CRII: CPS: Provably-safe Interventions for Human-Cyber-Physical Systems (HCPS)

人类与物理世界的互动越来越多地由自主调节，如飞机辅助飞行员，机器人辅助外科医生，汽车辅助司机。在这些系统中引入自动化是为了帮助人类并保证安全性和性能。然而，这样的保证很难提供，因为人类可能会误解自动化的意图或以意想不到的方式行事；法国航空（Air France） 447航班坠毁等悲剧性的例子表明，飞行员和自动驾驶仪之间的这种混淆可能会导致灾难性的后果。尽管某些应用可能有一天会实现完全自动化（例如，汽车已经可以在交通中自动驾驶），但在可预见的未来，与安全、问责制和不可否认性相关的法律和道德问题将确保人类和自主必须能够在许多此类系统中在多个层面上移交控制权。首席研究员（PI）建议在人类合作者存在的情况下灵活部署自治程度，以补偿任务或环境条件的变化。虽然PI专注于具体的机器人远程操作，但预期的结果将导致有利于人类在循环中的各种CPS的一般原则。为任何涉及人类的系统提供安全和性能保证是一个崇高的目标。PI建议通过以下方式实现这一目标：(i)将我们的努力集中在机器人远程操作的应用上；（ii）整合多个已建立的学术学科的发现，包括工程学科，如人的因素和控制理论，考虑人与动态物理过程之间的相互作用，以及科学学科，如神经机械运动控制和行为博弈论，解释人类如何在个体和群体中相互作用。通过开发预测建模和自动化干预的原则，为在城市地区分布的可证明安全的远程操作机器人提供运输、制造、远程医疗和应急响应服务铺平了道路。与目前不同的是，人类与自主性的目标或手段不兼容导致性能下降，从严重到灾难性，拟议的工作设想了一个未来，在这个未来，人类可以高度自信地安全地部署在社会的网络物理系统中。

Human interaction with the physical world is increasingly mediated by autonomy, as planes assist pilots, robots assist surgeons, and cars assist drivers. Automation is introduced in such systems to aid humans and guarantee safety and performance. However, such guarantees are hard to provide, since humans may misunderstand the automation's intentions or behave in an unanticipated manner; tragic examples like the crash of Air France flight 447 illustrate that such confusion between pilots and autopilots can lead to catastrophic outcomes. Although some applications may someday yield to full automation (e.g., cars can already drive themselves in traffic), legal and ethical concerns related to safety, accountability, and non-repudiation will ensure humans and autonomy must be capable of handing off control authority at multiple levels in many such systems for the foreseeable future. The principal investigator (PI) proposes to flexibly deploy degrees of autonomy in the presence of human collaborators to compensate for changes in task or environmental conditions. Though PI focuses on robotic teleoperation for concreteness, the anticipated results will lead to general principles that benefit a variety of CPS with humans in-the-loop.Providing safety and performance guarantees for any system involving humans is a lofty goal. PI proposes to achieve this goal by (i) targeting our effort on applications in robotic teleoperation and (ii) integrating findings from multiple established academic disciplines, including engineering disciplines like human factors and control theory that consider the interaction between people and dynamic physical processes, as well as scientific disciplines like neuromechanical motor control and behavioral game theory that account for how humans interact individually and in groups. The proposed work paves the way for provably-safe teleoperated robots distributed in an urban area to provide services in transportation, manufacturing, telemedicine, and emergency response by developing principles for predictive modeling and automated interventions. Unlike the present day, where incompatibilities in aims or means for humans and autonomy lead to performance degradation ranging from significant to catastrophic, the proposed work envisions a future wherein humans can be safely deployed amidst cyber-physical systems in society with high confidence.