Robot aided object segmentation without prior knowledge

In robot perception system, distinguishing objects from complex environment is a difficult problem if without prior information. In this article, we study three cases that a robot may encounter in real-world application, no movable object, one object, or multiple objects, and then provide an object segmentation strategy through manipulation for each condition. The result shows that this method can provide sufficient prior information for accurate objects segmentation from robot's observation. Through this unsupervised algorithm, a robot can learn objects around reliably.