Temporal Alignment and Demonstration Selection as Pre-Processing Phase for Learning by Demonstration

Abstract

Robots can benefit from users’ demonstrations to learn
motions. To be efficient, a pre-processing phase needs
to be performed on data recorded from demonstrations.
This paper presents pre-processing methods developed
for Learning By Demonstration (LbD). The
pre-processing phase consists in methods composed
of alignment algorithms and algorithms that select the
good demonstrations. In this paper we propose six
methods and compare them to select the best one.