Dr. Ibrahim Beklan Kucukdemiral

Glasgow Caledonian University

Portrait of Dr. Ibrahim Beklan Kucukdemiral

Reader in Control, Robotics and Applied AI School of Science and Engineering, Glasgow Caledonian University

Control that learns from data, and can still be trusted.

I study how learning-based methods can be given the guarantees that control theory provides. My work joins robust and predictive control with data-driven design, reinforcement learning, generative AI, large language models and expert systems, and applies them to power and renewable energy, biological and biomedical systems, and autonomous robots.

A feedback loop with learning inside it Block diagram. A reference signal enters a controller built on control theory, which drives a system such as a robot, an energy system or a biomedical process. The measured output is fed back to the input. A learning block uses input and output data to adapt the controller. The controller and the learning block sit together inside a dashed boundary labelled robustness and safety guarantees. Robustness and safety guarantees + − Control theory robust · predictive System robots · energy · health Learning from data DeePC · RL · GenAI & LLMs r u y data
The research in one picture: a controller that learns from data, kept inside a boundary of robustness and safety guarantees.

Current work

Nonlinear data-driven predictive control

My most recent work takes data-driven predictive control beyond linear systems, to machines whose dynamics are strongly nonlinear: aerial, legged and humanoid robots. The controllers are built from recorded data instead of a hand-derived model, and are designed to stay safe when conditions differ from that data.

  • Simulation

    Aerial robot

    Flight through an unmapped warehouse

    A quadrotor gathers a short burst of flight data at its launch point. A predictive controller built from that data then flies it past storage racks it discovers on the way.

  • Simulation

    Quadruped

    Walking over logs and ice

    Two controllers drive the same four-legged robot along a course marked by pylons, across stepping logs and then a sheet of ice, where they part company.

  • Simulation

    Humanoid

    Keeping balance on planks and wet ground

    A humanoid robot takes on the same kind of course: raised planks first, then wet, slippery ground.

Data-driven predictive controller Conventional baseline controller

Simulation studies from manuscripts in preparation. The methods will be described here once the papers are published.

Methods

Theory and learning in one toolbox

Classical control and modern AI are treated as parts of one design problem, chosen for what can be guaranteed as well as for what performs best.

All research themes →
  • Control theory

    Robust, optimal and predictive control: H∞ and LMI-based design, tube and event-triggered MPC, sliding-mode and adaptive methods for uncertain, delayed and constrained systems.

  • Data-driven control

    Controllers designed directly from measured trajectories, without first identifying a model: data-enabled predictive control (DeePC) and data-driven H∞ design with stability and constraint guarantees, now being extended to nonlinear machines such as aerial and legged robots.

  • Reinforcement learning

    Deep reinforcement learning for long-horizon decisions, where reward arrives late and credit is hard to assign, and learning-guided optimisation for large design problems.

  • Generative AI and large language models

    Where generative AI and LLMs add value in engineering systems (forecasting, design, operation and decision support), and how far their outputs can be relied on.

  • Expert systems

    Knowledge-based and probabilistic reasoning, from fuzzy inference to Bayesian networks, for diagnosis, reliability assessment and supervisory decisions that stay explainable.

The common thread

Robust and trustworthy by design

A method earns its place in a wind turbine, a treatment plan or a robot working beside people only if it keeps behaving well when the model is wrong, the data are noisy or a component fails. Four questions run through every project.

  • Robustness

    Does performance hold under model uncertainty, disturbances, delays and actuator limits?

  • Safety

    Are state and input constraints respected at every step, even when the controller is built from data alone?

  • Resilience

    Does the system detect and ride through sensor faults and cyber-attacks?

  • Deployability

    Is it data-efficient and light enough to run in real time on real hardware?

Applications

Where it is put to work

Three domains where a wrong decision is expensive, and where guarantees matter as much as performance.

Power and renewable energy

Wind and solar conversion, microgrids and smart grids, load forecasting, EV charging networks and cyber-secure power electronics.

Energy publications →

Biological and biomedical systems

Treatment as a feedback problem: data-driven dosing for cancer therapy, and powered prostheses that adapt to the wearer's gait.

Biomedical publications →

Autonomous robotics

Aerial, legged, ground, underwater and collaborative robots, and autonomous vehicles that keep their passengers comfortable.

Robotics publications →

Publications

Latest papers

The most recent entries from the full publication list.

All publications →
  1. The full list of journal articles, conference papers and books is on the publications page.

Leadership

Roles and service

Work for the control and robotics community, in the UK and internationally.

Chair
IEEE UK and Ireland Control Systems Chapter, 2025–2027
Steering Committee Chair
IEEE EUROCON 2027, Sarajevo
Editor
Results in Engineering; editorial board member of Ocean Engineering and the Turkish Journal of Electrical Engineering and Computer Sciences
Founder and lead
Robotics for Common Good Research Group, Glasgow Caledonian University
Peer review
EPSRC Peer Review College (UKRI); expert reviewer for EUREKA Eurostars and Innovation Fund Denmark
Research strategy
Thematic Leadership Group in Robotics and Autonomous Systems, Scottish Research Partnership in Engineering
Memberships
Senior Member of IEEE; Fellow of the Higher Education Academy; member of five IFAC technical committees

Background

Career in brief

I received my BSc, MSc and PhD degrees in Electrical Engineering from Yıldız Technical University (YTU), Istanbul, in 1997, 1999 and 2002. At YTU I was Assistant Professor (2004–2010), Associate Professor (2010–2015) and Full Professor (2015–2017). I joined Glasgow Caledonian University in 2017 and have been a Reader since 2021.

I was Glasgow Caledonian University's Principal Investigator on ROBMAN, an EPSRC project on robust robotic manipulation underwater with Heriot-Watt University and Subsea 7, and my research has attracted competitive and industrial funding. I have supervised six PhD students to completion and examine doctoral theses in the UK and abroad.

Teaching

Open textbooks and courses

Interactive course materials with full textbook chapters, worked MATLAB and Python examples, and hands-on simulation tools.

All teaching material →

Collaboration and PhD supervision

I welcome enquiries about research collaboration, PhD supervision and academic partnerships.

Get in touch