Spontaneous Reorientation for Self-localization

Abstract

Abstract. Humanoid robots without internal sensors (e.g. compasses) tend to lose their orientation after a fall or collision. Furthermore, artificial environments are typically rotationally symmetric, causing ambiguities in self-localization. The approach proposed here does not alter the measurement step in the robot's self-localization. Instead it delivers confidence values for rotationally symmetric poses to the robot's behaviour controller, which then commands the robot's self-localization. The behaviour controller uses these confidence values and triggers commands to rearrange the self-localization's pose beliefs within one measurement cycle. This helps the self-localization algorithm to converge to the correct pose and prevents the algorithm from getting stuck in local minima. Experiments in a symmetric environment with a simulated and a real humanoid NAO robot show that this significantly improves the system.