Regresar

A control-theoretic perspective on the design of distributed agreement protocols

Abstract:

We present a control-theoretic perspective on the design of distributed agreement protocols. First, we explore agreement-protocol analysis and design for a network of agents with single-integrator dynamics and arbitrary linear observations. One key contribution of our work is the analysis of protocols for networks with quite general observation topologies, including with multiple observations made by each agent. Another contribution is the development of techniques for agreement law design - i.e. for assignment of the dependence of the agreed-upon value on the initial states of the agents. Second, we explore agreement in a quasi-linear model with a stochastic protocol, which we call the controlled voter model. We motivate our study of this model, develop tests for whether agreement is achieved, and consider design of the agreement law. Third, we expose one application of agreement law design, to distributed maximum a posteriori probability (MAP) decision making and hidden Markov model (HMM) estimation. Finally, we provide some further thoughts regarding our control-theoretic perspective on agreement, including ideas for fault-tolerant protocol design using our approach. Copyright © 2006 John Wiley & Sons, Ltd.