But yes, industry uses optical technology in large CNCs all the time. Some homing switches are physical clicky switches, some are inductive proximity sensors, but an optical 'horseshoe' througbeam/fiberoptic sensor is the standard for highly repeatable sensor-based homing. Depending on your control system, homing to a hard stop by measuring motor torque can also be highly effective, then you don't even need switches.
I've personally worked on a number of CMMs (coordinate measuring machines, basically a CNC with a probe tip for checking that something was machined within tolerances instead of a spindle for actually cutting it) that use 90V DC motors and Heidehein glass scales for positioning. We calibrate them using laser interferometers; another optical technique - just with a single beam. Those same CMMs are being phased out across the industry in favor of optical measurement systems, just because they're faster.
Shane of the excellent "Stuff Made Here" Youtube channel recently made a big CNC painting robot that did optical tracking for a coarse positioning stage and used steppers for the local stage: https://youtu.be/osUTMnDFV30
The OP is coming from the CNC milling world where positional accuracy is more or less solved: you use cheap steppers or servos but expensive, precision-ground ballscrews; then you swear on Machinery's Handbook not to drive your machine too fast.
The real demons in dimensional accuracy come from things like spindle runout, deformation of the tool, and flexing and vibrations of the machine, fixture and workpiece (that are often different going in one direction than another!). Certain operations like drilling can't be corrected in real time. There are additional concerns like minimum amounts of material that can be removed with each pass - too little and you are just burnishing the workpiece.
Machinists actually do solve these problems in software when they generate toolpaths and fixturing.
For comparison, a typical ball screw, has positioning accuracy/resolution of around .02mm
It would be useful if CNC machines did a depth scan of the bed before starting to cut, to make sure that the machining plan didn't run the tool into a clamp or something. Not super high precision, just enough to check clearances.
Most of the headaches of CNC machining involve getting the workplace, tools, and clamps in the right place. Only then can the machine do its thing by itself. Help in that area would make CNC cutters more accessible to amateurs.
Some CNC machines intended for unattended operation have microphones or MEMS accelerometers listening to the cutting, to detect when something has gone wrong, like a worn-out tool. Monitoring spindle torque is common. Too little means the tool broke. Too much means the tool wore out.
It uses brushless motors and encoders.
Quite interesting, thanks for mentioning. They attempt to directly emulate stepper motors with brushless motors and encoders.
If you use a crude algorithm, maybe. But given what's possible with photogrammetry, I think you can use that as another sensor fusion input, with a kalman filter or something similar, and get even more precision at the output.
That's way more complicated when it comes to the algorithm, of course. But the idea was being able to plug more sensors and improve precision.
Optical tracking might be useful, but it also might not for the reasons I'm about to describe
CNC machines generally get "better" as they get bigger, not because they have stronger motors, but because they are more rigid. The problem with this CNC is not power, its rigidity. It will cut well enough for wood and plastic, assuming that the spindle is fast enough. However it will deflect significantly as the material "pushes back" against the cutting force.
In this machine, they are using belts an pulleys (which are contrary to common opinion not very elastic) The pulley used to keep the belt in tension are on plastic parts, which are elastic. (steel is also elastic, but much stiffer, so at this scale wouldn't flex as much, well not before the bolts bend)
With all this flex, it affects the accuracy during the cut. you might have felt when a drill bit binds up in a hole, and the whole drill is wrenched from your hand. That happens in a CNC when starting a cut, or plunging into a pocket.
Where big machines come in, is that they have huge mass, which allows for large rails/linear bearings, which are much more rigid. The more mass also means that when the end mill comes under load, the work piece gets the push back because its generally less mass than the gantry.
TL;DR:
yes you can use optical methods for homing, but its better to spend the money on more mass for the x/y/z assembly.
Has anyone tried to use cameras or a Valve Lighthouse (0.3mm precision), maybe with accelerometers and encoders, for tracking? That would allow the use of cheaper, faster, torquier, more efficient DC motors, as well as release the accuracy constraints for a lot of parts (depending on which part is being tracked).
The goal would be to trade hardware complexity and price for software complexity, since it's easier to re-purpose software (and something like lighthouse base stations has multiple uses, so the price could be shared between projects).