Agent rendezvous: A dynamic symmetry-breaking problem

Abstract

We consider the problem of a rendezvous (coordinated meeting) of distributed units (intelligent agents in network computing or autonomous robots). The environment is modeled as a graph, the node labeling of which may not be “common knowledge” to the units, due to protocol and naming convention mismatch, machine faults, status change, or even hostility of the environment. Meeting of such units is likely to be a basic procedure in the area of distributed “intelligent agent” computing and in the domain of coordinated tasks of autonomous robots. The crux of the problem which we present here and initiate research on, is the breaking of potential symmetry while the units dynamically move. The units are more intelligent (computing power, control and memory) than simple (traditional) pebbles or tokens, and our algorithms will make use of this capability for speeding up the convergence to a common place (e.g., we will allow units to meet exchange information and depart). We consider both randomized protocols and deterministic (but non-uniform) protocols; the problem is unsolvable by a uniform deterministic algorithm. The deterministic procedure employs ideas from design theory and achieves Õ(n) time, while the randomized methods are based on random walks and may achieve Õ(n) time where k is the number of agents.

Work partially supported by NSF grant CCR-93-16209 and CISE Institutional Infrastructure Grant CDA-90-24735.