151st ASA Meeting, Providence, RI

Robots That Listen:
Mimicking Human Hearing to Help Robots Find Snipers

Socrates Deligeorges - sgd@bu.edu
Aleks Zosuls - azosuls@bu.edu
David Mountain - dcm@bu.edu
David Anderson - da@bu.edu
Allyn Hubbard - aeh@bu.edu
Department of Electrical and Computer Engineering,
Department of Biomedical Engineering,
and Boston University Hearing Research Center,
44 Cummington St.
Boston, MA 02215.

Popular version of paper 2aSP1
Presented Tuesday morning, June 6, 2006
151st ASA Meeting, Providence, RI

Robots may become a soldiers best friend when dealing with snipers and other hazardous situations in urban environments.  Small, portable robots equipped with advanced sensors are being used for reconnaissance and surveillance missions to detect and locate hidden gunmen before troops expose themselves to hostile gunfire. A snipers location can be revealed once a weapon has been fired, if the robot is equipped with an acoustic direction finding (ADF) capability and can quickly point its optical sensors at the sound source.  Accurately localizing and identifying sound sources in urban environments is especially difficult, however, due to background noise and reverberation.

Many of the problems associated with working in complex acoustic environments have now been overcome using a new acoustic signal-processing approach based on models of human hearing.  This biomimetic (copying biology) approach replicates the signal processing that takes place in the human inner ear and brainstem. The system uses frequency and timing cues exploited by the auditory system. These features include time delays between microphones, level differences between microphones, and frequency profiles.
 

The black box on top of the robot head is the acoustic direction finding unit

The initial system of direction-finding algorithms was designed and tested using the EarLab simulation software (http://earlab.bu.edu).  The ADF electronics were designed and built in Boston Universitys VLSI and Neural Net Systems Laboratory (http://www.bu.edu/vnns) and mounted on an iRobot Packbot .  The system was programmed to detect and localize gunfire and orient the robots camera towards the shooter. Positioning errors in outdoor field tests with live fire were less than 2 degrees.  The robot also performed well during indoor testing in highly reverberant spaces such as the atrium of the Boston University Photonics Center using simulated gunfire.
 

The robot patrolling an urban environment

The performance of the prototype platform demonstrates the potential of the biomimetic approach and its applications to practical problems for commercial, civilian, and military acoustic processing.