Trained reference objects can be used on iPhone and Vision Pro. See how to create a single pipeline for maintenance, measurement, and training with AR.

Direct answer

In iOS 27, Apple added support for reference objects in the ARKit APIs. The company claims that machine learning training is platform-agnostic, allowing you to reuse the same reference object across iOS and visionOS. This favors a common recognition pipeline between phone and Vision Pro, but robustness depends on training coverage, scale, lighting, occlusion, part similarity, and field validation.

A reference object can serve two devices

Reuse reduces training duplication and helps maintain the same identifier between mobile and spatial experiences.

Dataset coverage defines robustness

Capture real views, lighting, orientation and states; reflective surfaces, nearly identical objects, and wear all need to enter the test.

Recognition does not prove perfect pose

Separate detection rate, stability, position error, acquisition time and behavior during occlusion.

Operation needs fallback

If the object is not recognized, offer instruction, repositioning, manual selection, or escalation without interrupting the task.

Nexus Reading

Use iPhone for quick outreach and validation and Vision Pro for hands-free tasks, while preserving the same asset taxonomy.

FAQ

Does the same trained object work on iOS and visionOS?

Apple claims that trained reference objects can be used on both platforms.

Does this replace QR codes?

Not necessarily. Natural objects improve fluidity, while markers can offer simpler identification in some environments.

Which business pilot is recommended?

A distinctive piece, with controllable lighting and measurable inspection, assembly or training task.

Essential guides to delve deeper into the decision

Primary sources and references

This editorial analysis was produced by Nexus from the official sources below, consulted on October 13, 2026. The text is original and interprets practical implications for companies.

Date reported by main source: WWDC 2026; analysis published on October 13, 2026.