// Cornell · Virtual Experience of Designed Environments · VR Research
Corner by Design
Does the angle of a corner wall change which artworks people look at? A VR gallery that measures attention through the eyes, automatically.
Overview
Corners are among the most overlooked places in any exhibition. Visitors drift toward walls they face head-on, and work hung in a corner quietly gets less attention. I asked whether that neglect is a fixed property of the room, or something we can design away.
As a solo project, I modeled a gallery with three corner-wall angles, built it as a VR experiment for Meta Quest Pro, and wrote the tools that detect where each visitor looks, log it automatically, and turn it into analysis. The same approach applies to any designed space where attention matters: galleries, lobbies, offices.
What I built
3 corner-wall conditions12 tracked artworksEye-tracking logger in UnityAutomated Python analysis
Question
01
Change one angle, keep everything else
Research shows that architectural corners shift where people direct spatial attention, but that idea had not been tested inside a real exhibition layout. I designed a within-subjects study: the same room, the same artworks, the same lighting. Only the display wall at the far corner rotates.
A
90°
standard flush corner, the baseline
B
45°
an oblique cut that turns the corner toward the room
C
30°
a shallower oblique angle
Each participant explored all three conditions in counterbalanced order. I measured dwell time on the far-corner artworks, and the spread of dwell time across all twelve works as a measure of how evenly attention was shared.
Build
02
From a model to an instrument
I modeled each condition in SketchUp, exported it to Unity, and turned the scene into something that could measure. Every artwork received its own collider, so a line of sight landing on a painting can be identified by name.
SketchUp model3 conditions
→
FBX exportgeometry + materials
→
Unity sceneMeta XR SDK
→
Collidersone per artwork
→
Quest Pro buildeye tracking on
Measure
03
Letting the eyes write the data
Instead of asking people what they looked at, I wrote a small system that records it. A participant enters their ID and condition on a VR start screen, walks the gallery freely, and presses one button to finish. Everything in between is captured automatically.
Gaze → artwork
Every frame, the logger averages both eyes from Quest Pro into one gaze ray and casts it into the scene. If it lands on a collider named Painting_, that artwork is being looked at. Low-confidence frames are discarded.
confidence = (left + right) / 2 // skip if < 0.5
ray = eye position → eye direction
hit.name starts with "Painting_" → log it
ray = eye position → eye direction
hit.name starts with "Painting_" → log it
Glance → fixation
A small state machine separates real looking from passing glances: gaze has to stay on the same artwork for at least 0.2 seconds to count as a fixation. It also records the first artwork each visitor notices, and when.
new target→held ≥ 0.2 s→fixation→target changes · saved
raw.csv
Every frameArtwork, gaze point on the canvas, visitor position, and confidence.
fixations.csv
Every fixationWhich artwork, when it started and ended, how long it lasted, and where the visitor stood.
summary.csv
Every sessionFirst artwork noticed, time to first look, total visit time, and fixation count.
Testing sessions in the studio. Each participant explored all three rooms on Meta Quest Pro, seated or standing, while the monitor mirrored their view.
From CSV to findings, in one command
A Python script collects every session, totals dwell time per artwork, tags the corner works, runs the statistics (Friedman tests across conditions), and generates the charts, including a top-down plan of every room drawn from the raw gaze points. It also merges the questionnaire, so what people said they noticed can be compared with what their eyes actually did.
Findings
04
What nine visitors’ eyes showed
Across nine participants, the corner artworks held attention longest when the corner was cut at 45°: an average of 3.53 s per visitor, compared with 2.56 s in the standard 90° corner and 2.78 s at 30°.
n = 9, within-subjects. The differences are not statistically significant (Friedman χ² = 0.27, p = .88). Medians follow the same order (1.20 s at 90°, 1.81 s at 45°, 1.59 s at 30°), but individual variation is large: one visitor looked long at the corners in every room, while three or four visitors per condition never fixated on a corner work at all. Treat this as a direction to test with a larger sample, not a result.
False equity of neglect
The 90° room had the lowest spread of dwell time across all artworks (SD 2.36 s, against 2.91 s at 45° and 3.62 s at 30°), which looks like attention was shared evenly. But an even spread can also mean nothing stood out, and the corner works received less time there than in the 45° room. A space can look statistically fair while quietly overlooking part of the display. (This difference is not significant either.)
Below conscious awareness
In the questionnaire, answered by the first five participants, nobody chose the 90° room as the one that gave the works the fairest exposure: two picked 45°, two picked 30°, and one saw no difference. The angle seems to register as a feeling more than as something people can point to.
The open floor wins
The artwork on the freestanding panel drew the most attention in every room, 7 to 11 s per visitor. Whatever stands in the open competes with everything on the perimeter, which matters as much for a lobby or an office as for a gallery.
Takeaway
A repeatable way to measure attention in space.
The same pipeline (model the space, log gaze automatically, analyze in one command) can test a layout before it is built: which display gets noticed, which desk is out of sight. A follow-up study would use a larger group and rotate the corner artworks between positions, so the angle is the only variable.