Ink palette¶

The installed palette is merged palette v9 (cad/merged-palette-v9), one printed 180 × 178 × 6 mm piece for the merged rig. It carries seventeen ink caps, a touch-trigger probe, a Raspberry Pi camera aimed at the probe ball, the drawing e-stop and an AprilTag on a flat seat. It is the only palette. It is independent of the robot URDF; its pose in the arm base is measured after every placement.

Source

Meaning

urdf/palette.urdf

The palette body and its fixed frames. Generated by the CAD build; do not edit

urdf/meshes/palette/merged_palette_v9.stl, .glb

Body mesh (visual and collision, decimated to 40,000 faces) and the viewer’s copy

config/palette.yaml

The seventeen slots, named after the URDF inkcap_* links, and cap sizes

config/palette_geometry.json

Revision merged_palette_v9, the URDF and body frame, the viewer mesh, the synthetic simulation pose, measured rim overrides (rim_z_m, none yet)

config/fiducials.json

targets.palette: AprilTag 36h11 ID 0, 46 mm black square

config/palette_load.yaml

Recorded cap contents

Frames¶

palette_root is on the probe axis 55.6 mm under the ball’s centre, where v11’s was, so the probe and camera frames are v11’s exactly and their calibrations carry over (runtime in cad/merged-palette-v9/design.json). +X points from the probe toward the Pi camera, +Y rearward, and +Z is up. The base’s top face is at Z −7 and its underside at Z −13.

Frame

Position (mm)

Meaning

palette_tag

114.5, −111.5, −7

The sticker’s seat centre on the base top; its turn measured (below)

palette_probe_ball

0, 0, 55.6

Centre of the probe’s 2 mm ball

palette_camera

122, 0, 55.6

The Pi camera lens; its +X points at the ball

palette_estop

128, −51, 37

The upward-facing e-stop’s panel; its +X points up

Caps¶

The caps stand in four front-to-back groups from the left edge toward the fiducial: three large, four medium, then two columns of five small (S1–S5 and S6–S10), each numbered front to rear.

Slot

Size

Centre X, Y (mm)

Floor Z (mm)

inkcap_large_1

large

-7, -132

-7

inkcap_large_2

large

-7, -104

-7

inkcap_large_3

large

-7, -76

-7

inkcap_medium_1

medium

16, -134

-4

inkcap_medium_2

medium

16, -114

-4

inkcap_medium_3

medium

16, -94

-4

inkcap_medium_4

medium

16, -74

-4

inkcap_small_1

small

35.5, -136

-4

inkcap_small_2

small

35.5, -120

-4

inkcap_small_3

small

35.5, -104

-4

inkcap_small_4

small

35.5, -88

-4

inkcap_small_5

small

35.5, -72

-4

inkcap_small_6

small

52, -136

-4

inkcap_small_7

small

52, -120

-4

inkcap_small_8

small

52, -104

-4

inkcap_small_9

small

52, -88

-4

inkcap_small_10

small

52, -72

-4

The inkcap_* frames are cap pocket floors, not rims. A rim is its floor plus the cap’s outside height from config/palette.yaml, which puts the large caps’ rims 7 mm and the medium and small caps’ 6 mm above palette_root. An entry in rim_z_m overrides the computed rim for that cap once it is measured.

config/palette.yaml states each cap by its outside: diameter_m across the rim, height_m from rim top to cap bottom, and wall_m for the rim, wall and base. Outside, the caps are 15 × 14, 11 × 10 and 8 × 10 mm (diameter × height; large, medium, small), in pockets of 15.6, 13.2 and 9.6 mm. The outside is what rests in the pocket and sets the rim height; the bore a tool enters is derived (ink_spec.CapSize.bore_diameter_m, .depth_m) and is a millimetre narrower. Every cap serves the arm that draws, the one stack.yaml arms drives; each placement still needs reach and clearance checks.

The palette tag¶

The same AprilTag 36h11 ID 0 sticker, 46 mm black edge in a 58.75 mm white square, sits flat on its 60 mm seat centred at (114.5, −111.5) mm on the base top.

The printed pattern’s turn on the seat is measured: the palette_tag frame’s +X runs along the palette’s +X (runtime.installed_tag in the design, tag_pose_status="confirmed" in the URDF). The overhead D555 found it as on v11 (2026-09-30): of the seat’s four quarter turns, only that one put the e-stop, the camera plate and the first large cap where its depth found them (2026-11-08). The probe’s thin stem does not resolve in the D555’s depth from the overhead mount. The station measures the palette only from a confirmed tag frame.

The previous v10 tag transform and palette placement are retired; no old camera measurement is transplanted to the new print.

Measuring the palette pose¶

The probe calibration measures the station afresh in every run, never from a stored pose: tatbot ros station places it in the arm’s base from the overhead D555’s view of the palette tag, through the arm’s registration (ros/README.md §7). The arm registration and palette placement must describe the current rig.

tatbot ros setup measures the drawing arm’s station, as does:

tatbot ros station

The ROS stack measures the station afresh from the D555, validates the views, and carries the full palette_tag transform from urdf/palette.urdf into the arm base. The subsequent calibration run enforces freshness and checks for station movement. The earlier camera-scan placement (tatbot vision palette and its viewer preview) is retired. A missing placement never falls back to the synthetic simulation pose.

Dipping¶

The needle cartridges dip. A dip aims by the station touch, the probe’s ball where the arm’s own touches found it, and the caps’ CAD offsets about it, not by the overhead fix alone (ros/README.md section 6).

The probe and the camera¶

The touch probe and the Pi camera are for tool-tip and station calibration at the ROS 2 stack’s palette station (ros/README.md section 7). The probe is normally closed, wired to the palette Pi’s GPIO27. The camera is an Arducam 64 MP autofocus module (OmniVision OV64A40) on the palette Pi’s CSI connector, aimed along −X at the probe ball 122 mm away. This mount now reaches the documented near-focus distance; focus and image scale must be checked on the installed camera. Measurements from the former 55 mm mount do not describe the new optical setup. The probe relay and the driver’s probe guard read the probe, and the probe calibration touches the tool’s tube end on its ball (ros/README.md §7). The camera watches the tip contact-free: tatbot ros calib sweep from joint-6 turns, and tatbot ros calib height for the ball’s height that dips aim by.

Retired¶

The eight-cap rack and the three-tip volcano fixture (cad/palette-v2) are retired, with their meshes and their 81 mm ID 22 tag, and so is everything built on the fixture’s three pits: the planted-tip refinement of the palette pose, cup teaching, and the per-arm hand-guided recipe with its kinematic refinement. None of their commands or calibration phases exists now. cad/palette-v2 remains only because the integrated-station concept (cad/integrated-station-v1) reads its geometry. The v11 touch probe replaces the pits for tool-tip and station calibration.