Comments on the Development of Spatial Databases to Support Autonomous Vehicles
Companies announcing new mapping initiatives seem to be crawling out of the woodwork. Why the sudden interest in mapping and what should we make of these new entrants in the mapping space?
The recent CES show was a hotbed for news releases on mapping initiatives related to the functioning of autonomous vehicles (AV). Toyota, for example, announced that, using automated cloud-based spatial information generation technology, it is developing a high-precision map generation system that, “…will use data from on-board cameras and GPS devices installed in production vehicles (around 2020).” In their press release on the topic Toyota noted that, in their opinion, this proposed constant collection of road data by multiple cars will offset the accuracy gained by using LiDAR (which they are not using) and will trump the infrequent, expensive road data collection process used by today’s mapping incumbents.
Ford, in turn, announced its own mapping initiative. Ford indicated that it will use a compact LiDAR sensor developed by Velodyne (at a mere $8,000 a unit) which it hopes will help it win the race to offer a fully AV. Ford appears to be confident that Velodyne’s Ultra Puck will enable its fleet of vehicles to create real-time, 3D maps surrounding their path through the environment.
Other companies making noise about the automotive market include Quanergy, which bills itself as the, “Future of 3D sensing and Perception – 3D LiDAR for ADAS, Autonomous Vehicles & 3D Mapping.” Panasonic is, also, developing a LiDAR capability for mapping, as well as enhancing its advanced camera imaging technology. In addition, let’s not forget GM and Mobileye’s exploration of advanced mapping with OnStar data, as well as GM’s collaboration with Lyft to build a network of AVs.
Why the interest in Mapping?
I am sure that many of the long-established automotive companies involved would indicate that they have always been interested in mapping. Of course, this type of statement would be broadly true, but most of their interest has been educational or licensing related – few automobile OEMs, Tier 1 or Tier 2 suppliers have spent any real money on developing mapping capabilities until now.
I suspect that the “move” to mapping is being driven by at least two fundamental issues. Data Integrity concerns are, I think, the prime mover behind the trend of in-house spatial data gathering. Building a service-related user community may be the second factor behind developing spatial data capabilities. We will discuss these two trends below, but first – a look at the profit motive.
It is my belief that the move into mapping by OEMs is not based on a direct-profit motive. Rather than making money on mapping data, or deferring costs by creating their own data, advanced mapping will be used indirectly to promote brand allegiance and expand brand reputation during a period in which individual car ownership may transition to some form of collaborative use of vehicles and their data networks, rather than purchase or lease of automobiles. Next, let’s acknowledge that developing and maintaining the spatial data that is critical for the operation of AVs may be a protective strategy to defend against potential injury lawsuits where it will be claimed that a specific AV did not perform as advertised.
Spatial Databases and AV
Let’s begin by acknowledging that the spatial databases supporting AV operation must fulfill at least two fundamental functions. One the one hand, the spatial database must be designed, populated and organized in a manner that supports the systems allowing the autonomous functioning of a vehicle within the surrounding environment. This means that the data needs to combine positional and contextual data about each vehicle and its surroundings, as well as information about other objects in its environment that are stationary or moving. Second, there is a need for a spatial database that can support the navigational needs of the autonomous vehicle when moving between two or more locations.
Perhaps a good way to think of this distinction is the difference in spatial knowledge required to know when to turn from one street to another, versus the spatial data that an AV would need to consider to be able to execute a turn at that location. The majority of the press releases at CES were concerned with creating accurate, reliable and up-to-date spatial databases that are fundamentally concerned with the methodology of AV operations rather than determining the routing characteristics between origins and destinations.
Data integrity and Spatial Databases in AVs
AV systems will demand enhanced mapping capabilities to ensure the proper functioning of systems designed to manage the performance and maneuvers of self-regulating vehicles. The associated spatial databases will become a critical component in the ability to create fully functioning autonomous vehicles that are reliably safe to operate and whose operation consistently reflects local road and transportation environments. In my opinion, it seems unlikely that vehicle OEM’s would be willing to cede the responsibility for setting the specifications for and collection of these mission critical spatial data to external suppliers of mapping data.
While collecting all of the spatial data required to support autonomous cars may be a task beyond the capabilities or interests of automobile OEMs, these companies may want to collect and control mission critical spatial data (e.g. road centerline data and other spatial data related to road and road environment characteristics) in a manner that fully supports their notion of how to best functionally and legally operate an AV. Conversely, routing related variables such as addresses and business listings would likely continue to be provided by third parties.
In today’s navigation and local search oriented markets companies are able to license mapping data from providers such as HERE and TomTom, but the specifications for these data are set by these two companies. Yes, the data format may reflect industry-based standards, but many of these standards reflect the influence of the data providers. In today’s era of rapid prototyping, innovative engineering and technological leapfrogging it may well be that the OEMs will generate data specification, data processing and data collection systems that are unique to the fundamental operation of their specific autonomous vehicle systems. In essence, these systems and the spatial data that help operate the AVs may be regarded by some companies as a distinct competency leading to a sustainable competitive advantage.
One major concern with the use of data from companies such as HERE and TomTom in AVs is that historically, as pointed out by Toyota in its press release cited above, their data collection, processing and distribution methods have been time-bound and not updated and disseminated at the speed that is expected to be required to operate AVs. Unannounced road closures, emergency road repairs, intermittent lane geometry changes, and the like need to be compiled with urgency and rapidly disseminated to a company’s AVs that may encounter these changes. It seems unlikely that TomTom and HERE can supply fleet-based, spatially comprehensive data in adequate enough time slices to support AV functions using their current data collection systems. Unless TomTom and HERE can provide (or partner with fleet owners who can provide provide) continuous, spatially comprehensive updates to their entire range of automotive customers, it is unlikely that they will be the beneficiaries of the emerging market for AVs.
While I agree that TomTom and HERE have significantly improved their data collection and distribution methodologies over the past few years, let’s not forget that in 2010 TomTom, HERE, Google and others in the map data business took months to recognize and reliably reroute traffic after a section of interstate highway was deconstructed in Providence, Rhode Island. Yes, you read that right. Even using live GPS traces indicating where cars were actually driving, these companies continued to route users over a section of interstate that no longer existed (see here http://blog.telemapics.com/?p=241 and here http://blog.telemapics.com/?p=248 for information on this topic). This kind of problem is not one that automobile OEMs or their lawyers will be willing to accept when AVs become a reality.
A final observation of some interest the notions expressed above is that HERE has recently been awarded a contract from the Geography Division of the United States Bureau of the Census (RFP CENSUS2015_GEO0227). The RFP stated that in addition to other requirements, “The Census Bureau is seeking to obtain complete and accurate datasets containing housing unit addresses with associated attribute and geographic coordinate data, and spatially accurate street centerline data to supplement and/or validate content of the MAF/TIGER System.”
In Amendment A0021 to this RFP the Census Bureau clearly stated that the street centerline data developed as a result of this effort (but not the address information) would eventually be released to the public. HERE’s willingness to supply road centerline data under this contract may speak volumes about the intentions of OEMS to collect spatial data.
Community -The advantage of size
Google search engineers have often stated that they are not smarter than anybody else, but note that when thinking about improving existing products or creating new products they accrue the advantage of having more data to analyze to properly define opportunities than most of their competitors. In a similar light, the larger automotive OEM’s could make their vehicles “spatially smarter,” by fielding a massive spatial data collection effort based on the number of vehicles deployed (directly (research fleets) or indirectly (sales)). Toyota, for example should be able to produce a more comprehensive and accurate spatial database to manage its autonomous vehicle performance than, say, Subaru. Ford, with its success in light-trucks, might be able to create more comprehensive data on rural routes than say a Volvo or KIA.
Purpose-built spatial databases produced by an OEM, in turn, could be used to analyze, tune and optimize the performance of every autonomous vehicle they produce. For example, instructions could be customized for individual vehicles, adjusted for their carry-loads, profiled for specific geographic areas/environments, optimized for local weather and the like. Without the mission critical spatial data to support the customization of their vehicles the OEM would be at a competitive disadvantage.
Spatial data is certainly an area where more can be better and the OEMs with the most sensor units/deployed vehicles could wind-up producing market leading, highly accurate, comprehensive spatial databases of road-related information. In turn this spatial data superiority could become a branding opportunity leading to more sales or more use of a company’s AVs. Although it would make complete sense to create an industry wide spatial database to support AVs I doubt that this will happen due to the need of players in the automotive industry to create sustainable competitive advantages. However, this may be an opportunity being explored by the new owners of HERE.
A Fly in the Ointment
The need for interoperability between autonomous vehicles suggests that one of the safest ways for these vehicles to commingle on the roads would be based on interpreting a common spatial database. As noted above this is unlikely to happen. It is possible that regulators will have the prescience and courage to demand it, although I do not see that happening in next decade and, perhaps, not at all.
If the AVs are in motion and adjacent autonomous vehicles from different OEMs do not have the same representation of the world around them, how will they efficiently maneuver in such a manner as to avoid being a threat to each other?
If companies creating AVs, or the systems for AVs, decide they need to license additional data to build their recipe for the spatial database required to support these vehicles, how will they deal with the issue of multiple representations that will most certainly be found when trying to integrate licensed spatial representations with the spatial representations they have collected using different systems?
Yes, they are all measuring the same world, but they will not be measuring it in the same manner or even with similar equipment. How closely their data models match is an open question. Given this possibility, how will the OEMs deal with issues of metadata mismatches or metadata that is incorrect or simply does not exist? The potential complexity of the multiple representation problem, in the context of spatial databases designed to support the operation of AVs, is stunning. However, it is only one example of a series of problems that commonly haunt applications based on extremely large spatial databases. Multiple instances of potentially non-interoperable, extremely large spatial databases distributed across numerous, comingled AVs sounds like the makings for a ripping good disaster movie.
It appears that automobile manufacturers (including the consortium that purchased HERE) have realized that spatial data may become a powerful benefit in the future – if they can find a way to make it their own. While mapping is a term that flows off the tongue and into strategic plans with ease, actually collecting, compiling and unleashing the power of spatial data in a beneficial manner takes more skill and know-how than many would-be practitioners may understand. Indeed, harnessing its power may require skills in spatial data handling and knowledge about the intricacies of mapping that have slowly withered as interest in these this disciplines has been eroded by the transformation of the spatial data world in its hot pursuit of GIS.
Major players interested in entering the spatial data/mapping market that is evolving to support AVs would do well to consider forming advisory boards where their strategists and data geeks can counsel with mapping and spatial data handling experts who might be able to give them an insight or two, or just maybe help prevent a catastrophic system failure.
Or, as a great sage once said, “you don’t know what you don’t know.”
Thinking about all this stuff has given me a headache – so I am heading to the shore to photograph birds – merrily driving my 1998 Corvette stick shift. I ordered the car purpose built and have been its sole owner. I finally managed to crack the 50,000 mile barrier this week. Do you know what is wearing out on this vehicle? The sensors (fuel pump sender sensor, air quality sensor, etc.). How about that, but no complaints from me.
Till next time.