Advancing Robotic Grasp Planning using Human Heuristics for Grasp Similarity
Published in Oregon State University, 2016
This thesis explores the use of human kinesthetic input to advance robotic grasp planning. It specifically focuses on identification of human heuristics of grasp similarity by extensive human-subject experiments. This heuristic is then utilized to develop a robust metric for identifying similar robotic grasps. The central hypothesis of this work is to speed up grasp planning and improve the reliability of the planned grasps by using the human heuristic for grasp similarity. Experimental results show that (i) new similar grasps generated from interpolation of human-provided grasps have similar performance (93.75% success rate) compared to original human grasps (97% success rate) and (ii) grasps with similar hand configurations are functionally similar when applied to large regions of everyday objects, but functionally similar grasp does not necessarily dictate similar hand configuration., Published in AAAI Fall Symposium 2016, https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/37720j09c
authors: Saurabh Milind Dixit
Authors: Saurabh Milind Dixit
Download Paper