*5.5. Integrity Monitoring*

The entire information flow between the localization subsystems and the Kalman filter passes through an integrity monitoring module to avoid possible undesired behaviors. Each time an observation is generated from any high-level subsystem, we compute the Mahalanobis distance between the prediction and the observation to reject outliers. The rejected observations are excluded from the Kalman filter correction step.

The integrity monitoring module also permits to correct the AMCL subsystem using the Kalman filter output, as we can see in Figure 7. This recovery procedure is especially useful to leave and return from/to the mapped area. The monitoring module will allow this correction step in cases where the AMCL variances exceed a certain threshold. To ensure the localization stability, this correction is applied only to the state vector of the AMCL, leaving the covariances as they are. This procedure helps to the convergence of the system when re-entering to a previously mapped area.
