Problems in decentralized control arise in several practical systems. Examples include cyberphysical systems, formation flight, and several other networked control systems wherein multiple agents try to achieve a common objective in a decentralized manner. Such situations arise, for instance, due to controllers not having access to the same information at the same time.
We consider optimal decentralized LQG control for a plant with nested structure controlled by two players receiving either state or partial output observations from the plant. A unidirectional one-step delayed error-free communication channel is assumed to exist between the two players, i.e., there is a unit delay in information transmission in one direction and no transmission in the other direction. One of the players has access to a subset of state variables while the other can only observe a noisy output. We show that the problem has a partially nested information structure, and thus a linear optimal control law exists. Using summary statistics, the computation of the optimal law is made tractable and a deterministic convex optimization problem is formulated whose solution yields the optimal control law.
This is a joint work with Dileep Kalathil and Professor Rahul Jain at USC.
Naumaan Nayyar received his Bachelor of Technology degree from the Indian Institute of Technology, Bombay in 2010. He joined the University of Southern California on the Provost's Fellowship to pursue his doctoral studies in Electrical Engineering. His research interests include broad areas in learning including stochastic, reinforcement and machine learning, decentralized control, and statistical applications in medicine. He is advised by Professor Rahul Jain in the EE department. –