人与机器人协同研究：认知反馈与团队训练

Human-robot cooperation is a promising topic in the future, which not only makes it possible for robot to assist human with space operations like object placing but also provides possibility for human and robot to work on complex failure diagnosis tasks cooperatively in the harsh space environment. Since interactive interface and information display cannot satisfy the need of human-robot team any more, discussing the human-robot cooperation enhancement at cognitive and behavior level from aspects of mutual awareness and learning is necessary. For this reason, two methods are proposed for human-robot cooperation enhancement: 1) cognitive feedback based on Lens Model, in order to improve human’s understanding on robots from the cognitive level; 2) team training based on cross training and machine learning, in order to improve the mutual learning and understanding from behavior level. The purpose is to build an efficient human-robot cooperation by two-way interactions. According to these two methods, we are planning to design two kinds of typical human-robot cooperation tasks which are space orientation and failure diagnosis, build a simulation experiment platform and a robot hardware platform, and analyze the effect of these methods on team performance by conducting human-robot cooperation experiments. With analysis on the mechanism of different methods and the process of cognitive development in human-robot cooperation based on different tasks, we are aiming at providing theoretical basis and methods to the design of human-robot cooperation tasks and robot intelligence. The results would have board engineering applications in various areas, such as aerospace.

人与机器人协同作业是未来发展趋势，在太空恶劣环境中机器人不仅可以协助人完成物品放置等空间操作，还能与人协同执行复杂的故障判断任务。信息提供和交互界面已不能满足人与机组成团队的需要，这就需要从认知和行为层面、从相互理解与学习的角度，探讨人机协同问题。因此，本项目以增强人与机器人协同能力为目标，从人理解机器人的角度，基于透镜模型的框架，研究认知反馈方法；从相互理解学习的角度，将机器学习和交叉训练相结合，研究团队训练方法；两种方法从认知与行为两个层面，通过双向交互筑建高效人机协同。结合空间定位操作和故障判断决策等典型人机协同任务，建立实验仿真平台与机器人硬件平台，通过人机协同实验，分析不同方法对团队绩效的影响，结合不同类型任务，分析各方法的作用机理以及人机协同的发展过程，为人机协同任务设计以及机器人智能设计提出理论和方法。研究结果对于航天等领域的人机协同设计及训练也具有实际工程应用意义。

航空航天、航海及核电等复杂系统领域中多是团队协同任务，而人与机器人协同作业是未来发展趋势。随着人工智能技术的发展，未来机器人不仅可以协助人完成物品识别及放置等简单操作，还能与人协同执行复杂的故障判断任务。信息提供和交互界面已不能满足人与机组成团队的需要，这就需要从认知和行为层面、从相互理解与学习的角度，探讨人机协同问题。本项目分别以空间站机械臂操作、核电站应急故障处置、船舶操作任务、灾后救援勘探等为任务背景，开展了团队任务分析、团队协同理论研究、认知反馈方法研究、团队训练方法研究、团队协同相关测评研究以及人机器人协同的应用研究。首先是结合各领域的团队协同任务，分别采用认知任务分析、VACP分解、任务复杂度度量等方法进行了分析，探讨了人机功能分配方法或团队模式，进而分别探讨了自动化信任、自动化通明度、任务难度、团队结构和时间压力等对团队绩效的影响，接着以增强人与机器人协同能力为目标，从适应性认知的角度，研究了认知反馈方法；从相互理解学习的角度，结合机器学习和交叉训练理论，研究了团队训练方法；两种方法从认知与行为两个层面，增强了人与机器人的相互理解与学习。项目最后探讨了团队工作负荷和情境意识的评价方法，并结合空间机械臂及电力机器人开展了应用研究。本项目的研究成果在航天领域的空间站机械臂设计、核电站主控室系统确认、船舶领域的团队工作负荷分析、电力领域的机器人设计等方面开展了应用，还可指导相关领域的机械臂遥操作设计、任务复杂度分析以及相关操作人员训练等。