A Robust and Adaptive Framework for Localization under Varying Sensor Modalities
Shaunak D. Bopardikar, Shuo Zhang, Alberto Speranzon · AIAA Guidance, Navigation, and Control (GNC) Conference · 2013
This paper proposes a modular estimation framework that enhances robustness and adaptability of legacy filters, such as the standard Extended or Unscented Kalman Filter (EKF/UKF), in the presence of sudden and unknown events such as sensor failures, unavailability, change of accuracy and out-of-sequence measurements. The framework comprises of three main outer modules wrapped around a core legacy filter: 1) Measurement gating, to detect large changes in covariance of sensors and reject their measurements until a new consistent covariance estimate is obtained; 2) Covariance estimation, to estimate the accuracy of each sensor at all times and; 3) Out-of-sequence processing, to perform optimal, in terms of minimum mean squared error, estimation when sensor measurements become out-of-sequence in time for any arbitrary arrival order. We apply this framework to localize a real mobile vehicle moving in an unknown environment, equipped with various suites of sensors. The contributions of this paper are two-fold. First, we demonstrate a successful integration of the out-of-sequence processing algorithm with the other two modules around a standard EKF. Second, we show via numerical simulations, that under randomly generated events of sensor failure, accuracy changes, unavailability or delays, our proposed approach performs significantly better than a simple baseline EKF. We report the results applied to simulated scenarios, as well as on multiple real data sets.