Adaptive Object Tracking with an Anthropomorphic Robot Head

Abstract

We present a system for model-free, adaptive object tracking with a redundant anthropomorphic robot head. The visual tracking system is based on the idea of "Democratic Integration" to fuse information from several cues in a biologically plausible, self-organized fashion. The motor control system utilizes the idea of "controller partitioning" to exploit complementary capabilities of redundant degrees of freedom in the neck and eyes. A biologically inspired vergence control mechanism ensures coordination of the robot's two eyes. We demonstrate the performance of the overall system in different active tracking scenarios.