true head direction / real posts
what the delayed picture shows

VR Latency Phase Plane (2D)

Top-down map of head yaw: why a late picture makes a headset uncomfortable. Twin of the 3D lab.
Motion-to-photon (mean)-
Peak world-swim error-
Peak head speed-
Vestibulo-visual conflict (RMS)-
Smear blur-
Discomfort index 0-1 (illustrative)-
Phase-plane loop area-
Time to finish task, session 20-
Head angle (true, blue) vs angle the headset shows (orange), last 2 s, degrees
Phase plane: true angle (x) vs shown angle (y), deg. A fatter loop means more lag; a straight diagonal means none.
Training time per session (s) with this headset (amber) vs ideal latency (green dashed)
The science
Honest scope: a model of three well-established VR engineering effects. Comfort thresholds and the learning constants are illustrative / typical values, not measurements of any product or trial.

Motion-to-photon latency is the time from a head movement to the photons that show it. The headset samples the pose once per frame and holds it, so the picture is on average older than the pipeline latency alone:

tau_eff = tau_pipeline + T_frame / 2 , T_frame = 1 / refresh

World-swim error. For a head turn theta(t) = A sin(2 pi f t) the image shows theta(t - tau). The mismatch is a sinusoid of amplitude

e_peak = 2 A |sin(pi f tau_eff)| ~ A (2 pi f) tau_eff (small tau)

Phase plane (new in this 2D view). Plot the shown angle against the true angle. A pure delay draws an ellipse of area pi A^2 |sin(2 pi f tau)|; zero lag collapses it to the diagonal. The orange rings on the plan are the vertical gap between loop and diagonal, turned into positions of the posts.

Vestibular conflict. The inner ear senses the true head velocity w(t), the eyes see w(t - tau). The RMS rate difference is

c_rms = sqrt(2) * w_peak * |sin(pi f tau_eff)| , w_peak = 2 pi f A

Sensory conflict is the standard explanation for simulator sickness. The discomfort index is D = 1 - exp(-(c_rms/c0)^2) with an assumed c0 = 50 deg/s: a teaching scale, not a medical measure. A frequently quoted rule of thumb is to stay under about 20 ms total latency (typical value).

Smear. A pixel lit for t_p while the head turns at w smears over w * t_p degrees; at an assumed 20 pixels/degree, blur_px = 20 w t_p.

Learning curve. Power law of practice: T_n = T_inf + (T_1 - T_inf) n^(-a), with T_1 = 300 s, T_inf = 90 s (made up), a_0 = 0.6, a_eff = a_0 (1 - D).

Simplifications: one rotation axis (yaw), no translation or eye movement, pure sinusoidal head motion, no motion prediction (real headsets predict ahead), no adaptation by the trainee.

Independent numerical checks
Made with 🟠 Blender