Outlier Accommodation in Sensor Rich Environments: Moving Horizon Risk-Averse Performance-Specified State Estimation

This paper considers state estimation for nonlinear systems when the measurements may be corrupted by outliers. The traditional approach to outlier accommodation utilizes the extended Kalman filter over a single time epoch only for those residuals that pass a Neyman-Pearson (NP) threshold test. This approach may utilize risky measurements that are unneeded to achieve a user-defined performance level, yielding unnecessary risk. Risk-Averse Performance-Specified (RAPS) state estimation is an optimization based approach that chooses a subset of measurements with minimum risk that satisfies an user-defined performance specification. This article extends the RAPS approach to moving horizon (MH) state estimation for nonlinear systems. The moving horizon approach allows past outlier decisions within the time window to be reevaluated in the light of new measurements. This paper includes experimental results for vehicle state estimation using Global Navigation Satellite Systems (GNSS) aiding an Inertial Navigation System (INS). The MH-RAPS performance is compared with MH-NP state estimation.