Silicon Valley Mapenings
Recently I decided to put a new “topical” category in my Google News App in hopes of the service finding articles on mapping and navigation. Doing so has netted me relatively few articles. In fact, the same group of titles that were presented initially hasn’t changed since coverage started. However, two articles from Reuters caught my eye. The title of the first article was “Silicon Valley debate on self-driving cars: do you need a map?” The second article reported on Uber’s acquisition of deCarta. More on both articles below.
Silicon Valley and self-driving cars
Before I contemplate this interesting, but misdirected article, I think we need to pay homage to Lewis Carroll who wrote Sylvie and Bruno, as well as Alice’s Adventures in Wonderland, the Hunting of the Snark and many other humorous stories. In Sylvie and Bruno, we find his character Mein Herr commenting on a country map that had been created at the scale of a mile to a mile. When asked if the map was used often, the response was “It has never been spread out yet,” said Mein Herr. “The farmers object: they said it would cover the whole country, and shut out the sunlight! So, we now use the country itself, as its own map, and I assure you it does nearly as well.” It is interesting to note that Lewis Carroll in 1895 more clearly understood the Silicon Valley debate than many of those who are, apparently, now involved in it.
One never knows if reporters reliably report the gist of the conversations they have when developing a story, or if they actually understood what was being said to them. Also, reporters have been known to exaggerate issues when writing headlines in an attempt to a) sell their story, and b) attract readers. Whatever the case may be, I think that some clarification is required.
First, the premise in the headline of the article is misleading. Of course you need a “map” for autonomous vehicle development, but in today’s world a map is merely a representation of the data in a spatial database.
It is an overly simplistic, but important generalization to note that while maps once were the primary sources used to create spatial data (i.e. digitizing the map to create data for a spatial database), the trend is now reversing. With the avalanche of GPS devices capable of reporting the location and distribution of geographic phenomena, data are now being poured (both by end-users and database creators) into spatial databases that are, subsequently, used to create maps.
I think the map/spatial database confusion is exemplified by the Reuters article. All autonomous vehicles, regardless of the specifications and design goals, will require some degree of spatial data to perform their basic function.
For example, consider that systems capable of routing vehicles between destinations do not require a “map” to perform this function. Instead they require a spatial database that can be utilized to calculate the path and vehicle maneuvers required to efficiently navigate from one location to another. The human driver is the one who prefers to see a representation of these spatial data in the form of a virtual map, presumably for reassurance that the vehicle has selected the correct destination, as well as to provide information on other objects along the path they are traveling.
The question being debated in the industry involves the type and amount of spatial content that will be required to meet the specifications for the vehicle-type that the spatial database is intended to support. Google, for instance, is reported to be developing a spatially comprehensive inventory that can be used to provide navigation guidance, as well as additional spatial detail required to meet Google’s specific design for vehicles supporting autonomous navigation. Other developers may choose to create a sparse model of spatial information and rely on real-time sensing and associated signal or image processing to deal with conditions that they may feel can be better solved as they are encountered.
The efficient, but minimal, sparse database for navigation requires complete and comprehensive compilation of street and road geometry on those thoroughfares where automobiles can be legally driven. In addition, this database would be populated with street names, addresses, political jurisdiction, directionality, information on Points of Interest, and a host of variables that would ensure that the system could recommend a path between origin and destination that was safe, legal, efficient and “userful” (that is data that might be useful and informative to those being transported by such systems). How much more spatial data is required for an autonomous car depends on the design of the vehicle and how “autonomy” will be provided to the vehicle.
In my opinion, the gating factor for the complexity of the spatial databases supporting autonomous cars will turn on the efficiency and accuracy of the result of processing specific spatial data based on sensed, real-time inputs, versus the efficiency and accuracy of relying on previously reported spatial data purposefully compiled or tailored for a specific use.
And Now, onto Uber
As you must know by now, Uber acquired deCarta. In the article cited, the reporters humorously declared deCarta to be a startup. For those who do not know, the company formerly was known as Telcontar (note the Middle Earth link) and founded in 1996 by engineers from ETAK.
What you might need to know about the acquisition is covered in detail by Marc Prioleau , formerly an officer at deCarta. After reading Marc’s analysis, I was still left with a nagging question. However, let’s get rid of the “yes, but…” statements first.
Yes, routing algorithms are difficult to create and appropriately tune.
Yes, spinning large volumes of spatial data to answer routing queries is complicated.
Yes, Uber can benefit from the knowledge of deCarta’s present team of engineers, as well as the knowledge of numerous talented engineers no longer with the firm as encapsulated in the deCarta software.
Yes, Telcontar/deCarta powered Google’s earliest entry into mapping and routing, as well as that of Yahoo and other well-known internet based mapping applications.
But wait, if deCarta was involved with Google, and Yahoo and others in creating routing and mapping systems, why didn’t any of them acquire the company? Why did many of the Telcontar/deCarta customers prefer to develop their own mapping platforms, rather than take the easier road of adapting the system provided by deCarta?
Of course, we will never know the truth of the matter, but I offer some insights for debate.
I suspect that deCarta’s investors were tired of holding the bag/investment and realized that a really big payday was never going to happen for this company. I do not have any insider information, but another suspicion I have is that deCarta’s revenue and customer bases were contracting. Next, while I am sure that deCarta staff is outstanding, the company has lost some of its best talent and most of those are not recent departures.
More importantly, it is likely that Uber is not going to be a commercial map publisher responsible for creating maps in the volumes of the interactions that Google serves. Rather, it seems likely to me that Uber wants deCarta for help with the analytics related to mapping and routing that will enhance its ability to locate, pick-up and service its customer base. In essence, Uber wants to enhance their logistics management efforts by establishing predictive abilities that would allow them to manage the contention for the drivers and vehicles needed to serve their customers based on time-of-day and location.
I am not of the school that thinks that Uber acquired deCarta before someone else could have done the deed, although Samsung was a possibility. Heck, the deed could have been done by another company any time in the last nineteen years of the “startup’s” existence. For Uber, the cost of deCarta was likely a rounding error in their budget. Indeed, it may have been more efficient for them to execute the deal than to analyze it.
And so it goes. They don’t call me Mr. Warmth without reason.