Robust Data Processing in the Presence of Uncertainty and Outliers: Case of Localization Problems

International audience; To properly process data, we need to take intoaccount both the measurement errors and the fact that someof the observations may be outliers. This is especially importantin radar-based localization problems, where some signals mayreflect not from the analyzed object, but from some nearby object.There are known methods for dealing with both measurementerrors and outliers in situations in which we have full informationabout the corresponding probability distributions. There are alsoknown statistics-based methods for dealing with measurementerrors in situations when we only have partial informationabout the corresponding probabilities. In this paper, we showhow these methods can be extended to situations in which wealso have partial inf0ormation about the outliers (and even tosituations when we have no information about the outliers). Insome situations in which efficient semi-heuristic methods areknown, our methodology leads to a justification of these efficientheuristics – which makes us confident that our new methods willbe efficient in other situations as well.