面向紧急任务的多智能体批次任务分配

Existing learning-based constructive task allocation methods require continuously generating a complete task allocation scheme before assigning tasks to agents, which fails to meet the real-time demands of large-scale urgent scenarios such as rescue or confrontation. To address this, a multi-agent batch task allocation method based on deep reinforcement learning is proposed in this paper. In this method, a policy model including an encoder, agent and task-node selection decoders, and a recursive embedding structure is designed that can generate a batch of partial task allocation schemes constructed by agent-task node pairs simultaneously according to the objective function’s optimality requirements. In online task allocation, agents no longer need to wait for the complete task allocation scheme before executing the tasks. The evaluation results showed that the proposed method improves the real-time performance, reliability, and cooperative capability of task allocation in urgent scenarios.