The Application of Audio Control in Social Robotics

摘要

Along with the development of the voice recognition technology, voice control plays an increasingly important role in daily life as it offers an intuitive interface that enables hands-free control of various devices, as a result blurring the sense of boundaries between human and robot. The rising demand for voice control in daily life has motivated this paper to develop a voice control system using machine learning models for the robot model LoCoBot WX250. In conjunction with robot hardware, this trained audio control system achieves the goal of improving the experience of interaction between human and robot. Based on this system, the LoCoBot WX250 can serve as a tour guide in museums by communicating with tourists via speaker and microphone. What's more, while talking to people, the robot can respond to certain instructions, such as navigation, change position and so on. This audio control system improves upon all previous similar projects in terms of the accuracy of intent classification and keyword recognition by implementing the state-of-the-art BERT (Bidirectional Encoder Representations from Transformers) model, so that it can have a smoother conversation with human. Furthermore, to accommodate the foreign visitors in the museum and tourist spot, the audio control system can be instructed to change the language, making it even more accommodating than a human tourist guide.