When you take attitude out of EKF you won’t get as good an estimate, but sometimes you do it to save on complexity.
The"Full" State Vector
Often, the full state vector actually has more than 12 variables in it. That's because in addition to estimating the typical 12 variables, we often also want to keep track of the IMU biases. Remember from the lesson on sensors that bias is typically modeled as a random walk, that is:
Estimating the bias in real-time like this is what's known as"online system identification".
NEXT
5. EKF Tradeoffs - Control
We can use the IMU measurements as control inputs instead of measurements to keep the state vector smaller and to avoid problems with low-cost hardware where the control signals don’t accurately reflect the control outputs.
Accelerometers when averaged over long time scales should point down. We can use that down reference to measure attitude. But only over long time scales.
Rate gyros give good estimates of angular rates, not noisy but are susceptible to drift. Not useful for long timescales.
4. EKF Tradeoffs - State
EKF State Vector
The "Full" State Vector
5. EKF Tradeoffs - Control
Additional Reading
6. Complimentary Filter
7. Complimentary Filter Math