This fixes train_filter() to find the most likely position from the new estimated state rather than the previous state.