Committee: 

Dr. Chernova, Advisor   

Dr. Hutchinson, Chair

Dr. Gombolay

Dr. Boots

Dr. Hermans

Abstract:

The objective of the proposed research is to show that functioning in diverse task settings benefits from robots to have a structured but flexible skill learning scheme. Motion generation to execute novel tasks requires robots to continuously expand their skillsets. In line of this, learning from demonstration (LfD) provides an effective way for robots to acquire new skills by leveraging human knowledge. Skills however can be vastly different, enforcing a range of constraints, requiring different representations to encode them. By enforcing appropriate structures, I propose to develop a family of learning from demonstration (LfD) approaches aimed at learning generalizable, feasible, and coordinated skills from a few human demonstrations.