Long-term robot motion planning for active sound source localization with Monte Carlo tree search

Abstract

We consider the problem of controlling a mobile robot in order to localize a sound source. A microphone array can provide the robot with information on source localization. By combining this information with the movements of the robot, the localization accuracy can be improved. However, random robot motion or short-term planning may not result in optimal localization. In this paper, we propose an optimal long-term robot motion planning algorithm for active source localization. We introduce a Monte Carlo tree search (MCTS) method to find a sequence of robot actions that minimize the entropy of the belief on the source location. A tree of possible robot movements which balances between exploration and exploitation is first constructed. Then, the movement that leads to minimum uncertainty is selected and executed. Experiments and statistical results show the effectiveness of our proposed method on improving sound source localization in the long term compared to other motion planning methods.