Guiding a robotic gripper by visual feedback for object manipulation tasks

This paper presents a novel object manipulation technique that could be adopted by any advanced mechatronic platform in order to perform demanding pick and place tasks. The ultimate goal of a robotics researcher is to provide an applicable manipulation solution that minimizes user's involvement. It has been shown that the best solution to this problem is provided by the introduction of sensors that allow an automatic or, at least, semi-automatic grasping of the targets. The proposed method relies on a vision-based framework that is responsible for several vital tasks that affect directly the manipulation process. The contribution of the paper incorporates a shape retrieval technique accompanied with classification and clustering algorithms that are utilized during the objects' pose estimation process. The experimental results obtained confirm the validity of the presented approach.