A Improved PI-Based Distributed Scheduling Method with Task Removal Prediction and Convergence Guarantee

Multi-task distributed scheduling is one of the key technologies for multi-agent systems, which is widely used in search and rescue, firefighting and other fields. The Performance Impact (PI) algorithm is among the most popular since it is efficient in terms of optimization. How to further improve the optimization of the baseline PI algorithm and solve its non-convergence is a problem worth studying. In this paper, a novel distributed scheduling method is proposed, which is obtained by integrating the task removal prediction and deadlock avoidance strategy to PI. First, a formal description of the classic search and rescue problem is introduced, together with its models given. Second, through a case study, it is found that PI may fall into suboptimal solution by relying on heuristic method to select tasks locally. To this end, the task removal prediction is proposed to be inserted in the task removal phase of PI, thereby improving the exploration properties. Third, since PI may suffer from occasionally falling into an infinite cycle of exchanging the same task, a deadlock avoidance strategy is designed to limit the number of times to remove and include the same task to guarantee the algorithm's convergence. Finally, with the application background of search and rescue, a large number of Monte Carlo simulation experiments are carried out from the aspects of optimization and convergence. The results show that, compared with the baseline PI algorithm, the proposed method can obtain a lower average time cost and the highest number of task assignments with the same number of iterations within the limited capability of agents. In addition, the convergence of the algorithm can be ensured.