A WiFi-controlled rover for a woodchip-covered allotment plot in North Yorkshire. First job: drive it manually along the rows to capture video and photos, then let AI flag plant health issues, pest damage, weeds and watering needs that get missed on routine visits. Longer term: it maps the plot once, then navigates to named beds on its own.
The equipment involved, grouped by which phase it's for.
One board replaces the motor driver, the WiFi module and most of the wiring loom that sank the previous attempt at this project.
All 8 motor wires are power only — speed and direction are handled internally by the board via PWM, no separate signal wires.
Each phase has to prove itself before the next one starts — the previous rover attempt stalled on component complexity, so this build adds one capability at a time.
Goal: drive the rover from a phone browser over WiFi.
Active: WAV-23730 + 4 motors + LiPo battery
Goal: control the rover via the Dabble app or a custom interface, over Bluetooth.
Active: same hardware as Phase 1
Goal: the rover patrols the allotment on its own, captures images, and returns an AI analysis report.
New hardware: Raspberry Pi 5 (on the WAV-23730's GPIO header), Pi camera module (forward-facing), HC-SR04 ultrasonic sensor, RPLiDAR A1 (SLAM), Pi 5 active cooler
WAV-23730 keeps handling motor control; the Pi 5 takes on navigation, camera and AI analysis. ROS2 on the Pi runs SLAM and waypoint navigation, the Pi camera captures stills every few seconds during a patrol, and each image is sent to a vision AI (Claude or GPT-4 Vision) for a plant-health read.
The WAV-23730 stays the motor controller throughout — Phase 3 just adds a Raspberry Pi 5 on top for everything that needs real compute: mapping, navigation and vision.
Motor control and navigation stay on separate boards deliberately — the WAV-23730 keeps driving reliably even while the Pi is busy mapping or waiting on an API call.
The plot surface is soft, uneven, loose woodchip throughout — the mecanum wheels on the current chassis may struggle to get traction on deep or loose sections, and that's the first thing to actually test in Phase 1 rather than assume.
Fallback identified: a tracked chassis (T101D aluminium, 12V motors, Hall sensors) if the mecanum wheels prove unworkable. The WAV-23730 board is chassis-agnostic, so it would simply move across if a chassis swap is needed — no rework of the electronics.
A browser-based map showing the plot from above, named waypoints, the rover's live position, route planning, a manual override, a mission log and the AI analysis report — this is the target shape once autonomous patrol lands.
Keep it simple — the previous rover attempt failed on component complexity, not on any single hard problem.
Test before adding — each phase gets validated before the next one starts.
Integrated solutions preferred — the WAV-23730 replaces a driver board, a radio module and most of the wiring loom in one part.
Arduino IDE for Phases 1–2, ROS2/Python on the Pi for Phase 3 — the simplest toolchain that fits each phase's job.
AI coding assistance used throughout — so effort goes into hardware validation, not boilerplate.