Nova pcb v6 power v2

Fennec — Autonomous Quadruped

Fennec is an autonomous quadruped I'm building on the NovaSM3 platform (by Chris Locke / novaspotmicro.com) — rebuilt around a unified Feetech STS3215 TTL servo bus, NVIDIA Jetson Orin Nano Super compute, and ROS 2 Humble, as a platform for Vision-Language-Action (VLA) model deployment, 3D SLAM, and autonomous mobile manipulation.

Status: 🔧 Phase 1 — hardware bring-up + sim training. Both PCBs ordered 2026-07-01 (JLCPCB) — power board v2 (Pattern B bus master, INA226 rail telemetry, Phase-4 arm rail provisioned) + logic board. Firmware critical-path green (safety FSM, watchdogs, servo limits, joint-map loader; rostest suites in CI). Legs (leg_v6) printing; chassis-integration + CAD fit-gate audits closed; measured leg geometry → URDF + IK green. 🎉 RL locomotion walks in sim (MuJoCo MJX + Brax PPO, 2026-07-16) — stair-climb curriculum in progress. See STATUS.md for the live blockers / next-actions board. Platform: Quadruped (12 DOF) — arm (6-DOF, Phase 4 future) on shelf Compute: NVIDIA Jetson Orin Nano Super 8GB Middleware: ROS 2 Humble Last updated: July 22, 2026

Design outline: docs/design-outline.md — the canonical top-level design document (architecture, doctrines, lane status). Start there.

Table of Contents

1. Project Overview
2. Differences from Stock NovaSM3
3. Hardware Architecture
4. Software Architecture
5. Power System
6. Network Topology
7. Bill of Materials
8. Build Roadmap
9. Pre-Assembly Test Sequence
10. Setup Guide
11. Open Decisions
12. References & Credits

Project Overview

This project is a heavily modified fork of the open-source NovaSM3 quadruped, redesigned to serve as a research platform for embodied AI. Stock Nova uses PWM hobby servos and a Raspberry Pi running custom locomotion code. This build replaces:

• 12 servos with Feetech STS3215 TTL bus servos for unified protocol, real-time joint feedback (position, load, temperature, voltage), and elimination of PWM wiring complexity. Six additional STS3215 (arm) remain on shelf for Phase 4.
• The Pi with an NVIDIA Jetson Orin Nano Super (67 TOPS sparse INT8 / ~33 TOPS dense; 8GB RAM tight but workable for VLA inference + ROS 2 + SLAM — see Phase 4 notes). Real-time servo bus is offloaded to a Teensy 4.1 (Pattern B); Jetson runs ROS 2 + perception only.
• The stock perception with Intel RealSense D456 (depth + RGB + IMU) and Unitree L2 4D LiDAR (360° × 96° FOV, 30m range)
• The stock locomotion stack with ROS 2 Humble (Jetson) + Teensy bus master (Pattern B) + Nav2 + RTAB-Map / POINT-LIO

The arm is carried over from a prior SO-ARM101 build, also Feetech-based. v1 build scope: quadruped only (12 servos active). Arm install + integration is Phase 4 future work — bus IDs 13-18 and the 7.5V arm-rail buck footprint (D42V55F7) are reserved on the PCB v6 redesign so the future install is a populate-and-go.

Goals

• ✅ Unified TTL servo bus across locomotion (12 servos v1; future-ready for 18 with arm)
• 🔧 ROS 2 locomotion via micro-ROS bridge to Teensy 4.1 (Teensy owns the Feetech bus in Pattern B; FE-URT-1 retained as bench fallback)
• 🔧 RL locomotion policy trained in simulation (sim/nova_mjx, MuJoCo MJX + Brax PPO) — ✅ walks in sim (2026-07-16); terrain curriculum + stair climbing + sim2real transfer in progress
• 📋 3D SLAM using LiDAR + visual-inertial fusion (POINT-LIO baseline, RTAB-Map comparison)
• 📋 Autonomous navigation (Nav2) on legged platform with stair climbing
• 📋 VLA fine-tuning and deployment for mobile manipulation tasks
• 📋 Open-source documentation of every design decision for community reuse

Differences from Stock NovaSM3

Why redesign the PCB (v5.2b → v6)

After the v3.2 architecture audit, the stock Nova PCB v5.2b can't host the upgraded power tree (Pololu modules, INA226 telemetry, hard-cutoff MOSFET, E-stop chain) or the Pattern A/B bus master selector. The redesign retains the spirit of the stock board (Arduino Nano aux slot, battery input geometry) and adds the new safety + power architecture. Full feature set in hardware/pcb-mods/README.md.

What carries over from Nova v5.2b

• Arduino Nano slot for aux peripherals — v3.5 cut: OLED + WS2812B only (PIR / ultrasonic / MP3 / MPU-6050 removed; D456 + L2 perception stack covers their roles)
• Battery input + reverse polarity protection geometry (replicated with MOSFET-based reverse protection instead of diode)
• Power switch, voltmeter, fuse (MRBF-30 in Blue Sea 5191 block, OFF-board at pack — ~9 kA AIC @ 16.8 V clears this pack's ~1.5–3 kA dead-short with 3–4× margin; ANL's 6 kA rejected, Class T superseded 2026-06-12)

What changes

• All 12 PWM servo output headers removed (Feetech bus is daisy-chain TTL, doesn't use PWM)
• XL4016 buck footprints → Pololu D42V110-class module footprints
• Add: SN74LVC125A half-duplex driver (Pattern B prep), INA226 ×3, E-stop chain, hard-cutoff MOSFET, MRBF-30 fuse OFF-board (Blue Sea 5191 at pack — F1 not on PCB), bulk caps at injection points, bus-integrity footprints (series R + ferrite beads), reserved arm-rail buck footprint

Hardware Architecture

Block diagram

┌────────────────────────────────────────────────────────────────────┐
│                  NVIDIA Jetson Orin Nano Super 8GB                 │
│                          (ROS 2 Humble)                            │
│   • Nav2, SLAM, VLA inference, sensor fusion                       │
│   • Sends /joint_commands to Teensy via micro-ROS over USB         │
│   • Built-in WiFi 5 (802.11ac/ab/gn) per NVIDIA P3766 spec sheet   │
└──┬──────────┬─────────────┬─────────────┬───────────────────────────┘
   │ USB-C    │ USB 3.1     │ Ethernet    │ USB (micro-ROS transport)
   │          │             │             │
┌──▼───────┐ ┌▼───────────┐ ┌▼───────────┐ ┌▼─────────────────────────┐
│ Realsense│ │ FE-URT-1   │ │ Gig switch │ │ Teensy 4.1 (BUS MASTER v1)│
│ D456     │ │ (debug/    │ │            │ │  • Real-time gait loop    │
│ (depth + │ │  bench     │ │ ◄─ L2 ─►   │ │  • UART → SN74LVC125A → bus   │
│  RGB +   │ │  fallback) │ │            │ │  • INA226 I²C reader      │
│  IMU)    │ │            │ └────────────┘ │  • E-stop GPIO sense      │
└──────────┘ └────┬───────┘                │  • micro-ROS over USB     │
                  ↓                        └──┬────────────────────────┘
              (JP_BUS_MASTER                  │ UART → SN74LVC125A half-duplex
               solder bridge:                 │ (active v1 path)
               default = B,                   │
               flip to A for ID setup)        ▼
                                  ┌───────────────────────┐
                                  │ 12× STS3215 (TTL bus) │
                                  │ per-leg: FL1-3 FR4-6  │

Servo configuration

v1 build = 12 active servos on a single daisy-chained TTL bus, 2 active servo-power voltage levels (7.5V leg, 12V hip) plus dedicated 12V/2.6A L2 LiDAR buck and 12V/3A Jetson buck. Bus master is the Teensy 4.1 (Pattern B) via a SN74LVC125A half-duplex driver. Bus IDs 13-18 and the 7.4V arm rail (D42V55F7 footprint) reserved on PCB v6 for Phase 4 arm install — populate-and-go.

Bus master pattern

Pattern B is v1 default. Both paths live on PCB v6 via solder bridge JP_BUS_MASTER.

• Pattern B (v1 active): Teensy 4.1 hardware UART → SN74LVC125A quad tri-state buffer (half-duplex driver) → TTL bus pads. Bare-metal real-time loop at 200-500 Hz. Jetson sends joint targets via micro-ROS over USB; Teensy translates to bus reads/writes. Survives Jetson restarts, kernel preemption, CUDA stalls, journald flushes — none of which affect bus servicing. Solder bridge defaults to B.
• Pattern A (bench / debug fallback): Jetson → USB → FE-URT-1 → TTL bus. Use for: initial servo ID assignment with the Feetech FD / SCServo SDK Python tools from a workstation (simpler than booting micro-ROS for ID setup), debug if Teensy firmware misbehaves, post-mortem inspection of bus traffic. Flip JP_BUS_MASTER to A.

Why B as default (decision history): Linux is not a real-time OS. USB-CDC latency on Jetson runs 1-10 ms typical, 50 ms+ under load. What Pattern B actually buys: bus-servicing isolation on the Teensy side — UART transactions to all 12 servos complete on time even when Jetson is jittery, so the bus doesn't time out and servos don't fault. The gait controller still runs on Jetson and publishes targets at 100 Hz, so Jetson's command rate is still Linux-bounded; the Teensy oversamples at 200-500 Hz against the last received target, holding it through Jetson stalls. A 100 ms Linux freeze becomes "robot pauses mid-step," not "bus dies and robot falls." v3.2 originally defaulted to A with "migrate if measurement forces it" — v3.3 flips to B-default because the migration cost is just populating one $1 IC + Phase 1 firmware work, and skipping the measure-then-migrate path eliminates a Phase 2 surprise. See Open Decisions row 5.

Phase 1 acceptance gate (revised): Pattern B's real guarantee is bus-servicing isolation on the Teensy side, not full RTT through Linux. Two mandatory criteria + one sanity check:

1. Teensy local loop tick jitter p99 <100 µs over 60 seconds (bare-metal — easy if firmware is right)
2. /joint_commands arrival rate ≥99% of 100 Hz target over 60 seconds (Jetson + uROS healthy)
3. (Sanity) End-to-end RTT median <5 ms, p99 <20 ms — Linux-bounded, not a hard pass/fail

If (1) misses → debug Teensy firmware (DMA vs ISR, UART config). If (2) misses → debug Jetson uROS / USB cable / CPU contention. If only (3) is high → accept: Teensy holds last command under Jetson jitter, robot stays stable.

Software Architecture

ROS 2 stack (target)

┌──────────────────────────────────────────────┐
│             Application Layer                │
│  • Behavior tree / mission FSM               │
│  • VLA policy node                           │
│  • Teleop bridge                             │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│            Navigation Layer                  │
│  • Nav2 (planning + control)                 │
│  • robot_localization (EKF)                  │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│              Perception Layer                │
│  • RTAB-Map / POINT-LIO (3D SLAM)            │
│  • realsense2_camera                         │
│  • unitree_lidar_ros2                        │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│            Locomotion Layer (Jetson)         │
│  • Gait controller (8-phase walk)            │
│  • 3-DOF-per-leg IK solver                   │
│  • Publishes /joint_commands                 │
│  • Subscribes /joint_states                  │
└──────────────────────────────────────────────┘

Key software components

Power System

Voltage rails

4S LiPo 14.8V nominal (12.8-16.8V)
   │  MRBF-30 fuse (off-board) · MOSFET reverse-protection · MOSFET hard-cutoff @12.4V · E-stop NC (servo rails only)
   │
   ├── Pololu D42V110F7  ──► 7.5V/10A ──► 8× STS3215 19kg femur/tibia
   │       (star injection at 4 points along chain, bulk caps near point of load)
   │
   ├── Pololu D42V110F12 ──► 12V/9A   ──► 4× STS3215 30kg hips ONLY
   │
   ├── Pololu D24V22F12  ──► 12V/2.6A ──► Unitree L2 LiDAR (1A, LC filter on output)
   │
   ├── Pololu D42V55F12  ──► 12V/~3A  ──► Jetson Orin Nano (barrel jack)
   │
   ├── [reserved D42V55F7] ► 7.5V/3-8A ──► 6× STS3215 arm (Phase 4 — footprint unstuffed)
   │
   └── UBEC 5V/5A ──► 5V rail ──► fans, aux 5V peripherals

   INA226 ×3 (leg 7.5V, hip 12V, Jetson 12V) ──► I²C ──► Teensy 4.1 ──► ROS 2 diagnostics
   13.0V comparator ──► Teensy GPIO ──► /battery_low ──► Jetson clean shutdown
   12.4V comparator ──► MOSFET ──► breaks battery feed (autonomous backstop)

Notes

• All rails share a common ground
• Jetson MAXN peak power ~25W → 2.1A at 12V, plus USB peripherals (D456 streaming ~2-3W, FE-URT-1 ~0.5W) → ~2.5A continuous worst case. Pololu D42V55F12 derates to ~3A continuous at 14.8V Vin (4.5A typ headline is at 42V Vin) → ~1.2× headroom. Modest, not generous; watch thermal under sustained VLA inference + RealSense stream + Nav2 planning. Min Vin 12V — set LiPo LVC alarm at 3.3V/cell = 13.2V to stay above dropout.
• Leg rail D42V110F7 (10A typ @ 42V Vin, derates at 14.8V Vin) sized for walking-gait avg 5-8A with bulk caps absorbing 25-40A impact transients near each star injection point. See docs/power-budget.md.
• Hip rail D42V110F12 (9A typ @ 42V Vin) sized for 4× 30kg hips only — sustained avg ~8A. L2 LiDAR was moved off this rail to a dedicated D24V22F12 buck (v3.4) because combined hip+L2 load was margin-thin under 14.8V Vin derating.
• L2 LiDAR rail D24V22F12 (2.6A / 12V / 36V Vin max) — dedicated buck for the LiDAR's 1A draw with LC filter on output. Clean power, no servo transient ringing.
• Battery low-voltage chain (ordered by trip point):
  1. 13.2V — 608AC charger LVC alarm beeps (user-facing warning)
  2. 13.0V — graceful-shutdown comparator → Teensy /battery_low topic → Jetson systemctl poweroff (clean SD unmount). ~30-60s window before hard cutoff.
  3. 12.4V — autonomous MOSFET hard-cutoff on battery feed (drops everything; Jetson should already be shut down per #2)
• E-stop kills leg + hip + L2 rail enables (LiDAR stops spinning) — Jetson rail stays alive for post-mortem debug + telemetry capture.
• L2 self-heats below 30°C ambient; ~30-60s delay before point cloud output on cold boots
• LC filter sits on the D24V22F12 output to clean any switching ripple before the L2 input (UDP packet loss is the failure mode)

Battery safety

• Always charge inside a LiPo safe bag
• Two 4S packs alternated; storage-charge to 3.8V/cell if unused >1 week
• Label packs (A/B) with first-use date; track cycle counts

Network Topology

Point-to-point — no switch on the robot (decision 2026-08-14).

┌──────────────────────────┐               ┌────────────────────────┐
│ Unitree L2 LiDAR         │     Cat 6      │ Jetson Orin Nano       │
│ Static IP: 192.168.1.62  ├────────────────┤ eth0: 192.168.1.2/24   │
│ UDP target port: 6101    │    (direct)    │ wlan0: DHCP from home  │
└──────────────────────────┘                └────────────────────────┘

• L2 → UDP port 6101 on the Jetson, direct on eth0
• Jetson's eth0 needs a manual static IP because L2 is not a DHCP client/server
• WiFi (built-in WiFi 5 / 802.11ac module on P3766 kit) handles dev SSH access + internet, leaving Ethernet free for the LiDAR. BT presence not explicitly listed in spec — verify on arrival.
• The 5-port switch is bench gear only — a dev laptop can join via the switch on the bench when wanted; it is not mounted on the robot.

Bill of Materials

Full BOM lives in BOM.md. High-level summary:

Build Roadmap

Superseded — this roadmap is the May-2026 plan kept for history. Live state: STATUS.md (boards fabbed + populated to stage 9, firmware links, legs printing).

Phase 0 — Pre-build setup (current)

3-week sequenced plan in docs/work-schedule.md. Front-loads leg-joint CAD + prints; reserves 2026-05-29 away-week for laptop-only PCB schematic work in KiCad.

• Define modified BOM (v3.4: Pattern B default, Pololu 4-buck split incl. dedicated L2, full safety, arm deferred)
• Validate component compatibility (Jetson power rail, Feetech bus, L2 ethernet)
• Pick LiPo charger → ISDT 608AC
• Power rail audit (XL4016 → Pololu D42V110-class)
• Confirm v1 scope = quadruped only (arm → Phase 4)
• Bus master decision → Pattern B (Teensy + 74HC125) v1 default
• Set up GitHub repo with this README and BOM

Week 1 (2026-05-15 → 2026-05-22):

• Leg CAD complete — V5 OpenSCAD original-shell-carve (hardware/cad/leg_v5/, STS3215 cavity in original NovaSM3 STLs). Pending: caliper on-hand servos + first-article leg-joint print.
• Install KiCad 8.x + Pololu library on laptop; cache datasheets for offline use during away-week
• Back up LeRobot Pi SD contents
• Create NVIDIA Developer account, download JetPack 6.x image

Week 2 (2026-05-22 → 2026-05-28): detailed checklist in docs/checklists/week-2.md

• Carryover from Week 1: caliper pass + first-article leg print (V5 leg CAD itself is done)
• ROS 2 Humble + librealsense2 + unilidar_sdk2 on Jetson (Jetson already flashed + persistent — Week 1 head-start)
• PlatformIO + TeensyDuino + micro-ROS Teensy firmware skeleton (compile-green)
• KiCad install + Pololu library (PCB v6 prep for Week 3)
• Bambu Studio + PA6-CF drier preheat

Week 3 — Away (2026-05-29 → 2026-06-05, laptop only):

• PCB v6 schematic captured in KiCad 9 (2026-06-03) — §1-§8 hierarchical, all 54 parts footprinted, pin-name wiring audit passed, ERC-clean (1 intentional warning: reserved arm rail). See hardware/pcb-mods/README.md
• PCB v6 layout + DRC + Gerber export in KiCad
• Backup: continue Teensy firmware / URDF / ROS 2 scaffolding if PCB stalls or finishes early

Week 4+ (back to shop):

• Bench-validate prints; submit Gerbers to PCBWay (~$60) — both boards fabbed at JLCPCB 2026-07-01 (~$203 all-in, see STATUS.md); continue Phase 1 hardware bring-up
• Order remaining parts (switch, Pololu bucks ×4, charger bundle, safety parts, accessories) — NVMe deferred

Phase 1 — Hardware bring-up (weeks 1-4)

• Flash JetPack 6.x to microSD; firmware update Jetson if needed
• Boot Jetson, complete Ubuntu setup
• Verify pre-installed WiFi works; install AX210NGW only if missing. Confirm BT presence.
• Install ROS 2 Humble + librealsense2 + unilidar_sdk2 (run from 128GB microSD; NVMe deferred)
• D456 standalone test (realsense-viewer)
• L2 standalone test (included 12V adapter + rviz2)
• Print parts: Bambu P1S + PA6-CF (dry 24h before each print)
• Bench-validate Pololu D42V55F12 / D42V110F7 / D42V110F12 / D24V22F12 / UBEC 5V (per BOM §12 step 2)
• Verify E-stop chain + MOSFET hard-cutoff @ 12.4V + INA226 I²C reads
• Write Teensy firmware (Pattern B critical path) — micro-ROS client + SCServo SDK port + SN74LVC125A TX/RX gating
• ID setup pass: flip JP_BUS_MASTER to A (FE-URT-1), assign servo IDs 1-12 (v1 active), label each. IDs 13-18 reserved for Phase 4.
• Flip JP_BUS_MASTER back to B (default). Verify Teensy firmware drives single servo, then full 12-servo chain.
• Acceptance gate (two mandatory + one sanity, see "Bus master pattern" section): Teensy tick jitter p99 <100 µs + /joint_commands arrival ≥99% of 100 Hz + RTT median <5 ms / p99 <20 ms. Pattern A is not a workaround if (1) or (2) miss.
• Assemble legs (redesigned for STS3215 dimensions)
• Assemble chassis, mount L2 on top-center riser
• Network setup: eth0 static 192.168.1.2; verify L2 UDP flow
• Arduino Nano: strip code to aux peripherals only
• Stand / sit test on hardware (no walk yet)

Phase 2 — Locomotion (weeks 5-8)

Two parallel tracks: the classical IK/gait stack below, and an RL policy trained in simulation (sim/nova_mjx — MuJoCo MJX + Brax PPO, procedural terrain curriculum, domain randomization bracketing measured servo behavior). The RL track walks in sim as of 2026-07-16; its deployment path is ONNX/TFLite-class policy export to the Jetson.

• Implement 3-DOF-per-leg IK solver (reference mogar/spot_micro)
• 8-phase walk gait controller on Jetson — publishes /joint_commands to Teensy via micro-ROS
• Stand, sit, walk validated on hardware
• Body stabilization feedback via D456 IMU (MPU-6050 cut per BOM v3.5)
• Calibrate LiDAR ↔ RealSense extrinsics
• EKF fusion: D456 IMU + L2 IMU via robot_localization (MPU-6050 cut)
• RTAB-Map 3D SLAM on walking platform
• POINT-LIO comparison (LiDAR-IMU only)

Phase 3 — Autonomy (weeks 9-12)

• Nav2 integration on legged platform
• Costmaps from D456 + L2
• Autonomous waypoint navigation
• Obstacle avoidance during walk
• Stair climbing test (gait variation)
• Telemetry dashboard

Forward-looking backlog (opportunistic, not scheduled)

Quality-of-life software + operational features captured in docs/notes-qol-features.md and docs/notes-virtual-view-autocal.md. Pick up during Phase 1/2 idle time or batch into a "QoL sprint" once v1 walk gait is stable. Suggested rollout order (per notes-qol-features.md):

1. Preflight health check (nova_ops) — bus ping sweep, E-stop, battery latch. v1 ships 3 checks; rest land opportunistically. Highest payback during Phase 1 servo bring-up.
2. make deploy for Teensy — build on laptop, flash over Jetson USB, no chassis open. Needed by mid-Phase-1 firmware iteration.
3. Always-on MCAP dashcam with incident freeze on E-stop / fault / manual trigger. Required before first walk attempt.
4. nova bringup launcher with profiles (bench / sensors / slam / walk / full / vla) — pay back at ~6 launch files.
5. Per-joint safety envelope in the gait controller — position/velocity/load/temp clamping. Paired with first Phase 2 gait controller commit.
6. Battery SoC widget — gated on adding a 4th INA226 on battery feed (PCB v6 footprint already supports it).
7. RGB status LED on Arduino Nano — Phase 2 polish.
8. Telemetry CSV writer → optional Grafana later (resist Docker on Jetson).
9. Bag replay harness — Phase 3 when SLAM/Nav iteration is the bottleneck.
10. Gazebo Fortress digital twin — Phase 4 prep for VLA dev without HIL.

Plus auto-calibration routines (nova_calibration package): IMU bias zero on boot, servo zero-position auto-detect via jig, camera↔IMU + LiDAR↔IMU + LiDAR↔camera extrinsics. Foxglove bridge over Tailscale for remote browser viewing of URDF + sensors + plots.

Phase 4 — Arm install + Manipulation + VLA (future, after Phase 3 stable)

Split into two sub-phases since hardware-arm-install precedes any manipulation software.

Phase 4a — Arm hardware install:

• Populate D42V55F7 arm-rail buck on PCB v6
• Install 6× STS3215 arm servos (carried from SO-ARM101), assign bus IDs 13-18
• Mechanical mount of arm to chassis (CAD pending)
• Full 18-servo daisy-chain ping test
• Bench-validate arm rail under realistic load

Phase 4b — Manipulation + VLA software:

• Arm joint state integration into unified URDF
• MoveIt 2 motion planning for arm
• COM-shift compensation in gait controller (arm extension + payload mass shifts support polygon — gait controller needs arm-state input)
• VLA model selection (OpenVLA / Pi0 / RT-2 class). Constraint: 8 GB RAM shared with ROS 2 + SLAM + RealSense full rate. OpenVLA INT4 fits but tight; budget memory carefully.
• Data collection harness for in-house fine-tuning
• On-device VLA inference (TensorRT-optimized)
• Mobile manipulation demo: navigate to object, grasp, transport

Pre-Assembly Test Sequence

Before assembling any subsystem into the chassis, validate on the bench. Catching a bad part on the desk is minutes; catching it after assembly is hours.

1. Jetson desk bring-up — Firmware update → JetPack flash → Ubuntu setup → SDK installs → individual sensor smoke tests (NVMe migration deferred — see BOM §1)
2. Power rail validation — Each buck/converter loaded with realistic current draw before being committed to the chassis
3. Servo bring-up — Single-servo bench tests before daisy-chaining; full chain ping before powered movement
4. Network smoke test — Static IPs, L2 UDP packet flow, simultaneous SSH-over-WiFi
5. Sensor smoke test — OLED + WS2812B via Arduino Nano USB-serial (PIR/ultrasonic/MPU/DFPlayer cut per BOM v3.5)

Full test sequence and acceptance criteria in BOM.md Section 12.

Setup Guide

📋 To be written as components arrive. Will include:

• JetPack 6.x flash + firmware update procedure
• NVMe rootfs migration (when NAND prices recover; see BOM §1)
• Built-in WiFi 5 setup + BT verification
• ROS 2 Humble + sensor SDK install scripts
• Feetech servo ID assignment procedure
• Network static IP configuration
• URDF + xacro for the modified platform
• Calibration procedures (servo zero-positioning, sensor extrinsics)

Open Decisions

References & Credits

• Original NovaSM3 by Chris Locke (cguweb) — GitHub · novaspotmicro.com · Discord — base platform, chassis geometry, parts list. PCB v5.2/v5.3 reference via PCBWay community share. Note: repo ships STL only, no STEP (re-model from STLs + datasheet + calipers for OnShape).
• mogar/spot_micro — ROS 2 reference for quadruped IK and gait
• discodyer/unitree_lidar_ros2 — ROS 2 wrapper for Unitree L2
• Feetech SCServo SDK — TTL bus driver
• NVIDIA Jetson AI Lab — JetPack documentation, VLA model deployment guides
• POINT-LIO — Unitree-recommended LiDAR-inertial SLAM
• Nova Discord community — Build gotchas, assembly tips

Project Status Log

This is a learning project as much as a research one. Decisions and tradeoffs are documented openly so others can fork and adapt.