Robot Dialog System in the Context of Hospital Receptionist and its Demonstration

Abstract

Task-oriented dialogue systems play an important role in human-robot interaction. Task-oriented dialogue systems help users achieve specific goals, and robots with systems are very effective in terms of reducing time and effort when replacing human workers. The conventional pipeline method, which has disadvantages such as a large amount of cost and time for development, has been applied to reception robots. For example, developers manually define responses that correspond to user input. Also, they use predefined robotic actions such as gestures and facial expressions. Recently, end-to-end learning of Recurrent Neural Networks is an attractive solution for the dialogue system. Based on the strengths of RNNs, we propose a social robot system in the context of hospital receptionists. We utilize Hybrid Code Network as an end-to-end dialogue system and extend it to select both response and gesture using RNN-based gesture selector. A user study is conducted comparing our proposed system with one of the existing methods, a rule-based approach. The empirical results show that the proposed method has an advantage in terms of dialog efficiency, which indicates how efficiently the user performed a given task with the help of a robot. In addition, there is no significant difference in experimental results between the proposed RNN based gesture selection and the rule-based gesture selection.