Weather node backup

05 — Dock interface

SKiDL-generated draft v0.1. NOT FOR FABRICATION — see blockers below.

Generate + check

# one-time setup (from repo root, shared with Boards 1-4)
& 'C:\Program Files\KiCad\10.0\bin\python.exe' -m venv .venv-skidl
.\.venv-skidl\Scripts\python.exe -m pip install skidl

# generate netlist + schematic, then KiCad ERC
.\.venv-skidl\Scripts\python.exe hardware\kicad\05-dock-interface\gen_schematic.py
& 'C:\Program Files\KiCad\10.0\bin\kicad-cli.exe' sch erc --output hardware\kicad\05-dock-interface\erc.rpt hardware\kicad\05-dock-interface\dock_interface.kicad_sch

Or, where make is installed (Git Bash / WSL / Linux): make kicad-board5.

Outputs: dock_interface.net, dock_interface.kicad_sch, erc.rpt. Verified 2026-11-04: SKiDL ERC 0 errors, KiCad ERC 0 errors (17 lib_symbol_mismatch warnings — SKiDL embeds its own symbol copies; cosmetic). The SKiDL log also prints "No footprint for PWR_FLAG" errors; PWR_FLAGs have no footprint by design and the files are still generated.

Circuit: external 14.6V charger -> J1 -> 10mOhm shunt (INA226, 0x41; 2-3A std, 5A max) -> spring contacts J2 (6 contacts) to the robot. D3 DECIDED 2026-11-04: the docked robot is the I2C master for this board's INA226 over two extra spring contacts (J2 pin4=SDA, pin5=SCL); no I2C pullups on this board - the single 4.7k pair on Board 3 serves the bus. Dock INA226 re-strapped 0x40 -> 0x41 (A0=VS) to avoid collision with Board 3's 0x40. CHG_ALERT is not contacted (robot polls; ALERT only on bench header J3). Contact 3 is dock-presence sense: tied to GND here, detected by robot MCP23017 GPB6/DOCK pullup - no sense processing on dock. Refill solenoid driver is Board 3 Sheet 5 circuit (IRLZ44N, 220R, 100k pulldown, 1N4007) on 5A fused tap; gate arrives on J2 pin 6 (REFILL_GATE_SRC, Board 3 MCP GPA3). Solenoid SPECIFIED (D7, 2026-11-04): 12V continuous-duty rated (or explicitly 15V-rated) solenoid - NOT 24V (won't pull in at 12.8V). With the rail at 14.6V peak the coil would dissipate ~48% over its 12V rating if held at 100%, so robot firmware MUST use hit-and-hold: 100% duty for ~100ms to pull in, then 60-70% PWM hold (coil heating scales with duty: ~0.65 x 1.48 ~= 0.96x rated - cool). PWM comes via the MCP23017 GPA3 over I2C, so use a low PWM rate (>=100Hz is plenty for a solenoid coil) and remember the 100k fail-closed pulldown turns it off when undocked/idle. J3 is bench/debug only. AprilTag mount mechanical only (BOM line, no circuit). Generated schematic uses net labels (mode="labels", same as Boards 1-4).

PCB layout (stages 1-2, scripted - DRAFT, NOT FOR FABRICATION)

dock_interface.kicad_pcb (100x70mm, 2-layer) is generated by make_pcb.py from dock_interface.net - placement, scripted routing, bottom GND zone:

& 'C:\Program Files\KiCad\10.0\bin\python.exe' hardware\kicad\05-dock-interface\make_pcb.py
& 'C:\Program Files\KiCad\10.0\bin\kicad-cli.exe' pcb drc --output hardware\kicad\05-dock-interface\drc.rpt hardware\kicad\05-dock-interface\dock_interface.kicad_pcb

Verified 2026-11-04: KiCad DRC 0 violations, 0 unconnected pads (cleanest board so far - not even silkscreen warnings). Render: board5-layout-stage2.png.

Placement rationale: charge path J1 -> R1 (10mOhm shunt) -> J2 spring contacts runs short and wide (2.5mm top copper) across the north edge for the 5A max charge current. The INA226 (U1) is rotated 90 degrees so its Vin+/Vin-/Vbus pads face north and take short Kelvin taps straight off the shunt pads, while SDA/SCL/ALERT/A0 escape south, away from the charge path. LM2936 (U2) sits west next to J1. The solenoid driver (F1 5A blade fuse, Q1 IRLZ44N, D1 flyback, J4) is grouped across the south edge, its drain return on the bottom layer, well away from the INA226 front-end. Bench header J3 is on the east edge; no I2C pullups on this board (the pair lives on Board 3, per D3).

Still NOT FOR FABRICATION: blockers 2 and 5 below remain open (no reverse-current blocking on the charge path - ideal-diode ECO still to add; spring-contact mechanics/ratings are not proven by this PCB), and footprints/pinouts (INA226 MSOP-10, LM2936 SOT-223 variant, blade fuse holder, J2 contact assembly) need human peer review.

Blockers before fabrication

1. Who reads the dock INA226? RESOLVED (D3, 2026-11-04): the docked robot reads it over two extra spring contacts (J2 pins 4/5) and the dock INA226 is re-strapped to 0x41. Remaining firmware duties: the robot must configure the dock INA226 (calibration for the 10mOhm shunt, any alert thresholds) at dock/power-up - CHG_ALERT itself is not contacted, so the robot must poll. Bench use via J3 needs pullups on the test jig. Spring-contact signal integrity (I2C over spring contacts: contact resistance, bounce at dock/undock, bus capacitance) must be verified on the bench - keep dock I2C at 100kHz initially.
2. Unfused charge path (now 5A max, not 15A): charge current is confirmed at 2-3A standard / 5A absolute max (2026-11-04), retiring the earlier 15A assumptions: spring contacts 1-2 and the Board 3 J_DOCK path need only be rated/routed for 5A (layout to confirm). Shunt now correctly sized at 10mOhm/1W (5A -> 50mV, 0.25W; INA226 FS 8.19A). Still open: only the solenoid branch is fused (5A); the charger -> contact path relies on the external charger's protection, and there is no reverse-current blocking - a docked robot battery can backfeed an unpowered/faulted charger through the shunt. Add an ideal-diode/blocking MOSFET (rated >=20V per the charge-path target) or document the charger's own reverse protection.
3. Battery chemistry CONFIRMED (D1, 4S LiFePO4, 14.6V charge) - the 14.6V charger is correct and consistent with Board 3's "~12V" pack — confirm. The refill solenoid is also assumed to tolerate the 14.6V rail; a 12V solenoid will run hot.
4. Regulator margin RESOLVED (D6, 2026-11-04): AMS1117 (15V abs-max) replaced by LM2936-3.3 (40V max input, 50mA - INA226 needs <1mA), meeting the >=20V charge-rail target. No TVS or reverse-polarity device is fitted on the charger input itself; the LM2936 protects only its own logic rail (main-path reverse current is blocker 2).
5. Mechanics dominate this board: spring-contact rating (5A working, 15A-class preferred), contact alignment/float, and the AprilTag mount are mechanical design items represented only by J2 and a BOM line. Docking misalignment that shorts CHG+ to the sense contact is a credible fault — review contact geometry.