An Incremental Feature Set Refinement in a Programming by Demonstration Scenario

Abstract

In transferring knowledge from human to robot using Programming by Demonstration (PbD), choosing features which can represent the instructor demonstrations is an essential part of robot learning. With a relevant set of features, the robot can not only have a better performance but also decrease the learning cost. In this work, the feature selection method is proposed to help the robot determine which subset of the features is relevant to represent a task in PbD framework. We implement an experimental PbD system for a simple task as proofing our concept as well as showing the preliminary results.