Inertial navigation and machine learning how?

Question

Physics

How, knowing the coordinates of the object (during training, and very rarely for calibration), create a neuron, which should, having values ​​from inertial sensors on the object, give out coordinates? What neuron to use, what structure... Knowing that everything needs to be tested, I will ask “In which direction to move, where to start, what to try”?
There is not one sensor, but the object is not homogeneous (there are moving parts), but we have the coordinates of the key points.

Answer 1

Damn, well, everyone is crazy about the National Assembly. Why shove IT everywhere.
Why can't you use integrators for your inertial sensors?

Answer 2

Actively engaged in this task for several years, though in a specific plane. If everything is simple in terms of architecture: recurrent NS. But the key point in the task will be filtering and eliminating errors. Those. just Neyronka may not work well. I saw works on dead reckoning for marine objects using neural networks, I saw works on filtering in navigation systems using neurons, there are works on using neurons to determine information failures in ANNs. But as I understand it, no one has inserted such a thing into "production". Gyroscopy and navigation is too conservative area.