Pupil steering

Pupil steering is a technique for moving a display's exit pupil to keep its light aligned with the pupil of a moving eye. It is studied in near-eye displays for augmented reality and virtual reality, including Maxwellian-view systems whose small viewing region would otherwise make the image disappear as the eye moves.

Reviewed 21 October 2026. Original optical papers and bounded project and conference abstracts About review dates.

Optical principle

In a Maxwellian display, the optics concentrate the image-carrying light at a viewing point near the eye's pupil. Steering changes that viewing point rather than illuminating a large area around the eye continuously. One implementation uses a stack of polarization-selective cholesteric liquid crystal lenses: switching a liquid crystal polarization converter selects which lens directs the image toward the eye. Each lens provides a separately designed optical path. A 2021 proof of concept demonstrated two such viewing points using two lenses and one polarization converter.

Several approaches enlarge the usable eyebox, but they distribute light differently:

Approach	Operation
Pupil replication	Produces several viewing points simultaneously. Splitting light among them reduces the light available at each point.
Pupil selection	Estimates the eye's position and activates the corresponding member of a set of viewing points.
Pupil steering	Moves the viewing region with the eye. Some research also uses the term for switching among discrete optical paths.

Position, orientation and pupil size

Ratnam and colleagues distinguished two-dimensional lateral steering, three-dimensional steering that also changes axial position, and five-dimensional steering that additionally controls tip and tilt. Eye rotation changes the ocular pupil's position and orientation, preventing complete alignment through lateral translation alone. Their five-dimensional configuration was idealized.

The relative sizes of the exit pupil and ocular pupil also matter. An underfilled system sends light through a smaller area than the eye's pupil, increasing depth of focus and tolerance to some alignment errors, while a very narrow beam limits resolution through diffraction. A critically filled system matches the pupil size but becomes sensitive to small misalignments.

Gaze matching and image quality

Moving a viewing point does not automatically preserve image quality. In their 2021 work, Xiong and colleagues modeled a conventional lens coupler whose beam incidence angle changed during steering. Larger changes produced coma and an enlarged focal spot, allowing some peripheral rays to miss the eye and reducing the usable field of view. Their alternative used separately recorded lens couplers for different viewpoints instead of changing the incidence angle on a single coupler.

Gaze matching also concerns the direction of the image's chief ray relative to the eye's viewing direction. Zou and colleagues' 2022 design used individually recorded off-axis liquid crystal lenses to match different viewing directions. The laboratory demonstration provided three viewing points corresponding to eye rotations of -16 degrees, 0 degrees and +16 degrees, with a central field of view of about 55 degrees. The authors captured the images with a camera positioned and rotated to represent the eye.

Research examples

The 2017 Retinal 3D research system combined pupil tracking with light-field projection onto the retina. Its authors described a dynamic eyebox and used a holographic optical element as the see-through image combiner. Their publication presented two prototypes and investigated focus-cue generation; it described a research display rather than a consumer headset.

NVIDIA's 2019 Foveated AR research combined a movable high-resolution microdisplay for central vision with a Maxwellian-view peripheral display. A traveling holographic optical element moved the peripheral display's nodal point to follow the eye, while an infrared camera supplied tracking information. The central display's location and focal depth also changed with gaze. This apparatus illustrates how pupil alignment and moving the high-resolution image region can operate together as different parts of one display.

Tracking and experimental limits

Ratnam and colleagues' 2019 optical simulations examined vignetting, contrast and resolution, including tracking and steering latency constraints. The schematic-eye model used monochromatic light and monocular viewing. Its proposed tolerances required perceptual testing; they were not measured human thresholds.

A 2024 SIGGRAPH paper by Lee and colleagues described a perceptual testbed with adjustable steering resolution, accuracy and latency. Its reported research comprised two studies, one examining saccadic suppression and another the interaction between eye tracking and pupil steering. The authors also described a model of artifact detection based on temporal changes in global luminance and human visual sensitivity.

The 2022 gaze-matched demonstration used monochromatic optics, manual polarization rotation and image-position changes in presentation slides to demonstrate its steering paths. Electronic polarization switching and a steering mirror were proposed for practical operation. Its camera results therefore did not establish the performance of a complete closed-loop, eye-tracked headset.