Abstract:

In this seminar, an online model-free reinforcement learning approach based on the actor-critic framework is applied to control a four-tank system. The four-tank system is a widely studied benchmark in process control, known for its nonlinear and coupled dynamics. While traditional controllers such as LQR require an accurate system model and full state information, the proposed method learns the control policy directly through real-time interaction with the system, without relying on a prior model. The seminar will introduce the basic concepts of actor-critic reinforcement learning, the design of the learning controller, and its application to regulate the water levels in the tanks. Preliminary simulation results will be presented to illustrate the behaviour of the learned controller.

Speaker Bio:

Mr. Mark Abdelmessih is currently pursuing an M.Sc. in Systems and Control Engineering in the Control and Instrumentation Engineering Department at King Fahd University of Petroleum and Minerals (KFUPM), Saudi Arabia.

He obtained an M.Eng. in Electrical and Computer Engineering with a focus on robotics from the University of Ottawa in Canada (2023), and a B.Sc. in Electromechanical Engineering from Alexandria University in Egypt (2016).

His academic background centers on robotics and control systems, with exposure to reinforcement learning, data-driven methods, system identification, optimal and adaptive control, estimation, and nonlinear control. He is currently developing his experience in these areas through research and practical work.