Self-driving cars are tantalizingly close, with real-world trials in progress around the world, but they still face several challenges before they can become mainstream – particularly when it comes to building the high quality maps that will be essential for them to drive safely.
Many automakers are working on their own mapping systems, or using ones from different vendors, but it would make much more sense if there was a way of pooling and standardizing all that information so that everyone can benefit.
HERE Technologies specializes in mapping, and hopes to make that goal a reality to accurately map the world in preparation for an autonomous future. We spoke to Sanjay Sood, the company's vice president of highly automated driving, to find out how.
HERE and now
HERE Technologies is owned by a consortium of carmakers (BMW, Audi and Daimler), which despite being rivals, are working together to usher in a new era of driving. It specializes in creating what it calls HD maps, which are detailed enough to provide autonomous cars with the data they need to make decisions safely, and are kept updated by collecting real-world data from drivers.
Four or five years ago, Sood says, there was skepticism among automakers about whether HD maps (or any maps at all) would actually be necessary for autonomous cars, which some suggesting that the car's sensors might be enough, but there now seems to be a consensus that those alone aren't enough.
"We've been in the business of HD maps since 2013," explains Sood. "We pride ourselves as an organization for pretty much inventing the notion of an HD map in conjunction with automated driving systems."
That work began with an automated driving demo for Daimler in 2013, which followed the Bertha Benz Memorial Route and marked the 125th anniversary of the world's first long distance road trip.
"At that point we were providing map content for infotainment use cases, or navigation use cases for Daimler, and together we said, hey, look, there's probably a need for a higher definition map – something that's way more accurate than anything we've ever created before – to assist the self-driving software to understand its location, its context, to be able to operate smoothly and safely," says Sood.
"So we built a prototype for them, and then we realized very quickly that this is a pretty key ingredient in a self-driving car stack. And so for the last five or six years now, we've been working on productionizing the technology. So taking it from a concept car to something that actually can scale and cover massive amounts of geography, because we think that's one of the one of the key components to successfully deploying the technology to be able to do global scale, or a continent scale."
Building the map
Building any HD map is a challenge, and keeping it constantly updated even more so, due to the changing nature of road layouts and features.
The traditional method of collecting map data is industrial capture – high precision data capture from aerial sources (like planes or drones), satellites, and terrestrial sources (specially equipped cars).
Another method crowdsourcing, capturing data from the sensors on production cars, or mobile phones. This data is much lower fidelity than that recorded using industrial capture, but if you have enough of it, you can aggregate it and use it to build a map.
"Now the good thing about that approach is that it's scalable," Sood explains. "It doesn't cost as much as building these custom mapping equipment systems. The downside, however, is that the data is relatively inaccurate because it doesn't have lots of hardware to support the accuracy of the collection. Also, it’s just not available widespread today."
HERE Technologies therefore uses a combination of both, creating maps using industrial capture and making small updates using crowdsourced data.
"So we've created this big base map out there that's highly accurate, and now you can take this low-fidelity data that comes from cell phones or production vehicles, and we pull it together among various car companies, and then we use that to help update the map," Sood says.
"So we're creating one global map that many OEMs can benefit from because they're all providing data to a pool, that helps the industry as a whole maintain and create an asset that can be used for autonomous driving.
"So I think our approach is very different than companies that are doing a pure ‘we're going to crowdsource this thing’, or other companies that are doing an industrial approach, but have no access to sensor data because they're not as closely aligned with the automotive industry."
Mapping the world also introduces some tough political challenges. Roaming the streets of LA with a sensor-studded car is one thing, but Pyongyang is a different matter entirely. To solve that problem, HERE has created a partnership called the OneMap Alliance.
"In certain geographies you have geopolitical issues that make it very hard to become a mapmaker on a global scale," says Sood. "For example, China, where you have to have a government license to map the roads and create these types of geospatial representations. We partnered with a company which is owned by Tencent, and what we do with them is essentially creating a global HD live map offering that includes China. We work very closely with NavInfo on ensuring that the map that they create in China aligns with our global specification."
"And what that means for the OEM, then, is that if they want to purchase or license map content, they can essentially license one map protocol that works across China and the rest of the world, so they don't have to build a separate system for China and one for the rest of the world. They can build one and it works seamlessly."
HERE has taken the same approach with South Korea, where limitations on taking mapping data overseas means all the processing is done in-country in partnership with the country's second-largest telecoms operator, SKT Telecom.
In Japan, HERE works with Pioneer to avoid issues with the country's limitations on the number of players in the market.
"This kind of global footprint [...] requires a level of scale that few companies can bring," says Sood. "HERE is one of those few companies that had the global footprint, and can form these types of alliances to really create a global offering for the industry."
Mapping a data-centric world
So which areas has HERE HD-mapped so far? Currently North America and Western Europe have the most coverage, since that's where most of the company's customers are based, but the Middle East and Southeast Asia are increasingly well documented too.
"Where we focus is essentially where we see this technology being adopted in production format first," says Sood. "Primarily with the customers we work with, it’s going to be North America, Western Europe. And then there are certain places like Singapore, Japan, where we have partnerships, China where we also have a partnership for HD maps, South Korea as well.
"So you have the advanced economies or growing technology economies in Asia, and then the traditional places where you see this automotive technology being rolled out, which is North America and Western Europe."
Of course, all that global data is only useful if it's consistent, so HERE is also working on establishing standards so that automakers worldwide can use its maps – and contribute, too.
"We're really about trying to open up and standardize so it's very easy for companies to operate in this data-centric world, and be able to exchange data between proprietary back-ends, and essentially be a global cloud like the IoT," says Sood.
Trust is still one of the biggest limitations for self-driving cars, but with a dependable global mapping system that's constantly updated with even small changes, they'll be a big step closer to winning our confidence.