With the motivation to optimize the cost and time of function development and calibration of the ECU, reinforcement learning (RL) technique, DDPG (Deep Deterministic Policy Gradient) has been applied to coordinated control of gasoline ICE air-system/combustion system manager. As an environment part of the RL system, surrogate ICE model using deep learning technique, simulates resembled real driving situation and it is coupled with ECU model including steady-state setpoints maps and intervention by RL-actions.