Home // ADAPTIVE 2014, The Sixth International Conference on Adaptive and Self-Adaptive Systems and Applications // View article

OfficeMate: A Study of an Online Learning Dialog System for Mobile Assistive Robots

Authors:
Steffen Müller
Sina Sprenger
Horst-Michael Gross

Keywords: online learning; dialog system; probabilistic planner

Abstract:
Service robots in the near future are supposed to live together with humans in their private homes for a longer time period. In this situation, experience and attitudes of the users change and thus, the robot has to develop its behavior, too, and it has to adapt to the user's way of interaction and the user's needs. The contribution of this paper is a probabilistic decision planner implementing the idea of online learning dialog strategies for a mobile service robot in long-term interaction. The planning system is part of a modular multi-modal dialog system and allows for an autonomous personalization of the robot's actual interaction behaviors. A model of observed transitions and user's rewards using mixtures of discrete samples is proposed for efficient inference in a factor graph model. The practicability of the dialog system and the rewarding mechanism have been evaluated in a ten day realworld experiment with 16 users.

Pages: 104 to 110

Copyright: Copyright (c) IARIA, 2014

Publication date: May 25, 2014

Published in: conference

ISSN: 2308-4146

ISBN: 978-1-61208-341-4

Location: Venice, Italy

Dates: from May 25, 2014 to May 29, 2014