Frequent on-site problems! How to solve AR drift issues

When trying to use AR (augmented reality) on-site, have you ever been troubled by issues like "the 3D model I displayed appears misaligned with the real object" or "after using it for a while the AR object drifts away from its position and floats"? In fact, in on-site AR usage such as outdoor construction sites or indoor equipment inspections, problems where AR content shifts from its intended position are frequently reported. When this phenomenon of "AR drift" occurs, the efficiency gains expected from AR are largely negated. So why does AR drift occur on-site? And what are the remedies? In this article we explain in detail the causes and countermeasures for AR drift that frequently occurs on-site, and finally introduce an approach using simple surveying with LRTK as a fundamental solution.

Contents

• Causes of AR drift and common on-site scenarios

• Basic countermeasures to prevent AR drift

• Fundamental solution: achieve "non-drifting AR" with high-precision positioning

• Easy high-precision AR using simple surveying with LRTK

• FAQ

Causes of AR drift and common on-site scenarios

There are several possible causes for AR displays shifting from the real-world position. One major factor is positioning errors. GPS in typical smartphones and tablets can have errors on the order of several meters. Therefore, if you try to display an AR object at a position from site maps or plans, the initial position can be off by several meters, causing the model to appear inconsistent with the real object.

Another important point is the surrounding environment and tracking accuracy. AR systems (the tracking technologies used in smartphones for AR) detect feature points from the camera image and estimate device motion to maintain the virtual object's position. However, if there are not enough distinctive patterns or objects in the surroundings (for example, a plain concrete wall or a featureless floor), the camera can easily lose position and the virtual object can slip or drift. Similarly, reflections from glass or water surfaces, dark locations, or highly dynamic environments (places with many moving people or vehicles) make the camera image unstable, degrading AR tracking accuracy and causing the object to jitter.

Also, errors in device heading (orientation) and pose estimation cannot be ignored. A smartphone’s electronic compass (magnetometer) can be disturbed near metal structures or high-voltage lines, causing the virtual model’s orientation to differ from reality. On sites where steel frames and heavy machines are common, compass disturbances can cause rotational misalignment in AR displays.

In addition, differences in data coordinate systems are a site-specific factor. If the coordinate system of construction drawings or design models does not match the actual surveyed coordinates on site, simply displaying the model in AR will not align it correctly. For example, if the drawing is drawn in a local reference coordinate system for each site, its origin and orientation will differ from the smartphone’s GPS or positioning data, so the model will not be displayed in the correct place as-is. If this coordinate mismatch is not corrected in advance, you will encounter troubles like "I placed it according to the drawing, but it appears misaligned on-site."

As described above, the background of AR drift involves various factors such as sensor accuracy, the surrounding environment, and data preparation. Especially on large outdoor sites, unlike indoor demos where the environment is stable, these problems become more apparent. So what concrete measures can prevent AR drift? The next section looks at basic points.

Basic countermeasures to prevent AR drift

To reduce AR content position drift, several basic measures and preparations are effective. First, as pre-data preparation, it is important to align the design data with the site coordinates. If drawings or models use a coordinate system different from the field survey coordinate system, perform coordinate transformation using known points on site (reference points) and adjust the data to match real space. For example, by acquiring the survey coordinates of several site reference points corresponding to points on the drawing and then translating and rotating the drawing data to align them, you can prevent major position mismatches when displaying AR. Also check the model scale and unit system. If the data are not created at the same scale as reality, the size will look off before even considering position, so unify units and set the model to actual scale.

Next, pay attention to device settings and handling. When using a smartphone outdoors, some apps prompt you to calibrate the device's electronic compass upon startup. Calibrate the sensors by moving the smartphone in a figure-eight before entering the site to get the most accurate heading possible. Also, when starting AR display, do not start walking immediately—slowly move the camera to look around the surroundings first to scan the environment. This allows the AR engine to recognize enough feature points and establish stable tracking. Avoid whipping the camera around, as sudden movements can disrupt tracking; instead, move slowly so the device can learn the surrounding scene.

Additionally, taking steps to improve environmental conditions is effective. In dark places, add lighting; in areas with many reflective surfaces, shoot from positions that avoid reflections; avoid times with heavy pedestrian or vehicle traffic to improve camera recognition. In feature-poor monotone locations, if possible, set up markers or targets. For example, placing a conspicuous patterned or colored target sheet on the ground, or attaching posters or markers (such as QR codes) to a wall, increases reference points the camera can use. Note, however, that outdoors these markers may be blown away or displaced by wind and rain, so marker installation is not always a permanent solution.

Stabilizing the device also helps prevent drift. Handheld operation tends to be unstable. If possible, mount the smartphone on a pole (a monopod or dedicated grip) to reduce camera shake and tilt. When using a pole, always operate with a consistent orientation (for example, facing north) to reduce heading variability and make rotational drift less likely. If orientation still does not match, comparing a linear object on-site (such as a road curb or building wall with a clear direction) to the corresponding line on the AR model and adjusting the model’s orientation until they are parallel is also effective.

If an AR object has significantly drifted, remain calm: stop moving, point the camera at the surrounding scene, and wait for tracking to stabilize. In many cases, remaining still for a few seconds and letting the device observe the environment will trigger re-recognition and correct the object position. If this does not fix it, use any in-app "reset alignment" feature and reload the model. Restarting from the initial state can eliminate accumulated errors prior to the reset.

By applying the above basic countermeasures, you can suppress many AR drifts. However, some drifts cannot be resolved unless the positioning accuracy itself is fundamentally improved. The next chapter explains the approach of resolving AR drift by means of "high-precision positioning," which is the key.

Fundamental solution: achieve "non-drifting AR" with high-precision positioning

In addition to the measures discussed above, if you can dramatically improve the positioning accuracy, which is the root cause of AR drift, you can greatly eliminate on-site position mismatches. The key here is utilizing high-precision GNSS positioning (RTK method) as an alternative to GPS. RTK (Real Time Kinematic) is a technique that improves satellite positioning accuracy by using correction information from a base station, achieving positioning on the order of a few centimeters. While conventional smartphone-built-in GPS could not avoid errors of about 5–10 m (16.4–32.8 ft), by using the RTK method even smartphones can obtain position accuracy of a few centimeters. As a result, AR-displayed models can be matched almost perfectly in real space, and virtual objects seen through the camera are fixed at the correct positions on the ground or structures.

With AR using high-precision positioning, the model will not move from its place even as the user moves, giving the impression of "projecting drawings or 3D models onto the site." Measures like placing ground markers or manual alignment mentioned earlier become unnecessary; simply pointing the device will align the virtual model with the real object. Because absolute position accuracy is ensured, users are freed from the recurring anxiety of "is the display accurate?" or the need to adjust a slightly misaligned model.

Of course, to use RTK GNSS positioning on-site, you need compatible receivers and services. A smartphone alone cannot perform RTK high-precision positioning, but by connecting a dedicated small GNSS receiver to the smartphone, you can upgrade your handheld device to support high-precision positioning. For example, a "smartphone RTK" system (e.g., LRTK) that combines a small RTK receiver attachable to a smartphone with a dedicated app makes it possible for one person to achieve centimeter-level positioning (cm level accuracy (half-inch accuracy)). By receiving correction information via the Internet in advance or utilizing satellite augmentation signals (such as Japan’s QZSS CLAS), you can obtain high-precision real-time self-position on the field.

Such high-precision positioning for AR is truly a shortcut to "non-drifting AR." In fact, the construction industry is advancing initiatives that combine high-precision GNSS and AR, and within the i-Construction movement advocated by the Ministry of Land, Infrastructure, Transport and Tourism, this is attracting attention as on-site DX. The next chapter introduces simple surveying with LRTK as a concrete solution to more easily realize such high-precision AR.

Easy high-precision AR using simple surveying with LRTK

Finally, as an on-site-friendly solution to utilize high-precision positioning, we introduce LRTK. LRTK (El-Ar-Tee-Kay) is a smartphone surveying system composed of a smartphone, a small GNSS receiver, and dedicated apps/cloud services. By simply attaching a small pocket-sized device to a smartphone, surveying tasks that once required specialized equipment and skilled personnel can be performed easily by one person. True to the name "simple surveying," work such as measuring coordinates of points on site or acquiring point cloud data via 3D scans can be completed with just a smartphone. Previously, achieving high-precision AR for construction required head-mounted dedicated equipment or positioning devices costing millions of yen; LRTK achieves comparable accuracy with only a smartphone and a small receiver, making it revolutionary from a cost perspective.

Using LRTK, you can obtain the centimeter-level positioning (cm level accuracy (half-inch accuracy)) described in the previous chapter in real time, so AR alignment headaches are almost eliminated. For example, if you upload CAD data or BIM models to the LRTK cloud in advance, you can display these models in AR on-site simply by pointing your smartphone, and their positions will match reality thanks to high-precision GNSS. You will likely be amazed as lines appear on the ground exactly as shown on the plans or structural models overlay perfectly onto the site scenery. The AR functions in the LRTK app realize "non-drifting AR displays" without repeated fine-tuning or placing markers.

Moreover, LRTK is not only for AR but also a surveying and design data integration solution. Acquired point coordinates, photos, and LiDAR scan point clouds are uploaded to the cloud immediately, allowing progress checking from office PCs and measurement of as-built conditions. Positions confirmed via AR can be recorded and shared as data on the spot, enabling you to correct discrepancies between the site and drawings immediately and preserve accurate construction records. For example, after driving a stake at a position indicated by AR, you can record the stake’s measured position with a single tap and share it on the cloud, enabling smooth workflows. Because surveying through construction verification are connected on a single platform, this becomes a powerful foundation for on-site DX.

By leveraging simple surveying with LRTK, you are not only freed from the stress of "AR drift," but you can dramatically improve on-site productivity and quality. If your site faces challenges in AR utilization, consider reviewing high-precision AR solutions using LRTK.

FAQ

Q1. What are the main causes of AR drift on-site? A. The primary causes are limitations in device positioning accuracy and tracking accuracy. Errors in smartphone-built-in GPS can cause initial display positions to be off, and unstable camera feature tracking can make virtual objects slide or drift. Outdoors especially, environments with few distinctive features and compass disturbances increase the likelihood that AR will appear displaced from its intended position.

Q2. How can I reduce AR drift? A. There are several measures. Align drawing data coordinate systems to field survey coordinates in advance, calibrate the device compass, and when starting AR, slowly look around to stabilize tracking. Additionally, add lighting in dark places, and place markers in feature-poor areas. If an object drifts, stop and let the camera re-recognize the surroundings, or use the app’s reset function to start over.

Q3. Do I need special equipment to achieve high-precision AR? A. Essentially you need an AR-capable smartphone or tablet, an RTK-GNSS receiver capable of centimeter-level positioning (cm level accuracy (half-inch accuracy)), and a dedicated app that links them (e.g., the LRTK app). While AR display itself is possible with a smartphone alone, RTK-based positioning is indispensable for high-precision alignment. Therefore, you need to connect a small GNSS receiver to your phone to acquire high-precision positioning information. The advantage is that you can achieve high-precision AR without the large-scale equipment previously required.

Q4. Can AR really achieve centimeter accuracy? A. If RTK positioning is properly performed and coordinate alignment with drawing data is correct, planar positions can be within an error of about ±a few cm. This rivals traditional surveying tools such as total stations. However, in smartphone AR, height errors due to device pose can be on the order of a few cm to 10 cm (3.9 in) (height projection is more affected for distant objects). Even if planar position is very accurate, there may be slight height discrepancies, so when checking critical heights it is advisable to verify on a perfectly flat surface, for example.

Q5. Will using markers prevent AR drift? A. Image markers or QR codes can allow precise alignment of models on site, but if the camera loses sight of the marker the model will drift again. Outdoors, markers are also difficult to manage and can be blown away or displaced. Therefore, marker methods are considered an auxiliary measure where GNSS cannot be used; using high-precision GNSS for alignment is generally more stable.

Q6. What should I do where GNSS cannot be received? A. RTK-GNSS cannot be used indoors or inside tunnels, so errors will inevitably accumulate over time. LRTK has an indoor positioning mode and can maintain positioning via inertial navigation for short periods, but drift increases over long durations. For large indoor areas, periodically resetting position using known points (pre-surveyed reference points) or correcting with QR codes or other markers is recommended.

Q7. Is LRTK easy for beginners to use? A. Basic operations are completed within the smartphone app and the GUI is intuitive, so even those unfamiliar with site work can quickly get used to it. Coordinate guidance and AR displays operate like a game by following on-screen instructions, so no specialist knowledge is required to handle them. However, understanding some principles of high-precision positioning (for example, "why sometimes Fix is not achieved" or "what a coordinate system is") helps diagnose problems if they occur. LRTK’s support site and manuals summarize such knowledge, so by learning alongside operation anyone can become proficient in a short time.

Q8. How should I start when introducing LRTK? A. First, obtain the LRTK app from the official site and try basic functions with a free trial account. Even with just your smartphone you can experience AR display and point cloud viewing using sample cloud data. A receiver is required for centimeter-level positioning, but rentals or demo units are available, so inquire if interested. When introducing the system, define which site tasks you want to use it for (e.g., stake-out, as-built verification), and introduce it in stages. With support from the LRTK team, start trials on small sites or limited workflows and gradually expand usage to achieve smooth adoption on site.