Gravar-mail: Linear Fitted-Q Iteration with Multiple Reward Functions