CPS: Breakthrough: Control Improvisation for Cyber-Physical Systems

受人类在日常生活中即兴发挥的方式的启发，这个NSF项目正在为网络物理系统设计创建算法即兴发挥理论。它正在开发一个数学框架，以工具为支持，以应对设计适应其操作环境中的不确定性和不断变化的要求的系统的挑战。此外，该框架与计算机科学和工程中的许多领域具有广泛的相关性。拟议工作的结果正在纳入教学，对加州大学伯克利分校的网络物理系统和正式方法课程以及在加州大学伯克利分校更广泛的推广计划下进行的本科项目产生了特别强烈的影响。此外，通过与行业合作伙伴的合作，该项目正在改善网络物理系统行业的验证和控制技术水平。设计过程中的不确定性，随着时间的推移而演变的子系统行为以及操作环境仍然是CPS设计的挑战。有必要设计自动控制器，即兴处理具有挑战性的情况下，作为一个熟练的人会。该项目通过一种基础方法来满足这一需求，该方法正在开发一种理论上合理的算法即兴创作定义，该定义也基于实践。它正在探索问题定义的各种变化，分析其计算复杂性，并在理论上可行的情况下设计有效的算法。此外，它正在开发新的应用程序来验证，用于模拟驱动验证和机器学习组件验证的新算法，以及控制，使用即兴进行随机机器人路径规划和自适应，基于学习的控制中的受控探索。总之，这种理论工作和实际应用的紧密结合，旨在打破网络物理系统科学的新天地。

Inspired by the manner in which humans improvise in everyday life, this NSF project is creating a theory of algorithmic improvisation for cyber-physical systems design. It is developing a mathematical framework, supported by tools, to address the challenge of designing systems that adapt to uncertainty in their operating environment and to changing requirements. Moreover, this framework has broad relevance to many fields in computer science and engineering. Results from the proposed work are being incorporated into teaching, with a particularly strong impact on courses at UC Berkeley on cyber-physical systems and formal methods, and on undergraduate projects conducted under broader outreach programs at UC Berkeley. Additionally, through collaborations with industry partners, the project is improving the state of the art in verification and control in the cyber-physical systems industry.Uncertainty in the design process, in the behavior of sub-systems that evolve over time, and in the operating environment remains a challenge for CPS design. There is a need to design automatic controllers that improvise to handle challenging situations as a skilled human would. This project addresses this need with a foundation approach that is developing a theoretically-sound definition of algorithmic improvisation that is also grounded in practice. It is exploring the full range of variations of the problem definition, analyzing their computational complexity, and devising efficient algorithms where shown to be theoretically possible. Additionally, it is developing new applications to verification, in novel algorithms for simulation-driven verification and verification of machine learning components, and to control, using improvisation for randomized robot path planning and for controlled exploration in adaptive, learning-based control. Together, this tight combination of theoretical work and practical applications seeks to break new ground in the science of cyber-physical systems.