My bad here regarding switchblade. It is a drone & loitering munition both. However, it does not change anything about other aspects. Its maximum speed is 100 & it takes at least 30mins to reach 50km which is long enough for nature's forces to act.

Let me give an example of MEMS INS of Honeywell:

View attachment 254107
Out of the 3 models, the first one is most accurate. So, I will take that:

It is a 6 directional IMU/INS with 3 direction of gyro & 3 direction of accelerometer.

Let us see the gyro deviation first:

Firstly, bias is the initial variation caused by various factors & can't be predicted : The gyro bias is 0.25 degree per hour standard deviation. Taking the standard deviation itself as the actual error, it would be 0.25degree of 0.0044 radian/hour.

Taking speed as 100kmph, distance 50km, time 0.5hr the deviation will be 0.0044*50*0.5 = 0.11km

Now, the random walk (Gyro ARW): 0.04degree/sqrt(hour). Since time is 0.5hr, sqroot is 0.71. So, the net ARW is 0.04*0.71 = 0.0284 degree or 0.0005 radian. Multiplying with 50km= 0.025km.

Net deviation in a direction is 0.135km

This happens in 3 directions. so we have to use formula sqrt(a^2 + b^2 + c^2) which gives minimum deviation of 0.235km

Next to accelerometer/velocity deviation:

The accelerometer bias deviation is hard to predict as it will depend on the maneuvering being done. As for velocity deviation, it is 0.03*0.71= 0.0213m/s

given the speed is 100kmph or roughly 28m/s, this will be ~0.6m/s. Over 30 minutes (1800s), the deviation would be nearly 1km.

The deviation from accel & gyro have to be added as sqrt( a^2 + b^2), we get 1.1km.

The operating conditions will be harsher than specified which can make its performance deviate even more.

Even assuming that the gyro & accelerator deviations are consolidated and recitified, we can expect at least 1% or 0.5km error at the least.

Since complex calculations & assumptions are involved, it is possible that I have erred. If there is any errors you can find from my side, let me know