Hardware specification

Every dimension, position, orientation and tolerance of the two published default designs, read directly out of the frozen runs behind the coverage viewer. It is written so that someone could set out to build this. Read the last section before you do.

Nothing here has been measured. This is a geometric design plus a first-order error budget, both simulated. No optical channel, mirror, coupling, tray or calibration has been built or tested. Several items a real build needs are not determined at all, and they are listed under What this specification does not contain.

Coordinate frame and units

All lengths are in mm and all angles in degrees. Every coordinate on this page is in the single scene frame shared by both jaws:

Positions are absolute in this frame, not relative to a part. The handle extends toward negative y; the tray sits at positive y.

Assembly overview

Both jaws reuse the same supplied handle exterior: 30.19 × 108.09 × 15.30 mm (width × length × height), closed volume 32383 mm³, bounds [-15.10, -147.87, -5.66] to [15.10, -39.79, 9.64]. The published handles are that exterior hollowed for the components and cut for the optical ports; no envelope volume was added (enlargement 0.00 mm³ upper, 0.00 mm³ lower).

Internal allowances used throughout: 1.00 mm pre-port wall allocation and 0.35 mm clearance between every component body and the cavity around it. Ports cut through the wall afterwards, so the remaining wall is thinner in places — measured values are in the tolerance section below.

Joins and interfaces

Handle-to-tray coupling

The tray seats on the handle through a Maxwell kinematic coupling: three balls into three grooves, six point contacts, exactly constrained with no over-determination. The error budget treats the whole handle-side optical assembly as one rigid body whose seating error is a single six-degree-of-freedom residual shared by both cameras and both light heads — that is what a kinematic coupling buys, and it is why the budget does not give each device an independent pose error.

Nothing about this joint has been measured. The 1 µm figure is the level at which the study ran; the same study also reports 5 µm and 10 µm cases, and the published error figures degrade accordingly.

Mirror seats

Every mirror is a cap sitting in a pocket cut into the tray, backed by tray material. The pocket is cut from the actual mirror solid plus its backing, so the seat follows the cap. Per-mirror seat points, seat normals and the available backing depth are in the mirror tables below.

Component mounting and wiring terminations

Cameras, light heads and the shared driver are represented as rectangular reserves inside the handle cavity, each with 0.35 mm clearance. Their harnesses are modelled as opaque swept cylinders of radius 0.50 mm entering each body by 0.15 mm. Status: provisional-opaque-routing-reserves. Actual brackets, fasteners, connectors, bend radii and strain relief are not designed.

Cameras

Two cameras per jaw, both inside the handle. A third was rejected: no orientation of a 5 × 5 × 8 mm body fits the free volume left in the cavity, and an independent third route would improve the camera-pixel term by only about 4 µm.

f, u and v are an orthonormal right-handed frame: f is the optical axis, u the sensor row direction, v the column direction. fx follows from the pinhole relation fx = (width/2) / tan(hfov/2), and the vertical field from 2·atan((height/2)/fx).

The layouts were searched with the 13 MP surrogate. The accuracy figures published alongside them assume the ×10 camera. Both are stated because they are different claims.

Structured light modules

Shared electronics

The driver strip is a volume reserve that blocks rays. It is not a designed board.

Mirrors

Upper · 30 mirrors

All 30 are R24 mm spherical caps, 6 × 6 mm. The normal points out of the reflecting face. Tilt is the angle between that normal and the tray +z axis; azimuth is measured in the xy plane from +x toward +y. u and v span the aperture and complete a right-handed frame with n. "Fed by" names the pupil whose route the mirror was placed to serve, which is a camera (C) on some mirrors and a light head (L) on others. The seat point and normal are the measured tray surface the backing sits against.

Download this table as CSV · complete layout JSON · tray STL · handle STL

Lower · 24 mirrors

All 24 are R24 mm spherical caps, 6 × 6 mm. The normal points out of the reflecting face. Tilt is the angle between that normal and the tray +z axis; azimuth is measured in the xy plane from +x toward +y. u and v span the aperture and complete a right-handed frame with n. "Fed by" names the pupil whose route the mirror was placed to serve, which is a camera (C) on some mirrors and a light head (L) on others. The seat point and normal are the measured tray surface the backing sits against.

Download this table as CSV · complete layout JSON · tray STL · handle STL

Optical ports and corridors

Each route from a pupil to a mirror needs material removed from the handle and tray along its path. These are real cuts: the remaining material still blocks rays everywhere else.

Actual convex hull cuts between finite source aperture and sampled curved mirror aperture; all remaining material stays opaque.

Cylindrical bores

The straight openings each pupil looks through, cut as capsules between the two points.

Per-mirror finite-aperture corridors

One corridor per mirror: the convex hull swept between the finite source aperture and the sampled mirror aperture, so the cut is exactly the volume that route needs and nothing more. Every triangle left after the cut still blocks rays. The hull vertices are in the layout and housing-spec downloads; the extent of each hull is given here.

Harness routing reserves

Capture sequence

This is a multi-shot design. The tray is re-seated rigidly between exposures and the shots are stitched; the published figures use 12 shots for the upper and 8 for the lower. Each pose is a rotation about the centroid of the required surface followed by the smallest translation that clears the anatomy, checked point by point against the rebuilt tray. The tables below are the full-resolution verification of each sequence, not the coarser search that planned it.

Upper

Full-resolution verification over the 7,988 required samples, area-weighted by region so the percentages are of required surface area rather than sample count, with unreachable points counted as not covered. Geometric means a camera route and a light route meeting at 10° or more. Resolved additionally requires the camera leg at 0.150 mm per pixel or finer, and it is the figure that distinguishes this design from the one it replaces. The coarser search that planned these sequences reported slightly higher coverage; these are the verified numbers.

After all 12 shots, 0.503 % of the required surface is still unreached: 16 samples in region 0, 16 samples in region 2, 8 samples in region 4. The pose envelope ceiling with unlimited shots is 99.70 %.

Lower

Full-resolution verification over the 6,000 required samples, area-weighted by region so the percentages are of required surface area rather than sample count, with unreachable points counted as not covered. Geometric means a camera route and a light route meeting at 10° or more. Resolved additionally requires the camera leg at 0.150 mm per pixel or finer, and it is the figure that distinguishes this design from the one it replaces. The coarser search that planned these sequences reported slightly higher coverage; these are the verified numbers.

After all 8 shots, 3.451 % of the required surface is still unreached: 19 samples in region 0, 76 samples in region 1, 113 samples in region 2. The pose envelope ceiling with unlimited shots is 98.29 %.

The pose envelope models the tray against the scanned anatomy only. Lips, cheeks, the floor of the mouth, the opposing arch, the maximum jaw opening and the insertion path are not modelled, and every one of them can only shrink this envelope. Stitching also assumes the motion between shots is known to the precision of the rest of the budget; recovering it from surface features instead would add a registration error that is not in any figure here.

Tolerances the accuracy depends on

These are the values at which the published error figures were computed. They are requirements on the build, not achievements of it. The right-hand column is the level the earlier study assumed; the published design needs the calibration column ×10 tighter.

Knowing 24 to 30 mirror positions to 1 µm on an assembled tray is itself an unsolved metrology problem, and 0.045 px centroiding is 3.3× better than the study assumed. Both are named as requirements precisely because they are hard.

Measured post-port wall thickness

Sampled normal-ray thicknesses on the actual post-port material, 12,000 area-uniform samples per jaw. These are ray lengths along surface normals, not a certified global minimum wall, and intentional openings have no wall at all across their aperture. The two jaws share a non-port minimum because they share the same supplied handle exterior, and the thinnest feature away from any port is the same feature in both. Port and cavity edges need mechanical refinement before fabrication.

Materials and thermal design

Unidirectional CF/PEEK is what removes the need to pre-heat the tray. The same laminate is about a hundred times worse transverse to the fibres, so the fibre direction is a functional requirement, not a preference.

Predicted performance of this specification

Merged over the shot sequence: each point keeps the best pose that reaches it, and a point no pose reaches counts as outside every threshold. Percentages are of the whole required surface. First-order propagated budget under the tolerances above — predicted, not measured.

Held-out single-pose results, the figures the coverage viewer draws.

What this specification does not contain

Read this list as the distance between this page and a buildable package. Each item is genuinely undetermined, not merely unstated.

• No measurement of anything. No optical channel, mirror, coating, coupling, tray or calibration has been built or tested. Every performance figure is simulation under assumed priors.
• No focus solution. R24 depth of field is about ±2.2 mm against camera-to-mirror legs spanning 32–101 mm. The design needs designed focal zones or focus bracketing across the shots, and no workable assignment of focus to pose has been demonstrated. This is the largest open item.
• No stitching or registration error. Merging the shots assumes the motion between re-seatings is known exactly to the precision of the rest of the budget.
• No transient thermal term. The thermal figure is an equilibrium bound; a partly warmed tray distorts non-uniformly and a single scalar correction cannot remove that.
• No mirror retention. Seats are geometric proximity of connected tray material. No adhesive, clamp, fastener or pull-off test is specified; every cap is backed by tray material without added bridges.
• No electronics design. PCB, processor, memory, power supply, battery, thermal path, connectors, cable packing, bend radii and strain relief are all outside this study.
• No fasteners or assembly sequence. How the handle closes, how the tray is retained on the coupling, how anything is serviced — none of it is designed.
• No sealing, sterilisation or biocompatibility. Nothing here addresses cleaning, autoclave survival, materials in contact with tissue, or ingress.
• No structural analysis. Wall thicknesses are sampled, not certified; stiffness, strength, drop and bite loads are unevaluated.
• No decoding or radiometry. The dot pattern, reflected contrast at short range, correspondence, finite entrance pupils, blur, noise and 940 nm camera/filter compatibility are all unvalidated.
• No clinical claim. Geometric access is not reconstruction accuracy and not diagnostic performance. Population fit and unscanned anatomy remain unestablished; results apply to the supplied scans in selected seated poses.

Sources

Every figure above is read from these published records:

Upper layout · Lower layout · Optical study record · Complete study data · compact physical validation · compact result validation · compact cone validation · compact delivery validation