Bayesian Additive Regression Trees (BART) is a relatively new machine learning model. An open research question is how to assess convergence of BART fit to real data. My research addresses this question by deriving a formal convergence criterion provided in a novel software implementation for fitting BART. Additionally, BART has been extended to two new application areas: 1) measurement error in predictors and 2) variable selection in the presence of mandatory and optional predictors.