Optimizing Programming by Demonstration for in-contact task models by Incremental Learning

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Sähkötekniikan korkeakoulu | Master's thesis

Date

2015-08-24

Department

Major/Subject

Space Robotics and Automation

Mcode

AS3004

Degree programme

SST - Space Science and Technology (TS2005)

Language

en

Pages

72+12

Series

Abstract

Despite the increasing usage of robots for industrial applications, many aspects prevent robots from being used in daily life. One of these aspects is that extensive knowledge in programming a robot is necessary to make the robot achieve a desired task. Conventional robot programming is complex, time consuming and expensive, as every aspect of a task has to be considered. Novel intuitive and easy to use methods to program robots are necessary to facilitate the usage in daily life. This thesis proposes an approach that allows a novice user to program a robot by demonstration and provides assistance to incrementally refine the trained skill. The user utilizes kinesthetic teaching to provide an initial demonstration to the robot. Based on the information extracted from this demonstration the robot starts executing the demonstrated task. The assistance system allows the user to train the robot during the execution and thus refine the model of the task. Experiments with a KUKA LWR4+ industrial robot evaluate the performance of the assistance system and advantages over unassisted approaches. Furthermore a user study is performed to evaluate the interaction between a novice user and robot.

Description

Supervisor

Kyrki, Ville

Thesis advisor

Montebelli, Alberto

Keywords

programming by demonstration, incremental learning, human-robot interaction, assistance

Other note

Citation