Confined Exploration for Fast Learning of Optimal Controller for a Class of Mechanical Oscillators
Location
Havner Center, Meramec Gasconade Room, 9:30am-11:30am
Start Date
4-2-2026 9:30 AM
End Date
4-2-2026 10:00 AM
Presentation Date
April 2:2026; 9:30am-10:00am
Description
This research presents a supervised online learning approach for optimal control of a class of nonlinear systems under stability constraints during the learning phase. A hybrid control architecture is developed by combining continuous-time adapive dynamic programming with impulsive supervised exploration. The approach introduces two Lyapunov sublevel sets to define an admissible operating region, where persistent excitation is enforced to ensure sufficient richness for parameter convergence. The algorithm uses an actor–critic structure, where the critic weights are learned online through a normalized gradient descent law. The learning scheme guarantees bounded system trajectories, and simulation results demonstrate both stability and online optimal learning.
Biography
Adebayo Oke received the National Diploma in Electrical Engineering from Yaba College of Technology, Nigeria, and a bachelor’s degree in Electrical and Electronics Engineering from the Federal University of Agriculture, Abeokuta, Nigeria. He obtained an MSc in Automation and Mechatronics from Saint Petersburg Electrotechnical University, Russia, where he worked on tracking control for robotic manipulators.
He subsequently completed a triple-degree Erasmus Mundus Joint Masters in Electric Vehicle Propulsion and Control (E-PiCo), studying across France, Italy, and Germany. His master's thesis focused on the identification of a state-dependent thermal digital twin for power modules, integrating physics-based models with machine learning for real-time prediction and optimization.
He is currently pursuing a PhD in Mechanical Engineering at Missouri University of Science and Technology, USA, specializing in control and robotics. His research interests include nonlinear and optimal control, sliding mode control, observers and estimation theory, robotics, intelligent energy systems, digital twins, and electric vehicles.
Meeting Name
2026 - Miners Solving for Tomorrow Research Conference
Department(s)
Mechanical and Aerospace Engineering
Document Type
Presentation
Document Version
Final Version
File Type
text
Language(s)
English
Rights
© 2026 The Authors, All rights reserved
Confined Exploration for Fast Learning of Optimal Controller for a Class of Mechanical Oscillators
Havner Center, Meramec Gasconade Room, 9:30am-11:30am
This research presents a supervised online learning approach for optimal control of a class of nonlinear systems under stability constraints during the learning phase. A hybrid control architecture is developed by combining continuous-time adapive dynamic programming with impulsive supervised exploration. The approach introduces two Lyapunov sublevel sets to define an admissible operating region, where persistent excitation is enforced to ensure sufficient richness for parameter convergence. The algorithm uses an actor–critic structure, where the critic weights are learned online through a normalized gradient descent law. The learning scheme guarantees bounded system trajectories, and simulation results demonstrate both stability and online optimal learning.
Comments
Advisor: Nilay Kant, nilaykant@mst.edu