Hybrid control systems: user-interface design

人与自动化的交互无处不在，发生在消费品、交通系统、生物医学系统、关键基础设施和科学研究平台中。人与自动化交互中的许多问题已导致昂贵、高风险和安全关键系统的重大故障，但几乎没有数学方法或工具来分析人-自动化系统。为了解决这一空白，我的研究帮助在系统建立、测试和部署之前识别人-自动化交互中的问题。我的方法基于混合控制系统，它同时建模(从物理过程)连续动力学和(从自动化模式-逻辑)离散动力学。几十年来，由人类和由自动化控制的系统(也称为半自动、混合主动或人工自动化系统)一直未能得到控制工程师的严格量化。混合可达性分析和控制器综合中的新技术模拟了人类上级的特权，但允许易于处理的控制计算。其结果是混合控制律，它为向用户显示的建议提供信息内容。我计划推广这些混合方法和工具，以验证和设计混合系统的用户界面，并应用于飞行管理系统、电网和人工监督的麻醉。这项研究的预期影响是增加人们对人类与自动化交互的信心，并减少已部署系统中的人类自动化故障。通过在实际系统部署之前识别数学模型中的此类错误，可以防止安全关键系统中的现实问题。数学和计算工具可以通过预测故障可能发生的位置以及人类操作员如何防止故障，在这些系统中创建新的信心和可靠性水平。

Human interaction with automation is ubiquitous, occurring in consumer products, transportation systems, biomedical systems, critical infrastructures, and scientific research platforms.  Problems in human-automation interaction have  resulted in major failures in expensive, high-risk, and safety-critical systems, yet few mathematical methods or tools to analyze human-automation systems exist.  To address this void, my research helps identify problems in human-automation interaction before systems are built, tested, and deployed.  My methods are based in hybrid control systems, which model both continuous dynamics (from physical processes) and discrete dynamics (from the automation's mode-logic). Systems controlled both by humans and by automation (also known as semi-automated, mixed-initiative, or human-automation systems) have eluded rigorous quantification by controls engineers for decades.  New techniques in hybrid reachability analysis and controller synthesis model the human upservisor's prerogative, yet allow for tractable control computations.  The result is a hybrid control law which provides the information content for advisories to be displayed to the user.  I plan to generalize these hybrid methods and tools to verify and design user-interfaces for hybrid systems, with application to flight management systems, the power grid, and human-supervised anesthesia. The anticipated impact of this research is to increase confidence in human interaction with automation and to reduce human-automation failures in deployed systems.  By identifying such errors in mathematical models, before actual systems are deployed, real-life problems in safety-critical systems can be prevented.  Mathematical and computational tools can create a new level of confidence and reliability in these systems, by predicting where failures might occur, and how human operators can prevent them.