A Data-Acquisition Robot for Acoustic Sensing & Localization (Robot Audition)

Robot audition platform for DRDO, Govt. of India

Work done with Dr. Sumohana Chhannappayya and Dr. K Sri Rama Murty (IIT Hyderabad)

(July 2021 - May 2022)

Overview

Developed a complete robot audition platform for robust and precise sound source localization and tracking in outdoor environments. The system integrates an 8-channel microphone array mounted on a custom-built all-terrain robot with a height-adjustable mast (up to 1.7 m).

Sound Source Localization

The localization pipeline uses the Generalized Cross-Correlation with Phase Transform (GCC-PHAT) algorithm for time-difference-of-arrival (TDOA) estimation between microphone pairs. For microphones \(i\) and \(j\), the GCC-PHAT function is:

\[ R_{ij}(\tau) = \int_{-\infty}^{\infty} \frac{X_i(f) \, X_j^*(f)}{|X_i(f) \, X_j^*(f)|} \, e^{j 2\pi f \tau} \, df \]

where \(X_i(f)\) and \(X_j(f)\) are the Fourier transforms of the signals at microphones \(i\) and \(j\), and \(\tau\) is the time delay. The phase transform normalization makes GCC-PHAT robust to reverberation and noise in outdoor environments.

The TDOA estimates from multiple microphone pairs are combined using Steered Response Power (SRP) to estimate the direction-of-arrival (DOA) of the sound source, scanning over a grid of candidate angles:

\[ P(\theta, \phi) = \sum_{i < j} R_{ij}\!\left(\tau_{ij}(\theta, \phi)\right) \]

where \(\tau_{ij}(\theta, \phi)\) is the expected delay for direction \((\theta, \phi)\) based on the array geometry.

Microphone Array Specifications

ParameterValue
Channels8
Array geometryUniform circular array (UCA)
Array diameter0.2 m
Sampling rate44.1 kHz
Mast height range0.5 - 1.7 m (adjustable)

Robot Platform

  • Custom-built all-terrain mobile robot with differential drive
  • Onboard HDD for live data storage from state estimation and sensor streams
  • 4G LTE communication for remote data streaming and robot control
  • Integrated IMU, GPS, and wheel encoders for state estimation
  • ROS-based software architecture for sensor integration and real-time processing
Data acquisition robot