Learnign high-level reasoning of human interactions in a scoial greeting using deep reinforcement learning