Acoustic surface perception from naturally occurring step sounds of a dexterous hexapod robot

Authors / Editors

Research Areas

Publication Details

Output type: Journal article

Author list: Ozkul MC, Saranli A, Yazicioglu Y, Vázques G

Publisher: Elsevier

Publication year: 2013

Journal: Mechanical Systems and Signal Processing (0888-3270)

Volume number: 40

Issue number: 1

Start page: 178

End page: 193

Number of pages: 16

ISSN: 0888-3270

eISSN: 1096-1216

Languages: English-Great Britain (EN-GB)

• View in Web of Science
•  | 
• View on publisher site
•  | 
• View citing articles in Web of Science

Unpaywall Data

Open access status: closed

Abstract

Legged robots that exhibit dynamic dexterity naturally interact with the surface to generate complex acoustic signals carrying rich information on the surface as well as the robot platform itself. However, the nature of a legged robot, which is a complex, hybrid dynamic system, renders the more common approach of model-based system identification impractical. The present paper focuses on acoustic surface identification and proposes a non-model-based analysis and classification approach adopted from the speech processing literature. A novel feature set composed of spectral band energies augmented by their vector time derivatives and time-domain averaged zero dossing rate is proposed. Using a multi-dimensional vector classifier, these features carry enough information to accurately classify a range of commonly occurring indoor and outdoor surfaces without using of any mechanical system model. A comparative experimental study is carried out and classification performance and computational complexity are characterized. Different feature combinations, classifiers and changes in critical design parameters are investigated. A realistic and representative acoustic data set is collected with the robot moving at different speeds on a number of surfaces. The study demonstrates promising performance of this non-model-based approach, even in an acoustically uncontrolled environment. The approach also has good chance of performing in real-time. (c) 2013 Elsevier Ltd. All rights reserved.

Keywords

Acoustic surface perception, Classification, Identification, Legged robots

Documents

No matching items found.