Chris Urmson: Is Lidar a Crutch? | AI Podcast Clips
00:00:00.000 | (gentle music)
00:00:02.580 | - You said LiDAR came into the game early on
00:00:10.400 | and it's really the primary driver
00:00:12.240 | of autonomous vehicles today as a sensor.
00:00:14.600 | So how important is the role of LiDAR
00:00:16.240 | in the sensor suite in the near term?
00:00:19.140 | - So I think it's essential.
00:00:21.060 | You know, I believe,
00:00:23.000 | but I also believe the cameras are essential
00:00:24.840 | and I believe the radar is essential.
00:00:26.440 | I think that you really need to use the composition of data
00:00:31.440 | from these different sensors
00:00:33.020 | if you want the thing to really be robust.
00:00:36.940 | - The question I wanna ask,
00:00:38.680 | let's see if we can untangle it,
00:00:39.880 | is what are your thoughts
00:00:41.200 | on the Elon Musk provocative statement
00:00:44.560 | that LiDAR is a crutch,
00:00:46.640 | that is a kind of, I guess, growing pains,
00:00:51.640 | and that much of the perception task
00:00:54.200 | can be done with cameras.
00:00:56.400 | - So I think it is undeniable
00:00:59.760 | that people walk around without lasers in their foreheads
00:01:03.680 | and they can get into vehicles and drive them.
00:01:06.200 | And so there's an existence proof
00:01:09.920 | that you can drive using passive vision.
00:01:13.900 | No doubt, can't argue with that.
00:01:17.040 | - In terms of sensors, yeah.
00:01:18.640 | - Yeah, in terms of sensors, right?
00:01:20.320 | So like there's an example that we all go do it,
00:01:24.520 | many of us every day.
00:01:26.600 | In terms of LiDAR being a crutch, sure.
00:01:31.600 | But in the same way that the combustion engine
00:01:37.480 | was a crutch on the path to an electric vehicle,
00:01:39.680 | in the same way that any technology ultimately gets
00:01:43.720 | replaced by some superior technology in the future.
00:01:48.800 | And really, the way that I look at this
00:01:52.120 | is that the way we get around on the ground,
00:01:55.840 | the way that we use transportation is broken.
00:01:58.300 | And that we have this,
00:02:02.280 | what was, I think the number I saw this morning,
00:02:04.120 | 37,000 Americans killed last year on our roads.
00:02:08.440 | And that's just not acceptable.
00:02:09.760 | And so any technology that we can bring to bear
00:02:13.840 | that accelerates this self-driving technology
00:02:17.240 | coming to market and saving lives
00:02:20.120 | is technology we should be using.
00:02:21.820 | And it feels just arbitrary to say,
00:02:25.240 | well, I'm not okay with using lasers
00:02:30.240 | because that's whatever,
00:02:32.200 | but I am okay with using an eight megapixel camera
00:02:35.120 | or a 16 megapixel camera.
00:02:37.280 | These are just bits of technology
00:02:39.040 | and we should be taking the best technology
00:02:40.760 | from the tool bin that allows us to go
00:02:42.640 | and solve a problem.
00:02:45.980 | - The question I often talk to,
00:02:48.040 | well, obviously you do as well,
00:02:49.520 | to the sort of automotive companies.
00:02:52.680 | And if there's one word that comes up more often
00:02:55.760 | than anything, it's cost.
00:02:57.000 | And trying to drive costs down.
00:02:59.680 | So while it's true that it's a tragic number, the 37,000,
00:03:04.680 | the question is, and I'm not the one asking this question
00:03:09.280 | 'cause I hate this question,
00:03:10.200 | but we want to find the cheapest sensor suite
00:03:14.360 | that creates a safe vehicle.
00:03:17.680 | So in that uncomfortable trade-off,
00:03:22.640 | do you foresee LiDAR coming down in cost in the future
00:03:27.640 | or do you see a day where level four autonomy
00:03:31.080 | is possible without LiDAR?
00:03:34.280 | - I see both of those, but it's really a matter of time.
00:03:37.260 | And I think really, maybe I would talk to the question
00:03:40.440 | you asked about the cheapest sensor.
00:03:42.240 | I don't think that's actually what you want.
00:03:44.840 | What you want is a sensor suite that is economically viable.
00:03:49.840 | And then after that, everything is about margin
00:03:53.880 | and driving cost out of the system.
00:03:55.780 | What you also want is a sensor suite that works.
00:03:59.840 | And so it's great to tell a story about how it would be
00:04:04.840 | better to have a self-driving system with a $50 sensor
00:04:08.760 | instead of a $500 sensor.
00:04:13.120 | But if the $500 sensor makes it work
00:04:14.960 | and the $50 sensor doesn't work, who cares?
00:04:19.200 | So long as you can actually have an economic,
00:04:24.440 | there's an economic opportunity there.
00:04:25.960 | And the economic opportunity is important
00:04:28.200 | because that's how you actually have a sustainable business.
00:04:32.200 | And that's how you can actually see this come to scale
00:04:35.560 | and be out in the world.
00:04:36.840 | And so when I look at LiDAR,
00:04:40.400 | I see a technology that has no underlying
00:04:43.320 | fundamentally expense to it, fundamental expense to it.
00:04:46.840 | It's going to be more expensive than an imager
00:04:50.520 | because CMOS processes or fab processes
00:04:54.800 | are dramatically more scalable than mechanical processes,
00:04:59.520 | but we still should be able to drive cost down
00:05:02.760 | substantially on that side.
00:05:04.560 | And then I also do think that with the right business model,
00:05:10.300 | you can absorb more,
00:05:12.000 | certainly more cost on the bill of materials.
00:05:13.880 | - Yeah, if the sensor suite works, extra value is provided,
00:05:17.040 | thereby you don't need to drive costs down to zero.
00:05:19.920 | It's the basic economics.
00:05:21.260 | (air whooshing)
00:05:23.920 | (air whooshing)
00:05:26.580 | (air whooshing)
00:05:29.240 | (air whooshing)
00:05:31.900 | (air whooshing)
00:05:34.560 | (air whooshing)
00:05:37.220 | [BLANK_AUDIO]