Interactive Perception for Deformable Object Manipulation

Interactive perception enables robots to manipulate the environment andobjects to bring them into states that benefit the perception process.Deformable objects pose challenges to this due to significant manipulationdifficulty and occlusion in vision-based perception. In this work, we addresssuch a problem with a setup involving both an active camera and an objectmanipulator. Our approach is based on a sequential decision-making frameworkand explicitly considers the motion regularity and structure in coupling thecamera and manipulator. We contribute a method for constructing and computing asubspace, called Dynamic Active Vision Space (DAVS), for effectively utilizingthe regularity in motion exploration. The effectiveness of the framework andapproach are validated in both a simulation and a real dual-arm robot setup.Our results confirm the necessity of an active camera and coordinative motionin interactive perception for deformable objects.