SGER: Discovery Driven Manipulation of NonRigid Objects: Representation, Sensing and Planning

点击翻译按钮获取中文摘要

IRI-9616131 Nandhakumar, N. University of Virginia $49,996 - 12 mos SGER: Discovery Driven Manipulation of NonRigid Objects: Representation, Sensing and Planning This award, in the SGER mode, funds an initial exploration of machine discovery of motor coordination skills through sensory input. Recent psychophysical studies have revealed that humans can discover proper motor coordination skills through sensory input for tasks in which they have had no previous experience. The PIs are currently developing a spatial Jacobian-based framework for the design of robotic, nonrigid object manipulation strategies which emulate this human ability. Unlike existing algorithms, the strategies derived using this framework do not require the use of a priori deformation models since the necessary object parameters are learned during the process. Since the spatial Jacobians of an object are spatially dependent,the critical obstacle in the framework's continued development is determining the amount of information which must be recovered in the learning or discovery phase of the algorithms for accurate reconfiguration. For general flexible objects, the spatial Jacobian sets will be quite large and not efficiently learnable. Thus, in this preliminary research we will determine what subsets of spatial Jacobians must be obtained by the process to move the object from an initial state to a final state within an acceptable positional error. An experimental investigation will also determine the possibility of creating a set of canonical shapes to reduce the configuration space size as well.