Best guesses for the values that waited for a calliper (2026-11-01, #170)
Cansin, 2026-11-01: “assume with best guess, I cannot measure”. Every value below is settled by a documented guess and
marked inferred where it is used; none waits for a measurement. Evidence order: the real build’s behaviour and photos
(the Mech runs every gesture and its calibrated servo ends, eyes/calibration.json, without binding), then the STEP and
STL nominals. Method: make sim‘s pose measure (model/engine/run.py measure) at each calibrated end, one variable
changed at a time, and the penetrating vertices’ extents to see which way each collision runs.
The lid crank at the closed ends (L -41, R -40)
Section titled “The lid crank at the closed ends (L -41, R -40)”- Symptom: the crank’s plain end 1.0 mm into the eye bracket’s outer edge from horn -32. Its inside vertices span x -27.94..-27.0 only: a plane overlap along the screw, not a swing into the edge.
- Cause: the model stacks the crank flat on the straight bar (x -27..-30); the bracket edge is x -27..-28. The bar and crank turn on the M2 screw under a nut that is not clamped (the joint must turn), so nothing holds the crank on the bar: where the edge meets it, it is pushed out along the screw.
- Guess: the crank rides 1.2 mm out (geometry.CRANK_PLAY_MM): 0.2 mm clear of the edge (1.1 is the least that clears), its stepped end 1.2 mm off the upper lid’s ear on the lid pin. The closed ends run clean; the free travel on that side now ends where the straight bar meets the bracket face (y -2.5) from -50, so the lid limit is servo 41..129 (horn -49..+39, geometry.LIMITS).
The eye horns’ tooth (L eye -32)
Section titled “The eye horns’ tooth (L eye -32)”- Symptom: the L eye horn 0.18 / 0.44 mm into the servo bracket and eye bracket at horn -32 (clean at -30, not at -31).
- The program centres each eye at the mean of its calibrated ends, servo 88 (eyes/eyes.py
_eye_angle), and the rest-front photo (docs/references/build-photos/eyes-rest-front.jpg) has both pupils straight ahead there; the rest-top reading less its camera tilt gave +3.2 each (docs/research/2026-10-31-shafts.md). The model’s neutral (eyes straight, the STL pose) is servo 88: geometry.HORN_AT_90 takes +2 for both eyes. The calibrated ends become -30 / +30, both clean.
The lid horns’ open end (L +52, R +45): accepted, listed
Section titled “The lid horns’ open end (L +52, R +45): accepted, listed”Superseded in part by #189 (below): the teeth are L +13.95, R +6.75, the ends L +52.95, R +45.75, the free travel 62 deg.
- Symptom: the lower lid’s back rim 1.3 / 2.0 mm into the servo bracket’s front face (z 18) from horn +39, across the lid’s width; the L upper lid 0.35 mm into the eyeball bar at +52.
- Rejected: a smaller lid tooth (the open ends sit 39 servo deg from 90 on both sides, which tooth 0 would put exactly
on the model’s contact). The calibrated close needs L <= 14.0, R <= 6.8 to shut the lids, and the rest photo
(model/test_traces.py
test_rest_matches_the_rest_photo: L -15.2, R -16.3) holds the teeth at L +13, R +6; tooth 0 would leave the rest lids nearly shut, which the photo does not show. - Rejected: servo gain. The 9 g servo turns 180 +-10 deg over 500..2500 us (docs/research/2026-10-31-electrical.md 1); even 170 deg brings the ends in by 3 deg, short of 13 / 6.
- Rejected: print tolerance. 0.2 mm at the rim is 1.3 deg of lid.
- Guess: the build runs these ends, so the model’s lid travel (88 deg free) is short of the build’s (93 / 85 servo deg) by a shape error in the lower lid’s back rim or the servo bracket’s front face that no tolerance in the sources reaches. Accepted and listed with that reason (model/test_report.py KNOWN_CALIBRATED), not fixed by inventing a part shape.
The jaw bar on the horn arm
Section titled “The jaw bar on the horn arm”- Was: 0.28 mm into the horn arm (the “0.58 across” overstated it), with 0.25 mm left between the bar and the jaw lug.
- Guess: flush seating. With both screws drawn home the bar fills the space between the lug’s outer face and the arm’s underside: moved 0.28 mm inboard (geometry.PT_BAR_PLANES jaw x 84.64 / 71.64) it touches both with nothing inside either. mating.INFERRED is empty.
STEP nominals taken as they are
Section titled “STEP nominals taken as they are”- Horn seat on the spline: 11.52 past the tab plane (geometry.SERVO_STEP horn_seat). Three independent fits already agree with it (docs/research/2026-10-31-shafts.md, Seat).
- Eye horn reach: the bar’s horn hole is 12.56 from the shaft against the STEP horn’s 12.75 hole; the pin stays 12.75 and the 0.19 mm is taken up by the bar self-tapping its r 1.05 hole (the mating check passes it, 0.2 allowed).
Other values
Section titled “Other values”- Servo lead length: 250 mm, the length 9 g micro servo listings publish as typical; Bambu publishes none (the STEP cuts the wires). model/electrical.py SERVO_LEAD_MM, the corridors’ limit (230 with the 20 mm slack).
- Receiver PWM header numbering: docs/references/build-photos/eyes-top.jpg shows four leads in the shield’s 2 x 2 header block, its silkscreen too small to read. Kept the STEP port order (geometry.RECEIVER_STEP). Any other numbering moves a lead’s end by at most 9.4 mm (the block’s diagonal); the tightest eye lead (r_lid, 218.5 of 230) has 11.5 to spare, so no choice changes a result.
- Bracket stem depth in the plate (stage 1): the build photos (eyes-lid-linkage-side.jpg, eyes-rest-front.jpg) have the bracket’s feet flat on the plate top and the stem out of sight in the plate. Superseded by #172: the stem is in the plate’s slot, not the pan recess, and the desk bracket’s stem fills the 5 mm plate, so geometry.STEM_AT stage 1 is z -2 (docs/research/2026-11-01-eyes-on-plate.md).
The joint-end overlaps (#189)
Section titled “The joint-end overlaps (#189)”The old engine skipped a lid or eyeball on its bracket pivot (model/engine/run.py PIVOT_PARTNER, now measured) and never measures the two lids (both move with the horn); #149’s MuJoCo report found them overlapping on the exact meshes at the LIMITS ends, which the build runs without binding. Measured per pair, one variable at a time, with the designer’s meshes swapped in as the cross-check:
- Lower lid into the eye bracket (1.86 mm at +39): our bracket dropped the designer’s relief on the band’s lower front edge (a 45 deg chamfer from y -7.5 on the front face to z 15.5 at y -10, both designer brackets; the designer’s lid is clean against it). Restored (model/printed/eye_bracket.py LID_RELIEF): clean to +38.
- The two lids at the closed end (1.39 mm at -49): a wrong limit. The lips meet at -30.72 and the shells cannot pass each other, so the closed end is the lips; Cansin (#189): the lids close quietly at lid_closed, edges just touching, no overlap. The lid teeth go to L 13.95, R 6.75 (inferred: the close lands at -30.75, 0.03 mm of lip), the top of the #134 window; the rest photo’s face-value lid reading sits 1.2 / 0.9 below, the side its camera tilt moves it.
- Superseded by #210 (below: the ears cross from +16.5, the open end is +16). The two lids (1.03 mm) and the upper lid on the eye bracket’s arm chamfer (0.78 mm) at the open end (+39): the lids’ ears meet at the back from +32, the upper ear the arm from +35, with the designer’s lids and bracket alike. No shape in the sources clears them, so the open end is the model’s free travel, +31 (geometry.LIMITS); the build’s open command (L +48, R +41.5) and calibrated ends stay past it, accepted with the servo bracket hit (KNOWN_CALIBRATED).
- Each eyeball into the eye bracket (0.77 mm at +-22): its back rim (z 5) meets the plate’s front face (z 13) from +-21, the eyeball turning 1.26x the horn (the bar’s 12.56 mm horn arm on its 10 mm one); the designer’s eyeball and bracket alike. The eye limits go to +-20 (free travel). Conflict, recorded: the program commands up to +-25 servo degrees and the calibration +-30, which would put the eyeball 1.5..2.5 mm into the plate in the model; the build runs them, so the four-bar’s gain or the eyeball’s back is off by more than any source’s tolerance (inferred, unsettled).
The lid ears past +16.5 (#210)
Section titled “The lid ears past +16.5 (#210)”meshes.judge tests vertices, so it missed the lids’ ears crossing edge through edge; MuJoCo’s tuned pieces (#190) found them. Measured on the exact CadQuery solids (common volume of the two turned lids, model/printed/test_lid.py), one variable at a time:
- The ears’ back corners (the 45 deg brow chamfer behind the pivot, r 4.5..6 from it) meet from horn +16.5 (lids +17.3 / -16.6, 34 deg apart) and cross 1.4 mm by +29; the upper lid reaches the eye bracket’s arm from +28. The lips meet at -30.72. Sections through the designer’s lids at the ear layers show the same corners crossing at the same angles, so it is not our redesign.
- Rejected: the pivot (both lids turn on the same pins, so only their relative angle matters); the tooth or the shared frame (#213: they move both ends together, the span stays); servo gain (180 +-10 deg, above); print tolerance (0.2 mm is 1 deg of relative turn at the corners).
- So the model’s free travel is -30.75..+16 (geometry.LIMITS; the close at the program’s lid_closed, the lips 0.03 mm in). Conflict, recorded: the build’s span from close to open command (72 servo deg, 153 deg between the lids in the model) is far past the parts’ 103 deg, and the model’s lids look fully open (model/lids.py) only at +27.2, past the ears. Inferred candidates, unsettled: the bars on a horn hole nearer the hub (8.75 mm gives 0.69x, which fits both ends; the build notes say the second-from-last hole) or the servo’s clutch (PG001, clutch protection) slipping at the stops (it would explain the drifting lid teeth, #213).
noggin v0.2.26-6-g09bf9b7 · 09bf9b7