The rise of the software-defined vehicle heralds an age of electric, automated and connected driving, with the user experience shaped by the digital cockpit. These elements could offer unprecedented levels of safety and convenience while slashing the sector’s carbon footprint. However, they also place unprecedented demands on location-based software and data.
Maps with real-time granular insights about routes, drawing on artificial intelligence (AI) and machine learning (ML) to process all the necessary data points, could accelerate the commercialisation of conditional autonomous vehicles and spur the arrival of fully autonomous vehicles. At the same time, electric vehicle (EV) drivers will need reliable information on charging point locations, specs, pricing and availability if they are to overcome range anxiety.
HERE Technologies has been actively transitioning from a mapping provider to a platform supporting all location use cases. Its Chief Product & Technology Officer Giovanni Lanfranchi believes that location is the number one element for a software-defined vehicle’s success. “We have kicked off some pretty key transformations in order to be ready [for the software-defined car] with far more than just a map,” he tells Automotive World.
What does the push towards a software-defined system mean for automotive location data, and specifically for HERE?
Multi-sourced location data is a key element for the software-defined vehicle, which requires a level of precision an order of magnitude up from the maps we as people use to navigate. We can automatically ingest—in real-time—several sources from the physical world and then conflate the raw data to produce one single semantically consistent digital representation of reality. The other important characteristic for location data is low latency, so all these processes are completely automated and performed through algorithms. A software-defined vehicle with a high level of autonomy requires a real-time representation of reality.
How does HERE stand on both of these metrics at the moment?
Our platform has around 500 million kilometres of vehicle probe and sensor data coming in every hour. In terms of latency, we can update a self-healing map with changes in the real world—such as road works or a change in the lane structure—within a matter of minutes. We can produce a map for an entire continent in less than 24 hours, starting from raw data.
On which tools do you draw to achieve that level of freshness?
We are extensively leveraging AI and ML techniques at multiple stages. It is very much a multi-source and multi-dimensional statistical set of problems we face. Let’s imagine you have a set of sources—sensor, satellite and probe data—and you need to build a set of roads. These three sources give you some information but they may have different levels of precision, age, or accuracy for a particular element. At the end of the day, you need to know where to draw your lines. You also need to provide a confidence indicator for automakers, because it’s never black or white. So, we have heavily invested in AI and ML, and we have the best data scientists.
How big is your team of data scientists?
There are more than 150 specialists with AI and ML skills for map making working on this content factory alone. Across the platform we have almost 1,000 people with skills in these areas.
Have you been working with automakers on this wider map-making transformation?
We are working in close partnership with some key OEM partners and hope to tell the world more about what we have been doing together in the months ahead. In fact, it was during workshops that we identified the importance of producing one single, semantically consistent representation of reality. This is very important for the software-defined car.
In order to have a software-defined car you need to have a software-defined map
Can you flesh out what ‘semantically consistent’ means in terms of mapping?
We want to be sure that when the passenger sees a map with a curve, the autonomous driving system sees the same curve with the same level of representation of reality. Traditionally, the in-vehicle infotainment (IVI) requirements are handled through one set of location maps but the autonomous high-definition map draws on a different set of schema. It is left to the automaker to mix and match what is important from an IVI user experience and what is relevant for a Level 2+ or Level 3 autonomous driving system. This brings additional complexity to the OEM and has a lot of user experience drawbacks.
Is this where your Map Object Model (MOM) comes in?
With this we have built a model of the world that is not just a representation of reality, as many competitors have, but that also reflects the relationships across the various entities. It’s semantically linked.
If a software-defined vehicle is on a bridge, there is a semantic attribute around the elevation, so the algorithm understands what type of driving policy is needed. We have achieved a semantic, consistent representation of the world, with the linkages, the arrows across the entities defined and updated in real time. This is extremely helpful for automakers from a safety standpoint. For example: about 60% of the speed limits in Europe are implicit; there’s no sign, but because you are near a school, you know what the limit is. These semantic relationships are captured in real time in our MOM, which is a future proof and extendible unified map content data model.
Generally speaking, how do you approach any gaps that may arise in terms of the data you gather?
We are able to forecast where we may have some data gaps and specifically go to an area to acquire—always in real time—more data to fill the gap before it becomes evident. This is what we qualify as campaign management.
‘Digital cockpit’ has become a buzzword today. How are you helping reduce the burden on OEMs as they pursue this vision?
We have developed a plug-and-play Lego block approach that enables OEMs to build the digital cockpit experience in a flexible way by combining a set of available building blocks. Or they can simply build on top, leveraging elements from a heterogenous ecosystem but retaining some level of IP and differentiation. For instance, we have an agreement with parking management company APCOA, which provides certain building blocks that an OEM can insert. Essentially, it provides details on an end-to-end parking experience including payments.
What about electrification? How does map data need to respond to the growing number of EVs on the road?
We are investing heavily around electrification and are aiming to be the world’s leading supplier of location data for EVs. We are looking at both static and dynamic attribution around charge points, including the pricing, discounts on offer, and power levels of stations. We have also developed a set of EV range services, taking range prediction up a level by incorporating details on vehicle speed, number and degree of curves in the road, slope of the road, and traffic conditions, etc.
We have worked on this strategy in collaboration with eight OEMs, and received positive reviews from all of them, especially on some use cases such as occupancy. EV drivers want to know whether there will be a place free for them once they arrive at a charge point, if all the chargers are working, if a specific location is compatible with their charger, and so on. We are doing these services not just for passenger cars but also commercial trucks.
Where are all of these advances taking HERE and its location data offering?
In order to have a software-defined car you need to have a software-defined map, which is basically what we have built over the past two years. This gives us an advantage not just in terms of speed but also in terms of quality and transparency, as the customers can see—attribute by attribute—the level of accuracy we provide.
This is a sponsored article and due to branding requirements, may not meet Automotive World house style
