CAREER: Cooperative Control and Decision-Making for Human-Agent Collaborative Teams

该学院早期职业发展（CAREER）计划资助解决了人类-机器人协作团队的控制和决策问题。该项目有两个主要研究方向。第一个重点考察了一个团队，该团队由一名熟练的人类和一名经过部分训练的机器人助手组成，该机器人助手能够无休止地重复工作，而不会感到无聊或疲劳。人类和机器人之间的工作负载平衡由人类对机器人的信任决定，这在数学上被建模为性能改进率和错误数量减少率的函数。创新的基于信任的算法将提供平衡的人类体验和有保证的团队绩效。第二个研究重点将创建新的规划策略，结合后悔的数学模型，这是人类理性决策的核心情感。基于后悔的自动决策辅助系统将为更自然的人机交互提供更人性化的决策。这两个项目的成果最终将使变革性的人机交互技术受益于美国经济和生活质量。该项目的教育举措将扩大代表性不足的群体在制造和机器人研究中的参与。技术方法需要将信任整合到合作控制中，并将遗憾整合到人类-智能体协作团队的决策中。为了探索信任的新的基本理解和实现有效的控制分配，第一个推力将开发新的动态信任模型的基础上定性的结果，从人的因素的研究，和新的基于信任的控制策略的切换系统建模的人-代理协作团队。将开发基于非保守多个李雅普诺夫函数的分析，以提供手动和自主模式的状态相关切换控制。第二个推力的特点是发展一个基于后悔的贝叶斯顺序决策框架，选择手动和自主模式之间的方式，使次优的决定将作出，以避免可能的遗憾作出错误的决定。这两个推力包括实验验证使用异构多机器人测试床和人形制造机器人与人在循环。这项研究的结果将促进控制理论和人为因素之间的新接口。

This Faculty Early Career Development (CAREER) Program grant addresses control and decision-making for human-robot collaborative teams. The project has two main research thrusts. The first thrust examines a team consisting of a skilled human with knowledge of a certain manufacturing or sensing task, and a partially trained robot assistant that is capable of endless repetition without boredom or fatigue. The workload balance between the human and robot is governed by the trust that the human has for the robot, which is modeled mathematically as a function of the rate of improvement in performance and the rate of decrease in number of mistakes. Innovative trust-based algorithms will provide a balanced human experience and guaranteed team performance. The second research thrust will create novel planning strategies incorporating mathematical models of regret, an emotion central to human rational decision-making. Regret-based automatic decision-making aids will provide more human-like decisions for more natural human-robot interaction. Results from both thrusts of the project will ultimately enable transformative human-robot interaction technologies benefitting the U.S. economy and quality of life. The educational initiatives of this project will broaden participation of underrepresented groups in manufacturing and robotics research.The technical approach entails the integration of trust into cooperative control and regret into decision-making for human-agent collaborative teams. To explore new fundamental understanding of trust and realize effective control allocation, the first thrust will develop new dynamic trust models based on qualitative results from human factors research, and novel trust-based control strategies for human-agent collaborative teams modeled by switched systems. Non-conservative multiple Lyapunov functions based analysis will be developed to provide state-dependent switching control of the manual and autonomous modes. The second thrust features the development of a regret-based Bayesian sequential decision-making framework that selects between the manual and autonomous mode in a way such that suboptimal decisions will be made to avoid the possible regret of making a wrong decision. Both thrusts include experimental validations using a heterogeneous multi-robot test bed and a humanoid manufacturing robot with human-in-the-loop. Results from this research will foster a new interface between control theory and human factors.