Spring School

Spring School on Machine Learning and Perception for Social Awareness

13-16 April 2026

Location: Catania, Italy

➡️ View Location on Google Maps

About the SWEET Spring School

The SWEET Spring School is a specialized training initiative organized entirely by the SWEET project consortium, though it remains open to external students and researchers. The SWEET Spring School is a specialized training initiative organized entirely by the SWEET project consortium, in conjunction with the TRAIL conference, and it remains open to external students and researchers. The core objective of the school is to deeply explore disciplines related to Social Awareness from the perspective of perception in Service Robotics.

The curriculum is designed to equip participants with a robust set of skills through a series of lectures and hands-on practical sessions. The focus will be on cutting-edge research methods for sensing and the development of machine learning algorithms essential for enhancing robots’ social perception capabilities.

Participation in the school is free of charge.

However, please note that all logistical costs, including travel and accommodation expenses, must be covered by the individual participants.

How to Apply

To apply for the Spring School, please submit the following documents:

  • A short Statement of Interest (max 1 page).
  • Your current Curriculum Vitae (CV).
  • A Letter of Support from your Tutor/Supervisor.

All application materials must be sent via email to: infosweet@unina.it

APPLICATION DEADLINE: February 28th, 2026

Acceptance Notification: March 15th, 2026

Progamme

Draft Programme

Speakers

Prof. Alessandro Vinciarelli

Full Professor at University of Glasgow

An Introduction to Social AI

The presentation introduces Social AI, the domain aimed at modelling, analysis and synthesis of human-human and human-machine interactions. After outlining the scope of the field (what Social AI is about and what are its most important scientific objectives), the presentation will cover its most distinctive aspects. Special attention will be paid to the interdisciplinary collaboration with Psychology and other Human Sciences because Social AI is about people as much as it is about machines. Finally, a few examples will provide a chance to develop familiarity with approaches, methodologies and research problems common to most Social AI works.

Alessandro Vinciarelli (http://vinciarelli.net) is Full Professor at the University of Glasgow, where he is with the School of Computing Science and the Institute of Neuroscience and Psychology. His main research interest is Social AI, the domain aimed at modelling, analysis and synthesis of verbal and nonverbal behaviour in human-human and human-machine interactions. He published 200+ works in international journals and conferences and he is or has been Principal Investigator in 15+ projects, including an EU funded European Network of Excellence (the SSPNet, 2009-2014, 6.2MEuros) and the UKRI Centre for Doctoral Training in Socially Intelligent Artificial Agents (2019-2027, 6MEuros, http://socialcdt.org). Alessandro co-organized 30+ international events as General Chair (IEEE International Conference on Social Computing 2012, ACM International Conference on Multimodal Interaction 2017, International Conference on Digital mental Health and Wellbeing 2026, etc.), Program Chair (ACM International Conference on Multimodal Interaction 2023, Affective Computing and Intelligent Interaction 2024) or other roles. In addition, Alessandro is co-founder of Klewel (http://klewel.com), a knowledge management company recognised by the IEEE as an exemplary impact story, and scientific advisor to Substrata (http://substrata.me), a leading Social Signal Processing company.

Prof. Ing. Igor Farkaš

Full Professor at Comenius Univeristy Bratislava

Machine learning for robot perception and control

Machine learning (ML) is a fundamental bio-inspired approach towards building AI systems optimized for various tasks, including robot learning. In the talk, we will present several ML methods based on artificial neural networks, useful for this domain. We will cover two paradigms: first, the supervised learning that can be exploited for building world models (latent representations) important for robots in order to represent the world. Second, the reinforcement learning paradigm used for robots in order to learn desired behavior (policy).

1. Self-Supervised Learning (SSL) for Robot Perception and Representation Methods that allow robots to learn from raw sensory data without manual labeling.
(a) Representation Learning: Contrastive learning = Learn state embeddings by distinguishing positive (e.g., time-adjacent frames) from negative samples.
(b) Predictive modeling (forward models, inverse models), Multimodal alignment.

2. Reinforcement Learning (RL) for Robot Control and Decision-Making = Methods that optimize behavior through trial and error.
(a) Model-Free RL – Value-based (e.g., DQN), Policy gradient (e.g., PPO, TRPO, A2C), Actor–critic (e.g., SAC, TD3), common for continuous-control robotic tasks (grasping, locomotion).
(b) Model-Based RL – Learned dynamics models (e.g., Dreamer) = Use predicted future trajectories for planning or policy learning. Hybrid planning + learning (combine model-based planning (MPC) with learned controllers).

Igor Farkaš is a professor of informatics at the Department of Applied Informatics, Faculty of Mathematics, Physics and Informatics at Comenius University Bratislava (UKBA). He obtained master’s degree in technical cybernetics (1991) and PhD in applied informatics (1995), both from Slovak University of Technology in Bratislava. He received a prestigious Fulbright scholarship (University of Texas at Austin, US) and Humboldt scholarship (Saarland University in Saarbrucken, Germany). His research interests span the related fields of artificial intelligence and cognitive science. More concretely, the areas include artificial neural networks and their analysis (explainable AI), cognitive robotics, language modeling, reinforcement learning (intrinsic motivation), abstract cognition, and more recently, human-robot interaction. Prof. Farkaš teaches courses on neural networks, computational intelligence, deep learning and cognition, and until recently grounded cognition. He coordinates the Centre for cognitive science at UKBA and serves as the main guarantor of the international Middle-European Interdiciplinary master’s programme in cognitive science (MEi:CogSci). More on http://cogsci.fmph.unibs.sk/~farkas.

Prof. Lorenzo Baraldi

Associate Professor at University of Modena and Reggio Emilia

Towards Reliable and Adaptive Multimodal LLM Agents

The rapid progress of multimodal Large Language Models (MLLMs) is reshaping how intelligent agents perceive, reason, and act within visually rich and interactive environments. As these systems advance, new challenges emerge in achieving coherent multimodal understanding, robust generalization, and transparent decision-making, also in embodied scenarios where perception and action tightly interplay.
In this talk, I will present research directions that strengthen visual grounding, reduce hallucinations, and improve reasoning in knowledge-intensive contexts. I will discuss how addressing missing or imperfect modalities leads to more resilient multimodal integration, how reflective mechanisms can enhance deliberation, and how structured representations (ranging from recurrent visual-linguistic processing to non-Euclidean embedding spaces) enable richer alignment between observations, instructions, and actions.
Further, the talk explores how agent behavior can be better aligned with user intent in interactive and embodied settings through personalization and improved modeling of contextual cues.

 

Lorenzo Baraldi is an Associate Professor at the University of Modena and Reggio Emilia, where he works on Deep Learning, Vision-and-Language integration, Large-Scale models and Multimedia. He teaches in the courses of “Computer Vision and Cognitive Systems,” Scalable AI, and Computer Architecture. He has authored more than 120 publications in international journals and conferences. Currently, he serves as an Associate Editor for Computer Vision and Image Understanding and Pattern Recognition and acts as an Area Chair for ICCV and major multimedia conferences. He is also a Scholar in the ELLIS society (European Laboratory for Learning and Intelligent Systems), where he coordinates the Modena ELLIS Unit. Since 2021, He has held the position of deputy director at the Interdepartmental Center on Digital Humanities at the University of Modena and Reggio Emilia.

Prof. Gabriel Skantze

Full Professor at KTH Royal Institute of Techonology

Conversational human-robot interaction

Conversational AI has evolved rapidly in recent years, driven largely by advancements in LLMs. However, significant challenges remain, particularly in spoken human-robot interaction. In this domain, the physical context is critical, interactions often involve multiple users, and the robot’s physical embodiment plays a key role. This talk will address specific challenges related to turn-taking, feedback, and conversational speech synthesis.

Gabriel Skantze is a Professor at KTH Royal Institute of Technology in Stockholm, Sweden, where he leads several research projects related to speech communication, conversational AI, and human-robot interaction. His research is highly interdisciplinary, encompassing topics such as computational modelling of turn-taking, feedback and gaze in interaction, language learning, and language grounding. He is the former President of SIGDIAL, the ACL special interest group on Discourse and Dialogue. He is also co-founder and Chief Scientist of the company Furhat Robotics.

Prof. Oya Celiktutan

Associate Professor at King’s College London

Multimodal Human Behaviour Modelling for Socially Assistive Robots

Robots are increasingly moving from research laboratories into real-world environments, where they are expected to provide companion care for older adults, support children with special needs, assist people in their daily activities, and offer services in public spaces. These applications require robots to interact naturally and effectively with humans, making robust perception, understanding, and adaptation to human behaviour essential.

Consequently, significant efforts have been devoted to advancing robots’ multimodal perception and interaction capabilities, enabling them to recognise, interpret, and respond to human actions with increasing levels of autonomy. In this talk, I will present an overview of our ongoing research on multimodal human behaviour understanding and generation for socially assistive robots.

In particular, I will highlight our work on enabling robots to comply with social norms, understand and imitate non-verbal behaviours, and explain their actions. I will conclude by discussing key challenges and open research directions in human–robot interaction.

Dr Oya Celiktutan is a Reader in AI and Robotics in the Department of Engineering at King’s College London, where she leads the Social AI & Robotics Laboratory. She is also the Honorary Robotics Lead at Guy’s and St Thomas’ NHS Foundation Trust.

Her research focuses on multimodal machine learning for autonomous robots and virtual agents that interact naturally with humans, including multimodal perception, human behaviour understanding and prediction, and socially aware navigation and manipulation. Her work has been supported by EPSRC, The Royal Society, and EU Horizon, as well as industrial partners.

She received the EPSRC New Investigator Award in 2020, and her team has won several awards, including Best Paper at IEEE RO-MAN 2022 and recognition at IEEE FG 2021 and the ICCV UDIVA Challenge 2021.

Prof. Antonio Chella

Full Professor at University of Palermo

Engineering Self-Conscious Robots: Cognitive Architectures and Mechanisms

This lecture addresses robot self-consciousness from the perspectives of computational and cognitive architectures, focusing on mechanisms that enable an artificial agent to represent, monitor, and regulate its own internal states and actions. Building on foundational theories of consciousness and self-awareness, the lecture examines how self-consciousness can be operationalized in robots through internal self-models, predictive processing, and global workspace–like control structures. Particular emphasis is placed on the role of embodiment and inner speech as recursive feedback mechanisms that support self-monitoring, decision-making, and behavioral coherence. Concrete robotic architectures and experimental case studies are discussed to illustrate how minimal forms of self-consciousness can emerge from the integration of perception, action, attention, and internal simulation. The lecture also reviews current approaches to evaluating robotic self-consciousness and highlights open theoretical, empirical, and ethical challenges in the design of self-aware artificial agents.

Antonio Chella is Professor of Robotics at the University of Palermo, where he founded and directs the RoboticsLab. His research focuses on cognitive architectures, artificial consciousness, self-awareness in robots, and socially assistive robotics. He has been actively involved in interdisciplinary research at the intersection of AI, cognitive science, and philosophy of mind. He has coordinated or participated in several national and international research projects. He is the author of numerous scientific publications on machine consciousness, inner speech, and self-aware robotic systems.

Dr. Anna Belardinelli

Principal Scientist at Honda Research Institute

“Do you get me?”- Multimodal communication for situated human-robot interaction

Intelligible, meaningful, and grounded communication is a fundamental requirement for successful human–robot interaction. However, such communication cannot assume symmetric cognitive and expressive capabilities between humans and robots.  Acknowledging this asymmetry, interaction design must focus on establishing common ground and fostering mutual understanding. Robots should therefore communicate their perceptions, reasoning processes, and action capabilities in an intuitive and explainable manner, potentially using modalities that differ from those employed by humans. Conversely, robots must be able to interpret human intentions by integrating verbal and non-verbal cues within their situational context. In this talk, these principles are illustrated through theoretical and experimental insights from cognitive science, alongside our own implementations and empirical studies of human–robot interaction in socio-physical settings. I’ll present work on AR-based learning from demonstration, attentive support in multiparty interactions, and the integration of language with complementary expressive cues.

Anna Belardinelli is Principal Scientist at the Honda Research Institute Europe, working on human-robot interaction. She has an interdisciplinary background, with interests at the intersection of Artificial Intelligence and Cognitive Science. Her research efforts have spanned visual attention in humans and machines, eye-hand coordination for manipulation and teleoperation, and computational models that support intelligent behavior in robots and interactive systems.

TRAIL Conference Keynote Speakers

In conjunction with the Spring School, the TRAIL conference will feature the presence and contribution of the following speakers:

Speakers

Prof. Angelo Cangelosi

Full Professor at the University of Manchester (UK)

Cognitive and Developmental Robotics: The Importance of Starting Small

This talk introduces the concept of Cognitive Robotics, i.e. the field that combines insights and methods from AI, as well as cognitive and biological sciences, to robotics (cf. Cangelosi & Asada 2022 for book open access). This is a highly interdisciplinary approach that sees AI computer scientists and roboticists collaborating closely with psychologists and neuroscientists. In particular, we will focus on Cognitive Developmental Robotics, which models the incremental learning and developmental acquisition of cognitive and sensorimotor skills in robots.
We will use the case study of language learning to demonstrate this highly interdisciplinary field, presenting developmental psychology studies on children’s language acquisition and robots’ experiment on language learning. One study focuses on the embodiment biases in early word acquisition and grammar learning. The same developmental robotics method is used for experiments on pointing gestures and finger counting to allow robots to learn abstract concepts such as numbers. We will then present novel developmental robotics models, and human-robot interaction experiments, on Theory of Mind and its relationship to trust. This considers both people’s Theory of Mind of robots’ capabilities, and robot’s own ‘Artificial Theory of Mind’ of people’s intention. Results show that trust and collaboration is enhanced when we can understand the intention of the other agents and when robots can explain to people their decision making strategies.
The implications for the use of such cognitive robotics approaches for embodied cognition in AI and cognitive sciences will be discussed. Moreover, the talk will discuss the approach of “starting small” (Elman 1993), and its repercussions for the current limitations of foundation models and LLMs/VLAs as cognitive models for language learning and understanding in AI and robots.

Angelo Cangelosi is Professor of Machine Learning and Robotics at the University of Manchester (UK) and co-director and founder of the Manchester Centre for Robotics and AI. He was selected for the award of the European Research Council (ERC) Advanced grant (UKRI funded). His research interests are in cognitive and developmental robotics, neural networks, language grounding, human robot-interaction and trust, and robot companions for health and social care. Overall, he has over 400 publications and has secured £40m of research grants as coordinator/PI/co-I, including the ERC Advanced eTALK, the EPSRC CRADLE Prosperity, the US AFRL project CASPER++, and five ongoing Horizon RIA and MSCAs grants. He is Editor-in-Chief of the journals Interaction Studies and IET Cognitive Computation and Systems, and in 2015 was Editor-in-Chief of IEEE Transactions on Autonomous Development. He has chaired numerous international conferences, including ICANN2022 Bristol, and ICDL2021 Beijing. His book “Developmental Robotics: From Babies to Robots” (MIT Press) was published in January 2015, and translated in Chinese and Japanese. His latest book “Cognitive Robotics” (MIT Press), coedited with Minoru Asada, was published in 2022 and translated in Chinese in 2025.

Dr. Alessandra Sciutti

Senior Tenure Track Researcher and head of the CONTACT (COgNiTive Architecture for Collaborative Technologies) Unit of the Italian Institute of Technology (IIT).

Toward Understandable Robots through Human-Inspired Cognition

As robots increasingly operate in close interaction with humans, interpretability becomes a central requirement for effective and trustworthy human–robot interaction. Interpretability should emerge from the robot’s embodied cognitive processes and from the dynamics of interaction. Drawing inspiration from human cognitive development, we adopt an embodied approach to interpretability in which robots acquire transparent and predictable behavior through memory, anticipation, and adaptation. By grounding robot cognition in models of human perception and action, robots can interpret non-verbal cues such as motion dynamics, timing, and effort, while simultaneously expressing their own intentions in ways that are intuitively legible to human partners. We show how robots can serve both as interactive partners and as experimental probes to investigate the mechanisms underlying human social understanding. Through embodied communication and developmental principles, interpretability becomes an intrinsic property of interaction, supporting mutual understanding, long-term adaptation, and trust.

Alessandra Sciutti is a Senior Tenure Track Researcher and head of the CONTACT (COgNiTive Architecture for Collaborative Technologies) Unit of the Italian Institute of Technology (IIT). She received her B.S. and M.S. degrees in Bioengineering and her Ph.D. in Humanoid Technologies from the University of Genova in 2010. After two research periods in the USA and Japan, in 2018, she was awarded the ERC Starting Grant wHiSPER (www.whisperproject.eu), focused on the investigation of joint perception between humans and robots. She has published more than 100 papers and abstracts in international journals and conferences, coordinates the ERC POC Project ARIEL (Assessing Children Manipulation and Exploration Skills), and has participated in the coordination of the CODEFROR European IRSES project (https://www.codefror.eu/).  She is currently Chief Editor of the HRI Section of Frontiers in Robotics and AI and Associate Editor for several journals, including the International Journal of Social Robotics and Cognitive System Research. She is an ELLIS scholar and the corresponding co-chair of the IEEE RAS Technical Committee for Cognitive Robotics. Her research aims to investigate the sensory and motor mechanisms underlying mutual understanding in human-human and human-robot interaction.

Dr. Thomas Weisswange

Principal Scientist at the Honda Research Institute Europe

Modelling Intended Impact of Assistive Interactions

Explainability is first and foremost grounded in social interaction. While it is important to research transparent algorithms, understand causal attributions and design expressive interfaces when creating explainable agents, the target will always be to understand how to achieve a certain effect on a human perceiver.  The need for an explanation only arises when part of the human’s world model is flawed. For deciding when, what and how to communicate, it is useful to incorporate inference of human cognitive processes into an agent’s behavior planning to assess the possible impact on this world model.

In this talk, I will present research across different applications that demonstrate how this can be approached and used for human-agent interactions. The talk will touch on work  incorporating predicted impact of robot’s and automated vehicles’ actions on human belief, behavior policies, mutual understanding, and situational cost.

Thomas H. Weisswange is a Principal Scientist at the Honda Research Institute Europe in Offenbach, Germany. He has a strongly interdisciplinary background covering bioinformatics, computational neuroscience, intelligent transportation, machine learning, technology ethics, and human-robot interaction. Thomas’ current research projects address robot interactions with groups, human-robot cooperation, theory-of-mind, and intelligent systems design.