From simulation to the yard: the plan, the parts and the honest business case
Chris went to watch TV and left one instruction: keep going. Make it real: a parts list for the yard, electrical and mechanical engineering, a business plan, an open-source path, a website, all in one monorepo. Here's what we researched and decided, and the one market fact that changed the plan.
The fact that changed the plan
A wire-free RTK robot mower, the kind that navigates by centimetre-level GPS with no boundary wire, now sells for about $669–1,099 (Segway's Navimow i105N, depending on promotions). Our sensor parts list alone, a Raspberry Pi 5 at $175 after the 2026 memory-driven price rises plus a u-blox ZED-F9P RTK board at €172–211 (≈ $195–235), is most of the way there before we've bought a motor.
So Mowbot can't win as a cheaper mower. The business plan (docs/business/business-plan.md) is built on what closed products can't
offer: you can read the code, there's no cloud, you can repair it, and it's a complete learning path. The order is
open project first, then education, then a blade-less developer kit. A certified consumer product comes only
with funding, insurance and a manufacturing partner.
The hardware plan: brain transplant first
The dangerous parts of a robot mower are the blade guard, the body and the battery, and they're mechanical rather than software problems. Cheap commercial mowers have already solved them and been through certification. So the first real robot is a used YardForce Classic 500 with the open-source OpenMower mainboard and our ROS 2 stack, the best-supported conversion target in that community. Stages:
| Stage | What | Blade |
|---|---|---|
| H0 | Simulation (now) | simulated |
| H1 | Pi 5 + RTK + IMU on a garden cart; survey the lawn boundary | none |
| H2 | Brain transplant: drive, map, dock | unplugged |
| H3 | After the safety checklist passes | on, supervised |
| H4 | Custom chassis, only if needed |
Engineering numbers, worked from physics
The electrical design (hardware/electrical/) sizes the drive: a 15 kg robot on a 15° lawn slope at 0.5 m/s needs about
52 N of traction and ≈ 4 N·m per wheel at 40 rpm, so geared or hub motors. The energy budget is about 55 W
average (with the blade as the big unknown), so two hours of mowing plus a 30% reserve is about 150 Wh: a 7S, 6 Ah
pack.
The most important circuit is the blade interlock: a series chain of normally-closed contacts (e-stop, lift, tilt, an MCU watchdog pulse, and the ROS request) feeding the blade controller's enable line. A cut wire reads as unsafe. A frozen program can leave a pin stuck high, but it can't keep toggling one. Software decides when to mow; hardware decides when it may.
The sim fed straight into the mechanical spec: the blade reaction-torque finding made "spin the blade up before moving" a firmware requirement.
What we couldn't verify
Honest gaps are written into the docs. We haven't read the paid safety standard, IEC 60335-2-107 (one summary says the blade must stop within 2 s of a lift; unconfirmed). We found no reliable used price for the YardForce. Free RTK corrections depend on your state. And the name "Mowbot" needs a trademark search before any public launch.
The monorepo now
src/ (robot software) · analysis/ · hardware/ · docs/business/ · blog/ · website/ · observability/.
The website is plain HTML, CSS and JS. Its hero is a canvas of the mower cutting stripes inside an RTK geofence,
drawn at the sim's real scale (40 px per metre, a 0.36 m blade swath).
Next: back to the robot. Phase 3 gives the simulated mower an IMU and GPS, and makes it work out where it is.