50.3.1 Choice of states for the IMU motion/kinematics model

Parent: IMU index
See also: Choice of model for the IMU motion model

According to  MKok 2017 , we can either

1. Use the full state vector
   • [+] knowledge about sensor motion is included in model
   • [-] large state vector
2. Or the partial state vector, where the inputs are the inertial measurements from the IMU

   • [+] process noise intuitively represents IMU noise. This is useful when we have no knowledge about the motion model.

   • [+] changes to acceleration and angular velocity will have a slightly faster effect (prediction not dependent on acc/ang.vel model)

The literature for EKF sensor fusion using IMUs mostly implement the latter option with IMU measurements as an input.  Solà 2017  uses the following states, inputs and perturbations (white Gaussian noise) in the  IMU kinematic equations .

Extending the equations to the error state dynamics, the error states are