TomTom, TeleAtlas and MapShare
Recently, I spoke with Patrick McDevitt, Vice President of Community Mapping at TomTom, who reached out to discuss the company’s view on the value of the MapShare and probe data gathered by users of TomTom’s Personal Navigation Devices.
Patrick has been around the world of mapping for quite some time, having earned his merit points with Tele Atlas and GDT. My interest in having the conversation was to explore the benefits that accrued to map database augmentation through the use of TomTom’s crowdsourced data and to explore the company’s strategy related to the use of these data.
I suppose you noted my use of the phrase “the company’s strategy” instead of indicating Tele Atlas, as it seems that Tele Atlas, as we know it, is slowly being absorbed into the corporate identity of TomTom. Indeed, if you look at the Tele Atlas website home page you will read that “Tele Atlas is the licensing business unit of TomTom N.V., the world’s leading provider of location and navigation solutions.” I will comment on the outlook for TomTom at the end of today’s blog, but want to separate that discussion, which includes thoughts that are solely mine, from my conversation with Pat, as we purposefully limited our discussion to issues related to MapShare and crowdsourcing.
I requested that Pat give me his “elevator pitch”, as a way of starting our conversation about Tele Atlas and its introduction to crowdsourcing. Here it is (without the elevator music) – Shortly after closing the acquisition of Tele Atlas, TomTom indicated to TA that they had a new source of mapping information called MapShare. TomTom thought MapShare would be quite helpful in updating the TA map database. In turn, Tele Atlas was skeptical that user input could be of significant benefit to map creation and updating process. However, that did not stop TomTom from saying “Here are your first million observations – let’s see what you can do with them.”
Previous to this, TomTom received community input in the form of map edits (corrections and augmentations) from users of MapShare who could indicate whether they wanted to share the data with the TomTom MapShare community. If so, TomTom attempted to verify the information and, if verified, it was added to the community (MapShare) version of the navigation database and distributed to users.
McDevitt indicated that the initial skepticism at Tele Atlas regarding the potential quality of the MapShare data soon transformed into intrigue over the data’s possibilities and eventually evolved to evangelism. While Tele Atlas understands that there are limitations to the data supplied by MapShare, they also believe that it is an exceptional tool for identifying change detection, especially for map updates that are hard to discover otherwise.
Once we had progressed through the elevator pitch, Pat indicated that it took Tele Atlas most of twelve months to understand how to use crowdsourcing to advantage the Tele Atlas map database. By 2009, TA felt that they knew how to use the crowdsourced data and added the technique to their standard arsenal of map compilation and updating tools.
Tele Atlas has two streams of community input. Active Community input, using MapShare, involves users purposefully interacting with the PND and providing input about errors found in the version of the map database resident on their device. Passive community input involves the user of a TomTom navigation platform agreeing to allow traces of the GPS positions and paths recorded by their TomTom PND to be anonymously uploaded through the TomTom Home software to the TomTom mother ship. TomTom Home is an application that resides on the user’s laptop or desktop computer that is activated when the user docks his navigation device to the computer. I asked Pat about the take rate and he indicated that “most” customers give their assent to allow the company to use their anonymous GPS data to improve the Tele Atlas map database. (Hmmm. I may be the lone exception.)
Pat acknowledged that Tele Atlas originally was hoping that the passive community data would help them to create street centerlines that were more accurate than those then in their database. However, Tele Atlas had concerns that potential problems with the quality of the GPS data collected (due to the performance characteristics of the GPS receivers and the influence of other factors such as signal scatter, foliage, urban canyons, etc.) could degrade the utility of the data. In fact, they were sure that the GPS readings collected from TomTom PNDs would be less accurate than the positional data captured by the Tele Atlas fleet of mobile mapping vans. What they had not appreciated, at the start of the venture, was that aggregating a massive number of data points collected over time would help them average out errors and create a database with an extremely high level of positional accuracy.
I was told that the company’s MapShare community input has now produced over two trillion GPS points that are being used to improve the quality of the Tele Atlas map database. In addition, Tele Atlas is now using the MapShare GPS traces in an attempt to create other kinds of map attributes, such as road gradients, road curvature and other high accuracy data that might be of interest to the Advanced Driver Assistance Systems (ADAS) community.
In a previous conversation I had with Chris Wilson, one of the thought leaders at Tele Atlas focused on ADAS, he indicated that the passive community data might have significant benefit as behavioral data for ADAS. For example, it might not be necessary to compute the shape of a curve and calculate a metric for a “curve speed warning” system, when, instead, you could provide a history of the speed that the curve could be driven safely, perhaps even by season or by driver profile, as recorded by tens of thousands of drivers.
Most people have an easy time visualizing what the paths of probes would look like on a map, but I thought that several of the illustrations used by Pat during our conversation might be of interest to you. Our conversation was augmented by an online presentation. The images shown during the online segment were low resolution JPEGs and I apologize in advance that my copies are even fuzzier. However, you should be able to glean the important issue, even though parts of the images will not be readable due to the size of the images. The images do not have legends that you can read, so go with the general notion that the brighter the colors, the more probe paths recorded.
The image above overlays a “heat map” of the paths of cars equipped with TomTom PNDs crisscrossing the transportation network in a section of Laguna Niguel near my home in California. The bright north-south lineation is the I-5 freeway connecting Los Angeles with San Diego. The east west arterial is Crown Valley Parkway, while the curved road in the lower left is the 73 Toll Road. On the right you can see the traffic patterns around a local mall. The remainder of the area is an industrial park with relatively light traffic.
In this example from Huntington Beach, California you can see that the probe data is capable of providing lane directionality along major streets. Some of the probe data is in very light colors that do not show up well on my copy of the image.
As you know, I beat on Tele Atlas (and everyone else) about their missing the I-195 realignment in Providence, Rhode Island. I asked Pat how TA had missed this change since they must have had units crossing the bridge before and after the realignment. He acknowledged that this was true and offered one of those explanations that cause management to shoot themselves in the ear! (Don’t tell me you haven’t seen those guys on the street with the big holes in their ears. Ex-technology executives, every one of them! Those hoops are just a disguise.)
The staff at Tele Atlas believed that the strength of the probe data was in aggregating the paths for a location over time in order to average out the errors and increase the accuracy of the observations. However, when I-195 was realigned, the massive amount of data that had been collected for the pre-realignment geometry caused the smaller number of traces recorded for the geometry of the “new” highway to be “averaged out”, leaving the old geometry king of the road, at least, in statistical terms. Below you can see the same location shown in an image that has been time-sliced. The probe data for the pre-realignment geometry is shown in brown and clearly visible, while the vibrant green color shows the new paths of the intersecting highways. Needless to say, time-slicing has now become part of the Tele Atlas strategy for interrogating their probe data.
TomTom continues to use imagery and other sources for revising their map base and has found that local probe data is a good way to evaluate imagery, indicating whether the imagery is current in terms of the transportation network or needs to be supplemented with other research sources. In the illustration below, you can see that the probe data shows a number of streets that do not appear on the reference imagery.
Don’t get the idea that Tele Atlas is focused on the probe data alone, as the next two illustration show reports from active community input, in which MapShare users have indicated that a road that used to be a four-way intersection is now a roundabout.
In the second “active” MapShare example shown below, users indicated that the TA map was incorrect and missing a street. Not only did the MapShare users note the error, but they also provided the correct name for the street. The probe data (passive community data – shown on the bottom right of the image) also provided evidence of the street. All information was verified by TomTom researchers before it was added to the database.
TA uses both active and passive community sources as change detection indicators to help the company’s researchers decide where to focus their efforts related to map updating and source gathering. When possible, Tele Atlas prefers to work with data mining and authoritative sources to remedy issue raise by MapShare input, but says it will deploy field research if required to resolve the issue to their satisfaction.
The final image shows the active MapShare community input for localities in California from the first four days in August, 2010. The dots show location where the reported information differs from that contained in the Tele Atlas database. While there is an obvious autocorrelation with population density, it is clear that active community input is a powerful tool for change detection.
I asked Pat about the volumes of data that were being generated through “active” community input and was told that so far in 2010 the top twenty countries, in terms of the volume of reports, have reported over two million change notices. Note, these are not passive GPS traces, but instances where someone has taken the time to tap the screen of their TomTom PND to capture the location of the change to be reported and taken the time the indicate the nature of the change. It is my understanding that since the TomTom MapShare program started the company has logged over 80,000,000 “active” reports by MapShare users. In addition, approximately 80,000 TomTom PNDs connect each month to upload “active community data”, although the number of units connecting to TomTom Home for updates and other downloads, is considerably larger.
Pat indicated that the Tele Atlas learning curve for understanding how to use the community data in an effective manner was quite steep. While Tele Atlas appears to have found benefits to using both active and passive community data, it appears to me that they prefer to work with passive data since its “pureness” would generally rule out malicious intent on the part of the contributor. When I asked about malicious intent, Pat responded that Tele Atlas has learned how to manage active community data and ferret out potential “problems” in MapShare inputs that appear designed to obfuscate information. I was surprised to learn that among the “tricks” they had found in the “active” data were instances of users living in neighborhoods plagued by speeders posting MapShare edits indicating that the speed limits around their homes are lower than officially posted. Their intent was presumably to have these bogus speed limits uploaded and provided to other TomTom MapShare users who might be driving too fast in their neighborhood.
McDevitt felt that the threshold to build out a map at Tele Atlas is falling, due the advantages provide by community input through MapShare. He gave the example that Tele Atlas did not have good access to comprehensive data in Romania, but was able to build a basic road network interconnecting major towns and numerous locations outside of the country through its MapShare customers who had driven through the country with their TomTom’s operating. The MapShare community took these data, edited them and helped to create a new map with some significant features, although the data was not comprehensive enough to add to the Tele Atlas navigation database.
One area where passive community data shines is in collecting data on traffic and travel time, which are areas that obviously benefit from probe data. In addition, Tele Atlas is working to determine whether they can extract lane count from the passive community data. They feel that their efforts are getting within striking distance of the results that they can derive from imagery and the lane data collected by their mobile mapping vans. Pat indicated that current test revealed an accuracy of 80% to 85% in lane counts for multi-lane roads using the passive community data uploaded by MapShare users.
I asked Pat his thoughts about the company’s mobile mapping vans and whether they still provided value. He indicated that the mobile mapping vans were now used for higher-end products and that the company had learned to be more purposeful in the types of problems they were deployed to resolve. Sending the mobile mapping vans to the field as a primary source to collect all road related data is just too expensive. Instead, the company culls all user reports and MapShare data to determine where deploying the vans is optimal and where surrogate sources will provide acceptable results.
I related my observation that for a period after the acquisition of GDT by Tele Atlas, the two companies seemed to speak with a forked tongue related to field work. I mentioned an occasion when the TA North America president issued a press release indicating the strength of the company’s data mining activities and their desire to go into the field only when forced to do so. Unfortunately, the same day, Tele Atlas corporate headquarters issued a press release indicating that they had just deployed a new generation of vans that would allow them to collect more field data a faster pace than ever before. Pat responded with the observation that, in the past, GDT (and Tele Atlas, North America) may have done too much data mining, but still feels that data mining makes sense in the United States and that the situation is now beginning to change in Europe (with programs such as Inspire, etc.,). However, he noted that the community input from MapShare has changed the game, as it provides a useful surrogate indicating the location of problem areas all allowing an evaluation of the best method for gathering the needed data.
When I asked about the competitive threat from OpenStreetMap, Pat replied that the “map geeks” at Tele Atlas (and he is one) celebrate map geekiness and OSM. However, he indicated that their clients are willing to pay for Tele Atlas data because of their need for verification and double checking of the data. He continued, noting that it was his belief that there are clients whose data quality needs can be best met by the high level of authoritative map data that Tele Atlas is able to provide.
I ended the conversation with a “pregnant” question, which was “ If Tele Atlas and TomTom depend on MapShare data to help increase the accuracy of their map base and reduce the expense of keeping it up-to-date, are they experiencing declines in the amount of data customers provide through MapShare?” Pat indicated that the company does not have a “dwindling amount of data” problem. He related that there was a continued interest by customers in all platforms. In addition, TomTom’s newer programs such as Traffic HD and TomTom Work (a logistics solution) are increasingly popular with customers worldwide. Pat noted that TomTom has experienced quite a bit of success getting devices into cars, as is shown by their collaboration with Renault (Carminat). He also noted that the TomTom iPhone device is not cannibalizing other TomTom products or services. In sum, the data just keeps rolling in and TomTom appears to have a tool that is the best of class in the industry.
Now you know some of what I heard from Pat McDevitt of TomTom
And Now For an Additional Perspective
I have long thought that TomTom’s MapShare should be a powerful differentiator in the navigation database market and that the quality of the Tele Atlas database should increase as some function of the number of active inputs and passive GPS points and traces contributed by the TomTom user community. My conversation with Pat McDevitt helped me to understand that it may have taken Tele Atlas longer to harness the benefits of community data than I had suspected. However, it seems to me that the evidence leads one to conclude that the company is now far along the learning curve posed by community sourced data.
While it seems clear that TomTom is figuring out how to use the data to benefit the accuracy and comprehensiveness of their database, it also appears that the popular reaction to the acquisition of Tele Atlas by TomTom seems to be turning antagonistic. Indeed, many industry analysts seem ready to conclude that the merger of the two companies has failed to produce the desired benefits.
I think these prognostications are premature. While it may have taken longer than thought to see the advantages of the merger, I believe that there is no other company better positioned than TomTom to benefit from owning Tele Atlas. While some financial analysts seems to feel that TA is available for the right price (what isn’t?), I can’t imagine that TomTom would feel that any price was the right price, unless they were planning to close the company and count money as a profession. However, I don’t think that is the plan, although it sounds good to me. MapShare has value to TomTom, but what would TomTom do with MapShare and no Tele Atlas? And what would Tele Atlas now do without MapShare? However, at TomTom’s present value someone might be enticed to buy the combined entity and regard it as a fair price for Tele Atlas. If this is the strategy, they had better be prepared to compete in the automobile navigation market or kiss their investment in the navigation hardware business goodbye.
While I appreciate that there are several companies out there who are potentially interested in being in the map business, most of them lack an understanding of the process required to build and maintain a navigation map database with competitive coverage and accuracy. In addition, most of the companies who think they are interested in owning a navigable map database company, are interested in owning maps that can be used to pimp their advertising business and have no idea what it is like to compete to provide systems and services to the navigation segment of the automobile market (including ADAS). On the other hand, if these companies are willing to abandon these markets, won’t NAVTEQ be happy? Well, this blog is already too long without going into potential acquirers, but look east, Far East (or west from California, where I write this blog).
Speaking of NAVTEQ, now that TomTom has shared some information with me, do you suppose NAVTEQ would be willing to give me a ride in one of their new, super-duper, Dilithium crystal powered mapping vans? Last January I received an email from one of NAVTEQ’s PR flacks (sorry – but it was not a person who worked for NAVTEQ) who had noticed my blog and invited me to take a ride with one of the company’s data research teams to see how they gathered data. I replied that I would love to take them up on their offer, but that for their own good, they should probably take this issue up with NAVTEQ’s senior management, as, from time to time, I had criticized NAVTEQ (and everyone else in the industry) just a tiny bit in my blog.
The response? I’ll bet you guessed. I received…silence (although, I was happy about that since I was told by my Mother that silence was golden). I waited a week and being something of a pest, wrote again asking for a response. Fate was cruel, however, and several days later I received an email rescinding the previous offer, but thanking me for my interest (even though I had not initiated the exchange). However, the author of the email did suggest it might be better to wait until later in the year when NAVTEQ’s new generation of vans was available. Hmmmm. I think I saw one of those new platforms parked outside the San Diego Convention Center at the ESRI show in… July. Perhaps they forgot about me? Do you suppose an offer is forthcoming? Man, Dilithium crystals – I can’t wait. Larry? Roy?
Speaking of ESRI, I am thinking about writing about their new Community Map program in the next month or two. I know that nobody will believe me, but I am beginning to thank my stars that ESRI exists and you should too. More about that claim in a future blog.