LOCALIZING A MOBILE ROBOT WITH INTRINSIC NOISE

Brian F. Allen (1) , Flavien Picon (1) , Sébastien Dalibard (1) , Nadia Magnenat-Thalmann (2) , Daniel Thalmann (3)

1 Institute for Media Innovation
2 MIRALab
3 Institute for Media Innovation

Robot localization is a key barrier to providing natural interaction between 3D virtual characters, human users and mobile robots. Knowing where the robot is, relative to a known world-frame, is essential to directed gestures, gazes and expressions between the robot and the other real and virtual participants in a telepresence system. The intrinsic noise of robots is a flexible and robust, yet under-examined, source for localization information. Sounds that likely localize the robot are identified and separated from background noises using a support vector machine. The resulting sound-bearing data is combined with robot odometry using a particle filter. Experiments conducted in a noisy office environment show substantial improvement over odometry alone.

Domaines: Robotique [cs.RO]

Brian F. Allen, Flavien Picon, Sébastien Dalibard, Nadia Magnenat-Thalmann, Daniel Thalmann. LOCALIZING A MOBILE ROBOT WITH INTRINSIC NOISE. 3DTV-CON 2012, Oct 2012, Zurich, Switzerland. ⟨hal-00732764⟩