Abstract
Capabilities in Heterogeneous Mu University of New Sout jen1 Introduction Groups of robots are often able to accomplish missions that no single robot can achieve by themselves. Robustness and flexibility are increased by the diversity of the robots, each contributing different capabilities. In highly unpredictable domains such as search and rescue, accurate predictions of the outcomes of a robot’s actions are virtually impossible. Approximate models and algorithms are required which help to estimate the outcome with highest possible confidence. Although many aspects of hetero-geneous multi-robot systems have been widely studied, few researchers explicitly formalize robot capabilities. A model of capabilities can prove very useful for describing and reasoning about robots’ diversity, task suitability and execution.This work presents a framework that formalizes a robot’s capabilities, abstracting from underlying robot architectures and providing a means to estimate a robot’s performance in a task. 2 Related work One key element in multi-robot systems is to assign tasks to robots such that a meaningful division of work is achieved. For estimating “expected quality of task execution”, utility is a widely used concept in multi-robot coordination. The goal of task allocation is to find robot-task assignments such that the overall utility is maximized [Gerkey and Matarić, 2004]. Many approaches to compute such a utility measure have been proposed, but only a few explicitly consider dif-ferent notions of robot capability. Most of this work relates capabilities to some kind of resource, e.g. sensors/actuators, processing capacities [He and Ioerger, 2003], [Chen and Sun, 2010], and/or software modules [Parker and Tang, 2006]. A capability can also be a simple subtask, for which each robot learns their suitability [Fua and Ge, 2005]. Such concepts are ultimately used in different ways to determine a robot’s utilitfor a task. Other research also formalizes capabilities relatinto robot components to infer what a robot can do [Kunze et al., 2011] or how to decompose a task into simple ’skills’ [Huckaby and Christensen, 2012]. Previous research has taken into account a robot’s intrinsic capabilities1 for estimating such utility values [Fua and 1 Intrinsic capabilities express what a robot can generall lift a rock); extrinsic factors specify task details (i.e., wei