They should be sweating, because if the other side can fire 100 rockets for $10k that are close enough to not immediately and obviously be off target, and you don't know whether a more expensive one with actual explosives is hiding within that barrage, you now have 100 targets to try to intercept, and suddenly your costs have gone up dramatically while the other sides costs has barely moved.

For a sub-minute flight the drift budget is actually pretty forgiving — a MEMS gyro drifts maybe 1-3 deg/sec, and if you're fusing with accelerometer data you really just need "which way is up" and "am I still pointed at the target." A $5 IMU can hold that for tens of seconds.

Where you're right is repeatability. Mil-spec works the same on launch 1 and launch 500 across temperature extremes. Consumer MEMS you'd need to characterize each unit individually — fine for a demo, impractical at any scale.

For different definitions of cheap though.

While the pure gyro/accelerometer stuff does suffer from major problems the improvements in SLAM using just cameras in the last 15 years are insane.

You can calibrate any sensor, its just a manufacturing step, and while cheap ones may be inaccurate and drift over time, I'm pretty sure the good enough ones (which cost tens of dollars, not fractions of a dollar) are accurate enough to work for the seconds-to-minutes flight time of a rocket like this.

But do they drift enough to hit girls schools?