Sound Source Tracking as a Heuristic for Frontier Exploration in Search and Rescue Using a Quadruped Mobile Robot

Abstract

Teleoperated field robots used in disaster events require teams of operators to act as both the decision making systems and perform the stressful work of searching for victims. Current disaster robots are limited in autonomy and perception because they do not incorporate the same combination of visual and audio based perceptive abilities as human rescue workers. This thesis describes the development of a system that integrates autonomous exploration agents for disaster scenarios with robots that have the ability to interpret their surroundings through sound. The thesis presents an autonomous frontier exploration system that uses sound source tracking of human voices as a heuristic for search. The system is capable of exploring previously unknown environments and locating a target sound source that emits human speech. Robotic audition performed at discrete and continuous intervals was evaluated, establishing that measurements taken while the robot was stationary did not translate to measurable performance improvements. The system was implemented on a Boston Dynamics Spot quadrupedal robot equipped with a custom payload comprised of a secondary computer for the robot operating system, a power supply, and a four-microphone array Field trials were conducted in a laboratory environment, along forest hiking trails, and in mock urban disaster scenarios. Across 64 trials, a success rate of 69 % and a success weighted by path length score of 0.42 was achieved. Failure taxonomy for teleoperated disaster robots is argued as inadequate for the classification of autonomous experimental disaster robots. The thesis provides recommendations about an encompassing failure taxonomy for autonomous disaster robots, requiring the classification of failures to be system architecture agnostic and to account for both terminal and non-terminal failures. This research demonstrates that sound source tracking tuned for the human voice is a suitable heuristic that can bring autonomous search agents used in disaster scenarios closer in capability to the humans they assist.