Integrating Graph-Based Vision Perception to Spoken Conversation in Human-Robot Interaction