Also just want to mention that, yes, integration errors accumulate when using intero-receptive sensors but if errors are small enough (white noise, various biases, sample rates, quantization, etc.) from the inertial sensors an odometry solution might be adequate until an extero-receptive sensor can localize the sensor within an external frame.
This can shift the discussion from solving a problem that has no solution (i.e. how do I integrate a signal with white noise without any error) to an engineering problem (i.e. what error parameters allow the odometry to be accurate within x% over some timeframe).
There was interesting work DARPA was sponsoring around the above idea that you can read more about here: https://www.darpa.mil/program/micro-technology-for-positioni...
>The end goals of the TIMU program are the demonstration of a single-chip IMU which maintains an accumulated position error of less than 1 nmi/hour with device volume of less than 10 mm3 and power consumption of less than 200 mW.
(My job is related to estimating location of things).
Today the international nautical mile is defined as 1,852 metres (about 6,076 ft; 1.151 mi).
That's the implied application, which is wrong.
I, for one, am glad to see people working to innovate along well-trodden paths.
(I worked for Trimble.)