Date: Tuesday, Oct 27, 12:30-1:20, PDL C-401

Speaker: Sasha Aravkin, Math, University of Washington

An L1-Laplace Robust Kalman Smoother

Robustness is a major problem in Kalman filtering and smoothing - it is desirable to have an algorithm that works well with Gaussian noise but also has reasonable performance when this assumption is violated. In this talk, we consider the case of heavy tailed observation noise, and show how interior point methods can be used to find a maximum a posteriori likelihood (MAP) Kalman smoother for a nonlinear model with Gaussian process noise and L1-Laplace observation noise. We test the smoother's performance on simulated data with contaminated normal observations, L1-Laplace noise, Student-t noise, and apply it to actual data for an underwater tracking experiment