Ask Your Question

Revision history [back]

Without having worked with this extended kalman filter package, in general it should be possible to remove drift over time by feeding the filter with positions measurements that correspond to observations of landmarks.

It would roughly work like this: -Run the EKF normally -Whenever a tag is spotted, augment the observation matrix to account for a low-covariance position measurement (For a linear kalman filter, this corresponds to adding a row -If multiple tags are visible, keep adding rows -When the tag is lost again remove the entry that corresponds to the tag that was lost

You could also add bias-states to the filter, which are basically integrator states that are summed with the velocity state to estimate a position. In this way, it is possible to reduce the drift of the filter somewhat (if the drift is constant), using IMU's this helped me a lot.

kind regards