Task parameterized robot skill learning via programming by demonstrations

Abstract

This thesis focuses on developing Programming by Demonstration (PbD) approaches for task parametrized learning. The presented work shows that this can be performed by utilizing very few demonstrations. The learned model can interpolate as well as extrapolate beyond the demonstrated ranges. Additionally the use of a Neural Network, by directly utilizing camera images, is also investigated for PbD. Lastly the issue of large variations in the teleoperated demonstrations is considered for PbD.