On the design of behavior-based multi-robot teams

Real-world applications that are ideal for robotic solutions are very complex and challenging. Many of these applications are set in dynamic environments that require capabilities distributed in functionality, space or time. These applications, therefore, often require teams of robots to work together cooperatively to successfully address the mission. While much research in recent years has addressed the issues of autonomous robots and multi-robot cooperation, current robotics technology is still far from achieving many of these real world applications. We believe that two primary reasons for this technology gap are that (1) previous work has not adequately addressed the issues of fault tolerance and adaptivity in multi-robot teams, and (2) existing robotics research is often geared at specific applications and is not easily generalized to different, but related, applications. This paper addresses these issues by first describing the design issues of key importance in these real-world cooperative robotics applications - fault tolerance, reliability, adaptivity and coherence. We then present a general architecture addressing these design issues - called ALLIANCE - that facilitates multi-robot cooperation of small- to medium-sized teams in dynamic environments, performing missions composed of loosely coupled subtasks. We illustrate the generality of this architecture by describing two very different proof-of-concept implementations of this architecture: a janitorial service mission and a bounding overwatch mission.