Is Your AR Navigation Misaligned? Causes and Solutions for AR Drift in the Field

Table of Contents

• Causes and challenges of AR drift

• Conventional countermeasures for AR drift

• The new on-site standard "LRTK"

• Drift-free AR navigation realized with LRTK

• The potential for simple surveying expanded by LRTK

• Summary

• FAQ

Causes and Challenges of AR Drift

In response to initiatives such as i-Construction promoted by the Ministry of Land, Infrastructure, Transport and Tourism, the use of AR (augmented reality) technology is expected on construction and civil engineering sites. By overlaying 3D models of blueprints and construction procedures onto the real scene via smartphones, tablets, and even smart-glasses-type AR devices, site personnel can intuitively share the finished image and issue instructions on-site. However, when AR is actually used on-site, there are often complaints that "the AR display is misaligned and problematic." Because the virtual objects displayed end up appearing inconsistent with their real-world positions, they are ultimately regarded as unreliable and avoided.

So, why do AR navigation and AR displays become misaligned on-site? The main causes are as follows.

• Device location error: The accuracy of the GNSS (GPS) built into typical smartphones has errors on the order of several meters. As a result, the device’s reported current position can be offset, and the placement of virtual models can be laterally displaced from the real object. Especially outdoors in wide areas right after launching AR, position offsets of 5-10 m (16.4-32.8 ft) can occur, causing the entire model to appear abruptly offset from reality. Vertical errors can also be large; with ordinary GPS, altitude can be off by nearly 10 m (nearly 32.8 ft). Consequently, models that are intended to be placed on the ground may appear to float in the air or be displayed as sunk below the surface.

• Device heading/pose error: If a phone’s electronic compass (magnetic sensor) is disturbed or the gyroscope has drift (accumulated error over time), the direction shown in AR will be off from reality. When the heading is incorrect, the virtual model will be rotated relative to the real world and will not align. Insufficient sensor calibration or magnetic noise in the environment can also be factors; if the model cannot be placed using a correct north reference, the whole display can appear skewed.

• Tracking failure due to the surrounding environment: AR apps estimate the device’s position by capturing feature points in the camera image, but tracking can become unstable depending on the surroundings. For example, in places with walls or floors that lack patterns or texture, at night or in dim conditions, or in environments with a lot of glass or water surfaces, the device can more easily lose track of its position, causing virtual models to gradually drift or suddenly jump. Without sufficient visual cues in the surroundings, AR cannot maintain alignment with the real world.

• Small errors in initial alignment: When placing a model in AR, the initial placement is sometimes done manually on-site by aligning to some marker. Even a small error in this initial alignment can produce inconsistencies across the entire site. In large sites, tiny angular or positional errors can manifest as large misalignments at distant locations. A model that appears to overlap correctly from one viewpoint may not match buildings or terrain when seen from another — an experience many users will recognize.

• Drift over time: AR displays can also drift gradually over time. This is caused by accumulation of internal sensor errors in the device or changes in lighting and structures that alter how feature points are detected. A model position that was correct shortly after starting may appear a few centimeters (a few in) off after the user has walked around for about 10 minutes.

• Mismatch between drawing data and coordinate systems: If the surveying coordinate system used on site does not match the coordinate settings of the design data used for AR, the entire model can be offset east/west/north/south or vertically. For example, if drawings were created in a local coordinate system based on an arbitrary reference point or orientation, they will not align with the site’s surveying coordinates as-is. Differences in unit systems (feet versus meters) or a shifted north reference on the drawing can also cause positional errors. No matter how accurately a device is localized, AR will be misaligned if the original design data do not match the site coordinates.

• Errors in the source data itself: If the drawing information or 3D model used for AR does not match reality, it cannot be overlaid accurately. With old drawings or data that have not been updated, the digital positions can differ from the actual site positions. In such cases, AR appearing “off” is to be expected — it is a problem of foundational information rather than the AR technology itself.

Because of the above factors, the AR 3D model displayed on-site can end up diverging from the real object (the so-called "AR drift"), resulting in cases where it is shunned as "AR is unreliable."

For example, if the position displayed in AR as "Dig a hole here" is actually off by tens of cm (tens of in), there is a risk of accidentally digging in the wrong spot. In the end, workers end up having to remeasure with tape measures and mason's line, and the AR deployment is ruined.

In recent years, devices equipped with LiDAR sensors and VPS (Visual Positioning Service) technologies have emerged, but it remains difficult to achieve stable, high-precision positioning across wide-area sites.

Conventional countermeasures for AR drift

Various techniques have been tried in the field to eliminate AR display misalignment. Let's look at some representative countermeasures.

• Install markers or QR codes: This method involves pre-attaching AR markers (image markers) or QR codes at the site and using the camera to read them as reference positions. It's convenient, but not realistic to rely on continuously at large outdoor sites. It can only be applied within the area where markers are placed, and outdoors they may peel off or become dirty from wind and rain, making stable operation difficult.

• Manual alignment using landmarks: This method compares obvious on-site landmarks (such as building corners or existing structures) with the model in AR and adjusts the position by eye. It provides some correction, but relies heavily on the worker's judgment and has limited accuracy. It's difficult to align with exactly the same precision every time, and variations between workers can occur.

• Reset whenever misalignment is noticed: This method re-aligns (resets) the model on-site each time the AR model appears misaligned. It can temporarily restore the correct position, but it interrupts work and is inefficient. Because it doesn't address the root cause, repeated resets are needed, which can become counterproductive.

• Pre-surveying and coordinate alignment: This method measures reference point coordinates at the site with surveying instruments and aligns the digital model to those coordinates. Accuracy improves, but it requires specialized surveying work and is time-consuming. Skilled personnel must be arranged, and there is also the risk that coordinate transformation errors will still leave discrepancies.

All of these measures entail additional costs and effort, and they often erode the inherent advantage of AR—its ability to be used easily and in real time on site.

As a result, many on-site staff may have half resigned themselves to “there’s nothing that can be done about AR misalignment.”

Note that carefully performing sensor calibration on the device (for example, waving the electronic compass to recalibrate it) can also be effective as a way to improve accuracy, but in busy on-site situations it is difficult to carry out thoroughly every time, and its effectiveness is limited.

The New On-site Standard "LRTK"

That's where LRTK (pronounced 'L-R-T-K') comes in. LRTK is a new solution developed to fundamentally solve the positioning drift problem in on-site AR usage. By attaching a compact RTK-capable GNSS receiver to a smartphone or tablet and obtaining high-precision positioning data in real time, it greatly reduces AR "drift". RTK stands for Real Time Kinematic, a method that uses correction information from a base station to reduce GPS positioning errors to the centimeter level (cm level, about half-inch accuracy). With LRTK, GNSS positioning errors that used to be 5-10 m (16.4-32.8 ft) can be tightened down to within a few cm (a few in).

The advantage of LRTK is that it makes it easy to achieve this dramatic improvement in positioning accuracy on site. It can be operated intuitively via a dedicated app so that personnel without surveying expertise can use it. As an initial setup, there is the task of registering the site’s reference coordinate system, but if the coordinate values of known points are available, it can be completed in a short time. Once the coordinate system is aligned, you only need to launch the app on site to immediately use high-precision AR. No complicated preparations or large-scale equipment are required; its simplicity—a regular smartphone plus a small device—enables survey-grade alignment, which is a major appeal of LRTK. Now regarded as “the new standard on site,” it is attracting attention at many sites struggling with AR misalignment.

LRTK-enabled "drift-free" AR navigation

When LRTK is implemented, on-site AR navigation becomes dramatically more accurate. Because the device's absolute position can be determined to within a few centimeters (a few in), virtual models can be displayed almost exactly in the correct location from the start. Initial placement errors—such as the several-meter (several ft) offsets that make content appear misplaced immediately after startup in conventional systems—are almost eliminated.

This makes "drift"—where the model becomes misaligned with reality—even less likely to occur when a worker walks around the site. Even if AR's self-positioning has some error, high-precision GNSS position corrections are always in effect, so virtual objects are less likely to shift or float away on their own. For example, in scenes where, with conventional systems, the model would start to look slightly lifted after moving about 10 m (32.8 ft), using LRTK keeps the model firmly anchored to the ground throughout.

Also, with LRTK, aligning the model to the site's coordinate system is smooth. By measuring a few known points on site (for example, the positions of boundary markers or structures) and assigning those coordinates to the corresponding points in the model data, you can align the drawing data with the site coordinates. Calibrating with multiple points further improves accuracy and allows you to correct the entire model's position, orientation, and scale. Thanks to LRTK's high-precision positioning, this coordinate alignment can be performed with high repeatability. Once properly aligned, even on large sites the AR display will consistently match the real world, and you will no longer be troubled by "drifting AR".

In other words, it is no exaggeration to say that LRTK is the first technology to allow AR to attain practical, in-field accuracy. Until now people may have thought "after all, AR is bound to be somewhat off," but LRTK overturns that conventional wisdom and is precisely what can realize a "drift-free AR navigation".

The potential of LRTK to expand simple surveying

The benefits LRTK brings extend beyond merely improving the accuracy of AR displays. It also brings innovation to on-site simple surveying. With high-precision GNSS enabling smartphones to perform like surveying instruments, some tasks that used to require professional surveyors can now be carried out easily.

For example, there are tasks such as recording the locations of buried objects and confirming as-built (post-construction shape), which involve measuring and recording point coordinates. With an LRTK-capable device, site personnel can measure points with centimeter-level accuracy (half-inch accuracy) and record them to the cloud themselves, without having to carry out surveying prisms or heavy equipment.

Also, the task of transferring coordinates from drawings to the field — a simplified version of staking or batter-boarding — can be carried out intuitively with AR. By displaying a virtual marking at the design location in AR and simply marking the ground while looking at it, high-precision positioning is completed.

In this way, LRTK enables on-site workers themselves to carry out tasks such as "measuring" and "establishing positions" to a certain extent. With labor shortages being widely reported, this also helps reduce the burden on surveyors and is expected to improve overall efficiency.

Moreover, because the data obtained can be shared digitally immediately, information sharing between the job site and the office is seamless. It can truly be said to be a tool that drives DX (digital transformation) in the construction industry.

The use of such simplified surveying can be applied not only to construction and civil engineering sites but also to any outdoor field that handles location information, such as inspection and maintenance of infrastructure facilities, agriculture, landscaping, and disaster response. LRTK proves effective in situations where you want to connect the site and drawings and visualize and record them in real time.

Summary

AR technology is a groundbreaking tool that expands possibilities on-site, but AR drift — the problem of "display shifting" — has long been a major barrier. Faced with misalignments caused by various factors such as the limits of device positioning accuracy, sensor errors, environmental factors, and coordinate inconsistencies, on-site teams have struggled to cope by installing markers and repeatedly resetting systems. However, those measures have not provided a fundamental solution and have even led some to give up on using AR.

LRTK revolutionizes this situation. By achieving absolute positioning with centimeter-level accuracy (half-inch accuracy) on-site, it aligns AR displays precisely with reality and makes "non-drifting AR" possible even as time passes or when moving across large areas. You no longer need to worry about "AR navigation drifting..." This enables confident use of AR on-site and greatly contributes to preventing construction errors and facilitating smoother communication. Furthermore, LRTK brings a new application to the field—simple surveying—making the bridge between digital and real more accessible.

By introducing LRTK, on-site positioning accuracy and work efficiency will improve dramatically. As a new standard in the i-Construction era, it can be said to be the optimal solution for improving productivity and digitalization in the construction industry. For more details, please also visit the [LRTK official site](https://www.lrtk.lefixea.com). If you have been troubled by AR misalignment, why not take this opportunity to evolve your site to the next stage with LRTK?

FAQ

Q: Why does AR display drift on site? A: The main causes are the limitations of device positioning accuracy and sensor precision, and the influence of the surrounding environment. A smartphone’s built-in GPS position information has an error of several m (several ft), and electronic compass errors and gyro drift also cause the heading to shift. Furthermore, in areas with few surrounding features, AR’s self-localization becomes unstable, and as a result the model no longer aligns with reality.

Q: If you use markers for alignment, can you prevent AR drift? A: Installing markers or QR codes provides a certain corrective effect, but it is difficult to use them routinely on large outdoor sites. They only work at the locations where the markers are placed, and there is the hassle of having to scan them with a camera each time. With LRTK, high-precision alignment can be maintained across the entire site without relying on markers, so that is a major advantage.

Q: Do you need any special equipment or expertise to use LRTK? A: Basically, you can use it with a smartphone (or tablet) and a compact RTK-capable GNSS receiver. The dedicated app is intuitive to operate, and you don’t need surveying expertise. You do need to register the site’s reference coordinates during the initial setup, but once you’re familiar with it, this can be completed in a short time. Even general site personnel should be able to master it with a little practice.

Q: How accurate is LRTK positioning? A: It depends on the environment, but you generally get accuracy on the order of a few centimeters (a few inches). While ordinary GPS has errors of several meters (several ft), LRTK achieves accuracy that is one-tenth or better of that. If you can receive satellites stably in a location with good visibility, reproducibility is such that measuring the same point repeatedly yields almost the same coordinates. Accuracy drops somewhat in places where radio signals are hard to reach, such as under elevated structures or under trees, but it is still far more accurate than conventional GPS.

Q: Can LRTK be used indoors? A: LRTK relies on GPS satellite signals for high-precision positioning, so it is fundamentally intended for outdoor use. It is currently difficult to use indoors or underground because satellite signals do not reach inside buildings or underground. However, there are cases where it can be partially utilized with some ingenuity, such as deriving relative indoor positions based on reference points measured outdoors. If it can be integrated with indoor positioning technologies in the future, its range of indoor use will expand.

Q: In what sites and applications can LRTK be used? A: It is expected to be useful not only at construction and civil engineering sites but in any outdoor scene that handles location information, such as inspection and maintenance of infrastructure, agriculture, and surveying work. Whether checking design data on site, considering the placement of structures with AR, or recording the locations of buried utilities, LRTK can be of great help whenever you need to link the field with drawings.