7 Causes of Positional Drift in AR for Buried Pipes | Field-Effective Corrections and Recalibration Procedures

In recent years, the use of AR (augmented reality) technology has progressed at underground infrastructure sites, including construction and civil engineering as well as water and sewer, gas, power, and telecommunications. Using a variety of devices—from smartphones and tablets to dedicated AR headset-type devices—it has become possible to visualize the locations of buried pipes and cables on site. By overlaying 3D models of buried objects and drawing information onto the real-world view seen through a camera, you can intuitively grasp "what is buried where beneath the ground." This is a revolutionary method for preventing construction mistakes and for safety checks, offering the advantage that even less-experienced technicians do not have to rely on the "instinct" of veterans. It also makes it easier for site personnel and clients to share the completed image and the positions of buried objects, contributing to a reduction in communication loss.

However, a common problem when using AR displays for buried pipes in the field is the phenomenon of the AR display position shifting. The virtual pipe model shown on the screen does not match the actual buried location, and in some cases it can appear displaced by tens of centimeters (several in) to several meters (several ft). For example, a water pipe that should be buried in the ground may be shown slightly to the side in AR, or its height may be incorrect so that it appears to float. If the display is offset, relying on AR to excavate where it indicates "the pipe should be here" can lead to a critical mistake where the actual location is different. In civil engineering and infrastructure work that require precise construction, errors in AR display are a serious problem that cannot be overlooked. If field staff lose confidence and think "AR can't be relied on," the digital technology that was introduced may end up unused.

Then, why does positional misalignment occur in buried-pipe AR displays? In this article, we explain the 7 main causes of positional misalignment and introduce field-useful correction methods and procedures for recalibration (readjustment) for each. By correctly understanding the causes and taking appropriate measures, you can align AR models of buried utilities precisely with reality and use them in on-site operations with confidence and safety.

Cause 1: Initial position offset due to the limitations of GNSS positioning accuracy

When launching AR for buried pipes on site, the entire displayed model can sometimes be displaced by several meters (several ft) from its actual position. In many cases, this is due to insufficient positioning accuracy of the device’s built-in GNSS (GPS). The GPS in typical smartphones and tablets is said to have a planar positional error with a radius of about 5–10 m (16.4–32.8 ft). When an AR app places a model outdoors based on that built-in GPS location information, the virtual piping model is displayed laterally offset from reality by that amount of error. Especially right after first starting AR in a wide, open outdoor area with good visibility, this positioning error can cause the entire model to appear noticeably shifted to the side.

Furthermore, errors in the vertical direction (elevation) cannot be ignored. With standard GPS the accuracy for height is particularly poor, and it can be off by around 10 m (32.8 ft). As a result, a model that is supposed to be placed on the ground may appear to float in the air, or conversely may appear buried. In addition, positioning accuracy can worsen further depending on the surrounding environment. For example, in urban areas with many high-rise buildings or in mountainous regions, satellite signals are easily blocked and positioning errors larger than usual can occur. When that happens, the initial misalignment of the model’s position can become even greater, and the virtual model may be displayed in a location far removed from the real object.

Cause 2: Offset due to inconsistent coordinate systems in drawing data

Large positional errors can occur when the coordinate settings on the design-data side for buried pipes and other elements displayed in AR do not match the site's surveying coordinate system. For example, drawings or 3D model data may have been created in their own local coordinate system (coordinates set with an arbitrary origin and orientation). If the reference points or coordinate system used on site differ from the references on the drawings, placing that model into real space with an AR app will not align correctly. Because the origin and orientation that should be matched are not aligned, the entire virtual model becomes offset by a fixed amount in the east–west, north–south, and vertical directions.

Similarly, differences in the scale or unit system of design data can also cause misalignment. If distances that should be measured in meters (ft) are handled in feet, or if the "north" on the drawings is offset from true north on site, attempts to align in AR will not match up. In particular, data for underground buried utilities are often drawn on traditional plan views using local dimensions, and in such cases you need to be careful when reconciling them with site coordinates. If you display design data that does not match survey coordinates directly in AR, it will appear offset on site no matter how accurate the device’s positioning is.

Cause 3: Inconsistencies due to visual alignment from a single viewpoint

When using a smartphone or tablet on-site to display AR, you may visually align the virtual model so that it overlaps with real objects in the camera view. For example, you might move an underground piping model to match the location of a manhole visible on the surface and adjust it so they appear to coincide. This approach is intuitive, but simply aligning from a single viewpoint can cause inconsistencies when seen from other angles. Even if the diagram and reality line up well from one spot, moving a short distance can make the model appear offset from buildings or terrain—many people have likely experienced this. This happens because relying only on visual adjustment from one location prevents achieving precise position and orientation alignment for the entire model.

Visual calibration by the human eye inevitably has its limits. Even if something appears to fit perfectly at one point, if the model is not placed at the correct height or depth a change of viewpoint will immediately reveal discrepancies. A lack of reference points used for on-site alignment can also be a cause of inconsistency. For example, even if you align the near corner of a building model to a single point on site, if the model is rotated even slightly compared with reality the far opposite edge will be misaligned with the site. To place an entire model in the correct position, orientation, and scale you should, in principle, verify and adjust it from multiple points and viewpoints; neglecting to do so will cause errors to become apparent later. There are cases where construction proceeds because things look correct from one direction, only to discover the misalignment from another direction later and require rework.

Cause 4: Model drift caused by accumulated AR tracking errors

A buried-pipe model in AR that was correctly aligned at first can gradually drift away from reality as the user walks around. This drift in model position over time and movement occurs because there are limits to the AR tracking accuracy of the device. The AR functions of smartphones and tablets estimate their position and orientation in real time using feature points in the camera’s imagery and motion information obtained from internal gyro and accelerometer sensors (IMU). This is fundamentally relative position tracking, and it cannot completely avoid the accumulation of small errors that occur with movement. Typical AR apps initialize the virtual space once at startup using the position obtained by the device’s GNSS as a reference, but subsequent movement is corrected only by relative position tracking from the camera and IMU. Therefore, as you walk around for long periods and over wide areas, small discrepancies accumulate and the model position that was initially correct gradually diverges from reality.

For example, a virtual piping line that initially aligned perfectly with ground markings may appear to hover a few centimeters (a few in) above the ground after walking around the area for 10-20 m (32.8-65.6 ft). Also, inadequate management of the anchors (reference points) that fix AR content can accelerate drift. In typical smartphone AR, you can set anchors based on a specific location to fix virtual objects in place. However, if the object serving as the configured anchor (for example, a marker placed on the ground or a distinctive surrounding object) becomes occluded from the camera or is moved elsewhere, the model will lose its reference and start to drift or float. In environments with large flat surfaces or featureless walls, AR is prone to lose track of its position, and the model can slip or slide even after a small movement. As described above, the tracking limits of a standalone device and anchor loss can cause the alignment of AR displays to degrade while moving.

Cause 5: Tracking failures due to the surrounding environment (feature points and lighting)

The accuracy of AR display is also heavily affected by the surrounding environment. To estimate its own position, a device depends on feature points in the environment captured by its camera, but depending on the site conditions there may not be enough feature points and tracking can become unstable. For example, a site where only monotonous concrete walls or floors without patterns or texture are present, a dark or nighttime work environment with poor lighting, or an environment with many highly reflective glass surfaces like mirrors—all of these prevent the camera from obtaining useful cues. As a result, the AR system may fail to detect position properly, causing virtual content to jitter or jump. Typical examples where a device easily loses track of "where it is" include wide open areas with few landmark objects, or scenes that show only walls painted pure white.

Also, there are environmental factors that can skew sensor accuracy. For example, if machinery that emits strong magnetic or radio noise—such as high-voltage power lines, large metal structures, mobile cranes, or generators—is nearby, the electronic compass (geomagnetic sensor) built into smartphones and tablets can be disturbed and fail to indicate the correct heading. Even a slight compass error can cause the orientation of a projected virtual model to shift, producing large positional inaccuracies. Similarly, interference from communication signals and vibrations can affect sensor behavior. In other words, when environmental factors degrade AR self-localization and sensor accuracy, on-site AR displays become unstable and prone to positional drift.

Cause 6: Unadjusted Device Sensors / Poor Calibration

Insufficient calibration of the various sensors built into smartphones and tablets is also a factor that causes AR display misalignment. In particular, the electronic compass (heading sensor) requires regular calibration, but when in a hurry on site people tend to neglect the adjustment. If the electronic compass is off, virtual models cannot be placed using the correct north reference, and the whole scene will appear rotated and misaligned. For example, if the compass heading is off by 5°, an object rendered 100 m (328.1 ft) away will be drawn about 8-9 m (26.2-29.5 ft) to the side. In addition, IMU sensors such as gyroscopes and accelerometers can contain slight errors due to temperature changes and aging. These are normally corrected by software, but it is best to perform a calibration once—such as immediately after restarting the device or when using AR functions after a long interval.

The accuracy of a device’s sensors can vary greatly depending on how they are used and adjusted. For example, before starting AR, holding the smartphone in your hand and moving it in a figure-eight motion makes it easier to correct bias in the electronic compass (the so-called compass calibration). Also, on devices equipped with a LiDAR scanner, looking around at the start and scanning the positions of the ground and surrounding structures helps stabilize spatial recognition and is effective at preventing vertical displacement of the model. If you omit these pre-adjustments, sensor errors can remain large, which will directly lead to reduced AR display accuracy and positional drift. It is important to make a habit of resetting and adjusting the device sensors as a small step before going to the field or starting work.

Cause 7: Errors and Insufficient Updating of Buried Pipe Data

Lastly, I will also touch on errors in the source data used for AR. No matter how much you improve device positioning and sensor accuracy, if the drawings and location data for the buried pipes themselves differ from reality, you cannot achieve an accurate overlay. What often becomes a problem on site is the reliability of old buried-asset ledgers and drawings. For water and sewer pipes, gas pipes, and the like that have undergone repeated repairs over many years, it is not uncommon for the positions recorded on drawings to differ from the actual buried locations. The fact that cases of “there shouldn’t be a pipe here according to the drawings, but one was found when we dug” continue to occur nationwide also reveals the uncertainty of existing infrastructure information.

When the input data itself contains errors or update omissions like this, AR simply displays the information it has been given, so it will be out of sync with reality. For example, if a ground-penetrating radar survey discovers a newly buried object but it is not reflected in the drawing data, AR will not display it as existing and it may be overlooked. Conversely, if a pipe that is shown on the drawings has actually been removed but remains due to a failure to update the data, a ghost-like, fictitious piping model will be displayed in AR and cause confusion. AR accuracy management depends on both the device side and the data side. On-site, it is essential to prepare up-to-date, high-precision buried-object data and to update information through surveying or investigation as necessary.

On-site corrections and recalibration procedures

Based on the above causes, we outline corrective measures and recalibration methods to minimize positional offsets of AR for buried pipes on site. With a few simple adjustments you can improve the accuracy of the AR display, so implement measures both before heading to the site and while working on site.

As an application of high-precision GNSS positioning, the basic starting point is to improve the positional accuracy of the device. Rather than relying on the smartphone’s built-in standard GPS, combine technologies that can achieve centimeter-level positioning (cm-level, half-inch accuracy) whenever possible. Specifically, it is effective to use high-precision location information enhanced by GNSS corrections such as RTK. In recent years, solutions have emerged where small external RTK-GNSS receivers are attached to smartphones to receive correction information in real time and raise positioning accuracy to within a few centimeters (a few cm (a few in)). By utilizing such devices, it becomes possible to almost entirely eliminate the initial model-placement offsets that used to be several meters (several m (several ft)). If absolute positional accuracy is increased, virtual models can be displayed in almost the correct location from the start, greatly reducing the risk of large positional displacement.

As a preliminary check and calibration of drawing coordinates, confirm and adjust in advance the coordinate systems of the drawings and buried-utility models you will be using next. Check whether absolute coordinates (latitude/longitude or plane rectangular coordinates) are properly set in the drawing data and make sure they are aligned so there are no discrepancies with the field survey coordinates. Even if survey coordinates are not included in the drawings, you can measure several known points on site (for example, the coordinates of boundary stakes or manholes) and use those values to correct the model on the drawing. Many AR apps have a calibration function that assigns coordinates obtained on site to corresponding points on the drawings and moves/rotates the entire model. By surveying a point such as a building corner or a point on the road centerline and telling the app where that point corresponds in the drawing data, you can align the model to the correct position even for drawings that use a different coordinate system. Because accuracy improves the more reference points you use for verification and calibration, it is ideal to perform alignment using two or more reference points whenever possible for positioning important structures. Also, if you attach the official survey coordinate system to the drawing data in CAD software beforehand, on-site adjustment work will be significantly smoother.

As a multi-viewpoint position check, whenever you place and calibrate a virtual model in AR, always verify its positional relationship from multiple viewpoints on the spot. This is because something that looks correct from one spot can reveal misalignment when viewed from a different angle. Even moving just a few steps to change your left-right or height perspective lets you check whether the model and the real object truly overlap accurately. For example, verify from other directions whether the buried pipe’s route line appears parallel to or aligned with the road curb or other structures, and whether the pipe end is positioned directly beneath a manhole. By using parallax for confirmation, you can ensure the overlay is not only visually convincing but genuinely precise. If you find any misalignment from another angle, fine-tune the alignment on the spot immediately.

To make effective use of anchors and markers, actively use any anchor-locking features in your AR app when they are available. A model that has been correctly aligned once will be less likely to drift later if you set an anchor (reference point). For example, if you can place a printed QR-code marker on the ground and register it as a reference point in the app, fix the model to that marker position. If you install markers in advance at reference positions on the plans (such as the center of an intersection or the corner of a structure) before going to the site, you can perform position alignment smoothly in the field. However, you must also ensure the stability of the reference itself—for example, secure paper markers so they won't blow away in the wind and take care they are not accidentally moved by workers. You can also use environmental landmarks such as surrounding buildings, utility poles, or trees as anchors. Choosing objects with distinctive patterns or shapes as references helps the camera keep track of them and stabilizes tracking. In recent years, systems have appeared that store anchor information in the cloud for sharing across multiple devices, as well as AR platforms that manage anchors by linking them to GNSS coordinates. Utilizing these features makes it less likely that the model's position will drift even when moving over a wide area.

To maintain AR accuracy on-site, calibrating the device sensors is also essential. Before starting AR, perform electronic compass calibration. Hold the smartphone level in your hand and slowly swing it several times in a figure-eight pattern to reset magnetic sensor bias (this action is widely known as compass calibration). Also, immediately after launching AR, move the device slightly up, down, left, and right to allow the camera to thoroughly survey the surrounding environment. Doing so makes it easier for the AR system to capture spatial feature points and stabilizes positional tracking. For devices equipped with LiDAR, we recommend walking once around the area to scan the surroundings before starting work so it can capture the shape of the ground and structures. If ground height is recognized correctly, you can prevent models from floating in midair or sinking below ground. When sensors are properly calibrated and warmed up, AR display stability and accuracy improve significantly.

As a measure to account for environmental conditions, efforts to improve the worksite environment also help prevent AR misalignment. When using GNSS, perform positioning in locations with as open a sky as possible, and avoid spots where signals are blocked, such as the canyons between high-rise buildings or dense tree cover. Strong direct sunlight not only makes smartphone screens hard to see but also reduces camera recognition accuracy, so use sunshades as needed or at least move into the shade during alignment. In dark sites, provide appropriate lighting with headlamps or floodlights so the camera captures sufficient information. Additionally, when working near large heavy machinery, be aware of sensor anomalies caused by magnetism and vibration. Occasionally check that the electronic compass is not drifting, and if you notice abnormalities, move away from the machinery and recalibrate. For long AR sessions, it is also effective to mount the smartphone on a tripod to prevent shifts in position, and to connect a portable battery to prevent malfunctions caused by low power.

Even with thorough precautions, the AR display can still drift on site over time or due to environmental changes. If you notice a drift during work, promptly perform a recalibration (re-alignment). First, check the device’s GNSS reception status and verify whether positional accuracy has degraded. If necessary, re-acquire correction information from satellites and try recalibrating the electronic compass. Then recalibrate the model position at multiple points on site. Measure reference points different from those used earlier and fine-tune the model position; in many cases this resolves the drift. The important thing is not to continue working in haste. If you leave it thinking “it’s only a little off, it’ll be fine,” it may lead to irreparable mistakes later. To prevent rework and accidents, if something feels off, calmly redo the positioning and alignment and keep the AR consistently overlaid in the correct position.

By combining and applying the measures above, you can significantly reduce AR display errors in the field and minimize the positional offset between virtual models and real-world objects. In recent years, smartphone performance improvements and advances in correction techniques have been remarkable, and the once-common notion that "AR is bound to be misaligned" is steadily being corrected.

Achieving drift-free AR with high-precision positioning using LRTK

A high-precision GNSS solution called LRTK is attracting attention as the trump card to fundamentally solve the positional-shift problem in buried-pipe AR. LRTK is a system composed of a compact RTK-GNSS receiver that can be attached to a smartphone and a dedicated app; simply attaching the device provides centimeter-level (cm level accuracy (half-inch accuracy)) high-precision position information via real-time kinematic (RTK) positioning. Weighing only about 125 g, it is highly portable and can be used easily by anyone without complicated operation or special qualifications. When combined with a smartphone, LRTK receives correction data from ground stations in addition to positioning signals from satellites, tightening the errors that were 5–10 m (16.4–32.8 ft) with conventional built-in GPS down to within a few centimeters (a few inches). In practice, the positioning accuracy when using LRTK has been reported as approximately ±1–2 cm (±0.4–0.8 in) in horizontal position and about ±2–3 cm (±0.8–1.2 in) in the vertical direction, allowing current position to be identified with dramatically higher accuracy.

If high-precision absolute positioning via LRTK is realized, the AR display accuracy of buried-pipe models will improve dramatically. For example, tasks that used to require placing QR markers or performing manual calibration at each site can be done simply by standing at the location with a smartphone if LRTK is available. Based on the precisely obtained current coordinates, the model on the drawings is automatically displayed in its designated position almost immediately. This eliminates the troublesome work of aligning positions and enables an AR experience without initial offsets. Even while the user walks around, LRTK continuously supplies centimeter-level position information (cm level accuracy (half-inch accuracy)) to the smartphone, greatly reducing the risk of virtual models floating or slipping mid-use. Even outdoor AR, where being off by tens of centimeters (tens of inches) used to be the norm, can be improved to a level of accuracy suitable for on-site construction by introducing LRTK.

LRTK can also be used on-site as a simple surveying tool. Although compact, it is equipped with a high-precision GNSS receiver, allowing you to measure arbitrary points on the ground and obtain coordinates. For example, you can measure local reference points or landmarks with LRTK and seamlessly import the results into an AR app with a single tap to update model positions. This enables field surveying and AR visualization checks to proceed in parallel, reducing work time and improving information-sharing efficiency within the team. From point coordinate measurement to reflecting them in drawing data and performing as-built checks via AR projection, being able to handle everything consistently with the compact configuration of a smartphone + LRTK is a major strength. It also supports cloud integration, allowing you to instantly share positioning data recorded on-site and model information placed in AR within the company.

By introducing LRTK to the field in this way, you can achieve drift-free AR in every situation, such as locating buried pipes, confirming pile positions, and sharing a preview of the completed design in advance. In particular, when visualizing underground buried objects, ensuring display accuracy can greatly reduce risks like “the pipe wasn’t where we thought it would be” or “we trusted the display and dug, damaging the pipe.” Compared to traditional methods that relied on paper drawings and visual estimates, high-precision AR using LRTK will dramatically improve site safety and work efficiency. It will also increase trust in AR technology and motivate all site staff to actively use digital tools.

If you are a site person troubled by positional drift in buried-pipe AR, it is especially worth trying this combination of high-precision GNSS and AR. LRTK makes it possible for anyone to perform near-zero-error, intuitive position verification with just a smartphone. Once you actually use it, you’ll surely be surprised to see the virtual model and the real object always aligning perfectly. In the coming era of construction sites where digital construction and infrastructure DX are increasingly important, positionally accurate AR technology will be a powerful tool. Be sure to incorporate high-precision AR positioning technology on site to realize safe, smart infrastructure management with “zero excavation accidents” and “zero construction errors.”