This paper considers the problem of estimating linear dynamic system modelswhen the observations are corrupted by random disturbances with nonstandarddistributions. The paper is particularly motivated by applications where sensorimperfections involve significant contribution of outliers or wrap-aroundissues resulting in multi-modal distributions such as commonly encountered inrobotics applications. As will be illustrated, these nonstandard measurementerrors can dramatically compromise the effectiveness of standard estimationmethods, while a computational Bayesian approach developed here is demonstratedto be equally effective as standard methods in standard measurement noisescenarios, but dramatically more effective in nonstandard measurement noisedistribution scenarios.
展开▼
