29:36.960
 that are essentially experts at safety

29:36.960 --> 29:39.040
 to try to discuss and show,

29:39.040 --> 29:41.800
 do you think the answer is probably no,

29:41.800 --> 29:44.360
 but just in case, do you think there exists a metric?

29:44.360 --> 29:46.360
 So currently people have been using

29:46.360 --> 29:48.200
 a number of disengagement.

29:48.200 --> 29:50.160
 And it quickly turns into a marketing scheme

29:50.160 --> 29:54.320
 to sort of you alter the experiments you run to.

29:54.320 --> 29:56.320
 I think you've spoken that you don't like.

29:56.320 --> 29:57.160
 Don't love it.

29:57.160 --> 29:59.720
 No, in fact, I was on the record telling DMV

29:59.720 --> 30:02.000
 that I thought this was not a great metric.

30:02.000 --> 30:05.360
 Do you think it's possible to create a metric,

30:05.360 --> 30:09.480
 a number that could demonstrate safety

30:09.480 --> 30:12.400
 outside of fatalities?

30:12.400 --> 30:16.640
 So I do and I think that it won't be just one number.

30:16.640 --> 30:21.320
 So as we are internally grappling with this

30:21.320 --> 30:23.600
 and at some point we'll be able to talk

30:23.600 --> 30:25.080
 more publicly about it,

30:25.080 --> 30:28.560
 is how do we think about human performance

30:28.560 --> 30:32.200
 in different tasks, say detecting traffic lights

30:32.200 --> 30:36.240
 or safely making a left turn across traffic?

30:37.720 --> 30:40.040
 And what do we think the failure rates

30:40.040 --> 30:42.520
 are for those different capabilities for people?

30:42.520 --> 30:44.760
 And then demonstrating to ourselves

30:44.760 --> 30:48.480
 and then ultimately folks in regulatory role

30:48.480 --> 30:50.760
 and then ultimately the public,

30:50.760 --> 30:52.400
 that we have confidence that our system

30:52.400 --> 30:54.800
 will work better than that.

30:54.800 --> 30:57.040
 And so these individual metrics

30:57.040 --> 31:00.720
 will kind of tell a compelling story ultimately.

31:01.760 --> 31:03.920
 I do think at the end of the day,

31:03.920 --> 31:06.640
 what we care about in terms of safety

31:06.640 --> 31:11.640
 is life saved and injuries reduced.

31:11.640 --> 31:15.320
 And then ultimately kind of casualty dollars

31:16.440 --> 31:19.360
 that people aren't having to pay to get their car fixed.

31:19.360 --> 31:22.680
 And I do think that in aviation,

31:22.680 --> 31:25.880
 they look at a kind of an event pyramid

31:25.880 --> 31:28.600
 where a crash is at the top of that

31:28.600 --> 31:30.440
 and that's the worst event obviously.

31:30.440 --> 31:34.240
 And then there's injuries and near miss events and whatnot

31:34.240 --> 31:37.320
 and violation of operating procedures.

31:37.320 --> 31:40.160
 And you kind of build a statistical model

31:40.160 --> 31:44.440
 of the relevance of the low severity things

31:44.440 --> 31:45.280
 and the high severity things.

31:45.280 --> 31:46.120
 And I think that's something

31:46.120 --> 31:48.240
 where we'll be able to look at as well

31:48.240 --> 31:51.920
 because an event per 85 million miles

31:51.920 --> 31:54.480
 is statistically a difficult thing

31:54.480 --> 31:59.440
 even at the scale of the US to kind of compare directly.

31:59.440 --> 32:02.280
 And that event, the fatality that's connected

32:02.280 --> 32:07.280
 to an autonomous vehicle is significantly,

32:07.480 --> 32:09.160
 at least currently magnified

32:09.160 --> 32:12.320
 in the amount of attention you get.

32:12.320 --> 32:15.080
 So that speaks to public perception.

32:15.080 --> 32:16.720
 I think the most popular topic

32:16.720 --> 32:19.520
 about autonomous vehicles in the public

32:19.520 --> 32:23.080
 is the trolley problem formulation, right?

32:23.080 --> 32:27.040
 Which has, let's not get into that too much

32:27.040 --> 32:29.600
 but is misguided in many ways.

32:29.600 --> 32:32.320
 But it speaks to the fact that people are grappling

32:32.320 --> 32:36.160
 with this idea of giving control over to a machine.

32:36.160 --> 32:41.160
 So how do you win the hearts and minds of the people

32:41.560 --> 32:43.600
 that autonomy is something

32:43.600 --> 32:45.480
 that could be a part of their lives?

32:45.480 --> 32:47.640
 I think you let them experience it, right?

32:47.640 --> 32:50.440
 I think it's right.

32:50.440 --> 32:52.720
 I think people should be skeptical.

32:52.720 --> 32:55.680
 I think people should ask questions.

32:55.680 --> 32:57.000
 I think they should doubt

32:58.040 --> 33:00.960
 because this is something new and different.

33:00.960 --> 33:01.960
 They haven't touched it yet.

33:01.960 --> 33:03.680
 And I think it's perfectly reasonable.

33:03.680 --> 33:07.360
 And but at the same time,

33:07.360 --> 33:09.360
 it's clear there's an opportunity to make the road safer.

33:09.360 --> 33:12.480
 It's clear that we can improve access to mobility.

33:12.480 --> 33:15.160
 It's clear that we can reduce the cost of mobility.

33:16.680 --> 33:19.520
 And that once people try that

33:19.520 --> 33:22.800
 and understand that it's safe

33:22.800 --> 33:24.480
 and are able to use in their daily lives,

33:24.480 --> 33:28.080
 I think it's one of these things that will just be obvious.

33:28.080 --> 33:32.280
 And I've seen this practically in demonstrations

33:32.280 --> 33:35.640
 that I've given where I've had people come in

33:35.640 --> 33:38.600
 and they're very skeptical.

33:38.600 --> 33:39.960
 And they get in the vehicle.

33:39.960 --> 33:42.640
 My favorite one is taking somebody out on the freeway

33:42.640 --> 33:46.080
 and we're on the one on one driving at 65 miles an hour.

33:46.080 --> 33:48.560
 And after 10 minutes, they kind of turn and ask,

33:48.560 --> 33:49.560
 is that all it does?

33:49.560 --> 33:52.160
 And you're like, it's self driving car.

33:52.160 --> 33:54.920
 I'm not sure exactly what you thought it would do, right?

33:54.920 --> 33:57.960
 But it becomes mundane,

33:58.920 --> 34:01.560
 which is exactly what you want to technology

34:01.560 --> 34:02.760
 like this to be, right?

34:02.760 --> 34:04.680
 We don't really...

34:04.680 --> 34:07.320
 When I turn the light switch on in here,

34:07.320 --> 34:12.040
 I don't think about the complexity of those electrons

34:12.040 --> 34:14.240
 being pushed down a wire from wherever it was

34:14.240 --> 34:15.880
 and being generated.

34:15.880 --> 34:19.120
 It's like, I just get annoyed if it doesn't work, right?

34:19.120 --> 34:21.440
 And what I value is the fact

34:21.440 --> 34:23.120
 that I can do other things in this space.

34:23.120 --> 34:24.600
 I can see my colleagues.

34:24.600 --> 34:26.200
 I can read stuff on a paper.

34:26.200 --> 34:29.240
 I can not be afraid of the dark.

34:29.240 --> 34:32.840
 And I think that's what we want this technology to be like

34:32.840 --> 34:34.160
 is it's in the background

34:34.160 --> 34:36.520
 and people get to have those life experiences

34:36.520 --> 34:37.880
 and do so safely.

34:37.880 --> 34:41.600
 So putting this technology in the hands of people

34:41.600 --> 34:45.800
 speaks to scale of deployment, right?

34:45.800 --> 34:50.360
 So what do you think the dreaded question about the future

34:50.360 --> 34:52.840
 because nobody can predict the future?

34:52.840 --> 34:57.080
 But just maybe speak poetically about

34:57.080 --> 35:00.600
 when do you think we'll see a large scale deployment

35:00.600 --> 35:05.600
 of autonomous vehicles, 10,000, those kinds of numbers.

35:06.360 --> 35:08.280
 We'll see that within 10 years.

35:09.280 --> 35:10.600
 I'm pretty confident.

35:10.600 --> 35:11.920
 We...

35:13.920 --> 35:15.920
 What's an impressive scale?

35:15.920 --> 35:19.000
 What moment, so you've done the DARPA Challenge

35:19.000 --> 35:20.240
 where there's one vehicle,

35:20.240 --> 35:22.000
 at which moment does it become,

35:22.000 --> 35:23.720
 wow, this is serious scale?

35:23.720 --> 35:27.960
 So I think the moment it gets serious is when

35:27.960 --> 35:32.040
 we really do have a driverless vehicle

35:32.040 --> 35:33.880
 operating on public roads

35:34.760 --> 35:37.760
 and that we can do that kind of continuously.

35:37.760 --> 35:38.640
 Without a safety driver?

35:38.640 --> 35:40.240
 Without a safety driver in the vehicle.

35:40.240 --> 35:41.320
 I think at that moment,

35:41.320 --> 35:44.160
 we've kind of crossed the zero to one threshold.

35:45.720 --> 35:50.000
 And then it is about how do we continue to scale that?

35:50.000 --> 35:53.720
 How do we build the right business models?

35:53.720 --> 35:56.040
 How do we build the right customer experience around it

35:56.040 --> 35:59.680
 so that it is actually a useful product out in the world?

36:00.720 --> 36:03.360
 And I think that is really,

36:03.360 --> 36:05.720
 at that point, it moves from a,

36:05.720 --> 36:08.960
 what is this kind of mixed science engineering project

36:08.960 --> 36:12.120
 into engineering and commercialization

36:12.120 --> 36:15.600
 and really starting to deliver on the value

36:15.600 --> 36:18.000
 that we all see here.

36:18.000 --> 36:20.680
 And actually making that real in the world.

36:20.680 --> 36:22.240
 What do you think that deployment looks like?

36:22.240 --> 36:24.920
 Where do we first see the inkling of

36:24.920 --> 36:28.600
 no safety driver, one or two cars here and there?

36:28.600 --> 36:29.760
 Is it on the highway?

36:29.760 --> 36:33.200
 Is it in specific routes in the urban environment?

36:33.200 --> 36:36.960
 I think it's gonna be urban, suburban type environments.

36:37.920 --> 36:38.920
 You know, with Aurora,

36:38.920 --> 36:41.560
 when we thought about how to tackle this,

36:42.400 --> 36:45.040
 it was kind of invoke to think about trucking

36:46.000 --> 36:47.760
 as opposed to urban driving.

36:47.760 --> 36:51.240
 And again, the human intuition around this

36:51.240 --> 36:55.360
 is that freeways are easier to drive on

36:57.040 --> 36:59.240
 because everybody's kind of going in the same direction

36:59.240 --> 37:01.560
 and lanes are a little wider, et cetera.

37:01.560 --> 37:03.280
 And I think that that intuition is pretty good,

37:03.280 --> 37:06.000
 except we don't really care about most of the time.

37:06.000 --> 37:08.360
 We care about all of the time.

37:08.360 --> 37:10.840
 And when you're driving on a freeway with a truck,

37:10.840 --> 37:15.840
 say 70 miles an hour and you've got 70,000 pound load

37:15.840 --> 37:17.840
 to do with you, that's just an incredible amount

37:17.840 --> 37:18.840
 of kinetic energy.

37:18.840 --> 37:21.440
 And so when that goes wrong, it goes really wrong.

37:22.600 --> 37:27.600
 And that those challenges that you see occur more rarely

37:27.760 --> 37:31.040
 so you don't get to learn as quickly.

37:31.040 --> 37:33.640
 And they're incrementally more difficult

37:33.640 --> 37:35.920
 than urban driving, but they're not easier

37:35.920 --> 37:37.400
 than urban driving.

37:37.400 --> 37:41.600
 And so I think this happens in moderate speed,

37:41.600 --> 37:43.840
 urban environments, because there,

37:43.840 --> 37:46.560
 if two vehicles crash at 25 miles per hour,

37:46.560 --> 37:50.040
 it's not good, but probably everybody walks away.

37:51.000 --> 37:53.680
 And those events where there's the possibility

37:53.680 --> 37:55.720
 for that occurring happen frequently.

37:55.720 --> 37:57.920
 So we get to learn more rapidly.

37:57.920 --> 38:01.320
 We get to do that with lower risk for everyone.

38:02.440 --> 38:04.280
 And then we can deliver value to people

38:04.280 --> 38:05.800
 that need to get from one place to another.

38:05.800 --> 38:08.160
 And then once we've got that solved,

38:08.160 --> 38:10.000
 then the kind of the freeway driving part of this

38:10.000 --> 38:12.440
 just falls out, but we're able to learn

38:12.440 --> 38:15.160
 more safely, more quickly in the urban environment.

38:15.160 --> 38:18.480
 So 10 years and then scale 20, 30 years.

38:18.480 --> 38:21.440
 I mean, who knows if it's sufficiently compelling

38:21.440 --> 38:24.320
 experience is created, it can be faster and slower.

38:24.320 --> 38:27.120
 Do you think there could be breakthroughs

38:27.120 --> 38:29.880
 and what kind of breakthroughs might there be

38:29.880 --> 38:32.360
 that completely change that timeline?

38:32.360 --> 38:35.320
 Again, not only am I asking to predict the future,

38:35.320 --> 38:37.280
 I'm asking you to predict breakthroughs

38:37.280 --> 38:38.280
 that haven't happened yet.

38:38.280 --> 38:41.800
 So what's the, I think another way to ask that would be

38:41.800 --> 38:44.240
 if I could wave a magic wand,

38:44.240 --> 38:46.640
 what part of the system would I make work today

38:46.640 --> 38:48.600
 to accelerate it as quickly as possible?

38:48.600 --> 38:49.440
 Right.

38:52.080 --> 38:54.080
 Don't say infrastructure, please don't say infrastructure.

38:54.080 --> 38:56.280
 No, it's definitely not infrastructure.

38:56.280 --> 39:00.520
 It's really that perception forecasting capability.

39:00.520 --> 39:04.760
 So if tomorrow you could give me a perfect model

39:04.760 --> 39:07.520
 of what's happening and what will happen

39:07.520 --> 39:11.400
 for the next five seconds around a vehicle

39:11.400 --> 39:14.480
 on the roadway, that would accelerate things

39:14.480 --> 39:15.320
 pretty dramatically.

39:15.320 --> 39:17.560
 Are you interested in staying up at night?

39:17.560 --> 39:21.680
 Are you mostly bothered by cars, pedestrians, or cyclists?

39:21.680 --> 39:25.920
 So I worry most about the vulnerable road users

39:25.920 --> 39:28.000
 about the combination of cyclists and cars, right?

39:28.000 --> 39:29.480
 Just cyclists and pedestrians

39:29.480 --> 39:31.880
 because they're not in armor.

39:33.240 --> 39:36.480
 The cars, they're bigger, they've got protection

39:36.480 --> 39:39.440
 for the people and so the ultimate risk is lower there.

39:39.440 --> 39:44.080
 Whereas a pedestrian or cyclist, they're out on the road

39:44.080 --> 39:46.520
 and they don't have any protection.

39:46.520 --> 39:49.760
 And so we need to pay extra attention to that.

39:49.760 --> 39:54.120
 Do you think about a very difficult technical challenge

39:55.760 --> 39:58.560
 of the fact that pedestrians,

39:58.560 --> 40:01.400
 if you try to protect pedestrians by being careful

40:01.400 --> 40:04.600
 and slow, they'll take advantage of that.

40:04.600 --> 40:07.560
 So the game theoretic dance.

40:07.560 --> 40:10.880
 Does that worry you from a technical perspective

40:10.880 --> 40:12.520
 how we solve that?

40:12.520 --> 40:14.600
 Because as humans, the way we solve that

40:14.600 --> 40:17.280
 is kind of nudge our way through the pedestrians,

40:17.280 --> 40:20.040
 which doesn't feel from a technical perspective

40:20.040 --> 40:22.320
 as a appropriate algorithm.

40:23.240 --> 40:25.960
 But do you think about how we solve that problem?

40:25.960 --> 40:30.960
 Yeah, I think there's two different concepts there.

40:31.400 --> 40:35.880
 So one is, am I worried that because these vehicles

40:35.880 --> 40:37.640
 are self driving, people will kind of step on the road

40:37.640 --> 40:38.680
 and take advantage of them.

40:38.680 --> 40:43.680
 And I've heard this and I don't really believe it

40:43.800 --> 40:46.000
 because if I'm driving down the road

40:46.000 --> 40:48.920
 and somebody steps in front of me, I'm going to stop.

40:48.920 --> 40:49.760
 Right?

40:49.760 --> 40:53.720
 Like even if I'm annoyed, I'm not gonna just drive

40:53.720 --> 40:55.200
 through a person stood on the road.

40:55.200 --> 40:56.440
 Right.

40:56.440 --> 41:00.440
 And so I think today people can take advantage of this

41:00.440 --> 41:02.600
 and you do see some people do it.

41:02.600 --> 41:04.200
 I guess there's an incremental risk

41:04.200 --> 41:05.920
 because maybe they have lower confidence

41:05.920 --> 41:06.760
 that I'm going to see them

41:06.760 --> 41:09.360
 than they might have for an automated vehicle.

41:09.360 --> 41:12.080
 And so maybe that shifts it a little bit.

41:12.080 --> 41:14.400
 But I think people don't want to get hit by cars.

41:14.400 --> 41:17.120
 And so I think that I'm not that worried

41:17.120 --> 41:18.800
 about people walking out of the one on one

41:18.800 --> 41:21.840
 and creating chaos more than they would today.

41:24.400 --> 41:27.040
 Regarding kind of the nudging through a big stream

41:27.040 --> 41:30.040
 of pedestrians leaving a concert or something.

41:30.040 --> 41:33.480
 I think that is further down the technology pipeline.

41:33.480 --> 41:36.920
 I think that you're right, that's tricky.

41:36.920 --> 41:38.560
 I don't think it's necessarily,

41:40.320 --> 41:43.360
 I think the algorithm people use for this is pretty simple.

41:43.360 --> 41:44.200
 Right?

41:44.200 --> 41:45.040
 It's kind of just move forward slowly

41:45.040 --> 41:47.600
 and if somebody's really close and stop.

41:47.600 --> 41:50.840
 And I think that that probably can be replicated

41:50.840 --> 41:54.040
 pretty easily and particularly given that it's,

41:54.040 --> 41:57.240
 you don't do this at 30 miles an hour, you do it at one,

41:57.240 --> 41:59.080
 that even in those situations,

41:59.080 --> 42:01.200
 the risk is relatively minimal.

42:01.200 --> 42:03.440
 But it's not something we're thinking

42:03.440 --> 42:04.560
 about in any serious way.

42:04.560 --> 42:08.000
 And probably that's less an algorithm problem

42:08.000 --> 42:10.160
 more creating a human experience.

42:10.160 --> 42:14.320
 So the HCI people that create a visual display

42:14.320 --> 42:16.280
 that you're pleasantly as a pedestrian,

42:16.280 --> 42:17.680
 nudged out of the way.

42:17.680 --> 42:21.960
 That's an experience problem, not an algorithm problem.

42:22.880 --> 42:25.480
 Who's the main competitor to Aurora today?

42:25.480 --> 42:28.600
 And how do you out compete them in the long run?

42:28.600 --> 42:31.200
 So we really focus a lot on what we're doing here.

42:31.200 --> 42:34.440
 I think that, I've said this a few times

42:34.440 --> 42:37.960
 that this is a huge difficult problem

42:37.960 --> 42:40.280
 and it's great that a bunch of companies are tackling it

42:40.280 --> 42:42.320
 because I think it's so important for society

42:42.320 --> 42:43.760
 that somebody gets there.

42:45.200 --> 42:49.040
 So we don't spend a whole lot of time

42:49.040 --> 42:51.560
 like thinking tactically about who's out there

42:51.560 --> 42:55.480
 and how do we beat that person individually?

42:55.480 --> 42:58.680
 What are we trying to do to go faster ultimately?

42:58.680 --> 43:02.600
 Well, part of it is the leisure team we have

43:02.600 --> 43:04.160
 has got pretty tremendous experience.

43:04.160 --> 43:06.400
 And so we kind of understand the landscape

43:06.400 --> 43:09.120
 and understand where the cul de sacs are to some degree.

43:09.120 --> 43:10.920
 And we try and avoid those.

43:12.600 --> 43:14.240
 I think there's a part of it

43:14.240 --> 43:16.240
 just this great team we've built.

43:16.240 --> 43:19.040
 People, this is a technology and a company

43:19.040 --> 43:22.280
 that people believe in the mission of.

43:22.280 --> 43:24.760
 And so it allows us to attract just awesome people

43:24.760 --> 43:25.680
 to go work.

43:26.760 --> 43:28.000
 We've got a culture, I think,

43:28.000 --> 43:30.440
 that people appreciate, that allows them to focus,

43:30.440 --> 43:33.080
 allows them to really spend time solving problems.

43:33.080 --> 43:35.880
 And I think that keeps them energized.

43:35.880 --> 43:40.880
 And then we've invested heavily in the infrastructure

43:43.520 --> 43:46.520
 and architectures that we think will ultimately accelerate us.

43:46.520 --> 43:50.640
 So because of the folks we're able to bring in early on,

43:50.640 --> 43:53.520
 because of the great investors we have,

43:53.520 --> 43:56.760
 we don't spend all of our time doing demos

43:56.760 --> 43:58.680
 and kind of leaping from one demo to the next.

43:58.680 --> 44:02.800
 We've been given the freedom to invest in

44:03.960 --> 44:05.480
 infrastructure to do machine learning,

44:05.480 --> 44:08.600
 infrastructure to pull data from our on road testing,

44:08.600 --> 44:11.480
 infrastructure to use that to accelerate engineering.

44:11.480 --> 44:14.480
 And I think that early investment

44:14.480 --> 44:17.320
 and continuing investment in those kind of tools

44:17.320 --> 44:19.400
 will ultimately allow us to accelerate

44:19.400 --> 44:21.920
 and do something pretty incredible.

44:21.920 --> 44:23.400
 Chris, beautifully put.

44:23.400 --> 44:24.640
 It's a good place to end.

44:24.640 --> 44:26.520
 Thank you so much for talking today.

44:26.520 --> 44:27.360
 Thank you very much.

44:27.360 --> 44:57.200
 I hope you enjoyed it.