面向人机团队的智能规划方法及实验研究

Abstract：A human-automation team (HAT) consists of human operators and automation systems that have mutual understanding and cooperate in decision making. The development of HATs is a recent trend in intelligence planning (IP) research, especially in the domain of high-tech and high-risk control systems. The keys to an effective HAT include the interchange of cognitive knowledge between the human and the machine and the machine's adaptation to the change of the human's cognitive state. The construction of IP algorithms with these capabilities is the major task of forming an efficient HAT. This study aims to develop IP algorithms that facilitate the forming of HATs, establish systematic indexes and methods for the evaluation of HAT performance, and validate the algorithms and methods with experiments. First, the proposed study will conduct cognitive task analyses on typical planning problems and identify the contents, timing, and methods for the dynamic interchange of cognitive information between the human and the machine. Second, the proposed study will develop IP algorithms that support the dynamic interchange of cognitive information, including algorithm representations for high-level human cognitive parameters, input ports and transformation functions for human cognitive factors, and algorithms for human comprehensible outputs. Third, the proposed study will establish systematic evaluation indexes of HAT performance from three aspects including algorithm, human, and team and develop systematic evaluation methods including cognitive-architecture-based modeling and experiment validation. Finally, experiments with unmanned aerial vehicle control and space vehicle control scenarios will be conducted to test and examine HATs using IP developed in this study. Results will be used to analyze the effects of different IP designs on HAT performance. Discussion and conclusion will be made about the principles of effective HAT.

人机团队指人与系统构成相互理解、共同决策的团队，是高新领域智能系统的发展方向。形成高效人机团队的关键在于人机之间的双向认知交互以及系统动态适应人的认知状态变化，而智能规划方法是实现人机团队构造的核心。因此，本研究首先针对高新领域典型规划问题进行认知任务分析，为人机的动态认知交互确定交互内容、时机和方式。在此基础上，开展支持动态认知交互的智能规划方法研究，提出具备高层控制接口的规划模型，确定人因要素的映射方法，设计对应的规划求解和解释生成算法，从认知决策知识嵌入、决策解释生成、动态适应认知状态变化三个方面支持人机动态认知交互的形成。针对以上人机团队，从算法、人和团队三个角度提出团队绩效评价指标，设计基于认知建模的仿真评价方法以及实验验证方法，结合无人机指控、航天器驾驶等任务规划问题，分析人机团队中人和机（规划系统）的交互层次、强度和团队绩效之间的关系，探索人机形成高效决策团队的规律和原理。

人机团队指人与系统构成相互理解、共同决策的团队，形成高效人机团队的关键在于人机之间的双向认知交互以及系统动态适应人的认知状态变化。围绕面向人机团队的智能规划算法及其实验验证，本项目重点开展了以下四点研究内容：（1）认知任务分析及认知建模：针对人控交会对接等典型任务进行了认知任务分析，并进行了认知失误建模，为人机的认知交互确定交互内容和方式。（2）动态认知交互的智能规划方法：从高层认知知识嵌入和利用、高效的知识学习、高认知压力下的在线规划方法等三个方面开展了支持动态认知交互的智能规划方法研究，提出了对应的规划模型与算法，并进行了实验验证及应用研究。（3）人机团队绩效评价：针对人机团队的绩效机理，研究了自动化水平及人的认知特征对人机团队绩效的影响机理和规律，从不同绩效评价角度提出了人控交会对接的绩效评价指标及模型，提出了机械臂遥操作的团队绩效评价指标。（4）人机团队实验研究：针对人控交会对接、机械臂遥操作等人机团队场景，设计实验验证方法，分析了人机团队中人和机的交互层次（不同自动化水平）、人的技能水平和团队绩效之间的关系。.在项目执行期内，项目组共发表论文18篇。其中，在《Computational Intelligence》、《Decision Support Systems》、《Frontiers in Psychology》等人工智能、人因工程相关领域知名SCI期刊发表论文11篇，发表EI期刊论文1篇，EI国际会议论文5篇，国内核心期刊论文1篇，另有论文6篇处于撰写和修改。项目资助培养青年人才1名，培养毕业（即将毕业）博士研究生5人，硕士研究生4人，研究成果支撑了项目负责人所在团队成功申请教育部“指控组织设计与优化”科技创新团队。.本项目探索了人与机形成团队的关键技术，建立了人机之间高效协同的“桥梁”，为人机结合的新型智能模式的发展提供了支撑，并在航天员训练、航母甲板调度、某大型传感器任务规划等国防相关领域等领域得到了推广与应用，保障了神九和神十等重大任务中的航天员训练，申请国防专利3项。