Decentralized task allocation for dynamic environments

Abstract

This thesis presents an overview of the design process for creating greedy decentralized task allocation algorithms and outlines the main decisions that progressed the algorithm through three different forms. The first form was called the Sequential Greedy Algorithm (SGA). This algorithm, although fast, relied on a large number of iterations to converge, which slowed convergence in decentralized environments. The second form was called the Consensus Based Bundle Algorithm (CBBA). CBBA required significantly fewer iterations than SGA but it is noted that both still rely on global synchronization mechanisms. These synchronization mechanisms end up being difficult to enforce in decentralized environments. The main result of this thesis is the creation of the Asynchronous Consensus Based Bundle Algorithm (ACBBA). ACBBA broke the global synchronous assumptions of CBBA and SGA to allow each agent more autonomy and thus provided more robustness to the task allocation solutions in these decentralized environments.