基于动作链的形式化任务协同规划  

Collaborative planning of formal tasks based on action chains

作　　者：刘泽森 李忠奎 国萌 Zesen LIU;Zhongkui LI;Meng GUO(College of engineering,Peking University,Peking 100871,China)

机构地区：[1]北京大学工学院,北京100871

出　　处：《中国科学：信息科学》2024年第11期2623-2641,共19页Scientia Sinica(Informationis)

基　　金：国家自然科学基金(批准号:U2241214,62203017,T2121002)资助项目。

摘　　要：基于形式化方法的多智能体任务规划因其丰富的任务形式和多样的系统功能而备受关注.然而,随着智能体数量增加,规划复杂度呈指数级增长,因此形式化方法在计算效率和集群规模上都受到了限制.已有的改良方法中,基于图搜索的方法对计算效率改进有限,只能处理中等规模的集群;而基于组合优化的算法忽略了智能体动作模型,无法处理大规模协同任务.本文针对以上局限,提出了基于智能体动作链的偏序任务分配算法,在保证正确性的基础上大幅提升规划性能.进一步针对环境不确定性和在线新任务,设计了自适应重规划算法和协作同步机制,确保任务执行过程中的鲁棒性.最后,通过数值仿真和对比验证了算法的有效性和可靠性.Multiagent task planning based on formal methods has attracted considerable attention due to its ability to handle diverse task specifications and complex systems.However,as the number of agents increases,the planning complexity grows exponentially.Existing search-based methods can only handle medium size fleets while integer program-based algorithms cannot incorporate collaborative actions.To address these issues,this paper proposes a novel planning scheme based on action chains,where tasks are assigned via partial order optimization.Furthermore,an online adaptation and synchronization algorithm is proposed to handle contingent tasks that are generated online and uncertainty during task execution.Extensive numerical simulations are conducted to validate the effectiveness and reliability.

关 键 词：多智能体任务规划 形式化方法 线性时序逻辑 在线自适应 偏序集 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]