An end-to-end sound source navigation method of indoor mobile robot

Abstract

At present, the study on sound source navigation mobile robot mainly be limited by the accuracy of sound source localization. On the one hand, the traditional sound source localization is susceptible to environmental interference. On the other hand, the sound source localization based on deep learning is limited by the datasets, and the accuracy cannot be guaranteed. This paper proposes an auditory perception and decision-making method based on deep reinforcement learning to effectively address this problem. Using the combination of the feature expression ability of deep neural network and the decision-making ability of reinforcement learning, the system environment perception and decision control are formed into a closed-loop system, and using the reward function of deep reinforcement learning, the sound source localization and navigation strategy are gradually improved during the interaction between the robot and the environment, and finally the end-to-end sound source navigation is realized. After a lot of simulation experiments, it is verified that the mobile robot can recognize the target information from the sound signal and accurately move to the target position.