ECR Webinar- AI meets Control
Date: 12th February 2025
Time: 10am – 1pm
Location: Online. The Zoom link will be sent before the event.
Registration: Register at this link
About the Seminar
The aim of this online event is to highlight the most recent research works at the intersection of control, Artificial Intelligence (AI), learning/data and their applications to real-life systems. The event will enable to connect researchers in general and early career researchers (ECR) in particular who are interested in the interplay field between control theory and AI. The event will split in three sessions. The first two sessions will highlight the recent research advances at the intersection of control theory and AI. The speakers will showcase their theoretical and applicative research in the field. The last session will be dedicated to learning about industrial interests and problems requiring the hybrid field of control and AI. The session will also open-up discussions for follow-up potential collaborations between researchers having an interest in combining control theory with data and learning from AI to solve particular research problems.
Agenda
| Session 1 | Research advances in Control, AI | |
| 10:00 – 10:10 | Welcome and introduction | Chair of the workshop:
Dr. Zehor Belkhatir (University of Southampton) and Lead for AI and data Grand Challenges Research Committees (GCRCs) Prof. Nabil Aouf (University of London) |
| 10:10 – 10:30 | Data-driven decision making and safety certificate synthesis | Prof. Kostas Margellos (Oxford University) |
| 10:30 – 10:50 | Application of Set Approximation in AI and Engineering | Dr. Morgan Jones (University of Sheffield) |
| 10:50 – 11:10 | Learning full body deformation of soft robots using end-to-end models | Dr. Thomas George Thuruthel (University College London) |
| 11:10 – 11:20 | Break | |
| Session 2 | Rapid fire talks | |
| 11:20 – 12:00
5 minutes talks (3 min + 2 min Q/A) |
Cooperative control in swarm robotics
Learning Nash Equilibria via Innovative Dynamics
Estimation and control of living neural circuits Online End-to-End Learning-Based Predictive Control for Microgrid Energy Management
|
Dr. Junyan Hu (Durham University)
Dr.Leonardo Stella (University of Birmingham) Dr. Jack Umenberger (University of Oxford) Dr. Thiago Burghi (University of Cambridge) Dr. Francesca Boem (University College London) |
| 12:00 – 12:10 | Break | |
| Session 3 | Industry meets AI and control and panel discussion | |
| 12:10 – 12:30 | Data/AI & Control Meet Industry | Mr. Hector Figueiredo, CEng, MIET, Capability Lead – QinetiQ, Autonomous Systems |
| 12:30 – 12:50 | Round table discussion | All |
| 12:50 – 13:00 | Closing remarks | Chair of the workshop: Dr. Zehor Belkhatir (University of Southampton) and Lead for Early Career Researcher (ECR) Group
Prof. Fulvio Forni (University of Cambridge) |
About the Speakers
Associate Professor Kostas Margellos
Kostas Margellos received the Diploma in electrical engineering from the University of Patras, Greece, in 2008, and the Ph.D. in control engineering from ETH Zurich, Switzerland, in 2012. He spent 2013, 2014 and 2015 as a postdoctoral researcher at ETH Zurich, UC Berkeley and Politecnico di Milano, respectively. In 2016 he joined the Control Group, Department of Engineering Science, University of Oxford, where he is currently an Associate Professor. He is also a Fellow of Reuben College and a Lecturer at Worcester College. He is currently serving as Associate Editor in Automatica and in the IEEE Control Systems Letters, and is part of the Conference Editorial Board of the IEEE Control Systems Society and EUCA. His research interests include optimisation and control of complex uncertain systems, with applications to energy and transportation networks.
Dr. Thomas George Thuruthel
Dr. Thomas George Thuruthel is an Assistant Professor in the Department of Computer Science at University College London. He received his B.Tech. degree in Mechanical Engineering from the Indian Institute of Technology Hyderabad, India, in 2013, followed by a Master’s degree in BioRobotics from Waseda University, Tokyo, Japan, in 2015, and a Ph.D. degree with Honors from Scuola Superiore Sant’Anna, Pisa, Italy. He served as a postdoctoral researcher at the University of Cambridge from 2019 to 2022. His research interests include modeling and control, dexterous manipulation, and soft sensing. Currently, he is focusing on the control of soft robots, design optimization of soft-bodied systems, development of novel soft sensors, and dexterous manipulation using visuo-tactile information.
Dr Morgan Jones
Dr Morgan Jones received the MMath degree in mathematics from The University of Oxford, England in 2016 and PhD degree from Arizona State University, USA in 2021. Since 2022 he has been a lecturer in the Department of Automatic Control and Systems Engineering at the University of Sheffield. His research primarily focuses on the estimation of reachable sets, attractors and regions of attraction for nonlinear ODEs. Furthermore, he has an interest in extensions of the dynamic programming framework to non-separable cost functions.
Dr. Junyan Hu
Dr Junyan Hu is an Assistant Professor with the Department of Computer Science at Durham University and a Fellow of Durham Energy Institute. Before joining Durham University, he was a Lecturer at University College London and a Postdoctoral Research Associate at the University of Manchester. His research interests include swarm intelligence, multi-agent systems, cooperative path planning, robust/adaptive control, with applications to autonomous vehicles and intelligent robotics. Junyan is a Senior Member of IEEE and a Fellow of Higher Education Academy. He serves as an Associate Editor for leading international journals and conferences in the field of robotics and control, e.g., IEEE Robotics and Automation Letters (RA-L), IEEE International Conference on Robotics and Automation (ICRA), and IEEE Conference on Control Technology and Applications (CCTA).
Dr. Leonardo Stella
Leonardo Stella is an Assistant Professor in the School of Computer Science, University of Birmingham. He completed his PhD degree on the application of game theory and control for bio-inspired collective decision-making in 2019 from the Department of Automatic Control and Systems Engineering, University of Sheffield, United Kingdom. His research interests are game theory, control, reinforcement learning, and multi-agent systems. Applications include machine learning for process parameter optimisation in materials science and multi-agent reinforcement learning in swarm robotics.
Dr. Francesca Boem
Francesca Boem received the MSc degree (cum laude) in Management Engineering in 2009 and the PhD degree in Information Engineering in 2013, both from the University of Trieste, Italy. She was Post-Doc at the University of Trieste with the Machine Learning Group from 2013 to 2014. From 2014 to 2018, she was a Research Associate with the Control and Power research group at the Department of Electrical and Electronic Engineering, Imperial College London, UK. From 2015 to 2018, she was part of the team at Imperial College working on the flagship EU H2020-WIDESPREAD-TEAMING project for the development of the EU KIOS Research and Innovation Centre of Excellence in Cyprus. Dr Boem was awarded the Imperial College Research Fellowship in 2018, and in the same year, she was appointed as a Lecturer in the Department of Electronic and Electrical Engineering at University College London, UK. She is also an Honorary Lecturer at Imperial College London. In 2022, she was awarded as PI the EPSRC New Investigator Award and an NNTI Joint Lab – Base Exploratory project in 2023. She currently is an Associate Professor with the EEE Department at UCL and the Director of UCL’s MSc in Integrated Machine Learning Systems. Her current research interests include distributed fault diagnosis and fault-tolerant control methods for large-scale networked systems, safety and security of cyber-physical systems, and learning-based control. Dr Boem is a member of the IFAC Technical Committees 1.5 (Networked Systems) and 6.4 (“Fault Detection, Supervision & Safety of Technical Processes – SAFEPROCESS”) and Associate Editor for the IEEE Systems Journal, the EUCA European Journal of Control, the IEEE CSS Letters, and for the IEEE Control System Society, IFAC and EUCA Conference Editorial Boards.
Dr. Thiago B. Burghi
Thiago B. Burghi is a post-doctoral research associate in the Control Group of the Department of Engineering at the University of Cambridge. In 2022, he received a 3-year Kavli Foundation research grant to study estimation and control algorithms for biological neuronal circuits. He completed his PhD in Control Theory in 2020 at the University of Cambridge. Previously, he received a M.Sc in Mechanical Engineering from the University of Campinas (Brazil), in 2015, and a double undergraduate degree in Engineering from ENSTA Paris and the University of Campinas, in 2014. His research interests lie at the interface between nonlinear control theory, system identification, and artificial and biological neural systems.
Abstracts
Title: Data-driven decision making and safety certificate synthesis
Data driven algorithms offer a natural framework to make decisions in environments affected by uncertainty, where uncertainty is represented by means of data. Neural networks constitute one class of such data driven decision making tools. However, the “learned’’ decisions are inherently random as they depend on the data used. In this talk we discuss how tools from statistical learning theory based on the notion of compression and randomized optimization offer a principled framework to analyze the robustness properties of these learned decisions. Our results build “trust on data’’, and accompany data driven solutions with probabilistic robustness guarantees that capture their generalization properties when it comes to new data, not included in the learning/training process. We review recent advancements in this area that allow to boost performance based on a data outlier removal procedure. We then show how this methodology can be employed to build what will be referred to as safety informed neural networks, that produce safety and reachability certificates for nonlinear dynamical systems, accompanying them with prescribed probabilistic guarantees with respect to their validity.
Title: Learning full body deformation of soft robots using end-to-end models
Learning-based techniques are becoming increasingly popular for developing models of soft robots for control purposes. Typically, they involve developing a mapping from the control space to the task space using real-world data. However, there have been limited works on using learning-based approaches for fully-body modelling of soft robots. This talk presents some of the latest work on learning full-body deformation of soft robots for static and dynamic cases, including interactions with the external environment.
Title: Application of Set Approximation in AI and Engineering
“This talk explores the application of set approximation techniques in AI and engineering, focusing on methods for approximating complex sets such as regions of attraction, reachable states, and feasible state spaces. Drawing from recent advances in sublevel set approximation using Hausdorff and volume metrics, we discuss numerical schemes based on Sum-of-Squares programming to approximate unions, intersections, Minkowski sums, and discrete sets. Applications in machine learning, such as one-class classification, and robotics, like collision-free path planning for Dubin’s car, illustrate the practical impact of these techniques in improving control theory and system design.”
Title: Cooperative control in swarm robotics
Abstract: Recent advances in computing, communication, and control techniques, as well as increasing computational power of embedded systems, provide a great opportunity to deploy networked intelligent robots in complex tasks for higher accuracy, safety and efficiency. However, current demonstrations of robot swarms are mainly restricted to controlled or structured environments, which significantly limits their deployment in complex real-world applications. The main challenge in designing practical cooperative robotic systems is to determine the individual agents’ controllers that enable the collective to achieve the desired global objectives. This talk will briefly introduce some recent works on theory and applications of swarm control from Junyan’s group.
Title: Learning Nash Equilibria via Innovative Dynamics
Abstract: Despite its groundbreaking success, reinforcement learning still suffers from fundamental issues, including instability and non-stationarity. Replicator dynamics, the most well-known model from evolutionary game theory (EGT), provide a theoretical framework for the convergence to Nash equilibria in stable games. However, this is not true in other settings, e.g., null-stable games, where these dynamics exhibit periodic behaviours. In contrast, innovative dynamics, such as the Brown-von Neumann-Nash (BNN) or Smith, can directly converge to Nash without the need for time-averaging. In this talk, I will present our multi-agent reinforcement learning framework based on innovative dynamics and their potential for fast adaptation to non-stationary environments.
Title: Online End-to-End Learning-Based Predictive Control for Microgrid Energy Management
Abstract: A novel Online Learning (OL) algorithm is proposed, integrating Recurrent Neural Networks (RNNs) and Model Predictive Control (MPC) in an End-to-End (E2E) learning-based control architecture, with application to the Microgrids energy management problem. The RNN predicts uncertain and possibly evolving profiles of electricity price, load demand, and renewable generation, which are then exploited in an integrated MPC optimisation problem to minimise the overall microgrid electricity consumption cost while guaranteeing operation constraints. To address the challenge of model uncertainty, a task-based loss approach is proposed by integrating the MPC optimisation as a differentiable optimisation layer within the neural network, allowing the OL architecture to jointly optimise prediction and control performance. The proposed methodology incorporates a specifically designed online version of the Stochastic Weight Averaging (O-SWA) and Experience Replay (ER) methods to enhance online learning capabilities, ensuring more robust and adaptive learning in real-time scenarios, as well as Transfer Learning (TL) capabilities.
Title: Estimation and control of living neural circuits
Abstract: Biological neurons have long been modelled in terms of nonlinear electrical circuits whose activity can, in principle, be regulated with the methods of control engineering. However, it was only recently that technological advances permitted us to meaningfully interact with living neural systems using real-time feedback control. Reliable closed-loop regulation of neural systems has the potential to revolutionize the treatment of neural disorders such as epilepsy and Parkinson’s, but achieving it poses major technical challenges. Neural systems are partially observed, nonlinear, adaptive and uncertain. In this talk, I will show how learning-based strategies combining LTI systems and artificial neural networks allow us to quickly obtain data-driven models of neural dynamics during an electrophysiology experiment. Using such models, I discuss ongoing work that aims to control neural dynamics by modulating the bifurcations underlying different regimes of neural activity. As a practical example, I discuss experiments in which such bifurcations are reliably triggered by changing the temperature of a neural central pattern generator.