SLHAP: Simultaneous Learning of Hierarchy and Primitives

March 2017

SLHAP: Simultaneous Learning of Hierarchy and Primitives

Authors:

Anahita Mohseni-Kabir, Changshuo Li, Victoria Wu, Daniel Miller, Benjamin Hylak, Sonia Chernova, Dmitry Berenson, Candace Sidner, and Charles Rich

Abstract:

In robot learning from demonstration (LfD), a human teaches a robot how to perform a task by executing the task himself. For complex tasks, such as the tire rotation shown in the accompanying video, this involves learning at two levels: the robot needs to learn the motion primitives and also how these primitives are combined into a hierarchy of steps to achieve the complete task. These two kinds of LfD have traditionally been studied separately. The contribution of this work is a novel humanrobot interaction paradigm, called SLHAP (for simultaneous learning of hierarchy and primitives), in which these two kinds of LfD are interleaved in a way that is natural for a human teacher. We have implemented a SLHAP proof of concept system in which an autonomous robot learns from a human teacher through a mixture of narration, in which the human speaks the name of a primitive when he executes it, and dialogue, in which the human answers the robot’s questions about how to group primitives into subtasks. The human’s motions are also tracked using a Vicon motion capture system.
@conference{Mohseni-Kabir-2017-126718,
author = {Anahita Mohseni-Kabir And Changshuo Li And Victoria Wu And Daniel Miller And Benjamin Hylak And Sonia Chernova And Dmitry Berenson And Candace Sidner And Charles Rich},
title = {SLHAP: Simultaneous Learning of Hierarchy and Primitives},
booktitle = {Proceedings of Companion of the ACM/IEEE International Conference on Human-Robot Interaction (HRI '17)},
year = {2017},
month = {March},
pages = {412},
}
Copyright notice: This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.