Fast Grasp Learning for Novel Objects
Dawid Seredyński , Wojciech Szynkiewicz
AbstractThis paper presents a method for fast learning of dexterous grasps for unknown objects. We use two probabilistic models of each grasp type learned from a single demonstrated grasp example to generate many grasp candidates for new objects with unknown shapes. These models encode probability density functions representing relationship between fingers and object local features, and whole hand configuration that is particular to a grasp example, respectively. Both, in the training and in the grasp generation stage we use an incomplete 3D point cloud from a depth sensor. The results of simulation experiments performed with the BarrettHand gripper and several objects of different shapes indicate that the proposed learning approach is applicable in realistic scenarios.
|Publication size in sheets||0.55|
|Book||Szewczyk Roman, Kaliczyńska Małgorzata, Zieliński Cezary: Challenges in Automation, Robotics and Measurement Techniques. Proceedings of AUTOMATION-2016, March 2-4, 2016, Warsaw, Poland, Advances in Intelligent Systems and Computing, vol. 440, 2016, Springer International Publishing, ISBN 978-3-319-29356-1, [978-3-319-29357-8], 919 p., DOI:10.1007/978-3-319-29357-8|
|Keywords in English||Grasp learning Probabilistic models Kernel density estimation|
|project||RobREx: Autonomy for rescue and exploration robots. Project leader: Zieliński Cezary,
, Phone: 5102, start date 12-12-2012, end date 30-11-2015, 513/1031, Completed
|Score|| = 15.0, 27-03-2017, BookChapterSeriesAndMatConf|
= 15.0, 27-03-2017, BookChapterSeriesAndMatConf
* presented citation count is obtained through Internet information analysis and it is close to the number calculated by the Publish or Perish system.