Uncertainties based queries for Interactive policy learning with evaluations and corrections