v1 Scout — the Scout case on the v0 Dingo mechanics

Current prototype: R20 (2026-10-01) — NOT PRINT-READY

Colourway "dark mode": a dark-grey shell, black armour (fenders, shields, handle, visor), signal-orange ear blades, badges, eyes and Dingo upper legs. Three PETG spools. No geometry change. See R20.

R19 (2026-10-01, colours superseded by R20)

Colourway "arctic / graphite / signal": a white shell, graphite armour (fenders, shields, handle, visor frame), signal-orange ear blades, badges and Dingo upper legs, ice eyes. No geometry change. See R19.

R18 (2026-10-01, colours superseded by R19)

Laid-back ears: the standing fins are gone. Each ear is now a graphite swept blade set 1.2 mm into a pocket in the head's flat crown, at most 3.5 mm proud, one print per side (the dark ear inlay is gone). See R18.

R17 (2026-10-01, ears superseded by R18)

Visor-wedge head and graphite shoulder shields: the Scout is the Dingo plus Schutzbleche. The shields continue the R16 fenders over the front of each front hip and stand in front of the knee servo horn, 42 mm link and cam; they bolt to the head (4 x M3x8 + 4 heat-set inserts, user-approved BOM change). The case budget is 590 g (user decision); the robot stays under 3.4 kg. See R17; R13-R16 are in docs/design/v1/.

R12 (2026-09-07, superseded)

Viewer repair: Scout now keeps full-detail, unquantized surface geometry in Sim, Assembly and the guide. The torn-looking head and handle were display corruption. All 17 STEP exports pass single-solid validity and BREP self-intersection checks; this does not resolve the manufacturing blockers below. Refresh localhost:8765.

The selected A1-F/H2 direction on the corrected detailed Dingo reference: larger eyes, compact rounded ear seats, reinforced/blunted lower head edges and a tucked rear skirt. The new detailed fit audit caught and eliminated 88 sampled rear-hip interferences that the simplified collision model missed. Original mechanics and controller stay unchanged.

!R12 detailed CAD in the simulated pose

Full workflow, evidence and blockers. make scout-workflow runs every diagnostic tier, including print orientations and G4 preflight, without changing goldens or pressing print. It currently exits nonzero on purpose: the handle adapters do not use the existing hole pattern, six electronics intersections cannot be resolved from invalid reference meshes, five parts need support validation, and hardware/module dimensions and goldens remain unverified. No print-ready release.

R11 implementation history (superseded)

Built after four visual choices: A → A1 → A1-F → H2. Softer head and sage face surround, shorter swept-back ears, a flowing Bezier handle and fuller lower body panels. The original Dingo mechanics, mounting datums and camera/audio carriers remain unchanged.

!Actual R11 CAD in MuJoCo

Selected concept, R11 changes and verification. Scout non-golden geometry checks, 15 assembly steps, walking and part settling pass; 130 tests pass. Case: 558.618 g / 560 g; robot: 3.3477 kg / 3.4 kg. Goldens need approval; actual hardware fit and handle strength are unverified. Not print-approved. make serve opens the rebuilt guide and walking viewer.

R10 implementation history (superseded)

Back to the supplied Scout R4 reference: a fuller soft-square head/face, taller alert ears, rounded front-hip arches, a fuller belly, dotted speaker grilles and compact handle roots. Original Dingo mechanics, camera/audio carriers, TPU paws and branding stay unchanged.

!Actual R10 CAD in MuJoCo

R10 changes, all views and verification. Scout non-golden geometry checks, assembly, walking, part settling and 125 tests pass. Case mass: 558.935 g / 560 g; robot mass: 3.3480 kg / 3.4 kg. Golden approval, physical hardware fit and handle proof loading remain open. Not print-approved. Run make serve for the rebuilt guide and walking viewer.

R9 implementation history (rejected oval direction)

A genuinely oval head outline and elliptical eyes-only face replace R8's corner-rounded housing. The raised head, swept body, handle, TPU paws and original Dingo mechanics are retained. Camera and chassis mounts stay put; the same speaker carriers sit 3 mm farther inward to keep their insert bores enclosed by the curved cheeks.

!Actual R9 CAD in MuJoCo

R9 changes and verification. Scout non-golden geometry checks, assembly, walking and 121 tests pass. Case mass: 559.713 g / 560 g; robot mass: 3.3488 kg / 3.4 kg. Golden approval, the existing loose-ear settling failure, physical hardware fit and handle proof loading remain open. Not print-approved. Run make serve for the rebuilt guide and walking viewer.

R8 implementation history (superseded)

A visibly raised rounded head, taller eyes-only face, swept shoulders and a tucked hind belly, following the R4 picture more closely. Main back z60, head crown z94; the shoulder rises only beyond the unchanged handle foot. Dingo mechanics, hardware attachment positions, camera/audio carriers, handle and TPU paws remain unchanged.

!Actual R8 CAD in MuJoCo

R8 changes and verification. Non-golden Scout geometry checks, assembly, walking and 117 tests pass. Case mass: 559.102 g / 560 g; robot mass: 3.3482 kg / 3.4 kg. Golden approval and one loose-ear settling check remain open, as do physical fit and handle proof loading. Not print-approved. Run make serve for the rebuilt guide and walking viewer.

R7 implementation history (superseded)

Rounder and more playful: a pebble-shaped visor, fuller cream eyes, swept round-tipped ears, elliptical side panels, a rounded camera rim and warmer sage/ivory colours. The original upper-leg prints are shown in sage; mechanics and hardware stay unchanged. R6's raised handle with local finger scoop and the TPU paws are retained.

!Actual R7 CAD in MuJoCo

R7 changes and verification. Scout geometry checks (except unapproved goldens), assembly and walking checks pass; 107 tests pass. Case mass: 549.238 g / 560 g; robot mass: 3.3384 kg / 3.4 kg. Run make serve for the rebuilt guide. Physical fit, module selection and handle strength remain unverified; two isolated cosmetic-part drop checks also remain open. This is not a print-approved release.

R6 implementation history (superseded)

The current model has a larger eyes-only face and ears, deeper central side skirts, a lower integrated handle and a slimmer camera flange. The detailed browser viewer's mesh-frame and pose-array bugs are fixed. Mechanics are unchanged.

!Actual R6 CAD in MuJoCo

R6 changes, evidence and remaining approvals. Assembly and both robots' simulation checks pass; 102 unit tests pass. Case mass is 543.409 g / 560 g, robot mass 3.3325 kg / 3.4 kg. Changed geometry goldens are deliberately not approved, and physical mounting/module dimensions and carry strength remain unverified. Run make serve to rebuild and open the repaired guide. No print-ready release is claimed.

R5 implementation history (superseded dimensions and results below)

What this is. The Scout R4 board (docs/design/v1/scout-r4.png) as sixteen parametric MOSS parts under parts/scout/ (22 printed instances), mounted on the unchanged v0 Dingo mechanics through this manifest (robot.yaml → make v1 → v1_scout.xml). Revision R5 (docs/design/v1/R5-IMPLEMENTATION.md, 2026-09-07) reshaped the whole case to the board: a tall rounded pill body (roof z 60, R28 crown), a head pod that envelops the front cap and is mostly screen (ivory bezel, black face blank, two cream eyes), tall leaning cat ears with orange inserts, an arched handle that grows out of the roof, charcoal mechanics, TPU paws. Case-only: legs, servo model, joint limits, stance, gait and controller are inherited from robots/v0_dingo/ and pinned by tests/test_v1_baseline.py.

What it is not. A printable release. Every chassis datum is derived from the decimated v0 collision mesh (lib/parts_db.DINGO_CHASSIS, unverified) and every electronics module is a placeholder (*_TBD), so most parts are print-blocked at G4 by design. make gate-assembly writes build/assembly/ASSEMBLY.md, whose first section lists exactly what to measure on the real Dingo before printing anything.

Provenance	mechanics: `robots/v0_dingo/robot.yaml` (Dingo, MIT); case: `parts/scout/.py` from `params.py` SCOUT_ + `parts/scout/_layout.py`; appearance target `scout-r4.png`
Inherited verbatim	`legs`, `servo`, `sim`, `stance`, `gait`, `conventions`; 12 leg-body masses, joint ranges, actuator gains (baseline test)
Added	`case:` 16 parts / 22 instances folded into the base inertial by parallel axis from OCCT; `mechanics_rgba` (charcoal render of the inherited meshes, visual only); `paws:` radius 0.013 m; `payload:` 110 g; `budgets.mass_hard_limit_kg` 3.4
Assembly	`assembly.yaml` (15 steps, bench sub-assemblies) → `make gate-assembly` → `build/assembly/{assembly.json, ASSEMBLY.md}`; the web viewer's Assembly mode reads the JSON
Files	`robot.yaml` (edit this) → `make cad && make v1` → `v1_scout.xml` (generated, committed)
Look	`make serve` → http://localhost:8765/ (Sim + Assembly modes, guide.html); `make record ROBOT=v1_scout` → `build/renders/v1_scout_*`; `tools/snap.py --all` → `build/snap/`

Gate ladder (`build/gates.json`, 2026-09-07, R5)

make cad G0 17/17 · make gate-geom G1 every scout part PASS (goldens re-created for the R5 shapes: that is the sanctioned golden change of this revision; the unrelated parts/mascot/ part fails watertight/genus and is not on the robot) · make gate-assembly G1.5 PASS (0 interferences, 15/15 steps mate + slide straight in, 16 internal fasteners OK, 10 chassis fasteners UNVERIFIED) · make gate-drop 17/17 · make v0 byte-identical · make sim (v0_dingo) G2+G3 PASS · make sim ROBOT=v1_scout G2 kin + G3 stand / walk2m / stepdown / sweep / friction / payload PASS · pytest --ignore=tests/test_incoming.py 66 passed (test_incoming.py belongs to the incoming-corpus work of 2026-09-07 and fails on its own fixtures).

Gate number	v1_scout R5
case mass (printed g, 22 instances incl. 4 paws)	523.6 / 560 budget
total mass (kg)	3.312 / 3.4 hard limit
CoM shift vs v0 (mm)	(+9.9, 0, +11.6)
belly clearance (mm)	136.6 (unchanged: lowest geom is still the chassis belly)

Mass — per part (printed grams, `build/cad/<part>.json`, R5)

Part	Material / colour	Each g	×	Total g	Budget g each	Print orientation
scout_saddle_front	PETG sage	61.1	1	61.1	105	upside down, roof on the bed; support the crown's first 8 mm, tabs, tongue lip
scout_saddle_rear	PETG sage	57.4	1	57.4	105	as above; badge pockets face sideways
scout_head	PETG sage	118.0	1	118.0	150	face down, open rear up; support ring under the R18/R11 face fillets
scout_rear_cover	PETG sage	40.3	1	40.3	60	upside down; support crown and tail rounds' first 8 mm
scout_handle	PETG sage	145.2	1	145.2	150	on its side (x-z on the bed), grip flank on plate supports
scout_handle_adapter	PETG charcoal	5.3	2	10.7	8	flat
scout_bezel_blank	PETG ivory	6.3	1	6.3	18	ivory face down, pegs up
scout_face_blank	PETG black	12.8	1	12.8	14	black face down
scout_eye	PETG ivory	0.4	2	0.8	1	flat
scout_camera_adapter	PETG charcoal	21.2	1	21.2	22	flange down, sleeve up; rear wall bridges 25 mm
scout_speaker_carrier	PETG charcoal	2.5	2	5.0	5	plate flat, ring up
scout_mic_carrier	PETG charcoal	0.6	2	1.2	2	plate flat, tube up
scout_ear	PETG sage	4.8	2	9.5	12	flat on its back, pocket up
scout_ear_insert	PETG orange	0.8	2	1.5	2	flat, orange face up
scout_badge	PETG orange	2.0	2	3.9	3	back on the bed, letters up
scout_paw	TPU charcoal	7.2	4	28.6	8	sole down, socket up
case total				523.6	560	`_scout_case` gate

The 560 g case budget (params.SCOUT_CASE_BUDGET_G, was 480 in R4) buys the R4 silhouette with air: the roof rose from 45.6 to 60 with an R28 crown, the head became a pod round the cap instead of a plate on it, and the ears grew from 25 to 44 mm. v0 (2.818 kg) + 0.56 kg stays under the 3.4 kg hard limit. The MOSS 70 g shell allowance reconciliation from R4 still applies unchanged.

Assembly (`make gate-assembly`, lib/assembly.py)

Every screw is declared once, on the part that carries its head (META.fasteners), and lands in the assembly.yaml step that places that part. The gate places every instance the way the emitter does (mirrors, poses), then checks: no part-to-part interference (manifold3d, ≤ 1 mm³); every part touches something placed before it (≤ 0.6 mm — a part whose nearest neighbour is the unverified chassis hull is reported chassis_unverified, because the G1 clearance gate holds it 2 mm off that hull); every step slides straight in along its insert vector; every fastener has a clear hole in the head part, a bore with material to bite in the into part, and a d8 × 60 mm driver path; chassis fasteners are UNVERIFIED, never OK. Head internals, the bezel stack and the ears are bench: sub-assemblies checked against their frame only, then move with it.

Design changes the gate forced in R5: the bezel is held by four glued d3.6 pegs on a diamond (BEZEL_TO_HEAD 88 × 51) — corner screws do not fit a 96 × 56 frame round an 80 × 44 window; the speaker/mic carriers are built in the head's tapered cheek frame and inserted from inside along the cheek normal; the camera cassette's flange sits on the face with its screws from the front; the speaker carrier's plate is as wide as its ring so the drop gate settles.

Sim: v0 Dingo vs v1 Scout (same gait, same gains, same servo model)

R4 numbers below; R5 re-ran green with the case at 523.6 g, total 3.312 kg, CoM shift (+9.9, 0, +11.6) mm, belly 136.6 mm (see the gate table above). Sections from here down are the R4 integration notes and still describe the construction ideas that survived.

Gate	v0_dingo	v1_scout	Note
total mass (kg)	2.8178	3.2203	+402 g in the base inertial (431 g case − 28.6 g paws, which sit on the foot geoms and are ignored by MuJoCo because the shank has an explicit inertial, as v0's 20 g foot already is)
CoM in base frame (mm)	(−8.0, 0, −31.0)	(−2.5, 0, −21.5)	**shift (+5.5, 0, +9.5),	10.98	mm** — forward and up
belly clearance (mm)	136.6	136.6	lowest base geom is still the chassis belly; the head bottom (z −42) is 169 mm up
self-collision, gait polygon + 20 mm	none	none	first hit at +30 mm is the chassis vs FR shank (−2.42 mm) on both robots: the case does not shrink the leg envelope
stand: tilt / height (m)	0.15° / 0.2086	0.21° / 0.2113	contact height +2.7 mm from the 13 mm paw (nominal +3.0)
walk 2 m (s) / drift (m) / peak q̇ (rad/s)	5.12 / 0.015 / 8.09	5.04 / 0.031 / 8.08
step-down 30 mm	survives, 1.41 m	survives, 1.41 m
mass × friction sweep	6/6	6/6
paw friction sweep μ ∈ {0.5, 0.85, 1.2}	—	3/3	each walks ≥ 1.5 m upright
payload +110 g at (0.17, 0, 0.01) m	—	stand 0.25° / 0.2112 m; 2 m in 5.12 s; q̇ 8.22	display + camera + 2 speakers + 2 mics (parts_db masses)

No gain, torque or gait parameter was touched: the baseline test would fail.

Contact-height change. The paw's flat sole is 13.03 mm below the foot centre (v0 sphere: 10.0), so every leg is effectively 3.0 mm longer and the standing base rises 2.7 mm in sim. The IK and stance are unchanged (they place the foot *centre*). This is reported, not agreed: the brief asks for agreement before treating it as unchanged mechanics. paws.radius_m is the single knob.

Collision proxies. Hoods and the handle use collision: slabs (one box per x-slab of the STL between slab_x_mm cuts) so the leg relief and the finger tunnel are not modelled as solid bricks; every other case part is visual-only. Head box: hood slab z 42..50 + face block.

Keep-outs, checked with numbers

chassis_clearance (G1, every rigid case part vs base_col.stl): saddles 2.4 mm, head 2.30,

rear cover 2.69, camera adapter 2.4, speaker carrier 2.22, mic carrier 4.0, handle 6.4; handle adapter 0.4 mm (tagged contact_chassis, sits on the top plate).

Part-vs-part: manifold boolean intersection of every touching pair (head/saddle, saddle/saddle,

saddle/rear cover, handle/adapters/saddles, bezel/camera/speaker/mic/ear vs head, badges vs saddle) = 0 mm³; tongue and lip fits 0.2 mm.

Legs: G2 self_collision samples 2000 poses with every foot anywhere in the gait polygon

widened by 20 mm (Cartesian), and self_collision_margin_mm records the largest clean margin. This replaced the joint-box ±0.15 rad sweep, which v0 itself fails at 2000 poses (own shank into chassis −0.27 mm, feet through each other −6.5 mm) and passed only by 300-pose sampling luck. To pass it the skirt got a leg relief: skirt bottom z 20 → 28 for |x| ≥ 95 on both saddles and ahead of x = −160 on the rear cover (the thigh tops sweep to |y| 85 above z 20 near the hips).

Print-blocked parts (G4 refuses; `make slice PART=…` exits 1 with the reason)

Part	Blocked on	Why it is unverified
scout_saddle_front, scout_saddle_rear, scout_rear_cover, scout_handle_adapter	`dingo_chassis`	every chassis datum (roof keep-out, plate top, switch stack, nose) is from the decimated mesh, not a measurement; tab/hole patterns are proposals
scout_head	`dingo_chassis`, `display_tbd`, `camera_tbd`	front-cap datums + no display or camera selected
scout_camera_adapter	`camera_tbd`, `dingo_chassis`	cassette envelope is a D435-class size guide only
scout_bezel_blank	`display_tbd`	no display selected
scout_speaker_carrier	`speaker_tbd`	driver not selected (18 × 4 placeholder, see deviations)
scout_mic_carrier	`mic_tbd`	breakout not selected
scout_handle	`heatset_m3`	the M3 insert has no `verified_by` (coupon-calibrated bore, no measured insert)
scout_paw	`dingo_foot`	the foot is the URDF mesh bbox; its physical attachment is unknown
scout_ear, scout_badge, scout_hand_coupon	—	printable now (ear and badge have no external dims)

Deviations from the implementation table (all in the source, none hidden)

Finger scoop 8 → 4 mm deep: base_col.stl has its top plate at z 37.6 over the whole saddle,

so a 2 mm-clear scoop floor cannot go below z 40 (params.SCOUT_HANDLE_SCOOP_DEPTH_MM = 4).

Head bottom −30 → −42 and camera slot −12 → −24.5, window centre 18 → 16.5: bezel

48 + slot 29 + a 2.3 mm bar + the R11 roof fillet do not fit in 80 mm, and the pocket top ran into the roof fillet (wall 1.18 < 1.2). Slot is 94 × 29 (camera + walls + fit), not 90 × 25.

The camera cassette stands 10.8 mm proud of the face (x to 226.8): a 25 mm-deep camera cannot

sit behind a 16 mm face with the chassis nose at x 197.6. The charcoal sleeve *is* the muzzle.

Head half-width stays 70: the front cap is full chassis width (81, not 74) to x 186.5, so the

head is a 6 mm hood over the cap plus a face block ahead of the nose; the cap shows 11 mm each side below the hood. Widening the head to 86 is the obvious next move (open decision).

Speaker placeholder 28 × 6 → 18 × 4 (parts_db.SPEAKER_TBD): the cheek cavity is 22 mm deep

between nose + 2 mm and the face wall; nothing larger clears both. The ring sits 2 mm behind the cheek (R2.4 inner corner fillet). Carriers screw onto boss faces at y 61 (speaker) / 63 (mic) from inside; MIC_TO_HEAD pitch 8 → 10, mic port z 32 → 35.

Interfaces corrected to what is built: BEZEL_TO_HEAD (60, 40), CAMERA_TO_HEAD (104, 0),

EAR_TO_HEAD tenon 3 × 6 × 4.4; SADDLE_TO_SADDLE bosses are only 4 mm tall (M3 × 4 short inserts, or raise SCOUT_SHELL_CLEARANCE_MM).

Rear cover has no step-down behind x −150 (hip-module hull corners at |y| 81, z 37.6 leave

0.2 mm); the switch opening is a notch open to the front edge (stack ends at x −128.4).

Head/saddle seam: the head carries a tongue *under* the saddle roof (top at 44 − fit) and its

roof steps 4.4 mm above the saddle roof; the saddle's x = 125 edge is plain.

Right-hand badge is the left one turned 180° about z (pose: in the manifest), not mirrored,

so robot dog v1 reads left-to-right on both flanks. Lettering is a 1.3 mm-stroke block font (every real font at this size fails the 1.2 mm wall gate).

The Dingo foot mesh's bbox centre is 3.9 mm above / 3.8 mm ahead of the foot *site*; the paw is

built on the site (as specified), so its visual sits 3.9 mm low / 3.8 mm aft of the visual shank tip. Sim contact (sphere at the site) is unaffected.

parts_db.DINGO_CHASSIS re-probed: x_max 198.6 → 197.6, front_cap_half_width 74 → 81,

rear_switch_x0/x1 −165/−132 → −158.4/−128.4, plus nose_chamfer_*.

Bugs found on the way (fixed, with tests)

lib/gates.mass_properties shifted build123d's matrix_of_inertia from the origin to the CoM —

but build123d already returns it about the CoM. Every off-origin part (all of Scout) got an indefinite tensor; the badge failed to load in MuJoCo. cube20 sat at the origin and never showed it. tests/test_gates_cube.py::test_mass_properties_inertia_is_about_com_for_offset_part.

make gate-drop placed each part's body origin at the base-frame origin (0.2 m from the part)

and judged "settled" on |qvel|, which a sub-gram plate on 2–3 hull contacts never reaches (numerical chatter). Now the body sits on its CoM, the floor is solref 0.005, and settled means pose drift < 5 mm and < 3° over the last second with the CoM within −10..+100 mm of the plane.

Open decisions (from the brief; this implementation could not resolve them)

Which Dingo assembly revision and physical attachment datums? Every tab, hole and adapter

is a proposal against a decimated mesh. Measure the chassis in hand → fill verified_by on dingo_chassis → the eleven blocks lift one by one.

Which display, camera, speaker and mic modules? DISPLAY_TBD 60 × 40 active, CAMERA_TBD

90 × 25 × 25, SPEAKER_TBD 18 × 4, MIC_TBD 10 × 12. The head is 90 mm long; a 28 mm driver or a flush muzzle needs ~10 mm more.

Which existing chassis fasteners can carry the handle? HANDLE_TO_CHASSIS is a 16 mm-pitch

M3 proposal through the top plate; no hole there is known to exist. The brief says stop for a scope decision if none does. No proof load is set, no carrying approval is implied; the handle's honest printed mass and profile come out of that test.

Does the finger scoop fit the internals? At 4 mm deep its inner floor is exactly on the

40 mm roof keep-out, 2.4 mm above the mesh's top plate. The real plate, standoffs and cables under it have not been seen.

Contact height +3 mm from the paw — needs the agreement the brief asks for; and the foot's

real attachment decides whether a socket cap is possible at all.

Head width 70 vs chassis 81: widen to the saddle's 86 so cheeks can drop beside the cap?

parts

display

section cut

assembly