Choosing a drive architecture for a low-profile wheel module

Every wheeled-robot drive architecture decision comes down to the same three constraints fighting each other: torque density, outer diameter, and how much of the budget and schedule you're willing to spend proving it out. Working on VARK's wheel module, I went through this loop more times than I'd like to admit before landing on a custom BLDC winding driving a planetary reduction stage, and it's worth writing down why, because the "just buy a hub motor" instinct is a strong one and it's wrong more often than it looks.

Start with the constraint that doesn't move

For a hot-swappable wheel module, the outer diameter is basically fixed before you pick a motor: it's set by the wheel size the platform needs for the terrain it's meant to cross. That's the constraint that doesn't move. Everything else (torque, current, thermal budget) has to fit inside a diameter you've already committed to.

Off-the-shelf hub motors are optimised for a different set of trade-offs: they're built for e-bikes, scooters, and general-purpose robotics platforms where the diameter is more negotiable and the volume is high enough to justify a fixed design. When I actually laid candidate hub motors against the module's target envelope, every single one forced a wheel diameter I wasn't willing to accept, either too large for the platform's footprint target, or too small to hit the torque figures the terrain demanded.

The custom path is a bigger bet than it looks

Deciding to wind a custom BLDC motor and design a planetary reduction stage from scratch is not a small decision, and I don't think people underrate the engineering time enough when they consider it. Winding design alone went through several iterations before cogging torque was low enough to be acceptable at low speed, a problem that simply doesn't exist if you buy a motor someone else already tuned. The planetary stage went through its own iteration loop, most notably after gear whine and visible wear showed up in gen-1 testing, which sent me back to the tooth profile and added a sealed grease cavity for gen 2.

None of that is free. If I total the hours against "buy a hub motor and accept a worse wheel diameter," the custom path cost months I wouldn't have spent otherwise.

Why it was still the right call

The reason it was still right: the wheel diameter is the module. A hot-swappable, multi-module platform lives or dies on whether the module envelope is right, because that envelope is the thing that gets repeated four, six, eight times across a chassis. Getting it wrong once and repeating the mistake across every module on the machine is a much worse outcome than spending extra months getting the drive right the first time.

The general version of this lesson: when a constraint is genuinely fixed (not "hard to change" but actually fixed by the rest of the system) it's worth asking whether the standard, low-effort component actually satisfies it before defaulting to it. Sometimes it does, and you save yourself months. Sometimes, like here, it doesn't, and the custom path, expensive as it is, is the only one that gets you a module you can actually repeat across a machine.