AI Mapping with MapScale® - Map Different

Simply put, MapScale® is capable of taking basic, 2D architectural files (most commonly in the CAD format) and, through our propriety AI model and patented indoor mapping technology, transforming those files into fully-fledged, interactive 3D maps, complete with points of interest and other labels, ready for use within an app, in a kiosk, via browser or through Pointr Express®.

MapScale® leverages AI in two core ways:

• When the CAD file is first ingested, we use AI, trained upon the thousands of maps that we've created over our years as a the market-leader in indoor mapping, to figure out what architectural features equal what map features. In its most basic form, this means translating basic features to the map - such as four walls and a door meaning a room. However, MapScale®'s AI has long since surpassed this sort of basic functionality: it can now, for example, differentiate between numerous different room types, interpret open areas within an indoor map, and identify important building features such as stairwells and elevators, as well as determining features that are present on the CAD file but that are unnecessary for inclusion on the produced map

• Once the map itself has been created, MapScale®'s Large Language Model (LLM) takes over in order to parse both the text data from the CAD file and the map itself to interpret, contextualize, and categorize different rooms, areas, and points of interest. It's at this stage where MapScale® also uses LLM data to help it work out the most likely map features based on the type of building being mapped - for example, it can look at two nearly identical rooms on two separate maps and determine one is likely to be a waiting room as the map it's working on is of a healthcare facility, and the other is a meeting room as it's part of a workplace

And of course, one of the key benefits of using AI and LLMs in this way is that the more they're used and corrected by human oversight, the better and more accurate they become. MapScale® has mapped billions of square feet already, and the more CAD files it ingests and the more maps it produces, the richer this virtuous cycle becomes.

For further details on AI mapping and how MapScale® leverages AI, check out this post.

The idea for MapScale® came about when a major US retailer contacted us explaining that their existing solution for providing maps for their thousands of stores was manually creating and updating them, one by one, using image editing software. This process was immensely time consuming and, due to the limited resources of the team working on the maps, it could take weeks or even months for an outdated map to be reviewed and updated, at which point that store would fall to the back of the queue again and the map would quickly once again be out of date.

We needed a way to address the needs of this client, as well as many other enterprise-size companies, who generally require a solution to be scalable to their needs to be considered viable. This is where the idea for MapScale® came from; finding a way to rapidly, consistently, and accurately transform the common data sources that most businesses had - such as CAD files - and turning them into maps.

Using AI simply enables us to leverage the vast amount of information we have accrued over years as the market-leaders of indoor maps and enhance our process, by taking work (such as building the map from the original file, then tagging individual room types and building features) that previously would have needed a considerable amount of human time and automating it. 

Though MapScale® can now accurately detect map features with a consistency of 95%, there are still edge cases we need to be mindful of, such as if MapScale® works on a CAD file with map features it's never seen before, or on a type of building that it's not had the opportunity to work on regularly before (as it has with workplaces, healthcare facilities, and more). 

For this reason, MapScale® makes use of a traffic light system whenever a map is updated via a new CAD file. This system works to detect the degree of changes between the old version of the map and the new one, and alerts the user if a significant amount of the map has changed between the two versions. This guards against instances such as the user accidentally uploading the wrong CAD file to replace a certain map.

MapScale® can also only work with the data it receives, even though it can often fill in the gaps in this data using AI. It's not unusual for MapScale® to create an almost-finished a map which just then requires some small human touches to finish things off - for example, adding rich details to some of the map's points of interest (like adding logos and opening hours on the listing for a restaurant in a shopping mall). In cases such as these, even when a new file is uploaded to MapScale® to update the map, the system remembers the human additions to the old map and retains them.

No two maps are the same, meaning it's impossible to put an exact number on the amount of time saved by using MapScale® compared to a human-created map. However, there's no question that when it comes to scale - having to update hundreds of maps simultaneously, for example - there's simply no comparison. MapScale® is capable of updating hundreds of maps in a matter of minutes, something it would take a human potentially days of work to complete.

There's no doubting that MapScale® is capable of feats of mapping speed that human-led manually mapping simply isn't. 

But how does it stack up against other AI-based tools? We believe that MapScale® is a one-of-a-kind tool with no direct competitors, so we can't compare it to another system exactly. However, by looking at other LLMs such as ChatGPT, we've been able to compare its ability to classify and identify mapping features. These results are published in our AI mapping benchmark.