Keynote speakers
Title: Engineering Digital Twins in the Age of AI
Abstract: Digital Twins, that is, digital representations of systems that operate alongside them, have become common for optimizing systems. Examples are numerous: they may optimize motion profiles of cyber-physical systems, improve scheduling of a logistics system, or smartly control a building's HVAC system. At the core of Digital Twins are the simulation models of the real system. They are the key to the what-if analysis necessary for making the offered smart services possible.
In recent years, techniques from the Artificial Intelligence field have found their way into the development and operation of Digital Twins. Black-box or surrogate models are replacing traditional white-box ones, classifiers and predictors support monitoring and anomaly detection, and AI-assisted workflows help with implementation and maintenance. This keynote explores these developments from the viewpoint of the modelling and simulation community, identifying both their strengths and weaknesses.
Dr. ing. Joost Mertens is a postdoctoral researcher at the University of Antwerp, Faculty of Applied Engineering in Electronics and ICT. His research interests comprise methods and techniques to model, deploy and operate Digital Twins, in particular their continuous evolution during operation. He specializes in Digital Twins of Cyber-Physical Systems, Systems-of-Systems and built environments.
Demo 1: A Digital Twin Approach to Closed-Loop Neurofeedback with GlassBrain
Abstract: GlassBrain is an interactive demonstrator exploring digital twins for real-time neurofeedback by connecting live electroencephalography (EEG) recordings to a three-dimensional virtual brain. Through individual calibration and selectable frequency bands, the virtual representation evolves in response to the participant’s cortical activity as they alternate between rest and simple hand movements. This human-in-the-loop interaction enables participants to observe their physiological activity and explore how visual feedback can support self-regulation. The demonstration will showcase the potential of physiological digital twins to support neurofeedback research and improve the interpretation of brain activity, while highlighting key challenges related to synchronisation, individual calibration, signal artefacts and uncertainty in EEG-based estimates.
Dr. Cédric Simar is a postdoctoral researcher in the Machine Learning Group at the Université libre de Bruxelles (ULB), specialising in robotic hand control and dexterous manipulation. His research combines machine learning, neuroscience and robotics to develop adaptive control methods, building on his Ph.D. work on decoding electrophysiological signals to translate muscle activity into intuitive finger movements for robotic hand control, with applications in prosthetics and teleoperation. His current research explores learning-based robotic manipulation, tactile feedback and digital twins to bridge simulation and real-world robotic experiments, supported by the WEL-T programme. He also lectures in Artificial Intelligence and Brain-Computer Interfaces at the University of Mons.
Demo 2: Decoding High-Dimensional Finger Motion from EMG Using Riemannian Features and RNNs
Abstract: Continuous estimation of high-dimensional finger kinematics from forearm surface electromyography (EMG) could enable natural control of hand prostheses, AR/XR interfaces, and teleoperation systems. We present an end-to-end framework for continuous EMG-to-kinematics regression using only consumer-grade hardware. The system learns to map muscle activity to continuous finger-motion commands, enabling intuitive control of multiple degrees of freedom for a single user through Riemannian features and recurrent neural networks. By leveraging physiological knowledge for feature extraction, our approach avoids the need for deep convolutional architectures, which are often considered state-of-the-art but typically require larger amounts of training data. During the demonstration, a participant who has previously calibrated the model will use EMG signals to control a robotic hand in real time. The demonstration will highlight the potential of EMG-based intuitive control while also illustrating the challenges associated with deploying such systems in real-world applications.
Martin Colot is a PhD candidate at the Machine Learning Group of the Université Libre de Bruxelles. His interests are in Machine Learning, Unsupervised Domain Adaptation, and Human Computer Interfaces including EMG and EEG applications.