Skip to content
noggin

Eyelid kinematics for the lid servos

Researched 2026-11-01. Background only: the gestures keep toon timing (#258, 2026-11-01-toon-eyes.md).

Builds on docs/research/2026-10-31-natural-eye-motion.md §1, which already covers Kwon, Sforza and Trutoiu.

Sources:

  • Down phase: 92±17 ms [V]. Other figures: 75 ms [E via V], ~100 ms [N].
  • Peak velocity: 565±297 °/s of lid rotation over a 21±6° amplitude [V], so ~27 amplitudes/s. [N]: ~180 mm/s on full blinks, 120 mm/s on partial ones. Peak velocity scales linearly with amplitude [E via W; N].
  • Where the peak falls: slow start, peak as the lid crosses the visual line, slowing before the lower lid [Doane via W]; [K] calls it parabolic. So a bell, peaking at ~40–50% of closing time (inferred; no source gives time-to-peak), not an early peak.
  • Closed hold: 10–50 ms [Ousler via W].
  • Up phase (to 95% open): 242±55 ms [V]. Other figures: 140–225 ms [E via V], ~220 ms [N].
  • Down:up ratio: ~1:2.5 in duration; peak speed 180 vs 60 mm/s, ~3:1 [N]. Opening starts fast, then a long decelerating tail [K].
  • Amplitude: often partial. 70 of 109 blinks were incomplete in [N]; spontaneous blinks stayed under 100% in [P].
  • Voluntary blink [V]:
    • Total 275±37 ms: down 88±13 ms, up 187±34 ms.
    • Amplitude 38±10°, about twice a spontaneous blink.
    • Peak 1402±159 °/s.
    • Less variable than spontaneous blinks.
  • Also: [S] 1.7× more full closures, faster; [K] close 76, hold 58±4, open 438 ms. Baker 2005 (abstract) found them ~18% longer, conflicting with [V]. All agree: faster, fully closed, longer hold.
  • Both eyes: in voluntary blinks one lid consistently starts earlier and moves further, and the asymmetry is stable per person (Kassem & Evinger 2006, https://pubmed.ncbi.nlm.nih.gov/16384962/).
  • Wink: I found no kinematic study. Primary work on winks is neural:
  • Wink, inferred: voluntary-blink speed, full closure, deliberate 200–400 ms hold; other lid narrows 10–30% in step. No source covers gaze or brow; a glance toward the winking side is style.

Symbols: A = closure amplitude, Tc = closing time, Th = hold time, τ = opening time constant.

closure(t) = A·(1 − cos(π t/Tc))/2 t < Tc (bell curve, peak mid-closure)
= A t < Tc + Th
= A·exp(−(t − Tc − Th)/τ) after; snap to 0 below 0.05·A
Tc Th τ (95% open at 3τ) A
spontaneous 80 40 80 [V] random 0.8–1.0 [N]
voluntary 80 60 [K] 60 [V] 1.0
wink (inferred) 120 300 100 1.0; other lid 0.2

Closure at each 20 ms step:

  • Spontaneous: 0:0, 20:.15, 40:.50, 60:.85, 80–120:1, 140:.78, 160:.61, 180:.47, 200:.37, 220:.29, 240:.22, 260:.17, 280:.14, 300:.11, 320:.08, 340:.06, 360:0
  • Voluntary: 0:0, 20:.15, 40:.50, 60:.85, 80–140:1, 160:.72, 180:.51, 200:.37, 220:.26, 240:.19, 260:.14, 280:.10, 300:.07, 320:0
  • Wink: 0:0, 20:.07, 40:.25, 60:.50, 80:.75, 100:.93, 120–420:1, 440:.82, 460:.67, 480:.55, 500:.45, 520:.37, 540:.30, 560:.25, 580:.20, 600:.17, 620:.14, 640:.11, 660:.09, 680:.07, 700:.06, 720:0

Lid position = level + (shut − level)·closure. These fractions are of servo travel. The lid bars make lid travel non-linear in servo angle (build-notes, #213).

4. Servo constraint (428 °/s, 22° rest to shut)

Section titled “4. Servo constraint (428 °/s, 22° rest to shut)”

The servo can match the human closing time, but not the human peak speed.

  • Peak speed: scaled to 22° of servo travel, a spontaneous peak needs 27 × 22 ≈ 580 °/s and a voluntary one 36 × 22 ≈ 800 °/s. Both exceed 428 °/s.
  • Duration: the human close averages 22/0.092 ≈ 240 °/s. The servo covers 22° in 51 ms, ~60–75 ms with a PWM frame and acceleration: inside the human 75–92 ms. Its flat speed replaces the bell, invisible at this length.
  • The 80 ms bell is at the limit: its peak is π/2 × 22/0.08 ≈ 430 °/s, and at 20 ms steps it is only 4 points.

What to do:

  • Close as a step, as blink() already does. If it looks harsh, send 0.5 and then 1.0 one frame later.
  • Interpolate only the opening and slow closes. The wink’s 120 ms bell peaks at ~290 °/s, which is fine.
  • Opening: τ = 60 starts at ≤370 °/s, within rating. The current (1−f)² over 300 ms is close to [V]; an exponential adds [K]’s long tail.
  • Time a real close on the Mech: the rating is no-load; lid friction and the slip clutch slow it.

Unverified: Evinger 1991 and Doane 1980 are cited second-hand, Baker 2005 from its abstract only, and I found no wink kinematics.

byYellow PineYellow Pine

noggin v0.2.26-6-g09bf9b7 · 09bf9b7