Active learning using mean shift optimization for robot grasping

When children learn to grasp a new object, they often know several possible grasping points from observing a parent's demonstration and subsequently learn better grasps by trial and error. From a machine learning point of view, this process is an active learning approach. In this paper, we present a new robot learning framework for reproducing this ability in robot grasping. For doing so, we chose a straightforward approach: first, the robot observes a few good grasps by demonstration and learns a value function for these grasps using Gaussian process regression. Subsequently, it chooses grasps which are optimal with respect to this value function using a mean-shift optimization approach, and tries them out on the real system. Upon every completed trial, the value function is updated, and in the following trials it is more likely to choose even better grasping points. This method exhibits fast learning due to the data-efficiency of the Gaussian process regression framework and the fact that the mean-shift method provides maxima of this cost function. Experiments were repeatedly carried out successfully on a real robot system. After less than sixty trials, our system has adapted its grasping policy to consistently exhibit successful grasps.

[1]  S. Gruber,et al.  Robot hands and the mechanics of manipulation , 1987, Proceedings of the IEEE.

[2]  Mohamed S. Kamel,et al.  A connectionist model of human grasps and its application to robot grasping , 1995, Proceedings of ICNN'95 - International Conference on Neural Networks.

[3]  Vijay Kumar,et al.  Robotic grasping and contact: a review , 2000, Proceedings 2000 ICRA. Millennium Conference. IEEE International Conference on Robotics and Automation. Symposia Proceedings (Cat. No.00CH37065).

[4]  Bernhard P. Wrobel,et al.  Multiple View Geometry in Computer Vision , 2001 .

[5]  Andrew Zisserman,et al.  Multiple view geometry in computer visiond , 2001 .

[6]  Dorin Comaniciu,et al.  Mean Shift: A Robust Approach Toward Feature Space Analysis , 2002, IEEE Trans. Pattern Anal. Mach. Intell..

[7]  Markus Lappe,et al.  Biologically Motivated Multi-modal Processing of Visual Primitives , 2003 .

[8]  Yiannis Demiris,et al.  Abstraction in Recognition to Solve the Correspondence Problem for Robot Imitation , 2004 .

[9]  Ruzena Bajcsy,et al.  Active learning for vision-based robot grasping , 1996, Machine Learning.

[10]  Antonio Morales Escrig Learning to predict grasp reliability with a multifinger robot hand by using visual features , 2004 .

[11]  Josef Pauli Learning to Recognize and Grasp Objects , 2004, Machine Learning.

[12]  Peter K. Allen,et al.  An SVM learning approach to robotic grasping , 2004, IEEE International Conference on Robotics and Automation, 2004. Proceedings. ICRA '04. 2004.

[13]  Ashutosh Saxena,et al.  Learning to Grasp Novel Objects Using Vision , 2006, ISER.

[14]  Helge J. Ritter,et al.  Experience-based and tactile-driven dynamic grasp control , 2007, 2007 IEEE/RSJ International Conference on Intelligent Robots and Systems.

[15]  Jan Steffen,et al.  SOM-based experience representation for Dextrous Grasping , 2007 .

[16]  Ruben Martinez-Cantin Active Map Learning for Robots: Insights into Statistical Consistency , 2008 .

[17]  Justus H. Piater,et al.  Probabilistic Pose Recovery Using Learned Hierarchical Object Models , 2009, ICVW.

[18]  Lawson L. S. Wong,et al.  Learning Grasp Strategies with Partial Shape Information , 2008, AAAI.

[19]  Nicolas Pugeault,et al.  Early cognitive vision: feedback mechanisms for the disambiguation of early visual representation , 2008 .

[20]  N. Kruger,et al.  Learning object-specific grasp affordance densities , 2009, 2009 IEEE 8th International Conference on Development and Learning.

[21]  Carl E. Rasmussen,et al.  Gaussian processes for machine learning , 2005, Adaptive computation and machine learning.