[jfoutz]

So, sure. I’ve written velocity guessers based on spinning rust disk accelerometers. It’s easy to build a poor quality system. But an afternoon hack can get you to a few percent accuracy.

Analog integration will improve your accuracy a ton. Multiple systems, averaged will improve accuracy even more.

Sure, there are problems. But tomahawk missiles use inertial guidance. Autopilots do it to. More data will give better answers. But don’t sell ig short. You can get very good answers.

[fyfy18]

I remember reading an article a few months ago (I guess here) about a precursor to GPS based navigation for cars that used inertial guidance. You needed to set the location when you started driving, but after that it just used inertial guidance to figure out your position.

I really wish smartphone navigation apps would use this, as I've ended up taking the wrong directions in tunnels or at the exits of tunnels (as it takes a few seconds to get a GPS fix).

[dmitrygr]

Real gyro != Mems gyro.

Real gyro is REALLY rigid in space. Mems gyro isn't. It causes errors after mere minutes of integration. INS systems in planes and gyros in missiles are REAL gyros. This isn't. This is more like a better mems system.

[jfoutz]

It seems like the jury is still out. Perhaps 10,000 of them work out better than a large system. It’ll take time for a lab experiment to turn into a production system.

You’re probably right in the short term, for an arbitrarily large value of short.