Fast Convergent Distributed Task Reallocation Method for Dynamic Multi-Agent Systems

Abstract

To address the problem that existing distributed task allocation methods for multi-agent systems hard to simultaneously achieve high allocation performance and fast responsiveness under dynamic changes in tasks and agent states, this paper proposes a task allocation method based on ownership-priority-driven task inclusion and consensus. In the task inclusion phase, a greedy-selection-based route estimation method is first designed to compute the marginal benefit of each task to different agents. Then, a new concept of task ownership priority is introduced to measure the likelihood that a task belongs to each agent. Each agent preferentially includes tasks that both increase its marginal benefit and are more likely to belong to itself, thereby reducing task inclusion conflicts while maintaining allocation performance. In the consensus phase, conflicting assignments are resolved using a marginal-benefit-increase strategy. By reducing invalid competition and repeated iterations, the proposed method achieves convergence with fewer task inclusion and consensus iterations, thereby improving responsiveness in dynamic scenarios. Simulation results demonstrate the effectiveness of the proposed method.