Faculty of Science, Technology and Medicine
Faculty of Law, Economics and Finance
Faculty of Humanities, Education and Social Sciences
Interdisciplinary Centre for Security, Reliability and Trust
Luxembourg Centre for Systems Biomedicine
Luxembourg Centre for Contemporary and Digital History
Luxembourg Centre for European Law
Luxembourg Centre for Socio-Environmental Systems
Embodied Intelligence for Safe Autonomous Systems
The Summer School on Autonomous Systems targets doctoral candidates and early-stage researchers in autonomous systems, robotics, artificial intelligence, and cyber-physical systems, providing advanced interdisciplinary training at the intersection of Digital Twin technologies, embodied and physical AI, distributed and data-driven intelligence, as well as safety and trustworthiness in autonomous systems. The program is organized around three core themes: enhancing efficiency, sustainability, and safety through digital twins and AI; enabling the integration of autonomous systems in complex environments; and advancing data-driven and distributed AI approaches for autonomous systems. It addresses the inherently physical nature of autonomy across land systems (intelligent transport and infrastructure), air systems (UAVs and aerial mobility), and space systems (robotics and satellites). A central premise of the Summer School is that autonomy is fundamentally embodied: intelligence arises from the closed-loop interaction between perception, actuation, physical dynamics, and environmental constraints.
Expected Learning Outcomes
Upon completion of the Summer School, participants will be able to position their research within the broader ATLAS research framework from a system-level perspective and design high-level architectures for Digital Twins of embodied autonomous systems. They will be trained to integrate sensing, perception, decision-making, and actuation within coherent AI-driven autonomous system architectures, and to apply distributed and federated learning approaches in networked and decentralized autonomous systems. Participants will also be able to consider safety constraints and explainability principles into system architectures, formulate cross-domain research concepts spanning Land, Air, and Space, and critically assess simulation-to-real gaps and validation strategies for physical AI systems.
Programme Overview
Each day of the Summer School is dedicated to a key theme, with embodied intelligence running as a common thread throughout. Day 1 explores Digital Twins and system architectures, Day 2 dives into data-driven and distributed intelligence, and Day 3 focuses on safety, validation, and integrating autonomous systems. The programme blends expert lectures, inspiring industry keynotes, hands-on cross-domain workshops, interactive research sessions, and wraps up with an exciting integrative system challenge that brings everything together. Participants may be grouped for three separate challenges, one for each domain: Land, Air, and Space. Alternatively, the challenge could focus on a single thematic area while being applied to one of the domains (Land, Air, or Space).
| Morning Sessions | Opening Session: ATLAS Ecosystem It offers a comprehensive overview of autonomous systems within the ATLAS ecosystem, presenting the scope of research across Land, Air, and Space domains. It highlights the key challenges, opportunities, and interdisciplinary connections in the development of intelligent, safe, and physical autonomous systems, setting the stage for the themes explored throughout the Summer School. Keynote: Digital Twins for Autonomous Physical Systems The keynote on digital twins for autonomous physical systems will explore how world models and digital twins are applied to vehicles, UAVs, and space robots, focusing on real-time synchronization between physical and virtual systems, physics-aware modeling, and accurate state estimation. It will also present cross-domain case studies that illustrate the practical implementation and benefits of digital twin technologies in complex autonomous systems. Lecture: Embodied AI Architectures for Autonomous Systems The lecture on embodied AI and closed-loop architectures will introduce the core principles of physical AI for autonomous systems, emphasizing how intelligence emerges from the interaction between sensing, actuation, and physical dynamics. It will cover perception-action loops, dynamics-aware and model-based learning, physics-informed AI agents, and the challenges of transferring models from simulation to real-world autonomous systems. |
| Afternoon Workshop | Workshop: Cross-Domain Design Lab In the Cross-Domain Design Lab, interdisciplinary groups work together to design a Digital Twin-enabled autonomous system, integrating a physical dynamics model, sensor and perception stack, control architecture, AI decision layer, energy modeling, safety monitoring, and a Digital Twin interface. Industry Panel: From Prototype to Deployment Industry experts discuss practical considerations for deploying embodied AI and Digital Twin systems, covering scalability, real-time constraints, data infrastructure, and lifecycle management. |
| Morning Sessions | Lecture: Distributed and Federated Learnng for Autonomous Systems (PCOG) This session introduces distributed and federated learning principles for physical autonomous systems, covering collective optimization, swarm intelligence, and privacy-aware learning. Participants will see how these methods enable coordinated intelligence across networked systems, from UAV fleets to multi-agent robotic platforms, including autonomous vehicles. Lecture: Multi-modal Perception for Autonomous Systems The lecture presents perception as a foundational element of embodied AI, emphasizing multi-modal sensor fusion, active perception, and scene representations and understanding. Participants learn how physically consistent state representations enable robust decision-making in dynamic, real-world environments. Lecture: Learning World Models for Embodied Agents This session presents the role of world models and latent representations in physical AI, showing how predictive models of the environment and agent dynamics can guide decision-making and control of autonomous systems. Topics include 3D/4D scene modeling, data-driven dynamics, and building connections with perception to action for autonomous physical agents. |
| Afternoon Workshop | Interactive Exercise: Designing a Learning-based Agent Participants are grouped to define the sensing, state representation, learning paradigm (supervised, reinforcement, federated, or hybrid), physical constraints, failure modes, and explainability requirements for a learning-based autonomous agent. PhD Students Workshop Doctoral participants present challenges from their thesis work receiving expert feedback to strengthen methodological rigor and cross-domain applicability. |
| Morning Sessions | Lecture: Foundations of Trustworthy AI in Physical Systems This lecture introduces the core principles of trustworthy AI in embodied autonomous systems, including robustness, reliability, accountability, and human oversight. It emphasizes embedding trustworthiness directly into the perception–decision–actuation loop. The session frames trust as a system-level design requirement for safety-critical autonomy. Lecture: Safety, Security and Explainability in Embodied AI This session explores how safety, cybersecurity, and explainability are implemented in AI-driven physical systems. Topics include formal safety layers, runtime monitoring, adversarial robustness, and interpretable decision-making. The focus is on co-designing learning components with verifiable safety and certification pathways. |
| Afternoon Workshop | ATLAS Grand Integration Challenge In this session, interdisciplinary teams design a next-generation autonomous system integrating at least one domain (Land, Air, Space) with Digital Twin infrastructure, data-driven and distributed learning, safety mechanisms, and explainability features. Panel Discussion Group presentations are evaluated by an academic–industrial jury based on scientific quality, system coherence, and integration of trustworthy AI principles. Awards are presented for the best projects in Land, Air, and Space. The program concludes with jury feedback, key takeaways, and a closing session. |
Venue
Black Box – Maison des Sciences Humaines
11, porte des Sciences
L-4366 Esch-sur-Alzette
Luxembourg
Transportation
Public transport is free in Luxembourg. The Belval campus is well connected to all transport hubs within Luxembourg, including direct train links from the central train station.
The closest stops to the Belval Campus are:
Train: Belval-Université
Bus:
– Esch-sur-Alzette, Porte des Sciences
– Belval, Porte de France
– Esch-sur-Alzette, Raemerich
– Belval (Université), Gare Routière
Plan your journey easily with Mobiliteit, Luxembourg’s official, comprehensive mobility platform provided by the Public Transport Administration.