CAREER: Hybrid Estimation and Real-Time Computational Algorithms for Networked Embedded Hybrid Systems

由于嵌入式系统和通信技术的进步，人们对交通系统、网络化机器人、传感器网络和生物系统等网络化嵌入式混合系统的应用越来越感兴趣。在上述应用中，典型的系统由一组子系统组成。每个子系统还拥有一台嵌入式计算机，并具有相互作用的物理和逻辑动力学以及分散决策等独特属性。网络化嵌入式混合系统所固有的是异质不确定性的存在。因此，网络化嵌入式混合系统比目前大多数控制理论所基于的连续动态系统要复杂得多。网络化嵌入式混合系统的复杂性在实时信息推理、控制和安全验证领域提出了重大挑战。由于这类不确定系统的复杂行为可以被建模为随机混合系统，因此，NSF职业生涯研究的目标是发展随机混合系统的理论、计算效率的数值算法和实验试验台，并将其应用于移动联网嵌入式系统(重点是空中交通管制)。为实现这一目标，本课题主要围绕以下几个方面展开研究：随机混合系统的实时混合估计与信息推理算法与分析方法。在计算效率方面，算法基于解析公式，而不是广泛使用的基于样本的估计算法；基于差分变换的计算高效的数值算法正在被开发用于混合系统的最优控制和可达集合计算。开发的算法正在包括自动驾驶飞行器在内的实验测试平台上进行验证。

Due to advances in embedded systems and communication technologies, there is an increasing interest in networked embedded hybrid system applications such as transportation systems, networked robotics, sensor networks, and biological systems. In the above applications, the typical system consists of a group of subsystems. Each subsystem also has an embedded computer and possesses unique properties such as interacting physical and logical dynamics and decentralized decision making. Inherent to the networked embedded hybrid system is the presence of heterogeneous uncertainty. Thus, a networked embedded hybrid system is much more complex than a continuous dynamic system, on which most current control theory is based. The complexity of a networked embedded hybrid system presents major challenges in the areas of real-time information inference, control, and safety verification. Since the complex behavior of such systems with uncertainties could be modeled as a stochastic hybrid system, the objective of this NSF CAREER research is to develop theory, computationally efficient numerical algorithms, and experimental testbeds for stochastic hybrid systems, with application to mobile networked embedded systems (emphasizing air traffic control). To achieve the objective, this project is focusing on the following topics: real-time hybrid estimation and information inference algorithms and analysis methods for stochastic hybrid systems. For computational efficiency, algorithms are based on analytical formulation instead of widely-used sample-based estimation algorithms; computationally efficient numerical algorithms based on Differential Transformation are being developed for optimal control of hybrid systems, and for reachable set computation. The developed algorithms are being validated on experimental test platforms, including autonomous air vehicles.