Robot Manipulation on Cluttered Scenes using 3D Visual Servoing
Smith Hall 100
Refreshments 4:15 pm
Talk 4:30 pm
This work focuses on the interaction of manipulator robots in complex environments using 3D sensors. In this kind of scenes, different free-form objects are located at arbitrary poses. In the worst case, the complexity of this scene includes: no shape-restrictions, shades, occlusion, cluttering and contact between objects. Scenes are sensed from a single point of view and, consequently, only partial information is available. An intelligent interaction (grasping, pushing, touching, etc) of a robot in such a scene is a complex task which involves several different research fields: 3D image processing, computer vision and robotics.
There are three main phases in this work: segmentation of the scene into their constituent parts, recognition and pose of the objects and robot planning/interaction. In the first and second phases, new 3D segmentation techniques, which extract the objects that compose the scene, and new object recognition and pose techniques, with the purpose of reconstructing the scene have been developed [1,2,3]. Taking into account the results of the previous phases, the robot interaction in the scene is carried out. For this, a grasp planner  and a path planner have been developed . The grasp planner is based on the solid representation model MWS (Modeling Wave Set) and solves the grasping problem on 3D free shapes. This implies reinforcing the connection between the fields of modelling and 3D-grasping. The path planner is a sample-based algorithm based on an adaptive discrete search and a non-random local sampling which implies computational time reduction. Cubic splines are used to generate smooth and continuous paths and also as heuristic functions. A motion planner is also being developed in order to determine the interaction of the robot on the scene.
All these algorithms are tested in a simulator environment developed specifically for this project. The integration of the systems of the experimental setup is based on a distributed system, using RT-LINUX for the hard real-time control of the robot and RT-CORBA for the soft real-time control of the rest of the experimental setup (3D system, planners, simulator, etc) 
1. A. Adán, A.S. Vázquez , M. Adán, P. Merchán, S. Salamanca, C. Cerrada (2005). “Objects Layout Graph for 3D Complex Scenes” IEEE International Conference on Image Processing (ICIP 2005). Genove, Italy (IEEE Signal Processing Society)
2. P. Merchán, A. Adán and S. Salamanca ” Recognition of Free Form Objects in Complex Scenes Using DGI-BS Models”. Third International Symposium on 3D Data Processing, Visualization and Transmission 2006. Chapel Hill, NC, USA.
3. M. Adán, A. Adán, A. S. Vázquez (2004), "Clustering Using WCC Models", Lecture Notes in Computer Science. LNCS 3138 pp 921-929. Publisher: Springer. (ISBN: 3-540-22570-6)
4. A. Adán, A. S. Vázquez, F.Molina (2005), "3D Grasping Solutions Through MWS Models" 12th Int. Conf. on Advanced Robotics, Seattle, Washington. IEEE Robotics and Automation Society (IEEE)
5. A.S. Vázquez, Roberto Torres, Antonio Adán, Carlos Cerrada,”Path Planning for Manipulation Environments through Interpolated walks” IEEE/RSJ International Conference on Intelligent Robots and Systems.IROS’06 Beijing China 2006
6. A. S. Vázquez, A.Adán, R. Torres, C. Cerrada. “Distributed Control System of An Experimental Robotic Cell with 3D Vision” 3rd International Conference on Informatics in Control, Automation and Robotics. ICINCO06. Setubal, Portugal
Andrés S. Vázquez received his BSc in Computer Science from the University of Castilla- La Mancha, Spain, in 2002. He has worked as a research assistant for UNED University from 2001 to 2002 in a 3D vision project. From 2002, he has been working as a research assistant for the ISA Computing Vision Group at the Computing School, Ciudad Real, Spain. Currently, he is pursuing a PhD degree in Computer Science at the University of Castilla–La Mancha under a FPI scholarship and has also been working as an assistant lecturer since 2003 in Control Theory and Introduction to Computing.
For appointments, please contact Ravi Balasubramanian (firstname.lastname@example.org)