Intelligent Control, Communication
and Robotics Overview

---

Optimal control approaches for uncertain systems use an approximation-based
solution by leveraging the approximate dynamic programming and reinforcement
learning. These approximate optimal control approaches assume continuous or periodic
feedback transmission and execution of the controller, leading to redundant usages of
resources.
Further, recent results from the machine learning community show the neural
networks are vulnerable to crafted attacks and can significantly change their
decision-making capabilities. In a learning control system, where the neural networks
are the backbone of the controllers, an attack on the controller learning signals may
lead to higher control cost, and eventually, instability. The research goals are to 1)
simultaneously optimize the communication and computational resource usages to
minimize the control, computation, and communication costs of large-scale
interconnected and multiplayer systems and 2) develop secure-by-design resilient
optimal adaptive controllers, which can perform near optimally, even under adversarial
attacks on the learning mechanisms.