Selective Attention by Perceptual Filtering in a Robot Control Architecture

Abstract

Modern autonomous robots must integrate multiple perceptual and behavioral modalities to be useful in our daily lives. Such integration is constrained by the limited onboard computing capacity of robotic platforms. To alleviate this issue, perceptual filtering, a selective attention mechanism, can be used to efficiently manage computing resources based on what the robot has to accomplish. This paper describes our implementation of perceptual filtering in a robot control architecture, implemented using robot operating system (ROS), and how it can dynamically optimize the use of the computing resources available on the robot. Our perceptual filtering mechanism is demonstrated and validated using a mobile humanoid platform integrating autonomous and teleoperated navigation, QR code recognition, face recognition, and sound localization capabilities.