Active perception for object manipulation

Abstract

The active perception makes an important step towards the integration of the sensing in the planning phase of the object manipulation. Closing the perception-planning loop iteratively, this helps to reduce the uncertainty in the sensing and planning offering better scene parsing and object manipulation applications. In this paper we show the active perception concept using an industrial like scene with different pipes which are sensed with a depth camera. The pose information from the camera is used for a 7 degree of freedom robotic arm mounted on a mobile base in order to perform the planning for object manipulation purposes. For the whole active perception pipeline we have successfully implemented and tested the scene parsing using object part decoupling and the planning with the robot arm.