Improving Robot Learning Using an Active Learning Approach in a Learning from Demonstration Framework

Abstract

Programming new abilities on a robot ought to take negligible time and exertion. One way to accomplish this objective is to permit the robot to ask questions. This idea, called Active Learning, has recently caught a lot of attention in the robotics community. We are interested in the potential of active learning to improve learned skills from human demonstrations in an HRI setting. In this thesis, I explore different types of queries proposed in the Active Learning(AL) Literature and apply them to Learning from Demonstration(LfD) problems. The central part of this work is to design a strategy for data selection for the query in order to avoid unnecessary and redundant queries, select different types of query that will help the robot learn better and explore how the incorporation of AL methods in LfD impacts a robot’s learning and performance.