AR drift in point-cloud overlays: How to suppress spatial misalignment?

Table of contents

• What is AR drift?

• Main causes of AR drift

• Problems caused by AR drift

• Conventional countermeasures and their limits

• Eliminate spatial misalignment with high-precision GNSS (RTK)

• Achieve stable point-cloud overlays with LRTK

• Expanded use cases with LRTK: enabling simple surveying

• Frequently asked questions (FAQ)

What is AR drift?

The use of AR (augmented reality)—where virtual objects such as design drawings, 3D models, or point-cloud data appear overlaid on the real world through a smartphone—is attracting attention on construction sites and similar environments. For example, overlaying a point-cloud scan of a completed structure with the design model on site makes it convenient to intuitively check for misalignments. However, when using AR in practice, users often notice that “the display is slightly misaligned with reality…?” Virtual lines or models may appear tens of centimeters off from where they should be, and the point-cloud overlay that was expected to match precisely may not align. This positional error in AR display is often called “drift,” a phenomenon in which virtual objects gradually diverge from the real object over time or as the device moves. Large misalignments undermine trust in the AR content and make it unusable for on-site inspections or work support.

Main causes of AR drift

Several overlapping factors cause positional errors in current smartphone AR. The main factors include the following.

• GPS accuracy limits: The positioning accuracy of smartphone-built-in GPS is generally said to be on the order of several meters, and in urban areas satellite signal reflections off buildings can cause errors of more than 10 meters. If the initial position is off by that much, virtual objects in AR will appear that far from their actual locations. Vertical errors can also be large, sometimes appearing to float meters above ground or sink below ground.

• Smartphone orientation sensor errors: Gyroscopes and electronic compasses (magnetometers) also have errors and drift (zero-point drift). Even a slight heading error can have a large effect on distant display positions (for example, if the heading is off by 2°, at 10 m (32.8 ft) ahead this can translate into several tens of centimeters of offset). Over long use, the gyroscope’s reference can shift and AR objects can gradually drift.

• AR mapping tracking errors: Smartphone AR estimates device movement by tracking feature points in camera images (VIO: Visual-Inertial Odometry), but this is not perfect. In wide movement ranges or environments with few distinctive textures or objects, self-position estimation can gradually deviate and the virtual model’s position drifts. In other words, tiny errors inherent in the AR engine accumulate and produce display misalignment over time.

• Environmental factors and coordinate inconsistencies: Magnetic disturbances from surrounding steel frames or machinery can throw off the compass and create heading errors. Also, if the coordinate system of the drawings or point clouds does not match the site surveying coordinates, the virtual model cannot be placed correctly in the first place, causing large positional errors. For example, if the design data’s reference point settings are incorrect, the AR display can deviate from reality by meters.

As described above, GPS errors, sensor errors, AR algorithm limitations, and coordinate inconsistencies combine to produce some degree of AR display misalignment that is difficult to eliminate entirely.

Problems caused by AR drift

If AR overlays remain misaligned, various problems arise when using them on site. First, the reliability of the display is compromised, so workers cannot trust the information shown in AR and end up relying on visual inspection or traditional measurement methods. For example, even if a construction element is actually placed correctly, if it appears misaligned in AR it creates unnecessary doubt about whether rework is needed. Conversely, real defects could be missed because of AR display errors.

Also, because low-accuracy AR is unusable in practice, implementations often remain “toy-like” on site and workers revert to tape measures or surveying equipment. If the misalignment is several meters it is unacceptable, but even errors of several tens of centimeters are unreliable in tasks that require precision, such as setting out pipe locations or checking as-built shapes. Site feedback often states, “AR that is off by several meters is unusable, but if the error is within a few centimeters it is sufficiently useful as a checking tool.” Put another way, only by suppressing AR misalignment to the centimeter level does AR become a practical tool.

Solving the AR drift problem and improving accuracy is therefore the key to full-scale on-site adoption of AR.

Conventional countermeasures and their limits

Before high-precision GNSS became available, various measures were used on site to cope with AR display misalignment. The main countermeasures and their challenges are as follows.

• Sensor calibration: Methods such as calibrating the smartphone’s electronic compass before use or placing it on a level surface to reset gyroscope drift. Basic actions like waving the device in a figure-eight or pointing the device to a reference direction to fine-tune reduce sensor errors as much as possible. However, this does not completely eliminate misalignment, and errors re-accumulate over long use or environmental changes.

• Alignment using markers or reference points: Placing QR-code markers around the site and reading them with the camera to calibrate the virtual model position, or manually aligning the model to known site reference points. This corrects the initial position but misalignment can recur in other areas when moving over a wide range. Placing markers or performing manual alignment each time is labor-intensive and unsuitable for quick checks.

• Use of simple surveying in combination: Instead of relying solely on AR, measuring coordinates of key points with a total station or GPS device and applying an offset correction to the AR model can be used. For example, measuring a reference point on the drawing in the field and entering the difference into the app to correct the entire virtual model. This can achieve high accuracy, but it is no longer “lightweight AR” and requires surveying work.

All of these measures have some effect and sites have experimented with AR using these tricks. However, they did not fundamentally solve the problem, and in practice AR remained at the level of “usable only with caution.” Manual adjustments and marker placement are time-consuming, and device sensors alone cannot maintain accuracy over long periods and wide areas. Conventional methods are insufficient to completely eliminate AR drift.

Eliminate spatial misalignment with high-precision GNSS (RTK)

A decisive solution to the AR drift problem is the use of high-precision GNSS positioning. Among these, the RTK (Real Time Kinematic) method dramatically improves positioning accuracy by correcting satellite positioning errors. While ordinary smartphone GPS has errors of 5–10 m or more, using RTK can measure current position with very high accuracy—horizontal positioning to ±1～2 cm (±0.4～0.8 in) and vertical errors within a few centimeters. By using correction information obtained from a base station to cancel out satellite signal errors in real time, errors of several meters can be reduced to a few centimeters.

If the smartphone can know its own position to centimeter-level accuracy, aligning virtual objects to real-world coordinates becomes dramatically easier. Because the device’s perceived “current location” becomes nearly accurate, placing models according to the coordinate data in design drawings or point-cloud models lets virtual and real objects align precisely in AR. What was previously difficult—overlaying drawings on the real scene with millimeter-perfect fit—becomes realistic.

Furthermore, if high-precision position information is constantly fed to the device, the AR engine’s drift (gradual deviation) can also be suppressed. Using absolute position coordinates obtained by RTK to periodically correct AR alignment automatically cancels small tracking errors. The result is stable displays where models do not float or shift as the user walks, without the user having to intervene. In short, RTK centimeter-class positioning can almost completely resolve AR positional error. Recently, solutions have emerged that make RTK positioning easy to use with smartphones, marking the start of an era where high-precision AR is available to anyone.

Achieve stable point-cloud overlays with LRTK

A representative solution that makes high-precision GNSS easy to use on site is LRTK. LRTK is an all-in-one positioning system consisting of a compact RTK-capable GNSS receiver that mounts on a smartphone, a dedicated app, and a cloud service, enabling centimeter-level AR display with a single smartphone. With an internet connection, correction data can be received from services such as VRS (Virtual Reference Station) using the Geospatial Information Authority’s reference stations, allowing RTK positioning without a dedicated base station. In areas with poor communication, a simple mobile base station can be set up to send correction data via radio, and within Japan LRTK can also operate with the CLAS signal from the Quasi-Zenith Satellite System (QZSS) to provide augmentation without internet connectivity. Because LRTK supports multiple correction methods, it can flexibly provide centimeter-class positioning depending on site conditions.

By projecting design data or point-cloud models stored in the cloud onto smartphone AR using the high-precision position and heading information obtained from LRTK, virtual objects appear at their true size and location without cumbersome pre-alignment work. For example, if the design model or as-built point-cloud data are pre-tagged with surveying coordinates (latitude, longitude, elevation), simply pointing an LRTK-mounted smartphone at the site will almost perfectly overlay that 3D data at the correct real-world location. There is no need to place alignment markers or perform manual adjustments each time. Even if users walk around a large site and view models from various angles, the model remains stably displayed in the correct position and orientation. Worries such as the model floating tens of centimeters when you walk to the site’s edge are eliminated, freeing users from AR drift-related stress.

LRTK is also designed for ease of use on site. The smartphone-mounted compact unit eliminates the need to carry large tripods or heavy surveying equipment. With just a smartphone, site supervisors and engineers can perform positioning and AR display on the spot, meaning many tasks no longer require a specialized surveying team. For example, baseline setting and as-built checks that formerly required outsourcing to a surveying company can now be done immediately by the team on site with LRTK. This reduces losses from outsourcing and scheduling, enabling a faster PDCA cycle on site. By enabling high-precision point-cloud overlay AR with only a smartphone, AR technology is beginning to establish itself as a practical tool on worksites.

Expanded use cases with LRTK: enabling simple surveying

LRTK is expected to be used not only for overlaying design data but also as a simple surveying tool. A new workflow called “smartphone surveying” is emerging, allowing tasks that previously required multiple people to be performed intuitively by a single person.

For example, in pre-construction pile layout (staking out) LRTK’s coordinate navigation function can display the bearing and distance to a specified coordinate on the smartphone screen. Workers can follow an on-screen arrow or virtual stake marker and arrive at the specified location with an accuracy of a few centimeters. Stake layout, which used to rely on manpower and experience, becomes something anyone can do accurately with a smartphone in hand. In as-built management, projecting the planned 3D model in AR lets you immediately check on site whether the finished shape matches the drawings. Whether fill material or a structure has reached the design elevation becomes obvious at a glance through the smartphone. If desired, point-cloud scanning with LRTK can be performed and compared with design data in the cloud to quantitatively verify quantities and shape differences on site.

LRTK also has a “geotagged photo” feature that automatically tags high-accuracy capture coordinates and camera orientation to photos taken with a smartphone and saves them to the cloud. When revisiting the same location later, icons representing past photos appear in the AR space, making it easy to check chronological changes or locate repair points. These features will significantly transform site management that has traditionally relied on manpower and intuition.

In short, LRTK is an all-in-one site DX tool combining AR visualization and positioning functions. It not only eliminates the stress of misalignment with high-precision AR but also enables seamless surveying, inspection, and record-keeping on site, promising dramatic improvements in work efficiency and accuracy. If you are interested, consider trying the new surveying and construction-management experience enabled by LRTK.

Frequently asked questions (FAQ)

Q: *What equipment is required for high-precision AR displays?* A: *Basically, a smartphone and a high-precision GNSS receiver (RTK-capable device) are required, along with a compatible AR display app. A representative approach is an RTK antenna that mounts on the smartphone with a dedicated app, as in LRTK. The smartphone itself does not have to be the latest high-end model, but it should be a device of sufficient performance that supports AR frameworks (ARKit or ARCore). For long-duration operation, a mobile battery is advisable.*

Q: *Can anyone use it with just a smartphone? Is special training required?* A: *Compared with conventional surveying equipment, operation is more intuitive, but it is ideal to receive basic training beforehand. Learning app operation, RTK fundamentals, and operational precautions in advance reduces the risk of confusion on site. The system is designed so that non-surveying specialists can handle it. In practice, site supervisors and construction managers have mastered it after a few hours of training and trial use and are using it effectively on site. Initially, it’s recommended to compare results with experienced surveyors and gradually expand the application range for smooth adoption.*

Q: *Do you need to set up an RTK base station every time? Can it be used where there is no network?* A: *It depends on how you obtain RTK correction information. If there is a nearby public reference station and internet connectivity, you can receive services such as VRS (Virtual Reference Station) from the Geospatial Information Authority on the smartphone and obtain centimeter-class positioning without setting up a dedicated base station (this requires the smartphone’s data connection). In areas with unstable communications, you can set up a simple base station (mobile station) and transmit correction data via radio. In Japan, if you use a receiver that supports CLAS (centimeter-level augmentation) from the Quasi-Zenith Satellite System, you can get correction information without internet connectivity. LRTK supports these multiple methods, allowing flexible operation depending on site conditions.*

Q: *How accurate is the AR alignment with reality in practice? I’m worried about errors.* A: *In favorable outdoor conditions, horizontal errors are typically within about 1～2 cm (0.4～0.8 in). Vertical errors can also be within a few centimeters, so in most cases you will not notice misalignment by eye. However, this assumes that the RTK fix solution is maintained and the device’s attitude sensors are properly corrected. Accuracy degrades in environments with poor satellite reception, and a miscalibrated smartphone compass will affect the display. We cannot guarantee 100% perfect alignment at all times, but in normal outdoor work it provides practically sufficient accuracy. It is important to verify AR accuracy on site against known points or landmarks and use it with occasional cross-checks. Even with some error, comparing with local references and making correction judgments on the spot makes AR robust enough for practical use and increases trustworthiness.*

Q: *Is AR overlay possible in dark or nighttime conditions?* A: *Positioning itself is obtained via GNSS, so measurement accuracy does not change at night. However, if the camera image is too dark the AR display becomes hard to see and visual tracking in the AR function becomes less reliable. For nighttime or dim sites, ensure camera visibility with lighting or use LiDAR-equipped devices which can be more stable in low light. Considering safety, it is preferable to perform surveying and AR checks during daylight when possible. In pitch-dark conditions, instead of relying on camera AR, switching to high-precision guidance by numerical cues (coordinates and distance guidance) from the GNSS is an effective alternative.*

Q: *What about places where GNSS cannot be used, like indoors or underground?* A: *Indoors or in tunnels where satellite signals cannot reach, absolute positioning by RTK is unfortunately not available. As an alternative, you can perform AR using a local coordinate system based on known points. For example, set reference points on the floor with a total station and manually input those coordinates into the app to serve as a substitute for the smartphone position, enabling basic indoor AR overlays. Research on indoor positioning using UWB (ultra-wideband) or Visual SLAM is also progressing. At present, achieving centimeter-level indoor alignment is not easy, but for some uses planar detection or marker-based AR may suffice. In short, where GNSS is not available, combining other positioning methods or reference-setting expands AR usability.*

Q: *I’m worried about cost—are these systems expensive?* A: *Compared with conventional surveying GNSS equipment or 3D scanners, a smartphone plus an RTK receiver is relatively affordable. Actual prices vary by model and service type, but costs can often be a fraction of a dedicated high-precision GNSS surveying set. Moreover, reducing outsourced surveying and preventing rework from construction errors can yield cost savings. Depending on site scale, savings in labor and time can lead to relatively quick return on investment according to some reports. There are also subscription-based products like LRTK that reduce initial costs and allow trial introduction. Starting with a small site to evaluate effects before scaling up is recommended.*