Nikhil:
I find it to be just as fascinating. I feel we are, the tip of the iceberg here, I feel there are so many applications of that and ways in which it can be utilized. City officials can finally start identifying how the city streets are being used. There’s so many ways in which this can help not only guide analytics but also help with design problems, et cetera. I’m really excited about the ways in which people will start using the data. Autonomous driving is one application, but I don’t think that’s the only application.

Daniel:
Hello and welcome to another episode of the Mapscaping podcast. My name is Daniel, and this is a podcast where I interview people that are doing really interesting things in the geospatial world. Today I’m talking to Nikhil, and he has built a system that lets you crowd source data for machine maps. It’s a really big idea and a really interesting conversation. I hope you enjoy it.

Daniel:
Hi Nikhil, thank you so much for coming along today and doing this interview with me. I’ve been really looking forward to talking to you. I think your company and what you’re doing is incredibly exciting. But maybe we should start off by learning a little bit more about you. Where do you come from, and where are you living now?

Nikhil:
Hey Daniel. I moved to the US from the India, I grew up in India, I grew up in Bangalore, studied engineering when I was there and moved to the US for Grad studies. This was in 2006. Moved to Pittsburgh, started my Grad program at Carnegie Mellon, was really attracted to their robotics program and joined there, got a Masters Degree there and moved to the bay area, continued Grad school at UC Berkeley. Hung out there for a few years, wrapped up my PhD and then worked in industry for a few years before founding this company, Mapper.

Daniel:
Okay. What was it that told you, hey, you should get out of robotics and into mapping?

Nikhil:
I still consider myself to be a roboticist and I’m still doing the robotics, and the interesting thing about what we are building here at Mapper is, we’re building not maps for consumers but maps the machines we call them machine readable maps. It’s very much mapping for robotics, mapping for applications of robotics like autonomous vehicles, drones, augmented reality, et cetera. I’m sitting in that intersection between mapping and robotics, which is my sweet spot.

Daniel:
Yeah, I see. Obviously if you’re building maps for machines, it’s a good idea to have a solid background in what machines are looking for. For those of us that are maybe new to the idea of machine maps, can you give us an idea of what a machine map would look, or is it something us humans could relate to at all? Or is it completely different from what we understand is maps?

Nikhil:
It is not, it is actually just more detailed and much more accurate and up to date. That’s basically what it is. The way we as people get ourselves situationally aware in any place is, we use regular maps like Google Maps or maps from open street maps, but then we have millions of years of evolution to guide with getting context. By just looking around, we can figure it out what things are. Machines on the other hand don’t have that. They need an extra layer of data, which is much more precise, much more up to date, layered on top of traditional maps in order to guide them through the world.

Nikhil:
The interesting thing about applications like autonomous driving , I kind of think of it as a world where there’s interaction between both cyber and people, cyber physical, so people and robots interact with each other. You could be driving a Honda and be at a traffic intersection where there’s another autonomous vehicle trying to navigate and negotiate through that intersection, and the level of information that that vehicle needs is just orders of magnitude higher than what we need in order to complete that seemingly simple task.

Daniel:
Okay. Maybe we should back up just a little bit here, because we started talking about machine maps and moved on to how they were different from people maps and now we’re talking about autonomous vehicles. What’s the connection there? Why autonomous vehicles? They need maps, can they not just make their own maps?

Nikhil:
That’s a really good question. Autonomous vehicles are a very interesting application of robotics because they are at the intersection of perceiving the world and understanding what’s around us, negotiating with other people and other vehicles on the road. It is a challenging problem, one of the most complex software engineering problems there is. Maps serve, for this application of autonomous vehicles, machine maps serve as a prior on the world, a prior on the environment, which guides them and helps them more quickly understand what’s around them. When we think about human maps, we traditionally just think about, you know, here’s a lane, sorry, here’s a road, there’s a body of water to the right, and there are a few buildings to the left and that is typically enough. That combined with a GPS position is sufficient for us to navigate the world.

Nikhil:
For a machine however, that level of information, while useful is not sufficient. It requires a much higher level of granularity, so the kinds of data that we encode, information that we encode in our maps includes things like where are the lane markings, where are the traffic lights, the traffic signs, where are the parking locations, what are the possible directions from which bicyclists might enter an intersection? All of this information, you can think of them as higher fidelity digital signatures that are layered on top of traditional map, and that’s what we consider a machine map. With this information, a machine, like an autonomous vehicle or your favorite autopilot system on a traditional consumer vehicle can function a lot more precisely, a lot more robustly, and that’s what we’re enabling for these kinds of use cases.

Daniel:
It’s so clear that you know a lot about is subject, and you’ve really spent some time invested in it and learning about it. It’s so clear from talking to you. I think what I found the most exciting about your work is the way you’re collecting this data. I think it’s a really unique approach. Could you tell us a little bit more about that?

Nikhil:
Absolutely. Let me begin by talking a little bit about how traditional mapping companies collect data and what they do with that data, and then explain what we’re doing. Traditional companies, these are companies like TomTom and [inaudible 00:07:25], Google maps, Google street view, et cetera, operate these very expensive vehicles that typically cost about a half a million dollars, they’re high end sensors, they have high end inertial devices on board, and they’re typically driven by two professional drivers. One driver who’s driving, the other guiding the driver, and they do all of this in order to collect data for the purpose of building consumer maps. They don’t need to drive very frequently, typically you find that they will remap any city once every six months to a year, sometimes even longer depending on what the city is. And all of this data is collected, they go back to their offices and they ship the data to data centers where they’re manually processed and conflated and then convert it into maps that we see on an iPhone or in our in dash display in a vehicle.

Nikhil:
This process is great for consumer mapping, however, it doesn’t work for machines. Machines, when I say machines, I mean applications like autonomous driving that acquire maps that are a couple of orders of magnitude more accurate and a couple of orders of magnitude more up to date than the traditional maps. You cannot rely on this old school way of acquiring data in order to build these maps. And so what we decided to do at Mapper is figure out a way to democratize this acquisition of data by leveraging vehicles that are already driving around on the streets. We work with an army of freelancers, be it the Uber drivers, Lyft drivers, construction workers, whoever wants to have a side hustle and make a little bit of extra money on the side, we can work with them.

Nikhil:
have created and designed a really low cost, 3D mapping system that can be easily installed in these vehicles, and as they drive around, they acquire the data with which we can build those maps that machines can utilize, essentially maps that are significantly more accurate and significantly more up to date than the old school mapping companies were doing with their [inaudible 00:09:40] vehicles.

Daniel:
As I said earlier, I think this idea is really exciting and incredibly innovative. I love the fact that I, tomorrow, if I wanted to, obviously if there was a need for this data as well, could get one of these mapping systems, your sensors and strap it to my car and drive around and be earning money at the same time. I love that idea of multitasking and of course we’ve seen this with Uber, everyone’s aware of that, and we’ve seen the idea of the gig economy, you know, moving into more and more different places. I think this is going to be an idea that’s really going to change the way we work in the future, but yeah. If we get back to mapping, do you think something like this would be possible without having open street map as a predecessor?

Nikhil:
That’s a really good question. I don’t think it would be possible. A lot of these ideas, I mean, I stand on the shoulders of giants here, and the people behind open street map who democratize just the creation of consumer maps to me was fascinating, and it just grew virally around the world, it was fascinating to me. There was one of the ways in which I got inspired building machine maps. It was sort of the combination of people’s desire to go map. People inherently want to think in a special manner, they want to acquire the data that will enable them to navigate the world more efficiently. On the other end of the spectrum, we have companies like Uber that started democratizing access to transportation, and that is where the idea for Mapper was born, leveraging both a freelance economy, as well as people’s inherent desire to create maps.

Nikhil:
In fact, some of the Uber drivers and Lyft drivers who are our mapping partners, they just love the job. They are very happy that they don’t have to deal with drunk passengers in the back, and they get to go and create some … One of them in fact called it digital infrastructure, which they really like to be a part of, they feel they’re part of something and that is something that makes me very excited about in the way in which we’re approaching this problem.

Daniel:
I think one of the really motivating things with open street map as you can see your edits, I can go out tomorrow and I can make a change in the map and of course if it’s accepted by the moderators I can see it. You know what I mean? I can go to openstreetmap.com, and I can see, hey I did that there. Is that possible with your maps or is it just sort of uploaded to the cloud and sort of mixed up with the other data that’s there.

Nikhil:
We actually leverage our drivers to not just build the map but also verify the map, so they get to see what they’ve created. Once they have driven and acquired data where no map exists and the data is uploaded, we can load it into maps, and then we download those maps back to them so that they can go back and drive those same routes again and verify that the maps are accurate. The best person to verify is the one who created in the first place, so they feel like they’re part of this. We close the loop with them, and that’s something that they take a lot of pride in.

Daniel:
Is that also a part of your ground truthing? I’m assuming that the more you drive the route, the more you confirm that yes, that building is there, yes, that sign actually exists, because now I’ve seen it two times, three times, four times or multiple times in my data. Is this part of the ground truthing as well for your data? Because I think this is going to be one of the big questions. It’s also one of the big questions surrounding Wikipedia when it first started up. How can we trust it? Can the crowd moderate to such a level that we can accept. And before we talked a little bit about the need for incredibly accurate data, is this I guess, is this a way that you test your data? Is this a way that you check your data?

Nikhil:
Absolutely. The Wikipedia example is spot on. In our case, we think that that is crucial for us to create a product that others can rely on. An autonomous vehicle manufacturer, or a car manufacturer or a truck manufacturer, if they’re going to start leveraging these machine maps, they want to treat them as ground truth. They want to treat them as the foundation on top of which those applications can exist. It behooves us to make sure that it is absolutely accurate. The way we do that is, we don’t think of our maps as a one and done map, it is a constant flywheel. As these drivers drive around, they’re acquiring the data with which we’re constantly verifying if the map was correct and if things have changed.

Nikhil:
For instance, we have a very well developed system for detecting changes in the scenario that this construction happening at a certain intersection and lane markings have been [inaudible 00:14:35] by a foot to the left in order to accommodate the construction. This is something that we can detect in our maps when the driver drives by and update the maps with that. Then two days later when the construction process is complete and the lane markings and moved back, we reflect those changes in our maps as well. We are essentially trying to create a digital layer that reflects the physical world as accurately as possible so that kind of ground truthing is crucial for our product.

Daniel:
This just occurred to me when you were talking about that, the constant updating of this map. This is going to be a such an incredible resource when you guys are at scale, when this is happening at scale around thousands, if not hundreds of thousands of people who are going to be contributing to this data set. I was just thinking the possibilities in terms of analysis for change over time in an urban setting are just incredible. They’re mind blowing.

Nikhil:
I find it to be just as fascinating. I feel we’re at the tip of the iceberg here. I feel there are so many applications of that and ways in which it can be utilized. City officials can finally start identifying how the city streets are being used. There’s so many ways in which this can help, not only guide analytics, but also help with design problems, et cetera. I’m really excited about the ways in which people will start using the data. Autonomous driving is one application, but I don’t think that’s the only application.

Daniel:
I’m not sure if you’ve seen this, but Google earth engine is a very simple web application, and I realized that we’re talking about a much more complex dataset. But it’s an interesting thing, if you go along, you can choose any spot on the world and you can see change over time, over the last 30 years. I’m assuming you’ll be capable of producing dataset like that when, you know, when this hits scale.

Nikhil:
That’s right. Our desire over here is that this, we’re capturing those changes with 30 seconds, not 30 years, right. The value of being able to create, I think of this is no longer as just a map, but it’s priors on the environments, it’s spatial priors which can help in a whole bunch of applications, of course, autonomous driving being one of them. And the more vehicles driving and contributing the data, the more up to date and accurate it is.

Daniel:
Could you imagine this kind of technology will exist in cars in the future as a standard? When we talk about the internet of things in your version of the future, does every car have a dashboard cam and some other built in sensors and is constantly updating and editing one sort of global source of truth in terms of a map for machines?

Nikhil:
Absolutely. I actually feel that that’s where this is going to go because, I think of our business as eventually becoming a network effects business. It’s like a LinkedIn or a Facebook, a network effects business for spatial data, for machine applications. Vehicles are already starting to have advanced sensors on them. Tesla, as an example, has eight to 10 cameras, the new General Motors vehicles have a whole bunch of cameras, and soon they’ll have more than just cameras and radar, they’ll have 3D lidar sensors as well. As they drive around, the acquisition of high quality data will become more and more commoditized, and they will be a natural sort of progression where this data is being conflated by single and, you know, federated by a single company, and I’m hoping M is that company that can enable that network effect and enable every single automobile or every single moving vehicle, anything that moves to have a better quality data sets to make decisions.

Daniel:
You mentioned Lidar there, and I’m sure you mentioned it earlier as well. Is this a part of your sensor set up for these drivers?

Nikhil:
We have a couple of setups. Lidars are 3D sensors that are essentially used for serving applications traditionally and now they’re being used quite a bit in automobile applications including autonomous driving. The sensors cost quite a bit of money right now because it’s still a cottage industry. No one has really figured out how to build these lidars at scale. There’s a lot of manual assembly that goes into each one and as it is out there, they’re a little bit cost prohibitive at this moment in time. What we have done is, we have built a solution that doesn’t just involve lidars, we also have a camera only solution with which we can update our maps, so that we can scale up acquisition of the data, and we’re not just relying on a lidar based source.

Nikhil:
For every single Alpha unit that has a lidar, we have a dozen Beta units that are camera only, so that way we can acquire data at a much higher frequency. In the future however, I see the cost of this hardware dropping, it is already starting to drop. In fact, the cost of these 3D lidars have dropped down by half in the past 18 months, and I anticipate that happening again over the next 18 months. It’s almost like the Moore’s law of Lidar.

Daniel:
This method that you’re describing of collecting data seems to be really focused on roads and, and until now we’ve only really talked about autonomous vehicles, and when we’re talking about autonomous vehicles, it feels very much like we’re only really concerned with traffic on the road. Is that the case or are you also thinking about other kinds of autonomous vehicles? I remember not so long ago there was a lot of hype about Amazon maybe starting to deliver packages via drones, and also I’ve heard about companies investigating, I’m not sure if anyone’s made a prototype yet, but investigating small robots that sort of travel in pedestrian areas and solving that last mile problem, so the trip to the supermarket would maybe become obsolete because we could order everything, it would be sent into a little robot and sent via these pedestrian walkways and things that we have to the house. Is that something that you guys are thinking about? And how would that work in terms of data collection?

Nikhil:
That’s a really good question. In fact, we already have customers for solving such middle mile and last mile problems for delivery. We also have companies that are, customers who are building autonomous vehicles in both public roads, as well as private areas like communities that are closed off. We don’t think about our maps as only focusing on public roads. We’re thinking of [inaudible 00:21:37], providing that digital layer, that digital infrastructure wherever anything moves. It could be indoors, it could be outdoors. We’re starting with outdoors because that’s where the market is large right now and that’s where we’re going to see the biggest benefit to society in the near term. Over the long term, I think it’s going to expand into sort of closed ecosystems, things like airports, et cetera, as well as indoors, like malls, perhaps personal robotics and homes, et cetera. These are all applications sectors and use cases that we are tracking, and in fact, we have customers in a few of these verticals already.

Daniel:
Well, I think you’ve honestly come up with a brilliant solution as far as I’m concerned. Not that I’m an expert in the field, but I love this idea that you’re able to collect data at scale like this, and I think that’s the kind of thing that needs to happen for this to be successful. In the past I’ve heard people talk about changing the infrastructure, saying we need to do a lot of changes to our infrastructure, especially here in Europe and the cities. Is that the answer to autonomous vehicles? Is it going to be a little bit of both? Do we have to actually change the infrastructure, the roads the vehicles drive on and have better maps, or can we solve the whole problem with just being more accurate about data collection and making really, really good machine readable maps?

Nikhil:
I’m a firm believer in the latter. That is my thesis that you shouldn’t have to change your city in order to bring in this kind of application. I feel like the technology and the product should be able to naturally fit into an existing society. Think of a lot of European roadways in the city, they were built for horse drawn carriages and then cars were built to that Spec. I think that the next generation of mobility should just follow that same continuum, that it should be built to an existing infrastructure because the cost of changing infrastructure is really, really high, and the return on that investment is not going to be very clear for a very long time. Instead, if city start out with not having to change their infrastructure, but can rely on just digital methods like mapping, et cetera, with which they can introduce these use cases, I think that’s just a lot more scalable and that’s a good way to penetrate the market and to test if people actually care, and if it actually benefits society or not.

Daniel:
I think too and I think anytime you’re introducing a new technology like autonomous vehicles, I think people aren’t going to accept less than what they have today. I don’t think we’re, as a society, as a culture, I don’t think we’re going to accept the fact that we have to rearrange our cities completely to serve robots. After all it works today, I mean you can argue of course how well it works in different countries, but it works. We get from A to B and we don’t have autonomous vehicles. I think whatever comes in the future, it has to be that next step or the same. I don’t think we’ll accept a step backwards.

Nikhil:
That’s right. I agree with that statement.

Daniel:
We’ve talked a lot about accuracy and data collection. What kind of accuracy do we need? Is there any industry standards?

Nikhil:
Typically in applications like this, these industry standards come about because of certain problems and then they keep evolving. The most important thing from an autonomous vehicle manufacturers point of view is that they know exactly which lane the vehicle is in, with very high accuracy. That naturally gives us a sense of, at the very least, what the ceiling on the accuracy should be. Typically if you want to be centered in a lane, you want to be plus or minus 20 centimeters from the center line of the lane as an example, and you’re still driving effectively. If you’re more than half a meter away from the center line of a lane, then you’re essentially drifting into a neighboring lane and can cause an accident. There are some general guidelines in order to sort of benchmark accuracy of maps, benchmark accuracy of the position of the vehicle on the maps and that’s where we can begin.

Nikhil:
But in the future, I actually think that the accuracy is only going to keep improving to the point where now you can start doing a lot of new interesting things. As an example, I don’t know, I grew up in India, like I said, and we have roads there, we have lane markings. A typical two lane street has four vehicles driving side by side. People don’t respect lane markings at all. It’s just that there’re so many more people in so many more cars. It’s complete disorganized chaos in some cities, but in some cases it’s organized chaos there. But at least when I look that it tells me that, if drivers can be more careful, cautious and capable, then the number of lanes can increase. And as the accuracy of the map increases, the accuracy of positioning increases, I think that existing infrastructure can accommodate more vehicles, and that would only be better introducing congestion. I think that starting with the existing guidelines is great, but that’s not a point to stop. It just needs to improve from there, moving forward.

Daniel:
I completely agree. I wonder sometimes when things are so new and as far as in the way I think about autonomous vehicles is, even though people have been working on it for many, many years now, I think in terms of as a society, as a culture, is still a new idea, and we still think of it as being somewhere far off in the future, and maybe our children will have autonomous vehicles. Personally, I hope it comes a lot sooner than that, but I think with a new idea, we tend to over engineer things a little bit if not a lot. When I think about roads for example, of course we need accurate data, no question about it, but like you were saying before, 20 centimeters is not, in the mapping world anyway, that’s not very accurate now. You want millimeter data if you’re planning a building, it has to be within a couple of millimeters.

Daniel:
I think maybe people are over-engineering things at the start and then we’ll sort of hopefully find a middle ground as we move on. How accurate does that data really need to be, because it’ll be a trade off in terms of accuracy and speed I can imagine, in terms of the calculations that need to be done. What are your thoughts on that?

Nikhil:
I agree with that, I do agree with that. I generally think of numbers, like trying to quantify position of a map or position of the position on the map. There has to be a reason to have that number, you don’t want your vehicle to be in the wrong lane is a great motivator. Now, in order to accomplish that outcome, which is that you are as close to the center line of the lane as possible, it might be the case the maps are millimeters in accuracy, your positioning is millimeters in accuracy, but when you combine the two, you might end up with integrated system that has centimeters in accuracy. This is what tends to happen, so trying to make each system as accurate as possible in isolation is great as long as you don’t end up boiling the ocean. I think it’s more important to look at the integrated output and guide what the individual components, how accurate each individual component has to be. That’s how we think about breaking down the accuracy of different aspects of maps, like the position of the lanes, that orientation of traffic lights, et cetera.

Daniel:
I think it’d be interesting too to, I guess at some stage this industry is going to start making comparisons to human drivers. What accuracy can a human driver achieve and therefore, and obviously take it a step further because like I was saying before, I don’t think we’ll accept anything less than what we have at the moment, it’ll need to be better. Are you aware of any studies that have done something like that? Looked at the accuracy of a human driver and said, okay, well if we take that up a notch that could maybe be a starting point for these standards for autonomous vehicles?

Nikhil:
I’m not aware of too many studies like that externally published, but I do know that this is exactly what autonomous vehicle companies are doing internally. They are comparing the machine, the software driver to a human driver, and they’re constantly trying to improve and understand how human driver would deal with the situation in order to improve the machine and likewise. This isn’t new though, it’s funny you were talking about how this is a completely new paradigm shift. If you really think about the transition from horse drawn carriages to cars, I found an old book that stated where, horse and Buggies companies were making statements about how these studies need to be done thoroughly before cars are sold, because the inherent difference between a horse drawn carriage and a car is if a driver falls asleep behind the wheel, a car can drive off a cliff, but a horse will know to naturally stop itself.

Nikhil:
There was already that thought process several hundred years ago. That study should exist between human drivers and at that time horses. Now it almost feels like there needs to be a new definition of horsepower, you know, maybe horse brainpower to compare against an autonomous vehicle. These studies are going to start becoming even more important. What we haven’t yet seen through the … The motivation for introducing autonomous vehicles is to essentially reduce the number of accidents, reduce number of motorway deaths. I do believe that it’s going to happen, but there haven’t been any systematic studies that can measure how the quantify human driver and compare that to autonomous system, but I think that’s a very good idea and I would love to see more autonomous vehicle companies put a focus on that because I think that will improve the general public’s desire to even get into one of these vehicles if they understand that.

Daniel:
I can imagine if I was a company investing heavily in autonomous vehicles, I would hold my cards pretty close to my chest. I wouldn’t go around telling everyone, hey, I’m doing this, Hey, I’m doing that. Maybe we talk a little bit before about standards, how do you know what kind of map to make for these different types of software? I’m assuming there’s different types of software, and they take different inputs and every company is probably, you know, making their own kind of software. How do you work with that?

Nikhil:
That’s right. Let me put it this way, there’s both fulfilling and challenging at the same time. This is like the early days of the internet, when there was no HTML, where there was no XML. This is a time when people are still trying to figure out whether the fork should be on the right side of the plate or the left side of the plate. There’re general specifications and guidelines that are being flushed out and as a mapping company at the stage that … The way we think of ourselves as a development focused company right now, autonomous vehicles are not yet in production. Then in the development phase, the whole industry is in a development phase and maps also being developed. By working with our partners and our customers, we are building up the tribal knowledge internally and coming up with a set of the general specification that can help the entire industry, help the entire ecosystem, and this is still an ongoing challenge, but we really like it because everyday we learn something new and once we learn it, we incorporate that learning into our maps and then all of our customers benefit as a result of that.

Daniel:
If you had to say what the one biggest hurdle of this entire industry is facing, what would you say? Is it cultural? Is it technical?

Nikhil:
From the autonomous vehicle perspective, it is absolutely technical. I don’t think it is cultural. I actually think people will get very easily accustomed to the comfort of a car that can go park itself once it drops you off at a restaurant. There are genuine pain points in driving that automation will remove. I don’t think that’s the issue. The issue truly is a technical one. It’s a very hard technological problem that when we have solved and made slow … It’s like pealing off each layer of an onion here, as we keep doing that and we solve the technical hurdles, I think that option will naturally happen.

Daniel:
We’ve talked a lot about autonomous vehicles now and the future, and I’ve really enjoyed that, I find this topic incredibly exciting. I can see it might take a little bit longer before I can just walk out to the end of my drive in the morning and the car picks me up, it takes me to work and drops me off and then fixes itself while I’m at work and that kind of thing. I look forward to that future, but it feels like it’s still a little bit long way away. If we come back to your system of mapping, because again, I find this really exciting, I think that it’s that perfect little middle step, if you understand what I’m saying. You’re going to make these maps and you’re going to make them at scale. Can you give us some examples of how it’s working today? Are you working with any particular companies? Or could you tell us maybe a little bit about how many drivers do you have? How much data you process on a daily basis? The coverage of the map?

Nikhil:
I can absolutely talk about a few of those. As an example right now we have built maps all over the world. We have maps in Japan, in Taiwan, in Europe, in countries in Europe, in the UK and several cities in the US. And we have proven this system to be functional in all of these different geographies. These are all maps we’re building for customers. We actually build maps where we have customer interest, we don’t proactively map anything, we actually do it once there is a customer for the map and like I said, it’s not a one and done map, it’s just a constant flywheel that we deliver, and so we have drivers in all of these locations.

Nikhil:
The kinds of customers that we have are across a few articles. In the primary, more than half of our customers that are developing autonomous vehicles, be it Robo taxi service, which is essentially an autonomous Uber, or a delivery robot that can Mosey along on a bicycle lane or on the side of the street to bring you your pizza. But we also have other more traditional businesses, we have vehicle manufacturers who are leveraging our maps to improve their driver assistance systems, so they currently sell vehicles that have driver’s assistance, like lane keeping and cruise control, and with our maps, these function a lot more accurately and robustly, so they leverage our maps for these applications.

Nikhil:
We also have traditional mapping companies who have started leveraging us to essentially update their existing maps, or to go and map areas where they have not yet acquired data before. These are the kinds of businesses that have come in, but we already have interest from internet of things, application companies, insurance companies that want to understand better how risk profiles change from a spatial perspective in cities and they’re just a whole number of new applications that come about when an infrastructure like ours exists. We are really excited about the possible ways in which this could be used, and every day I learn about a new application for our product that hadn’t thought of myself.

Daniel:
If we don’t think about navigation a second, in terms of usage of your maps, if you don’t think about navigation, what’s the most exciting bi product of your map, if that makes sense?

Nikhil:
That’s such a good question. Let me tell you that I am not someone who gets excited by any single use case. My focus and the reason I built this company, the reason I’m so excited about what we’re doing, is we’re an enterprise business to business focus company that is creating a platform, creating this infrastructure that can be leveraged by multiple applications. When I find that companies reach out and they say, hey, in fact we just spoke to a company based out of Japan that is interested in building very accurate 3D maps of their maps of their cities because they’re so earthquake prone, and they want to be able to track infrastructure changes for assessing where construction should be focused. This was not an application I could ever have thought about, and it got me so excited to figure out if we could work with them, and we’re in conversations now.

Nikhil:
What I find happening right now, as a result of building out this core capability, this core product, there are companies that hear about us, that read about us and they come to us and say, Hey, do you do X? Do you do Y? Some of them are really far out, and I don’t know if there’s an absolute reason to work with those companies that might not be a tight fit, but there are others who have figured out how they can leverage this new data layer for their internal processes or for their applications, and I couldn’t have thought of, and I think that’s what’s most exciting to me when people are coming in utilizing a product in ways that I didn’t even imagine. That is really exciting to me.

Daniel:
Okay. Now comes a couple of the last questions, the very difficult questions. If you had to say what the future of autonomous vehicles would look like in two years time and in five years time, what would you say?

Nikhil:
That’s a really good question. I’m an engineer by training, and there’s one thing you should know about engineers, engineers always over estimate what can happen in one year and underestimate what can happen in 10. That is always the nature of engineers, right. What that means is in the next couple of years, everyone who says they’re going to have autonomous vehicles at scale, everywhere, driving around, picking people up, dropping them off, it’s probably not going to happen. There’s going to be progress that’s made, but it’s probably not going to happen. But five to 10 years from now, I actually do truly believe that that can happen, so I’d like to think of the way this can actually progress to that level of production where there’s a number of these vehicles in multiple cities.

Nikhil:
It’s going to begin with a few cities where they’re going to prove out the use case, prove out the viability and the economics, and then start replicating this in other places. Once that replication begins, that’s going to be really hard to stop. It is going to be non linear, it’s not going to be linear, so over the next two years, I would imagine we’re somewhere similar to where we are today, but five years from now, I think the world has changed a little bit.

Daniel:
Nikhil, I want to thank you so much for taking the time to talk with me today. I’ve really enjoyed it. You’re clearly an expert in your field and I’m very excited about what you’re doing. I think the application that you built and the platform that you’re building is really going to change the world.

Nikhil:
Thank you so much, Daniel. I really appreciate the interview. I really appreciate the questions and look forward to hearing this and hearing your other podcasts as well.

Daniel:
Thank you so much. Just before you leave us, where can we go to learn more about you and the work that you’re doing?

Nikhil:
Our website, our company is called Mapper, M-A-P-P-E-R, and our website is mapper.ai.

Daniel:
Thank you so much. I’ll link that up in the show notes.

Nikhil:
Thank you so much. Take care.

Daniel:
That’s it for another episode of the Map Scaping podcast, don’t forget, show notes for this podcast and a full transcript are available at mapscaping.com. My name’s Daniel, and I want to thank you so much for tuning in all the way to the end. If you’d like to get in touch with me, it’s Mapscaping on Facebook and Twitter and Mapview on Instagram. I’d love to hear from you. Talk to you soon.