Quack core

Quack Core

The brain board for the Open Duck Mini: a Pi Zero 2W carrier with the 14-servo STS bus, a fused and switched 2S power path laid in stitched copper zones, USB-C charging with balancing, and the IMU.

Part of HeyPCB's open robot boards. Designed end to end with the HeyPCB method on KiCad 10 (headless pipeline, Freerouting 2.4.1); every number below is measured on the files in this repository. Mentions of ./run.sh, docs/rules/, blocks/ or parts/ refer to the heypcb-kicad SDK that produced them.

Not fabricated, assembled or powered.

Quack Core is a brain board for the Open Duck Mini (v2), the open-source miniature biped by Antoine Pirrone and contributors. It is a carrier for a Raspberry Pi Zero 2 W and it bolts onto the four existing trunk_top bosses in the duck's rear bay. It takes the 2S pack through a 20 A fuse, a 1 mOhm shunt and a back-to-back P-FET switch, drives the STS3215 servo bus, and charges and balances the pack from USB-C. It was designed with the HeyPCB method on KiCad 10 (10.0.6), by heypcb-kicad's headless pipeline with Freerouting 2.4.1, from the brief in board.json and the research spec robots/research/duck/spec.md (Option A).

This board has not been fabricated, assembled or powered. Everything below is what KiCad, the pipeline and the CAD scripts measured on the design files. Nothing here has been tested on hardware or in the robot, and nothing here contains or tests software.

Specs

• Pi Zero 2 W carrier. The Pi plugs in face-up on a 2x20 SMD socket (PM254-2-20-S-8.5, C3975165), 8.5 mm tall, on four M2.5 standoffs (not supplied) on its 58 x 23 mm hole pattern. The board's 5.1 V back-powers the Pi through header pins 2 and 4. Do not also plug the Pi's PWR micro-USB.
• Pack input: a vertical XT30 (J2), an SMAJ10CA bidirectional TVS (D1), a 20 A Littelfuse NANO2 (F1, 0451020.MRL) and a 1 mOhm 3 W 2512 shunt (R1). An INA226 at 0x40 reads the shunt and the pack voltage; its ALERT goes to GPIO4 and to the servo-kill latch.
• Main switch: two AON6411 P-FETs, source to source (Q1, Q2). The robot's existing backpack slide switch is re-used as a logic switch on J5 and carries about 80 uA. A reversed pack leaves both FETs off.
• Servo power: a third AON6411 (Q3) switches VSYS onto VSERVO. It is on by default: GPIO26 (with a 100 k pull-down) or a latched INA226 ALERT turns it off. VSERVO has 2 x 470 uF / 16 V bulk (C10, C11) and 10 uF plus an ESD5Z5.0 on DATA at each port.
• Servo bus: an exact copy of the Waveshare Bus Servo Adapter (A) / Feetech reference half-duplex driver (74LVC1G126 + 74LVC1G125 + MMBT3906) on the Pi's PL011 UART (GPIO14/15), with the spec's additions: 0 ohm source-select links, a J16 header for an external USB-UART, a DNP RTS link, a DNP DATA pull-up and four DNP test pads.
• Servo ports: four 5264-type 3-pin headers (HC-5264-3A, C2845815), DATA / VSERVO / GND: J6 left leg, J7 right leg, J8 neck chain, J9 spare.
• Pi rail: an LM74700 + AO3400A ideal diode with 470 uF hold-up behind the 5 A face/aux fuse (F2), then a TPS56339 3 A buck to 5.10 V (53.6 k / 10 k on a 0.802 V reference) and an XC6206 3.3 V LDO for the board's logic.
• Charger: BQ25887 2S boost charger from USB-C (J4, HRO TYPE-C-31-M-12, 5.1 k on each CC, a plain sink) through a 2 A PTC (F4), with cell balancing through the balance lead's mid tap (J3 pin 2). Input limit 1.48 A (750 ohm on ILIM). It taps the pack ahead of the switch through a 5 A NANO2 (F3), so it charges with the robot switched off. Orange CHG LED, /INT to GPIO16.
• IMU: BNO055 at 0x28 (fitted) and a BMI088 footprint at 0x18/0x68 (not fitted).
• Face link (the shared contract with the Quack Face board, spec 4.5): J12 JST-GH 12P top entry, J13 JST-XH 2P with the fused battery rail.
• Other I/O: foot switches on J10/J11 (JST-PH 2P, 1 k / 100 nF RC and ESD), a Qwiic port (J14, JST-SH 4P top entry, 3.3 V), green 5V LED, red LED1/LED2 on GPIO5/GPIO6.
• Board: 101 x 51.5 mm (CAD x -93.5 to -42.0, y -50.5 to +50.5), 1.6 mm, 4 copper layers. F.Cu signal with a GND pour, In1 a GND plane, In2 signal with a VSERVO pour, B.Cu signal with a GND pour, and the power path laid as stitched copper zones on F.Cu, In2 and B.Cu (see Power zones). Single-sided assembly. Four non-plated 2.7 mm holes for M2.5 screws on the bosses, and four more for the Pi's standoffs. The front edge is notched 0.8 mm between CAD y -28.25 and +28.25 to clear the trunk_top chamfer (see CAD fit).

Option A: the Pi moves into the body

Today the Pi Zero 2 W, its 5 V UBEC and the amplifier sit on the head's ceiling, and a harness through the 4-DOF neck carries 7.4 V, USB, I2C and the foot-switch wires down to the body (spec 1.4). Quack Core takes the research spec's Option A (spec 2.6): the Pi plugs into this board, in the body, next to the battery, the servo bus, the IMU and the foot switches. What that changes:

• Update the URDF masses before retraining. The Pi (about 10 g) and the UBEC (about 15 g, replaced here by the on-board buck) leave the head and join the trunk. Both figures are the spec's estimates. The walk policy was trained on the URDF's mass properties (head_assembly 352.6 g, trunk_assembly 698.5 g), so update both links, add this board's own mass to the trunk, and retrain (spec 2.6, 7). This board's mass was not measured.
• Camera. CSI would have to cross the +/-160 degree head-yaw joint. Either run a round CSI-to-HDMI extension through the neck (Arducam B0091, with a 22-to-15-pin cable at the Pi), or leave the camera off. The walk runtime does not use it (spec 2.6).
• Runtime: configuration only (spec 3.10). No Python changes:
  1. /boot/firmware/config.txt: enable_uart=1 and dtoverlay=miniuart-bt. The PL011 moves to GPIO14/15 and Bluetooth stays up for the gamepad. 1 Mbps is 48 MHz / 16 / 3 exactly.
  2. cmdline.txt: remove console=serial0,115200.
  3. A udev rule, KERNEL=="ttyAMA0", SYMLINK+="ttyACM0", so the hard-coded /dev/ttyACM0 in the runtime still opens the bus (or edit those defaults).
  4. Keep dtparam=i2c_arm=on. If the BNO055 reads glitch, apply Adafruit's i2c_arm_baudrate fix for its clock stretching.
  5. Keep the I2S amplifier overlay. Every GPIO number the runtime uses is unchanged.
• The neck harness becomes J12 (12 signals), J13 (the fused battery rail for the face board's own 5 V) and the neck servo chain on J8.
• Wi-Fi and Bluetooth move inside the PLA body, next to the cells. The spec estimates this as acceptable but slightly worse.
• No first-hand report of an STS3215 bus on a Pi's GPIO UART at 1 Mbps was found (spec 3.10). J16 and two 0 ohm links are the fallback: remove R17 and R18 and drive the same buffer from a 3.3 V USB-UART on J16 (1 GND, 2 TXD into the board, 3 RXD out of it).

Pinouts

Pi Zero 2 W header (J1)

The 2x20 socket's footprint is the mirror image of the Pi's header (socket pad 2n-1 meets Pi pin 2n), so the block wires each socket pad to the net of the Pi pin it meets. The table is in Pi pin numbers. Every GPIO number the walk runtime uses is the one it uses today (spec 3.8).

I2C1 carries the BNO055 (0x28), the INA226 (0x40), the BQ25887 (0x6A or 0x6B, see Check before ordering), the BMI088 footprint (0x18 / 0x68) and the Qwiic port. The board adds no I2C pull-ups: the Pi's own pull-ups on GPIO2/3 serve the bus.

Face link: J12 and J13 (the contract with Quack Face)

Both match the Quack Face board (commit 250de3c on robot-quack-face), whose harness runs pin n to pin n.

J12, JST-GH 12-pin top entry, BM12B-GHS-TBT (C2983891), spec 4.5 unchanged:

The two mounting pads are on GND. The spec named the side-entry SM12B-GHS-TB. The board's front edge sits 1.9 mm from trunk_top's rear wall, so a side-entry plug facing forward could not be inserted. The top-entry part has the same housing and pinout.

J13, JST-XH 2-pin (B2B-XH-A, C158012): pin 1 VBAT (VBAT_AUX, the pack after the main switch and the 5 A fuse F2), pin 2 GND. F2 also feeds the Pi rail. The spec's face worst case (about 2.6 A at the pack) plus the Pi (about 1.4 A) is below 5 A.

Other connectors

Positions are body centres, in the CAD frame of the print STLs (+x forward, +y the robot's left, mm), read off the final board.

J2's polarity is the footprint's: KiCad's AMASS XT30UPB-M footprint prints - beside pad 1, at the housing's chamfered end, and + beside pad 2, and the chamfered side of an XT30 is its negative contact. Pad 2 (+) faces F1; pad 1 (-) is at the rear edge.

J3 uses only the balance lead's mid tap. Its B- pin is the cell side of the pack's BMS: tying it to the board's ground would bypass the BMS's discharge FETs. So pins 1 and 3 are not connected.

Power path and what the copper carries

J2 XT30 ─┬─ D1 SMAJ10CA ─ GND
         ├─ F3 5 A ─ BQ25887 BAT (charger, J4 USB-C, J3 mid tap)
         └─ F1 20 A ─ R1 1 mOhm (INA226 0x40) ─ Q1 ┐ AON6411 back to back, gates on J5
                                                Q2 ┘
                                                 │ VSYS
              ┌──────────────────────────────────┼──────────────────────────┐
         Q3 AON6411 servo kill              F2 5 A ─ VBAT_AUX ─ J13 (face)    R8 / Q4 / Q5 gate drive
         (GPIO26, INA226 ALERT)                      └─ LM74700 + AO3400A ─ 470 uF ─ TPS56339 5.10 V ─ Pi pins 2/4
              │ VSERVO                                                               └─ XC6206 3.3 V
         C10, C11 470 uF; J6, J7, J8, J9 (10 uF + ESD5Z each)

The pack current runs in a straight column down the board's right side: the XT30 at the rear, then F1, R1, Q1, Q2 and Q3 one above the other, with Q3's drain at the front, beside C10 and J7. The FETs face source row to source row, so QS and VSYS are three short stubs each.

Net classes (from the brief; a class with a current is never narrowed by the routing ladder): BATT 3.0 mm (VBAT_IN, VBAT_F, VBAT_SH), FETLINK 1.2 mm (QS, VSYS, limited by the AON6411's 0.5 mm source leads), SERVO 2.5 mm with 2.2 / 1.2 mm vias (VSERVO), AUX 0.8 mm (VBAT_AUX, VPI_IN, 5V, BUCK_SW), USBPWR 0.6 mm (VBUS), CHG 0.4 mm (the charger's power nets). The tracks only connect the path; the current is carried by the zones below.

Power zones

After routing reaches 0 open, the pour stage lays the brief's board.zones: each a named net's copper over a region on chosen layers, above the plan pours, with every pad joined solid (no thermal spokes), and, where the zone has a stitch pitch, its own via lattice. Board coordinates are mm from the outline's top-left corner; every reason is also in board.json.

The gnd zone exists to join the XT30's - pad and the ports' GND pins solid. Any GND zone makes the plan's GND stitcher place nothing (it wants each via's ring inside every GND fill), so the gnd zone stitches GND itself; the plan's own stitching count reads 0. J7's GND pin sits between Q3 and J7's V pin, so on any one layer only one of the two nets can pass it: VSERVO reaches J7 on F.Cu and In2, and J7's GND leaves on B.Cu, down the right edge to the XT30.

Measured copper on the current path

copper.py (KiCad's Python) lists every track, via and pour of each power net on the final board and exports the copper (tracks, pads, plan pours and zones) as polygons. copper_cut.py (the robots CAD environment, it needs scipy) rasterises that copper on a 0.1 mm grid and finds, for each link of the path, the narrowest cross-section the link's whole current must pass: a max-flow / min-cut from the link's source pads to its sink pads, over every layer, with tracks, zones and via barrels together, in parallel. Outer copper is taken as 35 um, inner as 17.5 um (JLCPCB's standard 4-layer build), via plating as 25 um. Each cut is rated with heypcb_kicad.route.ipc2221_amps (IPC-2221, I = k dT^0.44 A^0.725), per layer, summed. Each link is cut a second time with the copper within 1 mm of its pads counted as the pads ("beyond the pads"): that rates the board's own copper between the parts, apart from the parts' fixed lands.

Read off the board (copper_cut.json, copper.json) from the full run with media. Before the zones, the same measurement gave 4.9-8.0 A at a 10 C rise for every link (the servo rail crossed one 1.2 mm via; the battery path was 3.0 mm tracks).

What this says about 15 A (the spec's design point: 15 A continuous, 30 A peaks; simulated walking 13.5-14.4 A, spec 3.9). At a 20 C rise on 1 oz outer copper:

• The battery side reaches it: VBAT_IN 18.3 A, VBAT_F 15.0 A, VBAT_SH 16.7 A. VBAT_F and VBAT_SH end on pads 0.9 mm from the next part's pads (fuse to shunt, shunt to Q1), so their narrowest section is the copper alongside those pads; the zones already surround them.
• The servo rail reaches it in total (24.4 A from Q3 to the four ports, GND return 23.7 A), and per side as the load divides: the left leg and neck 15.6 A, the right leg 8.2 A (VSERVO) and 10.3 A (GND), the spare port 6.7 A. The right leg is the narrow one: J7's GND pin sits in the path, so its VSERVO uses only F.Cu and In2 and its GND only B.Cu and In1. With the walking load divided over two legs (about 5-7 A each, estimated from spec 3.9), each leg's port has margin; a single leg near its servos' stall current does not.
• The two links between the back-to-back FETs do not: QS 8.4 A and VSYS 9.2 A. They end on the AON6411's three 0.5 x 0.66 mm source leads per FET. Whatever is poured, the copper that can touch those leads is about 3.7-4.2 mm wide on F.Cu (the slot to the drain pad is 0.67 mm, and no via fits inside the package's courtyard). That is a property of the 5 x 6 mm DFN package. These links are 0.7 mm long, lead to lead; IPC-2221's chart is for long conductors, and a sub- millimetre bridge between two FET packages sheds its heat into them, so this rating says little about the real limit. It was not measured. A part whose source is a wide tab (or two FETs in parallel per side) would raise it.

This board demonstrates about 15 A at a 20 C rise on everything except the 0.7 mm FET-to-FET bridges (8.4-9.2 A by this rule) and the per-leg ports (8.2-15.6 A). Nothing here is a measured temperature; measure Q1-Q3, F1, R1 and the XT30 under load before trusting a continuous rating. The path parts themselves are sized for the design point by the spec (3.9): F1 20 A, FETs of 40 A or more, the 3 W 2512 shunt (0.23 W at 15 A), the XT30. The 5264-type servo headers are rated 3 A per pin (spec 3.4), which a leg chain can exceed while walking, as on the Waveshare adapter the duck uses today. 2 oz outer copper would raise the outer-layer share of each figure by about 65 % (2^0.725).

The rest of the power copper, read off the board: VBAT_AUX 0.8 mm (117.9 mm, 4 vias of 0.8 / 0.4 mm, on F.Cu, In2 and B.Cu), 5V 0.8 mm (43.9 mm, F.Cu), VBUS 0.6 mm (45.8 mm, 2 vias, mostly on In2), VBAT_CHG 0.4 mm (30.6 mm, 2 vias). IPC-2221 at a 10 C rise on 1 oz outer copper: 0.8 mm 2.0 A, 0.6 mm 1.7 A, 0.4 mm 1.2 A; on 0.5 oz inner copper they carry less. The charger's input is limited to 1.48 A by its ILIM resistor.

Where the board differs from the spec

Everything else follows the spec's parts table (spec 6): BNO055, the Waveshare/Feetech bus driver parts, TPS56339, XC6206, LM74700 + AO3400A, AON6411 x 3, INA226, the 1 mOhm 2512 shunt, NANO2 fuses, BQ25887, HRO USB-C, XT30, HC-5264-3A, JST-XH, JST-PH. Part numbers came from live jlcsearch lookups on 2026-10-02 and are pinned in parts/lcsc_pinned.json with that date.

What it measured

A full run with media on 2026-10-03 (./run.sh --board boards/quack_core, KiCad 10.0.6, Freerouting 2.4.1). Gate numbers are from report.json. Anything marked read off the board was measured on kicad/quack_core.kicad_pcb with KiCad 10.0.6's pcbnew.

• The 8 DRC warnings: 7 lib_footprint_mismatch (J3, J4, J7, J8, J9, C10, C11: all seven sit at the board edge, where the silk clip dropped their strokes) and 1 silk_over_copper: the XT30 footprint's own "+" sits on the edge of F1's pad. The two via_dangling warnings of the run before the zones (the vias in Q1's and Q2's drain pads) are gone: the zones join them.
• Nothing on the board was edited by hand: every track came from Freerouting, every pour and stitching via from the pipeline.

Why the brief looks the way it does

Measured on the runs of 2026-10-02; each reason is also in board.json beside the key.

• The power column is locked part by part. XT30, F1, R1, Q1, Q2 and Q3 sit on one line, each pad facing the next. Q2 is offset from Q1 by one lead pitch (1.27 mm): aligned, Freerouting ran Q2's gate line through the 0.67 mm slot between Q1's source leads and drain pad and cut off Q1's source lead 1. Thin track keep-outs along those slots (q1_source_slot and two more) keep gate lines out without blocking the source net; a keep-out over the whole slot left 6 open.
• FETLINK is 1.2 mm. At 1.4 mm with 0.25 mm clearance the class could not reach Q1's own 0.5 mm source leads.
• The charger sits under the Pi with every satellite locked. Beside the USB-C, one of its analog pins (TS, ILIM, MID, CBSET, REGN) was left open on each of three builds; under the Pi with the resistors on the placer's shelves, ILIM and MID; with the 2.2 k LED resistor where the placer put it, under the TS / ILIM / MID pins, MID. Now the four resistors of those pins sit in a fan below them and the LED resistor right of the capacitors.
• J9's pin 1 is at board x 35.77 (its body centre at CAD y +12.2). In the rear-left corner its DATA line was left open once. At x 34.5 the placer put its 10 uF on the DATA side of the header, where the 2.5 mm VSERVO class could not reach it between the pins and the edge. 1.27 mm to the right, the placer put the capacitor beside the boss, and Freerouting routed all 312 connections itself.
• The routing was searched in parallel copies. Ten candidate briefs each ran from compose in their own --build directory, three or four at once. Three reached 0 open: one with router shorts left in (not eligible), one only after the last-mile router, and this one, which Freerouting routed alone. The board's own runs then rebuilt it from this brief and routed it the same way.
• The XT30's pads were swapped late. The first revision of quackcore_power_in put the pack + on pad 1. The silkscreen check showed the footprint's own "-" printed beside that pad: KiCad's AMASS footprint marks pad 1, at the housing's chamfered end, as the negative contact, which is the XT30's convention. The block now has GND on pad 1 and VBAT_IN on pad 2, and J2 is turned 180 degrees so the + pad still faces F1. Every number in this README is from the run after the swap.
• The power path is zones, not tracks (added 2026-10-03, once the pipeline had board.zones). The tracks still route first and every gate is unchanged; the zones then took every link from 4.9-8.0 A to the figures in Power path (10 C rise). What shaped them: one zone per net with distinct priorities (overlapping zones of one net at one priority are a DRC zones_intersect error); VBAT_F's zone follows the Pi standoff's keep-out circle, because a square notch there set its cut at 7.5 mm (14.0 A at 20 C) and the stepped edge at 8.3 mm (15.0 A); VSERVO stops at x 90.6 on B.Cu so that J7's GND pin keeps B.Cu ground (inside the VSERVO zone its only ground was In1's thermal spokes, 1.4 A); the GND zone stays off In1 (on In1 it left slivers of the plane, DRC isolated_copper); and the front band is not stitched, so In1's ground under J6 and J8 stays whole.
• The board's front edge has a 0.8 mm notch, from the CAD fit (see CAD fit).

CAD fit in the duck

cad/fit.py (the robots CAD environment: CadQuery, trimesh, manifold3d, VTK) places the pipeline's own STEP (fab/quack_core.step) in the CAD frame of the Open Duck Mini v2 print STLs, adds an envelope for each part whose KiCad 3D model is missing (named in fit.json, with its source) and a Pi Zero 2 W envelope on the socket, and measures against every body part: the print STLs and the URDF's servo cases, horns, cells, holder, BMS and power switch. Clearance is the smallest distance between points sampled 0.25 mm apart on the assembly and on the body, in both directions; a sample inside a body counts as interference. The board's bottom face resting on the four boss tips is the seating, checked separately. Numbers from cad/fit.json:

The front notch. The spec's rectangle put the board's front edge 0.11 mm into trunk_top's chamfer where the tongue meets the front wall (the edge sat at z -6.10 where the chamfer reaches -5.99, measured on trunk_top.stl). The outline is set back 0.8 mm along the front edge for |y| <= 28.25, which gives the 0.49 mm above.

B2, the USB-C opening (spec 5). fit.py cuts a 13.0 x 7.5 mm opening with 1.5 mm corner radii through the right-hand (-y) side wall of body_middle_top, centred on the receptacle at x -62.0, z -2.87 (spec: x -62.0, z -2.9), through y -51 to -56 (the wall is y -52 to -55). The receptacle's mouth is at y -51.45, 3.55 mm inside the wall's outer face. 286.6 mm3 removed; the result is watertight. It is cad/body_middle_top_usbc.stl (and a faceted STEP in cad/), Apache-2.0 with the change notice in cad/NOTICE.

Preview renders (headless VTK), written by fit.py to cad/: fit_iso.png (the board on the bosses, top shell hidden, the Pi as a translucent envelope), fit_top.png (front up, the Pi hidden, envelopes in orange), fit_section.png (a section at y -14 through two bosses, seen from the robot's right) and fit_usbc.png (the B2 opening with the receptacle behind it).

Envelopes, not vendor models: the XT30 (11 mm tall, estimated, no drawing read), the four 5264 headers (10 x 4.9 x 6 mm, LCSC row), the NANO2 fuses, the XFL4020 inductor and the USB-C shell (8.94 x 7.35 x 3.26 mm).

Check before ordering

The board passes ERC, the netlist round trip, DRC with schematic parity, DFM and the fab gate in KiCad. It has not been fabricated, assembled or powered.

• Current. At a 20 C rise on 1 oz outer copper the battery side (15.0-18.3 A) and the servo rail in total (24.4 A, GND 23.7 A) reach the 15 A design point; the 0.7 mm bridges between the FETs' source leads (QS 8.4 A, VSYS 9.2 A) and the right-leg port (8.2 A) do not, by IPC-2221. See Power path. Before a walking test at full current, measure the temperature of Q1-Q3, F1, R1 and the XT30 under load; 2 oz outer copper is the cheap margin.
• Pack BMS. The duck's 8 A / 10 A BMS is the first limit: the spec asks for a 2S BMS rated 20 A continuous or more, 40 A peak or more and 10 mOhm or less (spec 3.1). The XT30 is the pack connector, and the 2 x 470 uF bulk behind Q3 charges through the BMS each time the servos are switched on (Q3 turns on in tens of microseconds). Check the BMS's short-circuit trip against that inrush.
• 8.4 V on 7.4 V servos. VSERVO is the pack voltage, up to 8.4 V on a full 2S pack. The STS3215 is rated 4-7.4 V with an 8.0 V register limit (spec 1.3; whether shipped servos alarm at 8.0 V is an open item, spec 8). The BQ25887 charges to 8.4 V by default (REG00 resets to 4.20 V per cell). It runs with no host, with the robot switched off; to charge to 8.0 V, the Pi must write 4.00 V per cell to REG00 and disable the watchdog (REG05 WATCHDOG = 00) while it runs. Those settings then hold until the pack is unplugged.
• BNO055 axes. The chip is turned so that pin 1 is at its rear-left corner. By the data sheet's placement figure that turns chip +X toward the robot's left and chip +Y toward its rear, which is what raw_imu.py's remap needs for imu_upside_down = false (spec 3.6). This was read off a figure, not tested: check it against the figure (and on the bench, by tilting) before trusting the walk policy on this board.
• BQ25887 I2C address. TI's data sheet gives 0x6A in one place and 0x6B in another (SLUSD89B 8.3.11 and 8.3.11.5). Scan the bus once before writing driver code.
• Via in pad. route.via_in_pad_mm put a via in every SMD pad 2.6 mm or wider: Q1, Q2 and Q3's drain pads (1.2 / 0.6, 1.0 / 0.5 and 2.2 / 1.2 mm) and the charger's exposed pad (0.6 / 0.3 mm). All four now join a zone on B.Cu. Open holes in a FET pad wick solder: order these vias filled and capped. The 122 zone vias sit outside every part's courtyard and need nothing special.
• Envelopes. The XT30's 11 mm height is an estimate (no drawing was read), and five other parts are envelopes from catalogue sizes (see CAD fit). The tightest measured gaps are 0.49 mm at the front notch and 0.55 mm at the USB-C mouth before the B2 cut.
• Balance lead. Only the mid tap (J3 pin 2) is used. Check that the pack's balance plug puts the mid tap on pin 2 (the usual B-, mid, B+ order) and that it mates with a JST-XH header.
• No thermistor. TS is a fixed divider at 55.4 % of REGN, between the data sheet's 10 C and 45 C thresholds, so the charger always sees a normal temperature and never sees the pack's.
• Pi rail. The board back-powers the Pi through pins 2 and 4 at 5.10 V nominal. Never plug the Pi's PWR micro-USB at the same time. The Pi sits 11 mm above the board on a socket of 8.5 mm plus a 2.5 mm header spacer, on four 11 mm M2.5 standoffs (not on the BOM).
• Inner copper. The gerber job says 0.035 mm on every layer; that is KiCad's default stackup, not a fab quote. The copper ratings above assume JLCPCB's standard 0.5 oz inner layers.
• Part numbers and stock. The rows this board added were read live from jlcsearch on 2026-10-02. On that date the BQ25887RGER had 987 in stock. Stock and Basic / Extended class go stale: re-check before ordering. The BOM has 34 distinct Extended parts (about $102 in JLC loading fees).
• Silkscreen. Read off the board: the front carries PI ZERO 2 W, FACE PWR, UART, SWITCH, QWIIC, FOOT R, 5V, LED1 and CHG, and the D / V letters beside J9 (and G beside J6). The other words (BATT, L LEG, R LEG, NECK, SPARE, FACE, BALANCE, USB-C CHARGE, FOOT L, LED2 and 20A, which the run before the zones placed and this one did not) found no free spot among the parts and are not printed. The XT30 carries its footprint's own "+" beside the + pad, toward F1; its "-" falls past the rear edge and is clipped. The back carries the title and a legend: "servo ports 1 D 2 V 3 G: J6 L LEG J7 R LEG J8 NECK J9 SPARE" and "J2 XT30 1 - 2 + J13 FACE PWR 1 + 2 - J3 BAL 2 MID". Check the XT30 plug's wiring against that legend before the first connection.

Licence and attribution

• The board files here (board.json, the quackcore_* blocks, the scripts) are part of the heypcb-kicad repository.
• Open Duck Mini (CAD, STLs, URDF, documentation), by Antoine Pirrone (apirrone) and contributors, is Apache License 2.0 (github.com/apirrone/Open_Duck_Mini; Onshape document 64074dfcfa379b37d8a47762). cad/body_middle_top_usbc.stl is derived from its print/body_middle_top.stl; the change is stated in cad/NOTICE, and the licence is copied as cad/LICENSE_Open_Duck_Mini_Apache-2.0.txt. fit.py reads the other print STLs and URDF meshes from a local copy and does not redistribute them.
• Quack Core is a board for the Open Duck Mini. Apache-2.0 grants no trademark rights; no Disney or BD-X mark is used on the board or in these files.
• The Open Duck Mini runtime has no licence file, so none of its code is copied here. Pin and port assignments are used as interface facts.
• Data sheets were read for values and are not redistributed.

Files

• board.json: the brief. Blocks, net classes, keep-outs, placement (with a reason per lock), routing and silk settings.
• copper.py (KiCad's Python) and copper_cut.py (scipy): the copper measurement above. They write copper.json, copper_geom.json and copper_cut.json.
• cad/fit.py: the CAD fit, B2 and the previews; writes cad/.
• cad/body_middle_top_usbc.stl, cad/NOTICE, cad/LICENSE_Open_Duck_Mini_Apache-2.0.txt: the B2 part and its attribution.
• ../../blocks/quackcore_*.json: the 15 blocks written for this board. Each block's notes give the KiCad 10 symbol pinout it was wired from and the data sheet its values came from.
• The pipeline's outputs are in build/quack_core/ (not committed): report.json, the KiCad project, fab/ (gerbers, drill, BOM, JLC CPL, schematic PDF, STEP) and the media.

To rebuild, from the repository root:

./run.sh --board boards/quack_core                     # the pipeline, with renders and video
/Applications/KiCad/KiCad.app/Contents/Frameworks/Python.framework/Versions/Current/bin/python3 \
    boards/quack_core/copper.py                        # power copper -> build/quack_core/copper*.json
python boards/quack_core/copper_cut.py          # a Python with numpy and scipy
DUCK_CAD=<duck folder> python cad/source/fit.py   # CadQuery + trimesh; needs the Open_Duck_Mini STLs