Technology report: Augmented reality navigation 2021

Conventional navigation solutions rely largely on GPS signals. These are very suitable when accuracy at the granularity of roads is required. And this is also the service that these solutions offer: Navigation along roads. What’s more, these solutions are all map-based. This in turn is very suitable for road-to-road navigation, especially when the procedure is predictable and the options are limited, just like driving a car.

This is because the accuracy of GPS signals on commercially available devices such as smartphones is somewhere between 2 and 15 meters with good GPS reception. This means that I can find the highway entrance, but it is almost impossible to find a specific book in a library or a tool in a DIY store.

So there are two cases in which alternatives to conventional GPS navigation are required:

• If the GPS signal is too weak (e.g. indoors)
• When accuracy needs to be better

In the following, we show what technological means currently exist for improving these situations and what we can expect in the near future.

First, we would like to introduce the concepts of absolute and relative tracking, as these go hand in hand with the key issues in the choice of technologies. However, it should be said that these terms are not industry standardized.

With relative tracking, the position and movement of the user is tracked “locally” – i.e. in a detached system. The advantage here is that this tracking is much more accurate. However, there is no way of knowing where you are in a global context.

A good analogy is someone who can only look at the ground. He doesn’t know where he is in Zurich or where north is – but he knows very well that he has now taken 2 steps forward and then turned 15° to the right.

There is also another phenomenon with relative tracking: drift. After a minute of walking around, a person who only looks at the ground will no longer be able to tell exactly where their starting point was. The coordinate system therefore shifts and rotates gradually, which makes navigation more difficult.

First, we would like to introduce the concepts of absolute and relative tracking, as these go hand in hand with the key issues in the choice of technologies. However, it should be said that these terms are not industry standardized.

While relative tracking moves within a “loose” system, absolute tracking is clearly defined. Put simply, absolute tracking involves determining the exact rotation and position in the world. GPS and compass are traditionally used for this.

The advantage is that the system knows where the user is on earth at all times. This also means that there is no drift. However, as mentioned at the beginning, GPS data is not very accurate and therefore not suitable for all purposes. But more on this later.

ARCore and ARKit are the respective AR interfaces from Apple and Google. They make it possible to use AR applications on mobile devices in the first place. This is achieved by tracking smartphone movements in combination with the camera feed. The use of image trackers is also supported.

• free of charge
• easy to develop applications for it
• offline
• relative tracking is very accurate.

• Drift (as only relative tracking is used)
• Requires the implementation of a starting point for navigation (manual, image marker, GPS)

The technology works in principle and has been widely tested. However, it is only suitable for precise AR navigation to a limited extent, as drift has a negative impact on tracking. In addition, tracking is discarded if the user briefly leaves the app.

It must therefore be combined with additional tracking in order to determine the absolute position as well as the exact relative tracking.

These technologies are known and tried and tested. Likewise their shortcomings in relation to our intended use (inaccuracy), which have already been discussed above.

• free of charge
• offline
• No drift

• Very inaccurate (accuracy approx. 2 to 15 meters with good GPS reception)

Despite its inaccuracy, GPS is a good way of stabilizing navigation and mitigating the negative effects of relative navigation by using absolute positioning.

This technology also uses ARKit/ARCore for relative tracking. In addition, so-called “anchors” are stored online. Basically, this means that the technology can memorize “locations” (approx. 2x2m areas) and recognize them later. The big advantage of this technology is that it eliminates drift. This is because every time the smartphone comes across such an anchor, the smartphone can orient itself again (with an accuracy of approx. 10 cm), but a functioning Internet connection is required to recognize the anchors, which consumes approx. 5 MB per minute.

• free of charge
• Very robust
• Initialization elegantly solved

• Anchors must be set manually
• High data volume required

From the user’s point of view, Azure Spatial Anchors are currently the best way to navigate in AR. They combine relative navigation (by using ARKit/ARCore) with absolute navigation (by positioning the anchors). The biggest disadvantage, however, is that several anchors have to be set manually wherever navigation is required. This is not particularly suitable for outdoor applications. In buildings such as stores or train stations, however, the implementation can make sense.

As already mentioned, the greatest potential for improvement lies in improving absolute tracking. A few technologies are in the starting blocks for this, which we will briefly discuss below:

AR Geo Tracking is a solution that currently works in certain cities in the USA and is only available on iPhone. Put simply, the technology works in a similar way to Azure Spatial Anchors, except that entire cities have already been scanned. Once the technology is less restricted, it has enormous potential for AR navigation.

Google announced news in the field of AR navigation a few weeks ago. The new “Earth Cloud Anchors” project certainly looks very promising. It is still under development, but uses a combination of VPS (Visual Positioning System, i.e. the camera feed), Street View and machine learning. Combined with Anchors, this could result in a greatly improved version of “Google Maps Live View” that is also available for other applications.

Google announced news in the field of AR navigation a few weeks ago. The new “Earth Cloud Anchors” project certainly looks very promising. It is still under development, but uses a combination of VPS (Visual Positioning System, i.e. the camera feed), Street View and machine learning. Combined with Anchors, this could result in a greatly improved version of “Google Maps Live View” that is also available for other applications.

For our imaginary case of AR-based outdoor navigation, we currently recommend the use of Azure Spatial Anchors, as they offer the highest accuracy. However, they have the major disadvantage that they require manually set anchors, which does not always make sense or is even possible in terms of effort. In this case, you would currently still be limited to the combination of ARCore/ARKit and GPS. However, the examples of Apple and Google show that this situation is constantly changing and that much more will be possible in just a few months’ time.