ABSTR ACT
This research proposes a motion planning
framework for a robot equipped with a bioinspired
hearing system to enhance its sound source
localization, specifically for gas leak detection.
Similar to how animals use movements, such as
changing positions or rotating their ears, to precisely
localize a sound source, this framework enables the
robot to improve its estimation of both direction and
distance through strategic motion. The proposed
method uses the time difference of arrival (TDOA)
between two wideband optical microphones,
obtained through this motion, to accurately estimate
the direction and distance of a sound source. We first
introduce the underlying concepts and validate the
approach using experimental data collected with
a collaborative robotic arm. We then demonstrate
the implementation of the method on a quadruped
robot. The results show that the method can be
effectively deployed on mobile robots, achieving
directional and distance accuracy for gas leak
localization.
KEYWORDS: gas leakage, source localization, quadruped
robot, robotic dog, robotic inspection
Introduction
Gas leakage is a significant source of failure in various indus-
tries, including oil and gas, clean energy systems such as
hydrogen-based fuel cells, and the automotive sector. Early
detection of gas leaks is crucial to prevent catastrophic failures,
ensure safety, and reduce environmental impact. Gas leakage
detection and localization involve using a wide range of gas
sensing hardware, including metal oxide (MOX) gas sensors,
e-noses, and photoionization detectors (Francis et al. 2022).
Relying on human labor is demanding, while installing sensor
equipment can be expensive and often financially unfeasible.
As a result, there is growing interest in using mobile robots for
inspection, such as Boston Dynamics’ Spot (Guizzo 2019).
While the use of robots for industrial inspection is not
new, the application of mobile robots for inspection remains
an active research area. A review of various pipeline robots,
considering factors such as inspection accuracy, adaptability,
cost, and efficiency (Ambati et at. 2020), as well as the develop-
ment of inspection robots for nuclear power plants (Yamamoto
1992 Fu et al. 2006), demonstrates a long-standing interest
in this field. Furthermore, the use of robots has expanded
into other industrial applications, including waste deposits
and electric power cables (Jiang et al. 2005) civil engineering
tasks like inspecting building facades and concrete structures
(Ilyas et al. 2021) examining structures such as ships, wind
turbines, and aircraft (Almadhoun et al. 2016) and underwater
facility inspections (X. Wang et al. 2022). More recently, several
systems have been proposed for surveying nuclear power
plants to assess radiation levels (Abadi et al. 2023 Mitchell
et al. 2023). In the context of gas source localization, Francis
et al. (2022) provide an overview of gas detection principles
and review existing works that propose the use of mobile
inspection vehicles.
Since vertebrates became terrestrial 370 million years ago,
reliable information has been essential for making survival
decisions (Munoz and Blumstein 2020 Thewissen and
Nummela 2008). In our paper, we draw inspiration from the
remarkable evolutionary strategies employed by vertebrates in
their survival and adaptation within the animal kingdom. The
fundamental principle of aeroacoustics leak detection revolves
around the generation and propagation of sound waves when
gas escapes from a pipeline (S. Wang and Yao 2020). These
acoustic signals carry crucial information about the leak, such
NDTTECHPAPER
|
ME
GUIDED BY HEARING: ROBOTIC MOTION
PLANNING FOR PRECISION IN GAS LEAKAGE
LOCALIZATION
MOHAMMAD NASR†, NIHAR MASURKAR†, POURIA MESHKIZADEH†, AND EHSAN DEHGHAN-NIRI†*
*Corresponding author: nde@asu.edu
† Intelligent Structures and Nondestructive Evaluation (ISNDE), School
of Manufacturing Systems and Networks, Arizona State University,
Mesa, AZ 85212
Materials Evaluation 83 (4): 51–61
https://doi.org/10.32548/2025.me-04491
©2025 American Society for Nondestructive Testing
A P R I L 2 0 2 5 • M AT E R I A L S E V A L U AT I O N 51

as its magnitude, location, and potential impact on the pipe-
line’s integrity (Liang et al. 2013). The main advantages of using
acoustic signals for gas leak diagnosis include swift response
time, long range, and precise localization (S. Wang and Yao
2020). A key drawback of using acoustic signals is their reduced
sensitivity as the sensors move farther away from the source
due to significant attenuation (Liu et al. 2017). This attenua-
tion is even more pronounced with higher-frequency sounds
(ultrasound), which are often involved with smaller leakages.
One effective approach to mitigate this problem is by employ-
ing an array of microphones along with acoustic imaging
techniques (Li et al. 2021 Fluke 2021 Drives &Controls 2021).
However, this method generates a substantial amount of data,
which is not ideal for a near-real-time approach required for
creating a semi-autonomous platform and quick response.
Fortunately, this issue has been ingeniously addressed by bio-
logical systems in their evolutionary acoustic sensing systems,
using only two ears (binaural sensing). Instead of relying on
increased sensor numbers to enhance sensitivity, these bio-
logical systems evolved their external ear (pinna) shapes to act
as hardware for magnifying sound coming from specific direc-
tions. This evolution was further complemented by strategic
motion and motion control of the pinna and head as well as
their movement (Populin and Yin 1998 Fletcher 2014), allowing
them to dynamically change acoustic directionality and sen-
sitivities as needed. Inspired by these biological mechanisms,
we propose in this paper an acoustic sensing system that com-
prises two microphones, mimicking two ears and binaural
hearing, with a framework designed to optimize or strategize
movement to improve source localization accuracy.
Sound source localization (SSL) is a prominent topic in
robotics, with a detailed review of common methods provided
by Rascon and Meza (2017). While gas and air leak local-
ization is essential for industrial inspections and has been
explored in earlier research, many existing approaches rely on
fixed microphone arrays (Eret and Meskell 2012), handheld
devices (Liao et al. 2013) for manual detection, and statistical
time-domain features for pinpointing leaks in pipes (F. Wang
et al. 2017). Yan et al. (2018) used a four-element linear array to
identify multiple sources with the MUSIC algorithm within the
63–187 kHz emission range. Focusing on ultrasonic emissions
from leaks in pressurized pipes, Schenck, Daems, and Steckel
(2019) employed a peak search on a beam-formed spectrum
and use a 32-element microphone array. By leveraging
multiple poses of the robot and integrating simultaneous local-
ization and mapping (SLAM) techniques with the microphone
array, potential leaks are effectively localized. Most recently,
G. K. J. Fischer et al. (2024) introduced an autonomous robotic
system for comprehensive plant inspection, equipped with a
variety of sensors, including lidar, stereo, UV/IR/RGB cameras,
electronic noses, and five microphones. These sensors detect
factors like methane leaks, flow rates, and infrastructure issues.
The system was tested at a wastewater treatment site, achiev-
ing gas leak localization with a 50 cm error through acoustic
assessments.
Current robotic systems for industrial inspection rely on
various sensors, such as lidar, stereo, and depth cameras, to
map environments for navigation and distance estimation.
However, little attention has been given to optimizing the
robot’s movement and positioning to enhance leak localiza-
tion once an anomaly, such as sound generated by a leak, is
detected by the microphones. Typically, these systems use
microphone arrays to estimate the direction of arrival (DOA)
of sound, combined with techniques like SLAM and lidar to
locate the leak. While multi-microphone array techniques can
perform well from fixed positions, they often require more
complex hardware setups, increased costs, and larger physical
space. In contrast, the proposed method achieves accurate
localization using a simpler hardware setup (two microphones)
with lower computational demands. This is accomplished by
strategically leveraging the robot’s mobility, effectively trading
hardware complexity and computational power for movement.
Specifically, this paper explores how motion, combined with
a two-microphone array inspired by an animal’s external
auditory system, enhances sound source localization. In other
words, we focus on optimizing the robot’s movement strat-
egies to acquire samples that enable precise sound source
localization while maintaining a lightweight system configu-
ration. First, we examine the effect of motion using a fixed-
base collaborative robotic arm. Based on the insights gained
from this study, we propose a motion strategy to enhance the
robot’s source localization. Finally, we implement this strategy
on a quadruped robot to explore its effectiveness for mobile
platforms.
Method
This section provides a comprehensive overview of our meth-
odology, covering the sound source localization approach,
acoustic data acquisition process, robotic systems, and experi-
mental setup.
Source Localization Method
The time difference of arrival (TDOA) between two signals is
commonly estimated by calculating a cross-correlation vector
(CCV). The Pearson correlation factor, as shown in Equation 1,
provides the simplest way of calculating the CCV (Rascon and
Meza 2017): ​ ​
(1)​ CCV[τ]​ ​ =​
​_ ​_ ​Σ________________________​​)​​​​​2​−​]τ​​________________−_t[​​​​2a​​​(​​​)​​​​​1​−​]t[​​​​1a​​​(​​​​​____________t​​
√ Σt​​​(​​​a1​​​[t]​ ​ ​ − ​
​_
​ 1​​​​)​​​​2​ ​ ​ √ Σt​​​(​​​a2​​​[t ​ ​ − τ]​ − ​
​_
​ 2​​​​)​​​​2​​​​
where​​
a​1​​​ and ​​ ​ 2​​​ are the two discrete signals being compared,
τ is the time shift applied to ​​ ​ 2​​​ and ​ ​
_​ a​1​​​​ and ​​a​2​​​​ _​ are the mean values of ​​ ​ 1​​​ and ​​ ​ 2​​​ respectively.
Although computationally efficient and straightforward,
CCV-based TDOA estimation is highly susceptible to environ-
mental noise and reverberations, often leading to inaccurate
estimates (Brandstein and Silverman 1997). To mitigate these
challenges, the Generalized Cross-Correlation with Phase
ME
|
LEAKLOCALIZATION
52
M AT E R I A L S E V A L U AT I O N • A P R I L 2 0 2 5