Learning a Decision Module by Imitating Driver's Control Behaviors