Printed targets and mounts

Use this page to choose the target for a task, find its print files and match it to the detector configuration. Three different parts appear in this work:

Part	What it is for	Where to start
Rounded AprilCube	Estimate the pose of an object the robot will pick up	Cube profiles and print files
Dex3 dorsal marker	Observe the rigid hand during camera/arm calibration	Right/left wrist mounts
Torso ChArUco carrier	Investigate a camera reference attached to the torso	V7 prototype and fit limits

The cube and wrist-marker profiles are used in the documented manipulation and calibration work. The torso carrier is a later mechanical prototype: its full assembly and calibration accuracy remain unverified.

The physical target collection, photographed September 21: large and small marker cubes, a wrist-marker carrier and a ChArUco board. Use the sections below to match each part to its files and purpose.

AprilCubes

Runtime object	Edge / perimeter radius	Marker edge and dictionary	IDs
Small cube	40 mm / 3 mm	30 mm, 4x4_100	0–5
Primary large cube	60 mm / 3 mm	45 mm, 4x4_100	10–15
Secondary large cube	60 mm / 3 mm	45 mm, 4x4_100	20–25

Use the profile that matches the physical cube

Runtime profile	Detector IDs	Matched configuration
cube40-r3	0–5	40 mm profile
cube60-r3	10–15	60 mm primary profile
cube60-r3-secondary	20–25	60 mm secondary profile

For the 40 mm cube, the pinned AprilCube fork includes the multicolor 3MF, detector configuration and print settings. Use 100% scale. The documented starting settings are a 0.4 mm nozzle, 0.20 mm layers, four walls and 15–20% gyroid infill. Keep supports and the seam away from the marker planes, and remove support material without sanding the black/white marker surface.

The runtime profiles bind dimensions, detector configuration and grasp shortlist with hashes. For two-cube stacking, use the two distinct 60 mm ID ranges so the observations identify separate objects. The secondary profile shares the 60 mm grasp pool but uses its own detector IDs. A visually similar cube with another size or dictionary does not match these profiles.

For a new size or marker layout, use the pinned fork’s target generator and retain the resulting 3MF, config.json and mesh together. Review the generated dimensions and IDs before printing, then create a matching runtime profile and qualify its grasp pool. The runtime configuration is not a slicer project.

Why changing scale requires new grasps

The final 40 mm design replaced a 50 mm, 8 mm-radius design. Separately, the grasp-generation study used 45 mm geometry before the physical runtime was corrected to 40 mm. These are different experiments. New 40 mm proposals were generated and qualified; scaling the 45 mm atlas would change contact geometry.

The fork reports 48/48 successful synthetic views and 14 focused tests. These check generated geometry and detection. Physical print defects still need inspection: a damaged white cell or softened marker edge changes the image the detector sees.

The fork rounds the complete solid union rather than each voxel separately; the earlier per-voxel construction left seams in T-shaped parts. Marker planes stay flat. The printable 3MF and visual mesh share the rounded geometry; the MuJoCo collision boxes retain a conservative sharp envelope, as described in the cube design. Check the collision representation used by the task separately from the appearance of its rendered mesh.

For a replacement target, preserve the mesh, marker corner coordinates, dictionary, IDs, scale and print revision together. Inspect the printed edges and cell contrast before using its detections in a fit.

Dex3 dorsal markers

The wrist calibration targets attach to the rigid hand shells. They make the hand visible from the head camera without requiring a target held by the fingers.

Wrist-marker placement. The screw tab is toward the finger bases; the marker extends toward the wrist. This is a CAD illustration of the right-hand mount. Use the separately identified left profile for the other hand. Open the full-resolution CAD view.

Get the right print and detector pair

Hand	Print files	Detector profile
Right, ID 4	3MF, separate STLs and fit coupon	dex3_dorsal_aruco_target.json
Left, ID 5	3MF, separate STLs and fit coupon	dex3_left_dorsal_aruco_id5_target.json

Both use white PLA for the carrier and matte black PLA for the marker. In the 3MF, keep the two parts at their shared origin and original scale. The supplied orientation places the marker face on the bed and the support pads upward; the source specifies 0.20 mm layers, four walls, at least five top/bottom layers and no supports. The 0.6 mm inlay occupies three nominal layers.

Print the small m3_15mm_fit_coupon.stl first. The two screws should enter the shell holes without bending the coupon or forcing either screw sideways. Read the mount assembly instructions for the powered-off fitting sequence. The required hardware is two M3 × 8 ISO 10642 countersunk screws and a 2 mm hex key; washers are not part of that countersunk stack. Verify all three pads contact the shell and the marker clears the fingers. A square-on detector check should recover the intended ID and all four corners at a marker size of 0.040 m.

Dimensions and the first-print correction

The revision-2 carrier has a 50 × 50 mm optical plate and a 30 × 10 mm tab, giving a 50 × 60 mm outline and 4 mm plate thickness. It carries a 40 mm 6x6_50 marker: right ID 4, left ID 5. The quiet border is 5 mm and the inlay depth is 0.6 mm.

The two M3 holes are 15 mm apart and have a documented 3 mm blind depth. The specified M3×8 ISO 10642 stack leaves approximately 2.5 mm engagement and 0.5 mm reserve after a 5.5 mm stack. A longer M3×10 screw is not an equivalent substitution. Confirm the source drawing and the actual shell before reproducing the fit.

The first physical print exposed an important CAD assumption: its nominal 2 mm third support did not contact the shell. Accounting for the local tangent changed revision 2 to a 2.57 mm support with an approximately 1.994° pad-face angle. This is distinct from the drawing’s 1.143° mounting-surface pitch. A three-point support is only rigid when all three points actually seat.

Right dorsal is on negative palm Y; left dorsal is on positive palm Y. Copying the right-hand target transform to the left produced an invisible target in planning. Use the handed CAD transforms and a proper rotation.

Physical geometry and fitted geometry serve different purposes

A free calibration target transform can absorb FK error and place an effective marker inside the palm. Use the physical CAD target for visibility and occlusion tests. Use the selected fitted transform when evaluating its calibration model. Neither should silently replace the other.

Torso ChArUco carrier

Prototype status: the upper R2 coupons were tried with M6 × 25 screws, but that length did not engage through the lower coupons. Resolve the lower fit and measured engagement before treating the complete frame as ready to assemble. It is not the calibration reference used by the stacking demo.

The current design is V7, kept in robot-calibration-aprilcube-prototype. It uses two broad torso crossbars and two flat side plates. These support the accepted 210 × 300 mm multicolor carrier and its four M4 holes. The active board is 6 × 9 squares, with 30 mm squares, 22 mm markers and 5x5_50, covering 180 × 270 mm. V7 replaces the earlier frame whose small M6 boss broke during support removal/handling in PLA.

V7 CAD assembly. Two crossbars attach to the torso and two side plates carry the board. This shows the modeled arrangement; the full printed assembly has not been physically qualified.

Need	Exact source
Slice the current structure after resolving coupon fit	Four oriented structural STLs
Open or print the current structure	V7 CAD and printing guide
Inspect native CAD or exchange geometry	F3D and STEP files
Check the four shell interfaces	Compact R2 coupons
Understand what physically fitted	September 13 fit status
Edit or rerun CAD checks	Fusion and mesh script guide

The R2 coupons preserve the full contact patches, M6 bores and washer bearing planes while trimming the earlier coupons. Use the V7 print_parts/ and linked R2 coupon files. Pre-R2 meshes under original_reference/ are retained for geometric comparison, rather than as another revision to print.

The September 13 physical record reports that M6 × 25 worked with the upper coupons but did not engage through the lower coupons. Upper washer use and engagement depth were not measured; lower screw length remains unresolved. The earlier M6 × 20 assumption and rough 35/40 mm estimates are not a confirmed hardware specification. All current seats retain a nominal 12 mm stack at the screw axis. Full crossbar seating, assembly stiffness, board flatness and reinstall repeatability remain unverified.

For a fixed board, the geometric relationship is torso_T_camera = torso_T_board @ inverse(camera_T_board). V7 retains the nominal board center [245, 0, 127.5] mm in torso_link, with +21° pitch. The installed torso_T_board needs an independent physical check; low image reprojection error alone cannot establish it. Recorded CAD checks, including unobstructed target rays, do not qualify this fixture as a torso extrinsic standard. The V7 work has not supplied a replacement for the August wrist-based calibration bundle used by the stacking demo.

Evidence: AprilCube July 13–15 log; prototype dorsal-marker and torso-carrier documents; tabletop object-profile corrections. See sources and the media requests.

Follow the code

What to change	Code entry point	Responsibility
Runtime cube profile	tabletop_object.py	Bind cube dimensions, marker profile and task geometry.
Dorsal mount specification	dex3_dorsal_mount.py · Dex3DorsalMountSpec	Record dimensions and handed mounting geometry.
Physical marker transform	dex3_dorsal_mount.py · palm_T_marker_face	Compute the handed CAD marker pose.
Mount artifact	dex3_dorsal_mount.py · mount_manifest	Export geometry and metadata for the mount.
Current torso structure	build_g1_mount_v7_fusion.py · build	Construct the V7 crossbars and side plates inside Fusion.
Torso geometry checks	validate_g1_mount_v7.py	Export print orientations and check mesh, tool and optical clearance.

Checks and evidence to inspect

The fork’s synthetic views establish geometry/detection checks, while the first dorsal print supplies the physical third-support correction. The photograph above shows the separate parts; final mount seating, third-pad contact and print quality remain in the review queue. Hand installation and its cable routing are covered under hardware.