Repairable Robot Teams: Modeling, Planning, and Construction

Curt Bererton

Abstract

  To date, there has been no research which directly studies repair capabilities in robot teams. This work addresses this gap in the literature. To any robot or robot team designer the question of paramount importance is: ``How can I best design my robot team so that they achieve the best performance for the task or tasks that I have in mind?'' It is intuitive that the addition of repair capabilities should increase performance, but no one has ever looked at these effects. The effects of cooperative repair capabilities such as towing, cannibalistic repair, and multi-robot repair on the performance of a team of robots will be studied.

I study these effects in two different ways. The first is to model the entire repairable team using continuous time Markov models. This allows the calculation of a metric similar to the well know reliability metric of mean time to failure. The performance of teams with a wide variety of repair capabilities can then be compared using this metric.

The second point of view is to compare a variety of different teams performing a task. I have chosen to compare different teams using an adversarial task with a clear metric of performance. Robot laser tag provides a challenging task in which teams with a variety of repair capabilities and a variety of planning methods can be pitted against each other. Comparisons are performed using the obvious metric of ``Who wins more often?''. Two promising multi-robot coordination planning methods will be compared for use in teams of robots with repair capabilities. The first is a novel Markov decision based approach. The second is a Market based approach to planning. Contributions in both planning domains will be made.

The modeling and planning of repairable robots cannot only be done in simulation. If it were, one would never know if the truly important factors were modeled or accounted for during planning. To address this issue I have constructed several repairable robots including robots that repair and tow each other. The result is that important effects seen in using the actual robots are modeled and will be accounted for during planning.

The result of the thesis will be the first good look that anyone has taken at repair capabilities for robots. I will give the reader methods by which to model repair, comparisons of different planning methods for repair, and insights into the building of repairable robots.


Back to the Main Page

Charles Rosenberg
Last modified: Mon Feb 3 13:25:28 EST 2003