A Multi-Modal Approach for Natural Human-Robot Interaction

We present a robot that is able to interact with people in a natural, multi-modal way by using both speech and gesture. The robot is able to track people, process speech and understand language. To track people and recognize gestures, the robot uses an RGB-D sensor (e.g., a Microsoft Kinect). To recognize speech, the robot uses a cloud-based service. To understand language, the robot uses a probabilistic graphical model to infer the meaning of a natural language query. We have evaluated our system in two domains. The first domain is a robot receptionist (roboceptionist); we show that the roboceptionist is able to interact successfully with people 77% of the time when people are primed with the capabilities of the robot compared to 57% when people are not primed with its capabilities. The second domain is a mobile service robot, which is able to interact with people via natural language.