New Challenges in Multi-Agent Intention Recognition : Extended Abstract
New Challenges in Multi-Agent Intention Recognition : Extended Abstract
复制标题
多智能体意图识别的新挑战：扩展摘要
DOI：
--
复制
发表时间：
2002
期刊：
影响因子：
--
通讯作者：
J. Wendler
中科院分区：
文献类型：
--
作者：
G. Kaminka;J. Wendler
操作模式到下一个（作为任何这样的抽象模式的一部分，每个代理将执行不同的动作）。因此，代理人应该就团队所处的模式达成一致。在团队应用程序中，团队成员共同执行联合计划的协议确实很常见（Jennings 1995; Tambe 1997）。其次，在任务执行过程中，智能体之间的日常通信在某种程度上是可预测的，事实上，这种预测是可以学习的。通过消除不符合一致性关系的建模假设（即，代理不一致的假设），或者不符合预测的通信（即，假设代理在不发送预测消息的情况下切换模式），可视化的准确性平均增加到84%（在某些实验中高达97%）。此外，通过将建模算法仅限于这些假设，我们能够实现约
mode of operation to the next (as part of any such abstract mode each agent would carry out different action). Thus agents were supposed to be in agreement as to the mode the team is in. Agreement on the joint plan to be executed by team-members i indeed common i teamwork applications (Jennings 1995; Tambe 1997). Second, the routine communications between agents during task execution were somewhat predictable, and in fact such predictions were amenable to learning. By eliminating modeling hypotheses that did not conform to the agreement relationship (i.e., hypotheses where agents were not in agreement), or did not conform to the predicted communications (i.e., hypotheses where agents switched mode without sending a predicted message), the accuracy of visualization was increased to 84% on average (up to 97% in some experiments). Furthermore, by restricting the modeling algorithm to only these hypotheses, we were able to realize computational space and time savings of about