The technical hurdles between the drawing board and the first drive

The hurdle nobody photographs#

When people picture a build, they picture parts. In mid-August there were no parts — not one axle unit, no motors, no computer, not even a repository to keep notes in. Everything below is paper and a spreadsheet, and I want to be honest about that, because it changes what these numbers mean: they are design targets and decisions, not measurements. The vehicle did not exist.

The drivetrain: 32.3:1 was too slow#

The first real decision was how much reduction the drivetrain should carry. The axle units come from a large-scale hobby chassis and bring their own differential; the stock unit has a 3.23:1 reduction. Combined with the drive motor's built-in 10:1 planetary gearbox, that is 32.3:1 in total. Torque is generous, but the arithmetic said something uncomfortable: at the motor's rated 120 rpm output the wheel turns at 37 rpm — about 1.26 km/h on the 180 mm wheels. That is a crawl. It is fine for parking, and useless for crossing a production floor or leaving a room in a hurry. An inspection robot that cannot retreat briskly is a safety problem, not a performance one.

So the stock differential was dropped. The design now drives a bare 1:5 differential from a large-scale buggy, run without its own housing, inside a custom housing that adds a 1.5:1 spur stage. Total reduction: 15:1. At the motor's rated output the wheel turns at 80 rpm — 2.71 km/h, more than double, with a no-load ceiling around 4.5 km/h. The torque budget still holds: at the axle, 3 N·m × 1.5 × 0.9 ≈ 4.1 N·m rated, split evenly by the open differential to 2.0 N·m per wheel — 22.5 N per wheel, 180 N across all eight wheels. The price was a new part: the custom housing exists on paper, not in metal. The planning hands it to the build phase.

The pinion: buy it, then prove the mesh#

The motor output reaches the differential ring through a pinion, and the planning document is blunt about it: verify engagement against the real ring before running. A gear mesh is decided by the two parts touching, not by a drawing. The pinion was an easy trap: a printed one is a weekend job, and at this torque it is a wear part with a short life — a pinion failing mid-mission is exactly the failure you do not want. So it became a bought steel part on a small adapter for the motor flange. The verification is scheduled for the first assembled axle, which is why the whole thing is built to come apart.

The torque question: margin, not peak#

The drive motors are rated 3 N·m continuous and 7 N·m peak at the gearbox output (24 V, 2.5 A nominal, 7.5 A peak; 120 rpm rated, 200 rpm no-load). For a vehicle of 20–35 kg, the flat-ground demand is small: rolling resistance is a few percent of the weight, and the gradient term grows with the slope. Four drive motors, one per axle, push all eight wheels — the vehicle total is around 180 N rated and over 400 N peak.

The number that mattered most was not the peak but the margin at continuous torque. The controller commands 2 N·m per drive motor — two thirds of rated — and keeps a third in reserve on purpose. The real rolling resistance is unknown until the real tyres meet a real floor; margin is how the design survives that first contact. Ten motors in total: four drive, four steer, two on the camera head.

The battery#

The energy store is a 24 V 6S2P module built from 49.5 Ah cells — about 99 Ah in total. The module family originally powered a well-known compact electric car, and mine came second-hand from the Baltic region. That fits the philosophy: reuse what exists, keep the cost where a hobby project can carry it. The reference to battery production is real — it is what the robot is built for — but the cells come from a second-hand market, not from a factory floor. Sizing was endurance-first: approach, observe, listen, withdraw. The vehicle's appetite at walking pace is modest, and the headroom is the point.

The simulation that stood in for the bench#

With no hardware, the software had to be built against a stand-in, and the stand-in had to be honest. The rule that shaped everything: every subsystem without hardware simulates, and every simulated value is clearly marked as such. That made the loop fast and honest — the control loop, the watchdog and the operator interface were all exercised before a single motor was connected. When the prototype arrives over the next months, the software that meets it will have already driven a virtual hall thousands of times.

What these hurdles had in common#

None of them could be solved by buying something. They were decisions, made on paper, with margin left for reality. The next hurdle is the good kind: the first axle, assembled, pinion meshed with the real ring — the sentence from the planning document finally tested against metal.