Regarding use of IMU value in predict step of EKF

[adityag6994]

Hi,

Great wrok!
I have a question regarding the use of sensor values in the prediction step of the Extended Kalman Filter (EKF). Could you clarify the logic behind incorporating IMU sensor data in this step or refer me to relevant literature on how it is applied?

            f_k = imu[0:3] * GRAVITY_MAGNITUDE
            self.f[k] = f_k
            dw = omega_k * dt                      # Attitude error
            # nominal state motion model
            # position prediction 
            self.Xpr[k,0:3] = self.Xpo[k-1, 0:3] + Vpo.T*dt + 0.5 * np.squeeze(self.R.dot(f_k.reshape(-1,1)) - GRAVITY_MAGNITUDE*e3) * dt**2

https://github.com/utiasDSL/util-uwb-dataset/blob/833bc611c0e472927d9afa803e36e8bfb992bd2c/scripts/estimation/eskf_class.py#L69C1-L74C141

Thank you!

[Williamwenda]

"Quaternion kinematics for the error-state Kalman filter" from Joan Sola is a great reference for inertia navigation. You may find Section 5, specifically Section 5.4, helpful for your understanding IMU-driven system kinematics.

Regards,

Wenda