Production LiDAR capture app affected by ARKit world-tracking drift on iOS 26.4+ (re: thread 827240) mitigation guidance and fix timeline?

Hello, and thank you for taking questions this week.

We develop DeepWalk, a production App Store application that captures LiDAR depth, RGB, and ARKit camera poses while field technicians walk continuous outdoor routes (sidewalk and infrastructure assessment). Recorded ARCamera.transform values are used downstream for 3D reconstruction, so world-tracking accuracy directly determines our output quality.

Problem: Beginning in early April 2026 and coinciding with the iOS 26.4 rollout across our device fleet, we observe directional, accumulating pose drift in recorded trajectories during normal walking motion.

The behavior matches the regression reported in forum thread 827240 (world-tracking drift on LiDAR devices on iOS 26.4+, absent on non-LiDAR devices running identical code). Link to existing post here: https://developer.apple.com/forums/thread/827240

Our configuration is minimal: a default ARWorldTrackingConfiguration with frameSemantics = [.sceneDepth, .smoothedSceneDepth] and no other options set, running on a bare ARSession with a Metal renderer (no SceneKit/RealityKit).

We measure drift directly from per-frame ARCamera.transform values, so the displacement appears in ARKit's pose output itself rather than in any rendering layer. The drift accumulates with walking distance and does not occur when the device is stationary.

We've observed this across a large variety of devices (iPhone 15/16/17 Pro/Pro Max) on all iOS versions 26.4 or greater. We've also observed this on iPad Pros with updated firmware. Thread 827240 additionally reports that disabling autofocus, HDR, and plane detection and varying scene reconstruction and world alignment did not resolve it.

Question Because depth capture is our core function, we cannot avoid engaging the LiDAR scanner, and we're not aware of a public API that adjusts its contribution to world tracking. We'd be grateful for guidance on three questions:

  1. Is this a known issue in iOS 26.4 and later? Did the LiDAR contribution to world tracking change in that release?
  2. For apps that require sceneDepth, is there any recommended session configuration, video format, or capture-side practice that reduces the drift in the interim?
  3. Is a fix planned, and if so, should we expect it in an iOS 26.x update or the iOS 27 cycle? This affects how we manage our deployed device fleet.

We're happy to provide trajectory datasets, additional sysdiagnoses, or anything else useful. We have quantified drift-per-distance data across OS versions.

Answered by Vision Pro Engineer in 892467022

Hello,

Thank you for your report. The observation you are describing is a known issue and under investigation. To help us investigate further, I encourage you to file a feedback report with detailed information about how to reproduce. Please also include a sysdiagnose from a device that exhibits the issue and, ideally, a data recording taken with Reality Composer. Thank you.

Hello,

Thank you for your report. The observation you are describing is a known issue and under investigation. To help us investigate further, I encourage you to file a feedback report with detailed information about how to reproduce. Please also include a sysdiagnose from a device that exhibits the issue and, ideally, a data recording taken with Reality Composer. Thank you.

Thanks so much for taking a look here and getting back to us! We'll work to submit a feedback report including the sysdiagnose + data recording from Reality Composer ASAP.

In the mean time, is there anything we could try to mitigate the impact of this issue on our end?

Thanks again!

We are seeing the same ARKit world tracking drift behavior in our outdoor AR application.

We filed a Feedback report: FB23200349

We are seeing this drifting issue with Pix4DCatch when scanning objects using LiDAR assisted photgrammetry. Feedback report: FB23375411

Our team has resorted to bulk purchasing used iPhones from Ebay to mail to our users. This might be the only viable workaround for certain business models.

This has been a very disruptive issue for us and we're actively transitioning our business away from iPhone based scanning. Feedback report sent by our team.

We are still seeing this issue on iOS 26.6.

We have continued testing our outdoor ARKit application on LiDAR-equipped iPhones, and the world-tracking drift is still reproducible for us on the latest iOS 26.6 release.

Our Feedback Assistant report FB23200349 is also still open.

Has there been any progress on the investigation since the last update? In particular, has this issue been addressed in any current iOS 27 beta, or is a fix planned for an upcoming iOS 26.x / iOS 27 release?

This remains a significant issue for our application because stable outdoor world tracking is a core requirement for us.

Update from our side:

Apple replied to our Feedback report (FB23200349) and indicated that they believe the issue was fixed in iOS 26.6.1.

We have since done additional outdoor testing on newer iOS versions.

Our current result is:

  • The continuous drift issue that we were previously able to reproduce appears to be resolved on newer iOS versions.
  • However, we still observe some tracking/alignment inconsistencies during outdoor AR use, especially when changing perspective.
  • We therefore would not yet consider the overall outdoor AR tracking behavior fully resolved.

We are continuing to test and have shared our findings back through Feedback Assistant.

We are still seeing this on iOS 27, so from our side it does not look resolved yet.

Context: we ship a 3D scanning app where the user walks around a subject while an ARSession runs. The same app build behaves differently depending on the device family:

  • LiDAR devices (iPhone Pro models): pose drift during the capture. In our results it shows up as a metric scale error of roughly 3–5% that changes from one scan to the next, so we cannot calibrate it out.
  • Non-LiDAR iPhones: no such issue with the same build and the same capture procedure.

Configuration: ARWorldTrackingConfiguration, scene reconstruction off, with the sceneDepth and smoothedSceneDepth frame semantics on LiDAR devices.

An earlier reply in this thread mentions that Apple indicated, through Feedback FB23200349, that the issue was believed to be fixed in iOS 26.6.1. We could not find any mention of it in the release notes from iOS 26.4 up to the iOS 27.3 beta, either as a known issue or as a resolved one, and what we observe on iOS 27 suggests that at least part of the problem remains.

Could someone from Apple confirm:

  1. whether a fix has shipped, and in which release;
  2. whether the remaining drift on LiDAR devices on iOS 27 is a known issue;
  3. whether there is any recommended configuration or capture practice to reduce it in the meantime?

Thanks