Multi-agent task planning and allocation

Abstract

We study the problem of multi-agent task allocation and planning with local communication between agents. Centralized planning algorithms for task allocation limits the range of the mission and the fixed location where the agents are required to communicate act as single failure point. Decentralized algorithms are thus required to increase the mission range and to remove single point failure. In this report, we propose a better solution for two existing decentralized algorithm for task allocation which is CBAA (Consensus Based Auction Algorithm) and CBBA (Consensus Based Bundle Algorithm). The CBAA and CBBA algorithms are based on auction algorithm and consensus based decision strategy among local communication between agents, and proven to converge to a conict free assignment with greater than 50% optimality. Our solution build on these algorithm, and uses an incremental approach to find better solution. The algorithm utilizes a gradient ascent algorithm strategy to create a disturbance in task allocation assignment and repeat the incremental solution approach to look for a better solution. The approach is validated using simulations. Four local communication structure between agents are simulated, and varied randomly with increase in number of agents and tasks. The result with comparison among existing and our solution are reported with graphs.