But ultimately the control system has nothing more than input voltage to the galvanometers to steer them, and depending on the source of inaccuracy they may or may not be able to overcome it.
Fast, close-to-linear response over a short range can be cheap; the overall linearity you get from your $1000 gadgets is unnecessary. Fixed calibration targets could help compensate for broad nonlinearity and also drift from, e.g., temperature variation. Varying focus as you home on the target with a spiral path forgives a lot of initial inaccuracy. The system could refine its response curves with each kill, to home in faster; calibration targets might not be needed if unfocused illumination is forgiving enough.
It doesn't need a 100% kill rate in the first 10 minutes. Indeed, in a usefully big volume it gets plenty of attempts on each target. So, there is plenty of time for the system to tune itself to its own hardware.
The reason I ask is because after seeing the OP video (which is remarkable) I am left thinking that in terms of getting a cheap MVP to the market maybe a ballistic system of some sort would be preferable.
Take the same approach of using a high speed camera but using a pneumatic launcher shoot a handful of grains of sand at high speed. There are obvious downsides to this approach but the risk to human health is lower. Thoughts?
Sand grains suffer from a large surface area relative to their mass, so their speed would drop off very quickly. I have one of those bug-a-salt guns and they kinda suck at anything more than 10 feet (also bugs seem to be pretty resilient to that kind of damage). Obviously you could make one that is more precise but I feel like there would be a lot of new variables to deal with.
Some kind of beam steering ultrasonic setup might work as well (although probably expensive) If you can get 10-15 transducers to pop off a precisely timed 5W burst of say 60kHz sound such that they constructively interfere at the critter, you might be able to get them to disassemble in air without any moving parts on your rig.
Maybe you only deafen them, so they die of old age without mating.
I once priced out an array of half-inch-sized ultrasonic transducers. Seems like they were astonishingly cheap... like under $.50 each? Plus $.50 microcontrollers and a drive transistor for the spike. Maybe an FPGA running parallel convolutions?
I would expect a sandblast to be worse to get in your eye than a flash of light. So, you would still need good large-body sensors. You would need to lead the target, another complication, and it would be hard to know how close you had got. The ones that miss would keep going and come down somewhere, although not fast. The range would be very limited because of how fast the grains would slow down -- drag goes by the area, but kinetic energy by the mass, hence volume, so smaller projectiles lose it very fast.
To me the largest engineering challenge is actually the identification step, where you'd need to get sufficient return to positively identify the critter. Would be interesting to see how they pull this all together.