Direct integration of IMU will result in massive error incredibly quickly.
What I am wondering is, how do you use an IMU over a small time intervals to aid in pose estimation?
In particular, suppose we have used Kalman filter for the first time step t_0 and got a fused estimate t_1, what happens next with the IMU?
You don't continue to integrate it, so what do you do?
Do you zero it? Doing so will only then give you acceleration over the current time step. How does this get combined into a velocity estimate?