Plenary Sessions

Plenary speaker

Ming Cao

University of Groningen, The Netherlands

Abstract

As robotics enters the era of “Physical AI,” where AI systems must perceive, reason, and act in the physical world, a central challenge is how to bring the power of generative AI into robotics while maintaining rigorous safety guarantees. This is especially important in control and robotics, where safety often means keeping system states within a desired safe set at all times.

However, many modern data-driven methods rely on black-box neural networks, whose lack of transparency makes formal safety analysis and reliable real-world deployment difficult. This talk explores a new direction that connects generative AI with rigorous control theory: using data to generate explicit symbolic mathematical expressions. Unlike black-box models, symbolic expressions are transparent, computationally efficient, and well suited for formal verification.

We focus on the discovery of control barrier functions, which act as algorithmic “shields” that prevent robotic systems from entering unsafe states. By using transformer architectures guided by reinforcement learning, our framework can automatically generate valid and human-interpretable symbolic safety rules. Through robotic examples, the talk will show how generative AI models and classical control principles can work together to support mathematically rigorous, transparent, and safe autonomy.

Biography

Ming Cao has since 2016 been a professor of networks and robotics with the Engineering and Technology Institute (ENTEG) at the University of Groningen, the Netherlands, where he started as an assistant professor in 2008. Since 2022 he is the director of the Jantina Tammes School of Digital Society, Technology and AI at the same university. He received the Bachelor degree in 1999 and the Master degree in 2002 from Tsinghua University, China, and the Ph.D. degree in 2007 from Yale University, USA.

From 2007 to 2008, he was a Research Associate at Princeton University, USA. He worked as a research intern in 2006 at the IBM T. J. Watson Research Center, USA.

He is an IEEE Fellow. He is the 2017 and inaugural recipient of the Manfred Thoma medal from the International Federation of Automatic Control (IFAC) and the 2016 recipient of the European Control Award sponsored by the European Control Association (EUCA).

He is a Senior Editor for Systems and Control Letters, and is or has been an Associate Editor for IEEE Transactions on Automatic Control, IEEE Transaction on Control of Network Systems, IEEE Transactions on Neural Networks and Learning Systems, IEEE Transactions on Circuits and Systems, IEEE Robotics & Automation Magazine, and IEEE Circuits and Systems Magazine. He is a member of the IFAC Council. He is the General Chair of IFAC World Congress in 2029 in Amsterdam.

His research interests include autonomous robots and multi-agent systems, complex networks and decision-making processes.

Plenary speaker

Dario Bauso

University of Palermo, Italy

Abstract

Generative Adversarial Network (GAN) training is routinely hindered by numerical instabilities and parameter-space oscillations that compromise convergence and reproducibility. Modern regulatory landscapes and operational standards designate smart grid energy infrastructure software as high-risk, mandating absolute telemetry privacy, rigorous technical documentation, and explicit model interpretability.

Opaque, deep “black-box” architectures are inherently ill-suited for this environment, as they are prone to memorizing rare operational states—causing catastrophic infrastructure privacy leaks—and offer zero visibility into how synthetic profiles are derived.

This paper addresses these critical vulnerabilities by introducing an interpretable, white-box framework that provides a closed-form mechanical characterization of GAN dynamics, mapping high-dimensional minimax optimization trajectories onto a second-order mass-spring-damper mechanical system. Linearizing the GAN Jacobian near equilibrium isolates two distinct dynamical regimes: latent-mediated weight interaction and direct bias-bias coupling.

We prove that parameter trajectories follow equivalent Newtonian equations of motion, where structural learning rates define a virtual mass, game-theoretic cross-interactions act as adversarial spring constants, and player curvatures function as physical damping coefficients. This mechanical analog yields an explicit scaling law for the system’s natural frequency, proving it scales with the geometric mean of the learning rates and is strictly bounded by the generator’s activation sensitivity.

We validate these boundaries via spectral density analysis of training trajectories over a 128-dimensional synthetic industrial smart grid telemetry data manifold, transforming complex energy informatics limit cycles into a predictable, transparent dynamical system.

Biography

Dario Bauso has received the Laurea degree in Aeronautical Engineering in 2000 and the Ph.D. degree in Automatic Control and System Theory in 2004 from the University of Palermo, Italy. Since 2005 he has been with the Dipartimento di Ingegneria, University of Palermo (Italy). He was with the Jan C. Willems Center for Systems and Control, ENTEG, Faculty of Science and Engineering, University of Groningen (The Netherlands), where he was Full Professor and Chair of Operations Research for Engineering Systems.

Since 2018 he has been a guest professor at Keio University, Japan. His research interests are in the field of Optimization, Optimal and Distributed Control, and Game Theory. Bauso was an Associate Editor of IEEE Transactions on Automatic Control from 2011 to 2016, of IFAC Automatica from 2015 to 2021, of IEEE Control Systems Letters from 2016 to 2021, of Dynamic Games and Applications from 2011 to 2022, and is Associate Editor of Journal of Dynamics and Games since 2019.

Invited speaker

Mariagrazia Dotoli

Participation format to be confirmed.

Research interests

Optimization of supply chain management and traffic control in smart cities, fuzzy control systems, and the use of Petri nets in modeling these applications as discrete event dynamic systems.

The talk title, abstract and biography will be announced when available.