Parkopedia started out a decade ago with the aim of mapping the complex parking landscape in major cities, giving users of a single view of all the nearby car parking options.
CEO Eugene Tsyrklevich explained to Techworld how far the business has come over the past few years in terms of data science and why it is betting big on helping car manufacturers with the challenge of autonomous parking.
"When the company started big data didn't exist, the iPhone had just launched, deep learning as a term didn't exist, so we couldn't have predicted any of this stuff. What we did know was parking was and still is a problem.
"Fundamentally we still allow people to find and pay for parking around the globe, but how we do it and go about that is changing and continues to change."
You may have already used Parkopedia without realising. It provides parking information onboard newer vehicles from the likes of Volvo and Volkswagen (VW), as well as within Apple's own iOS maps app.
The startup has amassed a trove of parking data since being established in 2007, from fairly static data on various car parks (address, capacity, height restrictions, opening hours, price, etc.) to an increasingly dynamic data set provided via APIs from partners, which includes availability at any car park that has sensors fitted to its barriers, plus GPS and traffic flow information from the car manufacturers themselves.
Parkopedia started out with people on the ground surveying and taking pictures of car parks and storing the data on racks of on-premise servers. Now it has been shifted to Amazon Web Services. It now holds petabytes of parking information in the cloud, including millions of images, alongside the dynamic data feeds from its partners.
Here the startup can leverage some of the elastic compute power available through AWS to do more advanced processing of this data, increasing the reliability of the service for users.
This is where the data science comes in. The Parkopedia team of four PhDs has developed predictive algorithms to give drivers an indication of availability at a car park ahead of arrival. Tsyrklevich says that at a car park with sensors installed, and which is feeding this data to the Parkopedia app, is able to predict availability to 95 percent accuracy.
However, Tsyrklevich notes "the parking industry is not a very sophisticated industry" and most of the time this sort of technological infrastructure isn't in place. He explained: "The reality though is that most car parks don't have that because it is expensive. Particularly for street [parking] there isn't a huge appetite from local authorities for this, so we have to fall back on other data sources."
As part of partnership arrangements with car manufacturers Parkopedia receives information from connected navigation systems. "That data is already used to create traffic maps. We are using the same data from the same systems but instead of working out how quickly they are moving we work out how they are moving, so are they circling an area to find a space? We use that to figure out if there is a space available or not," he said.
"That is just one example of how we leverage machine learning in a relatively dumb domain of parking."
Tsyrklevich's focus now is on autonomous parking.
"While everyone is concentrating on the exciting and hard part of navigating a motorway or a city, very few companies today are talking about what happens when a vehicle gets to the destination and what happens then," he said. "That for me is where we step in and grab this last mile, or 500 metres, of routing after you are dropped off."
In short, Tsyrklevich wants Parkopedia to be the data source autonomous vehicles use to understand where to park when there is no GPS coverage - such as in an underground or multi-storey car park - and then to help it navigate to the space. This is what the company proved with Volkswagen and presented at the Consumer Electronics Show (CES) in Las Vegas this month.
Commenting on the project, Parkopedia's COO Dr. Hans Puvogel said: "The self-parking car of the future will need excellent data to find the right parking spot, transaction capabilities to book and/or pay for it and indoor navigation capabilities."
Using a 3D laser scanner Parkopedia creates a 3D map of inside a parking garage to the centimetre, including every wall, ramp and pillar. The car then uses its own sensors to navigate the space, as it would with GPS coverage, to navigate to, and eventually into, the parking space. If a car park has sensors over every space this could then be used to indicate which spaces are free for an autonomous vehicle.
Parkopedia is still a small company though, and Tsyrklevich realises that this sort of project has its risks. He believes that autonomous driving - and parking - is a question of when not if, but "there is such a thing as being too early".
"At this point we need to answer a few questions, some are technical regarding if it's possible for a vehicle to navigate fully autonomously without GPS coverage, and we have proven that out with VW.
"The next question is when will the vehicle of the future actually require this? As you can imagine, going out and laser scanning car parks is not a cheap endeavour so it is expensive and time consuming and while we have people on the ground globally, we would rather not do work at that scale and expense unnecessarily."
This means that although this is a big project for Tsyrklevich he is proceeding with caution: "We are continuing to work on the autonomous valet parking but we need to work out when the car industry is ready."
"That is a huge project, we have proven some of the concept but it is a massive area trying to figure out how to do this better and at scale, not just at one car park but across 75 countries."