00:00hi and welcome to the a 16z podcast I'm
00:03Hannah and we're here today talking
00:05about the evolution of cartography how
00:07map making is fundamentally changing in
00:09the age of autonomous vehicles with
00:11whale oh see oh oh and head of product
00:14of deep map which creates HD maps for
00:16autonomous vehicles and David Rumsey map
00:19collector of one of the largest private
00:21paper map collections in the world now
00:23at Stanford and of the largest online
00:26digital map collection we talked about
00:28how the tools we use have changed from
00:30sextants to measure the Stars to
00:32computer vision and lidar as well as how
00:34we think about what maps actually do and
00:37are used for maps have really evolved
00:39from incredibly primitive technology
00:42right like sextants and stars really to
00:46to now I guess satellite and camera and
00:49laser and lidar and beyond so has the
00:52nature of mapping itself fundamentally
00:54changed or do we use maps differently
00:58now or is it the same just more than it
01:00has been in the past well I see I never
01:03referred to sextants as primitive the
01:07lidar of the day sextants opened up the
01:11whole world and created the ability to
01:13make scale maps and to map essentially
01:17beyond where you were when would you say
01:20the beginning of sort of the golden era
01:22of mapping was well it was really around
01:25the time of Columbus because the whole
01:27discovery of North and South America
01:29produced a phenomenal amount of desire
01:33to map it to possess it one of our
01:36favorite globes is the it's called the
01:38earth Apple which was by Martin Behaim
01:41the Year Columbus discovered America
01:431492 and there's no North and South
01:46America on the globe amazing of course
01:48there's a huge Japan and China even
01:54though we know Columbus sailed West in
01:57order to go to China seeing it on the
01:59globe without North and South America
02:01makes it so real so that's when maps get
02:03really powerful well it's interesting
02:05because it's capturing an entire time to
02:08write it's a representation of our all
02:10our knowledge at the time well that's
02:13that globe has information that you
02:16cannot get any other way so if the bones
02:18of the maps or once you know longitude
02:22and latitude and medians what are the
02:24bones of maps now today where are we at
02:27in the state of mapping mapping has
02:29always been almost a frontier of
02:31technology people who can make really
02:34good maps for that time tend to be the
02:37most technically advanced and
02:39mathematical advanced well because it's
02:41always about processing information
02:43right it's always about getting the
02:45right and for me it's about getting the
02:46right information and then do it in a
02:48very incredibly the most precise way at
02:52the given time 300 years ago even 500
02:56years ago their understanding of the
02:59environment is very limited but somehow
03:02they need to predict into areas that
03:04they cannot visually see and today you
03:06know back to your question you know how
03:08what are the backbones of today's
03:10digital maps well first of all we have a
03:13lot of sensors that can enable us to see
03:16far right you know there are satellites
03:18up in there there you know airplanes up
03:21in the air even the cars on the road a
03:24lot of them have a lot of sensors as
03:26we have radars we'll have cameras and
03:29and so on so a lot of these digital
03:32information is already getting collected
03:34and we can use that information to help
03:36us create digital maps and it's and
03:39they're being created all the time it's
03:40Cod the Maps are constantly being
03:42redrawn that's actually a very
03:44interesting question because as of today
03:47that process is actually not happening
03:49we are collecting a lot of data but the
03:52information that is actually being
03:54shared for map creation purpose is very
03:57small so if you think about the maps
03:59that you actually use today right there
04:02are maps that you have on your phone
04:03there are maps that are probably in your
04:05car navigation system as well
04:07those maps were not created from the GPS
04:11unit that you have on your car or the
04:13camera system that you have in your car
04:14and whatnot they were created by special
04:18what we call mapping survey fleets
04:20so companies like you know for instance
04:23Google will build these special fleet
04:27special sensors mounted on these special
04:29cars they would hire human drivers to
04:33drive this road collect little data and
04:35then they do a lot of offline processing
04:37and and the end of day some digital map
04:40of you know cities would be produced and
04:43they get inserted into the cars they get
04:47stored in a cloud and then sent to your
04:49phone but even though it's using new
04:51tools it's sort of an old fashioned
04:52model of thinking about map making right
04:54like you said one pioneer out to map
04:57everything and it's a static thing
04:59absolutely we in addition to collecting
05:02maps I do collect the tools of map
05:05making oh yeah and for mapping the West
05:08there would be you know typically
05:11officers in the Army 1870 with a we have
05:16a big wooden board that they would keep
05:18on their arm as they drew the map from
05:21their horse sort of like you can imagine
05:37the length of time for that drawing to
05:41get from that person onto the map on the
05:44Pony Express which they thought was
05:49revolutionary you know that it was very
05:51fast so the speed has really increased
05:53stright the mapping speed is what's
05:55increased yes but the nature of it
05:57hasn't really changed so much or you
06:00would you say know one thing that has
06:02changed drastically from paper maps is
06:05that maps are dynamic now mm-hmm
06:07maps are changing you know with paper
06:09maps you have a sense of scale the scale
06:11is printed on to the map and it's fixed
06:14but with digital maps the scale is
06:18fluidly fluid and variable and the
06:21accuracy just depends on the size of the
06:25resolution mapping as a field has
06:27evolved or reinvented itself many times
06:30throughout history and now we're
06:32reinventing it again what are those
06:34moments methods of printing particularly
06:37shape mapping so the first maps were
06:41block you could only get you know a
06:44hundred good impressions so maps were
06:46held by you know people who were rulers
06:50who were wealthy and so on then it
06:51changed to copper engraving you could
06:53make five hundred impressions copper
06:55engraving to lithography in the
06:57nineteenth century thousands of maps and
06:59now chromolithography and then
07:01eventually digital distribution over the
07:03web so maps have become ubiquitous so
07:06the access to the max the access has
07:09been the most profound change the actual
07:11depiction of space I think are the goal
07:15to depict space has been remarkably
07:18consistent its democratizing the map you
07:21know not only the printing or the access
07:24of the map has changed over time but if
07:26you look at the general trend maps tend
07:28to get more and more accurate over time
07:30just in general right we're already
07:33doing that a lot of people thought that
07:35you know mapping has been solved with
07:36today's commonly available navigation
07:40maps because they seem very accurate for
07:42kids we all have access I already have
07:43access they'll like they're updating
07:46that's right you get a real-time traffic
07:47yeah under Google Maps app and so on and
07:50so forth but if you actually dive deeper
07:53Maps today including the digital maps
07:55are actually built for human consumption
07:58only ok so whether or not you are
08:01looking at say Google Maps or you know
08:03Apple maps ways and and so on these are
08:06maps purposefully built to be first of
08:08all very easy to interpret by human
08:10beings and easy to use for navigation
08:13purpose then the vacation system is
08:15telling the humans how to navigate that
08:17car according to some very simple
08:19instructions it's also in assuming that
08:21there is a human with human capabilities
08:23dangerous interpretate yes that's
08:26and well we are the new period we're
08:29gonna entry to is or what's needed to be
08:32built for a self-driving space is
08:35actually Maps build purposefully build
08:38for robotics systems so now we have to
08:41democratize beyond the human that's
08:43right so now we have a totally different
08:46set of demands that we need out of our
08:49maps when we start looking at autonomous
08:51vehicles we don't we don't have enough
08:53right now to help make sure
08:55autonomous car can predict exactly where
08:59something is and be safely in real-time
09:01right so what has to change about
09:02mapping to serve that we need maps that
09:06are easy to interpret by robots deep map
09:10is focusing on self-driving cars but you
09:12can't generalize it to any robot that
09:14needs to roam around the physical world
09:16yeah even though robots can outperform
09:19humans seeing certain aspects in other
09:21aspects they are intelligence wise
09:23humans are actually much smarter things
09:26people tend or humans tend to take for
09:28granted such as stopping at the right
09:31place at our intersection watching for
09:34the right traffic signal or making a
09:38split last-minute decision to avoid a
09:40raccoon in the road that's right yeah
09:42these decisions are very hard for robots
09:45to make and as part of the
09:48decision-making process the mapping
09:51becomes a very critical component of
09:55helping the robots to make the right
09:57decisions because they're essentially
09:58reading the whole world around them
10:00through that that's right what does that
10:02mean for the kind of map we're now gonna
10:04need in the future of autonomous
10:06vehicles the maps that are purposively
10:08field for self-driving purpose are
10:10usually called high-definition maps or
10:13height or HD map for short they
10:17specifically refer to the maps that have
10:19extra that have extremely high precision
10:21and we're talking about centimeter level
10:23accuracy or precision because the robots
10:26need very precise instructions on how to
10:30maneuver themselves and know how to
10:32navigate themselves around the 3d space
10:35right a few centimeters makes a really
10:37big difference when there's a curve
10:39there absolutely I mean people tend to
10:42think why do I need to say 5 centimeter
10:44or 10 centimeter accuracy when I'm
10:47driving down the road in most cases you
10:50know the tolerance for error might be
10:53higher than that but then they're gonna
10:55be cases where if you know you're
10:57driving on you know I don't know the
11:00road to Tahoe there's literally cliffs
11:02on one side and there's really no room
11:04for error or any error so the map needs
11:09precise anding used to contain a lot of
11:12information that again humans may take
11:15for granted so not only we need to know
11:18where the lanes are where the road
11:20boundaries are we also want to know
11:22where the curbs are how high the curbs
11:25are if it's five centimeters in HD
11:28mapping that's approaching something
11:30that's been sort of a holy grail for
11:32mappers forever which is what we call
11:35the one-to-one map it's the map of the
11:37world as big as the world
11:38Wow Jorge writes a short story about the
11:41cartographer that made a map that was
11:43one-to-one and the problem he couldn't
11:45unroll it it made me realize that HD
11:49mapping is a one-to-one mapping and yet
11:51we don't unroll it we go through it it's
11:54so interesting because I think of like
11:56maps on some basic level as being
11:59condensing a huge amount of information
12:00and like taking one element of that
12:02information and showing and simplifying
12:05one INT down into one picture of like
12:08you know borders or topography or land
12:12versus sea but now you're talking about
12:14actually way more than one element it's
12:16exploding the simple map to as full as
12:21comprehensive as possible the map needs
12:23to describe every little thing on the
12:25road and used to describe a lot of
12:27hidden things that you don't typically
12:30see on the map like what like for
12:33instance the speed limit on a road that
12:35you see you want to tell you know encode
12:38into the map whether or not this Lang is
12:41allowed to go straight at intersection
12:43or is required to make a left turn or
12:45right turn David am i right that it was
12:47only one sort of one boiling things down
12:49to one dimension or is it actually much
12:52more multi-dimensional than I'm thinking
12:54I think a little more multi-dimensional
12:55because it it isn't just about scale
12:58it's also about comprehensiveness
13:01so the first atlas was published in 1570
13:05by abraham ortelius and it was actually
13:07the best-selling book in Europe years
13:10and years and years and the reason it
13:12was so important to people was there
13:15were maps of the entire world in the
13:17book always been important to capture as
13:22as much as possible now these were not
13:24obviously not anything like HD maps or
13:27even digital maps but they were covering
13:30the whole earth people love on old maps
13:33the notion of terra incognita which is
13:36put over the centre of Australia or the
13:40literally I don't think it is gone
13:46really if we think about the oceans I
13:48know ways not talking about submarines
13:53symmetry mapping is is is a complete
13:58asymmetry mapping being we are at a
14:06whole new frontier right of trying to
14:08process all that information into an
14:11areal map now you are you're really just
14:14exploding what the map is the map
14:18well it's also an immersive map so what
14:24kind of tools allow you to do that how
14:26are we actually starting to explode this
14:27map to map reality and the very high
14:30level obviously there's the hardware
14:31component as well as sulfur component
14:33the hardware components are more visible
14:35because if you look at you know even a
14:38picture of a self-driving car you will
14:40quickly recognize this a self-driving
14:41car mostly because it has a lot of
14:44sensors typically around its rooftop and
14:47all these sensors are useful for map
14:50creation and map update purpose we use a
14:54combination of different type of sensors
14:56that includes cameras lidar GPS IMU it
15:02which is a unit that tracks the movement
15:04of the car and the radars as well are
15:08you using everything you possibly can or
15:11have you made strategic choices about
15:12what the best combination and why those
15:15tools are we try to actually make good
15:17use of all the sensors that are going to
15:20be mounted on a let's say a typical
15:23self-driving car the reason is that
15:25self-driving car number one requirement
15:28is used to be very safe yeah we want to
15:30make sure we can take advantage of all
15:32the sensors in case when the cars is
15:36is running and one type of a sensor may
15:38fail or make it blocked we can actually
15:40switch between different type of sensors
15:42in those as well so why isn't just
15:47visual visuals really important but I
15:50mean for the same reason way is saying
15:53today in the past they would use other
15:55methods triangulation and taking
16:01mathematical calculations that they were
16:04doing essentially by hand which are now
16:06done just instant you know either you
16:08measure with a ruler or with I don't
16:10know horses you're trying to get some
16:15kind of distance matter
16:17through triangulation with camera data
16:20you can do some triangulation and get
16:22distance estimate for sure but when
16:25you're talking about extreme high
16:27precision measurement cameras oftentimes
16:30it's just not enough
16:31and lasers gives you a precise
16:34measurement of depth or distance in the
16:373d space if we're talking about sensors
16:40that are mounted on the self-driving car
16:43lighter and the cameras are always
16:45working together generally speaking
16:47they're all running very fast multiple
16:49times per second as the car is driving
16:52for instance at fairly high speed I
16:54don't know seventy miles per hour on the
16:57highway you were the car is generating
17:00or collecting a lot of data as it's
17:02driving at the high speed we suppose the
17:05cameras the lidar the radars and all the
17:08other sensors do you have a little like
17:10map making bubble basically that you're
17:13traveling around the world in also
17:15reading the map it's consuming the map
17:17and creating the map SRAM
17:18time and it's not one bubble with lots
17:21of bubbles well that's what's
17:22interesting right because maybe because
17:24we were always consuming and creating
17:27the map at the same time would you say
17:29or is that is that different well there
17:31often was a time lag though yeah and you
17:34know the the the sources of information
17:37long ago would be all the way from
17:40sextants that we talked about but also
17:42itineraries you know people would walk
17:44or go by horse and say how many hours it
17:47took from this town to the next and they
17:50what the distance was yeah very so it
17:53was never as instant as it is today what
17:56fascinates me about the 21st century
17:59mapping is it's so open you know way
18:03saying that the cars on the road will
18:06actually share information with each
18:08other sharing information and the
18:10history of mapping has been really
18:13important in moving accuracy ahead and
18:16not every state in the world believed in
18:22sharing information for instance at one
18:24time in the period of explorations the
18:26Spanish made extraordinary maps but they
18:30did not share them they viewed the
18:31information as totally proprietary and
18:34highly valuable but their competitors
18:36for instance in the Pacific Northwest of
18:38the United States the English were
18:40releasing all of their map information
18:43as soon as Vancouver explored the area
18:45around Seattle all that information went
18:48to press in Britain and was widely
18:50circulated so it's their names that
18:53lasted and it was their settlers that
18:55came so the whole notion of open content
18:58goes back a long time as being a very
19:01successful business model generally
19:06those who hold information really tight
19:09it may have short term value but long
19:12term it does not have nearly as much
19:14value as sharing so I think that's one
19:17very important themes in terms of the
19:19history of Car Talk I love that so we
19:21talked a little bit about the sensors
19:23are there ways that you gather the
19:24information other sources that are
19:27perhaps widely available that you guys
19:28are pulling in as part of this the other
19:30part is obviously the software and then
19:32the software piece is really interesting
19:34because as you mentioned each car is a
19:37little bubble cloud data and obviously
19:39the software used to be there to power
19:41the hardware to collect recorded data
19:44and then this information somehow used
19:46to be shared yeah so where are you right
19:50now with your one-to-one map of reality
19:53are you already how long does it take to
19:57get to get that kind of depth of map I
20:00mean is it just one car driving through
20:03enough one soldier on a horse I don't
20:06think one soldier on the horse is
20:08sufficient a single car cannot map the
20:10entire world the reason is that when we
20:13think about how the we not only think
20:16about how the map gets created we also
20:18think about how the map needs to be
20:20maintained and updated and updated to
20:22reflect the changes on the road so
20:25instead of having one soldier with one
20:27house you know I don't know if driving
20:30down the road once we design for
20:33aggregated effect multiple cars driving
20:36down the same road and we aggregate all
20:38that data from multiple drives multiple
20:41cars together and the more cars you have
20:43on the road or the more data that we
20:46collect from the higher quality the map
20:50a constellation a constellation yes many
20:52bubbles see you're dealing with enormous
20:54amounts of data that you have to
20:56instantly update right receive and
20:58process and send what are some of the
21:01challenges are the ways that you think
21:02about dealing with that cost is
21:04definitely a big challenge because you
21:06don't want to send tons and tons of data
21:08over cellular network so the way that we
21:11think about how to solve the problem is
21:13to almost categorize what information
21:16that we to share in real time and what
21:19information we don't need to share in
21:21real time that can be to David's point a
21:24slight lag that's right
21:25interesting how do you qualify those
21:27different types of well things that
21:29really impact driving behavior needs to
21:32be shared and then distributed to other
21:36cars that may be be affected such as you
21:40know accident on a road right or maybe
21:43you know you know ranked last rainy
21:44season they're adding a barrier there
21:46were a lot of landslides not trees
21:48falling down on the road how about
21:50faster things like a deer runs across
21:51the road you can but I'm not sure if
21:54that information is very valuable to
21:57share with the car behind you it's true
21:58right yeah there are certain I mean as
22:01much as like there might be more yeah
22:05that's possible or if you have I'm I'm
22:08just thinking out and out if you have a
22:10delivery man who is spilling out I don't
22:16yes lower than a second yes yeah but as
22:21a human driver if you needed to respond
22:24to that scenario in the real time and
22:25the car behind you would respond in a
22:28similar fashion that information should
22:31be shared as quickly as possible if it's
22:34something that needs to be in the
22:35broader driving ecosystem instantly
22:37right let's say you know there's a new
22:41road construction happening which may
22:43take quite some while to actually
22:45complete and as the cars are going over
22:48that stretch of road
22:50you can keep collecting the data till
22:52the change actually gets into or once
22:56the change is actually in effect and
22:58then you distribute that information
23:00more broadly to all the cars it's so
23:02interesting that it's sort of exploding
23:03the time dimension of map to you know
23:06it's like becoming a living organism
23:08it's a very much organic process the way
23:12I think about it so now you have all
23:14this data that we're collecting in these
23:16deep maps what happens to those where do
23:19they live and how do they how long do
23:21they live for you know are we storing up
23:23this map of the world that will you know
23:27that hat is recording it in a way we've
23:28sort of never seen before there are a
23:30lot of data being collected now and
23:34there are gonna be a lot more data being
23:35collected right in the future once you
23:37have more self-driving cars on the road
23:39this is where a cloud infrastructure
23:42comes into place of course it's the
23:44automated storage place and it's also
23:46the place where 99% of computation is
23:50done for map creation and map updates
23:53but then each individual car will have
23:56its own I don't know memory cells as
23:58well as its own intelligence another
24:01term people tend to use is called edge
24:04computing essentially each car will
24:06carry some storage and some computing
24:09power so that it can make its own
24:11decision independently right if you know
24:14it's completely off lying it should be
24:16still fully functional so the maps are
24:18constantly being called by the rest of
24:21self-driving software stack such as the
24:24perception system such as the planning
24:26and control system again many times per
24:28second the interpretation the inter
24:31and as you come across a change a change
24:34detection module needs to kick in and
24:37say that looks different from what my
24:40map tells me now I need to decide if
24:43that's an important enough change to be
24:45distributed in real time or you know
24:49through the edge compute process that
24:50the computers on the cars need to decide
24:54whether or not I need to actually maybe
24:56enter active data collection mode and
24:59then share that data with the cloud so
25:01the decision-making is separate from the
25:03map it's completely two separate
25:06functions so on a very high level the
25:09self-driving sulfurous that consists of
25:11four components the first component is
25:14called the perception system you can
25:17think of it as the eyes of humans it's
25:20trying to see what's around the car and
25:22figuring out for instance do I see a
25:25human crossing the road
25:27or do I see the signal light being red
25:29or green another piece is called
25:32localization and localization module
25:34basically tells the car where you are in
25:38the 3d space and what's actually around
25:40you and says ok you are a hundred and
25:44fifty three centimeters from the next
25:46stop line and here's a crosswalk of this
25:50with you need to do XY and Z this car
25:52and that's what the planning and control
25:54module will kickin and say ok I'm gonna
25:57slow down right and then make a full
25:59stop at the next intersection and then
26:02the first component is the mapping
26:03component and the map income you can
26:06think of that mapping is having
26:08tentacles into all these three things
26:10that I just mentioned earlier because
26:12for instance for localization mapping
26:16and localization actually work very very
26:18tightly together constantly comparing
26:20it's where you are in the map needs to
26:23know what is supposed to be here and
26:26then it can tell the difference right if
26:28I know they're supposed to be
26:30intersection and the crosswalk and then
26:33I see a moving object around there like
26:35crossing it that's probably a pedestrian
26:37right that seems like something really
26:38new to me that a map can
26:40only have a have represent what's
26:43actually there but what a sort of shadow
26:45map of what it thought was there you
26:48know what it was was once there before
26:50or might be it might have been there on
26:52Thursday you know but isn't there now
26:54it's a living thing I think it's a real
26:56revolution and mapping I've never seen
26:58anything like it and of course I want to
27:00collect it well let's what is it like
27:03what does it look like to the car does
27:05the car see something does a passenger
27:06see anything has David seen what this
27:09map looks like that he wants to go oh
27:11yes I guess data is data at the end of
27:19the day how we visualize is up to you
27:22know how we want to consume it yeah it's
27:24like it was the not the knowledge
27:26everyone visualized the knowledge in a
27:28different way right in the Renaissance
27:29and in the digital map you have even
27:32more options and color-coded you can
27:34apply different lifestyle so on and so
27:36forth but generally you have two
27:39audiences now the humans and the robots
27:41the whole robots don't really need to
27:44see anything they need to consume the
27:47data and they consume the data through
27:49API sand and so on so what's the point
27:51of having a visual representations and
27:53at all then those are for humans as part
27:57of the map creation and update process
27:59we will have human moderators to check
28:02the computer output to see you know
28:05there are always ambiguities in the real
28:08world there are certain intersections
28:10that are so complex that are hard even
28:13for humans to interpret the traffic
28:15rules yeah dysfunction Junction so we
28:22have visualization tools to help us as
28:25part of the map creation and maintenance
28:27process but again theta is theta and
28:31theta can be useful in so many different
28:34ways there are so many aesthetic choices
28:36that have go into map making which I'm
28:39sure part of why you love maps so much
28:42it's an art so what kind of aesthetic
28:44choices did you guys make or how are you
28:47you know that when the human is looking
28:49at this this this deep map that you guys
28:52have been thinking about map
28:54and maintenance I think for us at least
28:57functionality kind of trumps over the
28:59beauty aspect because we need to make
29:02sure the quality of map is very very
29:04high otherwise our human operators won't
29:07be able to work as effectively so I
29:10think productivity and the efficiency is
29:13about design for but when we are
29:16displaying the maps for other purpose
29:18you know for simulation for virtual
29:21reality then you know depending on which
29:23use case we are looking at we can make
29:26the right no design choices oh say it
29:29see it's not difficult to change maps
29:31have always been also a representation
29:33right of law in some form Li of sort of
29:37territory and rules and where people are
29:39allowed and where they are how do you
29:41guys think about some of the legal
29:43issues apps have have always been very
29:47sensitive in law their claims and
29:49counterclaims the Sea of Japan is it did
29:53you know others want to call it the sea
29:55of Korea or maps have always had a
29:57function and administrative function you
29:59know a taxing function a governing
30:01function Wars awful as they are produce
30:04enormous amount of mapping and I've
30:07collected and I fought over mapping I
30:11got a three-volume Atlas that was from
30:161786 that had been captured from a
30:18French vessel and it was a fight between
30:21the British and the French right off of
30:23Brittany in France during the American
30:25Revolution and the British captain in
30:27his notes felt that the three atlases he
30:31got were more valuable than the ship you
30:34know it's just in those days that kind
30:37of knowledge was very very powerful one
30:40of the problems today is of course it's
30:42so malleable yeah it can be
30:44counterfeited it can be falsified it's
30:46it isn't printed so you you know people
30:49will put out false claims on maps and
30:52and they circulate and yeah and then you
30:55don't know what it's real so how do you
30:57guys make sure that that you protect
31:00that information those three valuable
31:02atlases on the ship it's gonna be a big
31:04challenge for HD mapping because
31:07as of today even government regulations
31:09are not clear even though digital maps
31:11are widely available today but there are
31:13countries who heavily heavily regulate
31:16their geospatial data you know China is
31:19one sauce career is one there is
31:21actually a whole bunch of other
31:22countries who just like vain yeah this
31:26is Li forbid exporting geospatial data
31:31you cannot move data out of South Korea
31:33about the boundaries and the roads of
31:37South Korea streets so that problem
31:39actually already exists today with just
31:41the normal navigation maps when it comes
31:44to HD maps you can imagine all sorts of
31:46sensitivity because you have these cars
31:49with very high resolution lidar and then
31:52cameras constantly scanning the street
31:54right and possibly see into people's
31:57private driveways as well so being able
32:00to protect people's privacy and fulfil
32:04some government regulations from
32:06security point of view we can do a lot
32:08of encryption and so on to make sure
32:09that data don't accidentally leak out
32:12having said that the future remains to
32:15be seen US government is fairly active
32:18you know trying to post protect and
32:21advance the self-driving technology and
32:24make sure privacy is people's privacy is
32:27well protected and it's I'm out
32:29literally week by week that's right and
32:31the other governments are doing the same
32:33we will I think in a few years of time a
32:35lot of dust will settle and we'll see
32:38what is the nobles from the technology
32:40side what we can do as well as from
32:43regulation side what the government is
32:45willing to do for instance who should
32:47own the data as the driver of the car
32:49or as the private homeowner do I own the
32:53data around my private Hortonville hard
32:55make or the car maker or does the map
32:57maker or show the government own the
32:59data what's the classic thinking about
33:01that in terms of maps was it the person
33:02who drew the map that owned the data or
33:04was it the person who was hired to draw
33:06the map I think all these issues just
33:13speak to the power of maps and they're
33:16just getting more powerful all the time
33:18because they they embody all of these
33:21kinds of sensitivities they tell history
33:24and they record our lives and
33:26particularly these now these living
33:29dynamic maps which we've never had
33:31before I love this that you talk about
33:33the maps as capturing a moment in
33:35history too because if you're capturing
33:38this kind of this depth of map and this
33:40representation of kind of immersive
33:43worlds that we've never had before
33:44it is capturing an incredible moment of
33:47time do you think about the historic
33:49value of capturing the world like that I
33:52was saying to way earlier we are coming
33:55over here please hold on to the data
33:58because it's like capturing it's like
34:06before we had our conversation I to be
34:09honest I didn't because you know when it
34:11comes to the value is to have it in the
34:14present just always about freshest it
34:17should be that's exactly what you should
34:19be doing but at the same time you should
34:21you know preserve I mean I used to be an
34:26art book publisher and one of my
34:27favorite books on the list at Princeton
34:29was the Barrington Atlas Oh which is
34:31just a wonderful representation of what
34:34the world used to be and you can think
34:35about people doing these kinds of
34:37looking backwards in an in totally
34:40different way last thing you know when
34:42we think about terra incognita and then
34:44it becomes terra kognito right it opens
34:46up all sorts of new population shifts
34:49and trade routes and products and
34:51economies we've talked a lot about the
34:53autonomous vehicle what do you see some
34:54of the other sort of unexpected
34:55consequences of mapping out the world
34:58like this possibly being essentially
35:00when we are creating the eight HD maps
35:03we're creating a digital infrastructure
35:05of a cities road network and they do try
35:10to create a digital infrastructure today
35:12because almost every city in the US has
35:14a GIS department they send out a
35:16surveyors they map the the road the
35:20Public Works Department oftentimes to
35:22the same this process is static because
35:25it's very costly to do is very
35:27labor-intensive and can't repeat these
35:29surveys all the time on the other hand
35:34you have cars with the rice sensors
35:37running on the road they're constantly
35:39updating the road so the process of
35:41actually having a central government
35:43having to maintain that social
35:45infrastructure can suddenly be solved
35:47with people with cars or just cars and
35:50this information is tremendously useful
35:53beyond just self-driving having seen
35:56maps evolve what strikes you as some of
35:59the possible consequences of thinking of
36:00maps of it these exploded Maps
36:03well everybody becomes a mapper which is
36:06very powerful everybody takes ownership
36:09and because they contribute as they're
36:11driving to making the great map I think
36:15it will have an effect on people to be
36:17part of the knowledge creation system
36:19I'm not sure exactly how but I think
36:22it's part of empowering people and
36:25historians and for historians to have
36:28access to you know what people were
36:31doing 50 years earlier I mean you can
36:34imagine like a Barrington's atlas that
36:36you would literally be submerged into
36:38and it's almost too much to see patterns
36:41in this vast also read it just for human
36:47patterns it's always been something
36:48we've wanted so badly was to be able to
36:51do you know optical character
36:53recognition in old maps but it's
36:55impossible to do because the text is
36:57going up and down and they're different
37:00fonts and we love them but with the deep
37:06maps really they can they are data to
37:11begin with so it completely gets over
37:14that particular issue and I see deep
37:17maps being made in rivers and seas we're
37:19opening up a whole new frontier of
37:22mapping that is mapping at a very very
37:25high definitional level which creates a
37:28whole new kind of map when you map at
37:30that level of accuracy you are making a
37:33living map a map that is full of so much
37:37information that we've never had you're
37:39really mapping the world one to one and
37:41that's sort of the Holy Grail so there's
37:44still tons of terra incognita that needs
37:48and it's all in front of us I think it's
37:50a big big paradigm shift when we all
37:54become map makers we're all map makers
37:56what a wonderful note to end on thank
37:58you both so much for joining us thank
38:00you too thank you very much