37.6 Retinal projection🔗⧉

Stand back and look at what every display in this part has quietly assumed. The stereoscope, the VR headset, the lenticular screen, and even the holographic panel each put light in front of the eye and rely on the eye's cornea and lens to form the retinal image, bending that light to a focused image on the retina. The display owns the photons; the eye owns the focus. That division is so natural we rarely name it, but it is the source of half this part's troubles. It is why a headset must place its screen at some fixed optical distance and then fight the vergence–accommodation conflict of Chapter 37.3, 3D displays with accommodation; it is why a person with a cataract or a strong refractive error sees every one of these displays through the same blur they see the world through. The eye's optics sit in the path, and they are not always cooperative.

Retinal projection asks the obvious next question: what if the display formed the retinal image itself, and reduced its dependence on the eye's accommodation? Instead of illuminating a screen the eye must focus, aim a thin, controlled beam of light through the pupil and onto the retina directly, painting the picture on the photoreceptors the way a scanning electron beam once painted a CRT. The beam still passes through the cornea and lens; controlling its pupil footprint reduces defocus sensitivity rather than bypassing those optics. This idea runs from a clever always-in-focus eyepiece to a display that addresses the retina cell by cell.

37.6.1 The Maxwellian view: focus set by the display, not the eye🔗⧉

The trick that makes retinal projection work is more than a century old and is named for James Clerk Maxwell. In ordinary vision, light from a scene point diverges as it reaches the eye, and the lens must bend that diverging cone down to a point on the retina: its power has to match the divergence, which is exactly what accommodation adjusts and what a wrong prescription gets wrong. A Maxwellian view arranges the optics so that the display's light instead converges to a tiny focus right at the plane of the pupil, and then diverges again past that point to spread across the retina. Because every ray of a given image point passes through the same pinpoint at the pupil, the map from "which direction the ray was going" to "where it lands on the retina" is much less sensitive to the eye's focus when the illuminated pupil region is small. Changing focus or refractive error then causes less blur over a useful range. Diffraction, aberrations, scattering, and alignment still limit sharpness (Figure 37.6.1⛶ fullscreen).

That single property is the whole appeal, and it cuts two ways. As an augmented-reality display it means a bright, always-focused overlay that never asks the eye to refocus between the virtual and the real. As an assistive device it is more striking still: reduced defocus sensitivity can help some viewers with refractive impairment. A narrow beam may avoid a localized lens opacity, but it still traverses the ocular media; widespread scattering or retinal disease is not bypassed. Clinical benefit depends on the condition and device. The catch is geometric and unforgiving. Converging all the light through a pupil-sized point means the eyebox (the region your pupil may occupy and still see the image) is itself only about that size. Move the eye a millimeter and the beam misses the pupil and the image vanishes. Practical retinal-projection systems therefore live or die by pupil tracking and by steering the convergence point to follow the eye, and even then the field of view and the tolerance to motion are hard-won. The trade-off is a pupil-sized eyebox.

37.6.2 From the virtual retinal display to laser eyewear🔗⧉

The idea became a device at the University of Washington's Human Interface Technology Lab in the early 1990s. Thomas Furness and Joel Kollin's Virtual Retinal Display Kollin 1993 (Virtual Retinal Display) did exactly what the name says: it scanned a low-power beam of light in a raster directly across the retina, modulating its brightness pixel by pixel, so a complete image was drawn on the photoreceptors with no screen anywhere in the system. Where a CRT sweeps an electron beam over phosphor and a projector throws pixels onto a wall, the virtual retinal display swept light directly across the living retina. It established the template every later system follows: a light source (soon lasers), a fast scanner to steer the beam, and Maxwellian-view optics to send it through the pupil.

Modern implementations trade the bulky scanner for a MEMS micromirror and use three (red/green/blue) semiconductor lasers for a full-color, collimated beam. The clearest commercial example is QD Laser's RETISSA eyewear QD Laser RETISSA (2018), which pairs an RGB laser picoprojector with a MEMS mirror and a Maxwellian optical relay to project directly onto the wearer's retina. Because the image is focus-free, RETISSA is marketed both as an AR display and as a low-vision aid, and several clinical studies use it precisely because it can present a legible image to eyes that refractive correction cannot help. The same retinal-projection concept surfaced in consumer AR prototypes such as Intel's Vaunt and the North Focals glasses, which reflected a low-power laser off a holographic combiner in the lens and into the eye to float a small, always-sharp heads-up display in an ordinary-looking pair of spectacles. Across these systems, the main advantage is a bright, in-focus overlay in a glasses-sized form factor. The main limitation is a tiny eyebox, a modest field of view, and the need to keep the beam threaded through the moving pupil.

First, retinal projection is not, by itself, a 3-D display: a single Maxwellian beam delivers a focus-free 2-D image to one eye. It earns its place in this part because it avoids the vergence–accommodation conflict: there is no screen plane to accommodate to, so the display's accommodation demand is weakened. This does not reproduce natural depth-dependent blur or guarantee comfortable viewing. The important distinction is that retinal projection is focus-free, not focus-correct. It removes the demand to accommodate rather than presenting a wavefront the eye can accommodate to, the way Chapter 37.5, Holography does. It sidesteps the problem; holography solves it. Second, retinal projection and holographic near-eye displays are cousins (both control the light delivered through the pupil), and the holographic near-eye work of Maimone and colleagues Maimone, Georgiou & Kollin 2017 can be read as the wavefront-reproducing member of the same family, shrinking the space-bandwidth budget to an eyebox rather than a room.

37.6.3 The frontier: writing to individual cones🔗⧉

Every system so far still projects an image, a raster of spots on the retina, blurred together by the eye's own sampling into a smooth picture. The frontier pushes past the image to the sampling grid itself. The enabling technology comes from vision science, not displays: adaptive-optics retinal imaging, which corrects the eye's own aberrations well enough to resolve single cones in the living eye, and can even classify each cone by its type (L, M, or S) from how it responds, the technique Roorda and Williams introduced in 1999 Roorda & Williams 1999. Once you have a map of the retinal mosaic that says which cone at which location is a long-, medium-, or short-wavelength cone, a new possibility opens: don't just paint light on the retina, deliver it cone by cone.

That is what Oz does, a 2025 system from a Berkeley-led team (Fong, Doyle, Wang, Ng, Roorda, Sabesan and colleagues) Fong et al. 2025 (Oz / "olo"). Oz uses an adaptive-optics scanning laser ophthalmoscope to image and classify roughly a thousand cones in a patch of retina, tracks the eye's constant fixational tremor at cellular scale, and then fires tiny laser microdoses (hundreds of thousands per second), each aimed at a specific, chosen cone, correcting for the eye's motion so the light lands on the cell it was meant for. This is retinal projection taken to its literal endpoint: the display no longer forms an image the retina samples; it sets the activation of the mosaic directly, cone by cone. And with that control it can do something no ordinary display can. Recall from Chapter 3.2, Perceptual color and trichromatic vision that the cone sensitivities are non-orthogonal: the M curve is buried under L and S, so an ordinary spatially uniform spectrum cannot strongly stimulate M cones without also stimulating the overlapping L population. This is spectral overlap, distinct from optical spill between neighboring cones. Because Oz stimulates selected cones directly, it is not limited by that constraint: it can activate only the M cones, a combination of cone signals that lies outside the natural gamut because no spectrum could ever produce it. Observers see a blue-green of impossible saturation the team named "olo," which can be brought into the ordinary matching gamut by desaturating the Oz stimulus with added white, and the same rig can present arbitrary cone-by-cone imagery and video, not just a single hue (Figure 37.6.2⛶ fullscreen).

The Maxwellian eyewear is real and shipping, but constrained by a small eyebox, limited field of view, and one focus-free plane. Oz is a laboratory instrument, not a product: it works over a tiny patch of retina, requires a room of adaptive optics and a dilated, tracked eye, and demonstrates a principle, programmable control of photoreceptors at population scale, rather than a wearable display. The safety, field-of-view, and eyebox problems that limit ordinary retinal projection are only harder at the per-cone scale. But the trajectory is clear and it is the fitting close to this part: displays began by throwing light at a screen and asking the eye to make sense of it, and they end by addressing the eye's own sensor cell by cell, the most direct answer to the flatness this part set out to defeat, and a display that can, for the first time, evoke a color outside the gamut of ordinary spatially uniform spectral stimulation.