I wish I had this summary quite a few years ago. I ended up buying aeronautical software textbooks and trying to glean the information from those to figure out how to integrate multiple IMU inputs.
I learned that real world IMU use in navigation is pretty ugly and has to be supplemented due to constant errors. This can be particularly difficult when flying. In my case I used other sensors, such as a GPS. But a strictly IMU-only approach to lengthy navigation is, AFAIK, impossible.
> I learned that real world IMU use in navigation is pretty ugly and has to be supplemented due to constant errors. This can be particularly difficult when flying. In my case I used other sensors, such as a GPS. But a strictly IMU-only approach to lengthy navigation is, AFAIK, impossible.
Yeah IMU error can be annoying unless you shell out $$$, but this issue occurs with many other kinds of sensors as well.
One common trick is to treat the IMU error as another state variable that is evolving over time. It won't eradicate the compounding error, but it will reduce the amount of calibration necessary.
Depends on the quality of the IMU. With a MEMS IMU (what’s in your smart phone and what’s cost effective for most consumer or industrial products today) yes, it’s impossible. Too much drift as a result of process variance in MEMS manufacturing and the “runtime” noise of the IMU itself. With a precision machined IMU, like something you’d find in a submarine or pre-GPS aircraft, totally possible. There’s a YouTube video (would have to dig for the link) showing a military plane in the 1950s flying transcon NYC to LA solely on IMU dead reckoning and arriving with 1/8 mile drift or something ridiculous. The IMU probably weighed a few hundred pounds and cost many hundreds of thousands, if not millions of dollars, though.
> a strictly IMU-only approach to lengthy navigation is, AFAIK, impossible
This is true when using very-low-cost (i.e. consumer MEMS) sensors but as a general statement it's false; many military and civilian space- and aircraft systems are solely or primarily inertial.. even ~50 year old passenger jets can make 12-13 hour flights with very high inertial accuracy - enough to find the dest airport runway without a radio beacon (though probably not enough to automatically land on it).
For anyone interested in the technical history of inertial nav, I'd highly recommend "Inventing Accuracy" by Don MacKenzie.
There are good IMUs that can be used for dead reckoning (and, if an old professor is to believed, watching the thermal expansion of a building over the course of a day) but they're hard to come by and extraordinarily expensive.
I learned that real world IMU use in navigation is pretty ugly and has to be supplemented due to constant errors. This can be particularly difficult when flying. In my case I used other sensors, such as a GPS. But a strictly IMU-only approach to lengthy navigation is, AFAIK, impossible.
Either way, good write up.