Building better swarms through competition: lessons learned from the AAAI/RoboCup Rescue Robot competition

A common problem in a disaster situation, whether man-made or natural, is the need to quickly identify locations of potential victims and potential hazards to human rescuers, with minimal personnel on site at the beginning of the emergency. Robots can be used to supplement available personnel and to assist them in their search. However, it is important that the robots used not pose a danger to rescuers and victims through collisions or causing structural collapse and that they not get in the way of rescue efforts. Another problem in a disaster area is that environmental conditions (e.g., rubble, scrap metal, dust, smoke, and fire) can make it difficult for robots to maneuver, reduce effectiveness of sensors, and hamper communications. We believe the best approach to developing an effective robotic assistant to human rescue teams is through swarms of low-cost, small robots. The robots should be low cost to eliminate concerns about not recovering them after a disaster. They should be small so they don't hinder or pose a risk to humans (rescuers can kick them or step on them if they are in the way and they will be too small to move objects or injure a person through collision). And we believe a swarm should be used to maximize the area covered while maintaining enough contact between robots to maintain a reliable communications link. A swarm of small robots programmed with simple behaviors can achieve these goals while keeping costs down. This paper describes the efforts we have undertaken at Utah State University in developing urban search and rescue robot swarms for the American Association for Artificial Intelligence and RoboCup Rescue Robot competitions. Both of these competitions make use of the standard urban search and rescue test course developed by the National Institute of Standards and Technology (NIST). Our initial efforts to try to determine the bare minimum competence required for a rescue swarm are described, as are our efforts to improve communications between robots and a base station. Our current efforts to implement a flocking behavior in a robot swarm are described, as is the most recent modular, behavioral robot design we are working on. The development of a graphical user interface for use by rescuers to interface with the swarm is also described. Finally, we discuss the direction we think this research needs to take in the future.