Combining GPS with accelerometer and gyroscope is sometimes called "sensor fusion" - one common way of doing it is with a "Kalman filter" and if you google that you'll find a wealth of information.

If you've got plenty of money, companies like https://www.oxts.com/ have working systems available off the shelf. They're used for things like measuring the dynamic performance of cars.

Such systems will sorta work indoors but not very well. The gyroscope/accelerometer will gradually drift; when you're outdoors with a clear view of the sky, the GPS can compensate for that but when you're indoors that isn't possible. How fast the drift happens depends on how much money you spend to get the highest performance gyroscopes and accelerometers.

It's also possible to integrate other sensors, but that's application-dependent; for a car going through a tunnel, an odometer is a great addition, for a smartphone not so much.

GPS-denied navigation is the search term. It works well enough for driving through a tunnel, because eventually you get a known location to correct previous estimates. For a fully-indoor environment it's very difficult to get good results and the technique depends on the use-case.

"Indoor GPS" using bluetooth beacons is technically possible but the space is filled with proprietary protocols/apps that try to make money by serving ads. It unfortunately doesn't seem to be evolving into a more useful extension of google maps.

For indoor, there's UWB, ultra wide band. Can also use Bluetooth or even lights blinking at different (high, non-humanly-visible) frequencies to do indoor positioning.

For indoor you will prefer UWB[0] where you can set up base stations yourself that offer a service pretty much like GNSS signals (i.e., you can passively receive them from 4+ base stations and turn those pseudoranges into a position, provided you are told where they are and kept updated about their clock drift relative to each other (e.g. by them listening to their neighbors and piggybacking 2-way-ranging sessions on top of the beacons they already broadcast, each offering the clock offset to their neighbors in the data payload of their own beacon)), with the added benefit that by using a shared-key CSPRNG to generate the bitstream of the ranging code instead of a fixed known sequence, you can get authenticated ranging where MITM attacks are limited to artificially inflating the range (i.e., a trigger of "has to be close enough to the door handle that e.g. a pair of cameras looking at the areas right in front of either side of the door have it in view" can't be faked with a wireless MITM).

There are some links/papers in the publications section[2] of [1].

[0]: https://www.firaconsortium.org/ [1]: https://developer.android.com/develop/sensors-and-location/s... [2]: https://developer.android.com/develop/sensors-and-location/s...