2025 Scientific Symposium (Symposium)

All current and former employees and partners of the Max Planck Institute for Intelligent Systems are welcome to attend this event. If you have any questions, please contact Eva Lämmerhirt, Institute Management Officer, at eva.laemmerhirt@tuebingen.mpg.de

Schedule
Tuesday, February 25, 2025
Time	Activity
9:00	Talks (25 min. talk, 5 min. Q&A)
9:00-9:30	Stefan Mönch Future Energy Conversion: From Integrated Circuits to Caloric Cooling Abstract and speaker’s biography >> Abstract Two emerging technologies for a clean and efficient energy future are explored: First, wide-bandgap semiconductor-based power integrated circuits for efficient and compact electrical energy conversion, and second, electrocaloric polymers and ceramics for solid-state heating and cooling. Lateral gallium nitride (GaN) power semiconductor technology enables the monolithic integration of high-voltage power conversion stages with driving circuits, sensors, control and logic into one power integrated circuit (ICs): For electromobility, 1200V GaN transistors with integrated added functionalities are under development as enablers for more efficient, compact and cost-effective bidirectional DC battery chargers, traction inverters, and 48V on-board networks. Demonstrated designs include a low-voltage three-phase motor inverter IC, a closed-loop dc-dc buck-converter and bidirectional blocking semiconductor switches with reduced area requirement. The electrocaloric effect – a reversible temperature change induced by electric field change in special polymers and ceramics – is introduced as an emerging heating and cooling technology. The potential for next-generation, efficient and solid-state heat pumps and refrigeration systems is discussed, emphasizing the critical role of over 99% efficient electrical charge recovery circuits and converters. Biography Dr.-Ing. Stefan Mönch is a Juniorprofessor at the University of Stuttgart, Germany, since 2023, and a project leader at the Fraunhofer Institute for Applied Solid State Physics (IAF), Freiburg, Germany, since 2017. He has a M.Sc. ('14) and Ph.D. ('21) degree in Electrical Engineering with a focus on highly-efficient and integrated wide-bandgap power electronics for electromobility and renewable energy conversion. He contributed to over 80 publications, supervised over 30 student theses, was technical program chair for "GaN Applications" at the 2024 IEEE WiPDA conference, and recently received the 2023 IEEE JESTPE First Place Prize Paper Award on ultra-efficient power electronics.
9:30-10:00	Thomas Buchner Rethinking Robot Intelligence Through Mechanical Design Abstract and speaker’s biography >> Abstract Traditional robotics relies heavily on computational solutions, yet nature shows us that intelligence can emerge from mechanical design itself. By rethinking how we manufacture and actuate robots, we can achieve adaptive behaviors through physical architecture rather than complex control algorithms. I believe mechanical intelligence will complement computational approaches, opening possibilities for next-generation robotics that better mirror the elegant solutions found in biological systems. I discuss specific examples of how advances in multimaterial manufacturing enable the creation of sophisticated robotic systems with integrated sensing and actuation. In addition, I will present a system based on energy efficient electrohydraulic artificial muscles that adapts to its environment through features inherent to the structure and actuator. These developments pave the way for agile autonomous robots outside of structured environments. Biography Thomas Buchner is a physicist with a PhD candidacy in Mechanical Engineering working with Prof. Robert Katzschmann at ETH Zurich. After completing his bachelor's degree in physics at the University of Heidelberg, he gained practical experience as a state-certified farmer and then pursued a master's degree in Applied and Engineering Physics at TU Munich. During this time, he studied at Tsinghua University, China and successfully conducted his Master’s research at MGH/HMS and MIT, USA. Currently, he focuses on developing mechanically intelligent robotic systems that combine soft and rigid components for enhanced adaptability and efficiency.
10:00-10:30	Antonia Georgopoulou Making Sense of Soft Robots Abstract and speaker’s biography >> Abstract Senses allow a biological organism to perceive and interact with their environment. Living organisms are equipped with several types of sensory receptors, a key aspect of survival, adaptability and the evolution of life on earth. Soft resistive sensors are developed to mimic the natural sensory receptor response by functionalizing soft polymers with conductive species. The resulting sensors can be used to detect different stimuli in soft robotic applications; Piezoresistive sensors can be used as mechanoreceptors for proprioception and tactility; thermoresistive sensors can detect the changes in the temperature, attributing to the robot thermoreception; chemoreceptive sensors can detect the presence of vapors and solvents, resembling the sense of olfaction. Nociception, the function of detecting damage in biological systems, can be mimicked by piezoresistive and chemoreceptive sensors. One approach to develop multimodal soft robots is to combine multiple specialized receptors selective to a specific sensory stimulus. To avoid stiffening the soft robot, multi-material 3D printing allows the integration of sensors in the form of single threads that introduce functionality to the robot without impairing its motion. Another approach to multimodality is the use of single receptors that are capable of detecting multiple stimuli, a feat possible with ionogel materials, which are soft hydrogel polymers functionalized with ionic species. To realize selective tactile perception with multimodal ionogel receptors, training recurrent neural networks is necessary to achieve sensory stimulus classification when multiple stimuli are simultaneously present. The localization of the stimulus as well as the calculated prediction of the sensing is more accurate when crosstalk between different sensory information is present. Taking these considerations into account, multi-sensory soft robots can obtain a better perception of the internal and external state for improved functionality and increased autonomy. Biography Dr. Antonia Georgopoulou received her degree (M.Eng.) in Chemical Engineering with distinction from the University of Patras in 2017. She received her degree (M.S.) in Biomedical Engineering from ETH Zürich in 2019. In 2022, she received her Ph.D. in Engineering Science at Vrije Universiteit Brussel with highest distinction. She first studied tissue engineering and biofabrication before joining Empa – The Swiss Institute for Materials Science and Technology in the department of Advanced Materials and Surfaces working on the topic of sensor development for soft robotics. In 2023, she was awarded the prestigious fellowship Women in Science (WINS) and joined the soft materials laboratory at EPFL and the NCCR bioinspired materials to develop electronic skin with selective multi-sensing capabilities inspired by natural sensory neuroreceptors. Her research focuses around bioinspired and biohybrid materials and structures, additive manufacturing and functional materials for soft robotic applications.
10:30-11:00	Break
11:00-11:30	Hwayeong Jeong Intelligent Interaction Between Robots and Nature Abstract and speaker’s biography >> Abstract As robots transition from factory settings to interacting within the open, unpredictable environments of the nature, they face the same complexities that humans encounter daily. Humans are incredibly skilled at navigating and responding to diverse interactions with nature, using intelligence and adaptability. This talk explores how soft robotics can enable robots to interact intelligently and safely with the natural world. I will describe my contribution on bio-inspired embodied intelligence and the design of physically intelligent soft robots, incorporating fluidic circuits for advanced control. My work focuses on developing customizable systems that enable robots to sense, actuate, and respond to stimuli in an intuitive and comprehensive manner. I will also explore the use of electromyography (EMG) for enhancing human-robot interaction, providing insights into how humans interact with their environment and robots. Scalability and deployability have been consistently considered alongside the core focus of the research, leading to the integration of digital additive manufacturing for more flexible and efficient soft robot designs. This presentation will showcase how these advancements are enabling the creation of adaptable, intelligent robots that can thrive in dynamic environments, working seamlessly alongside humans and nature. Biography Dr. Hwayeong Jeong is a postdoctoral researcher at the Reconfigurable Robotics Lab (RRL) at the Swiss Federal Institute of Technology (EPFL). She completed her bachelor’s (2017), master’s (2019), and PhD (2023) in mechanical engineering at KAIST. Her master’s thesis was recognized and funded as a doctoral-level project, and was patented in both Korea and the US. Her PhD research tackled key challenges in soft robotics, particularly the limitations of flexible controllers and scalable production, through the development of a soft pneumatic demultiplexer with a reproducible manufacturing process. Her research focuses on robot-environment interaction and haptics, with an emphasis on understanding human sensation and movement intention, and leveraging this understanding to enable robots to interact more intuitively and comprehensively with the world. She received various honors and awards, including the Future Talent Award in 2024 from the Korean Federation of Women’s Science & Technology Associations and the IEEE RA-L Best Paper Award in 2022.
11:30-12:00	Patricia Capsi-Morales Sensorimotor Primitives and Softness for Human-Centered Assistive Robotics Abstract and speaker’s biography >> Abstract Compared to their biological counterparts, state-of-the-art artificial limbs remain limited in flexible motor skills. However, modularity is present in both structural and computational components of control architectures with a functional purpose. Indeed, literature commonly assumes that the remarkable versatility and adaptability in human motor control results from employing modularity as the key organizational principle of the central nervous system. This talk explores how sensorimotor primitives can be integrated into the design of human-centered robotic systems. The control of a system, which is conventionally thought to be handled by the brain or a controller, can partially be outsourced to the physical body and the interaction with the environment. Advances in robot technology and synergy-based approaches allow researchers to investigate how the motor, sensory, and cognitive functions might be integrated into meaningful architectures. By leveraging soft robotics, complexity reduction techniques, and neural decoding, we can develop assistive devices that dynamically adapt to their environment, enabling natural and intuitive interfacing. This talk explores how these innovations are transforming prosthetics and human-machine interaction, enhancing dexterity and their seamless body integration. Biography Patricia Capsi-Morales obtained her PhD in Information Engineering at the Italian Institute of Technology and the University of Pisa, under the mentorship of Prof. Antonio Bicchi, Prof. Giorgio Grioli, and Dr. Manuel Catalano. Her thesis, "Neuroscientific and Soft Robotic Principles for a New Generation of Natural Bionic Limbs," was awarded the IEEE Italy Section 2022 PhD Thesis Award for New Challenges in Energy and Industry – Technology. During her PhD, she conducted visiting research stay at Imperial College London with Prof. Dario Farina, exploring real-time neural decoding for advanced bionic control. Later, she collaborated with Prof. Marco Santello in Arizona State University on a long-term clinical at-home study. Her work has been published in top-tier journals and has led to multiple patents. In her last work, published in Science Robotics, she combined her robotics and neuroscience background to propose a natural bionic interfacing through the direct connection between postural and neural synergies of spinal motoneurons. She played a key role in the SoftHand Pro Team's silver medal win at Cybathlon 2020 Global Edition. Recognized as a promising early-career researcher, she was invited to the EU-US Frontiers of Engineering 2022 symposium. Currently, she is a Senior Research Fellow at the Technical University of Munich (TUM) in the School of Computation, Information, and Technology, under the supervision of Prof. Cristina Piazza. She specializes in human-machine interfacing and the development of assistive and rehabilitation robotics. Her research focuses on the mechanical design and neural control of upper-limb prostheses, leveraging soft robotics, myoelectric control, and AI-driven approaches to enhance human-robot interaction. She aims to develop intelligent, user-centered prostheses and robotic manipulators that seamlessly adapt to human needs, ultimately enhancing human capabilities in real-world applications.
12:00-12:30	Philipp Müller Multi-Modal Behaviour Analysis for Effective Human-Machine Collaboration Abstract and speaker’s biography >> Abstract As artificial intelligence systems become increasingly integrated into our daily lives, enabling effective collaboration between humans and machines is more important than ever. For such collaboration to be seamless and intuitive, machines must develop a deeper understanding of human behaviour - how we communicate, express emotions, and direct our attention. In this talk, I will present my work on advancing the state of the art in human behaviour analysis, with an emphasis on the value of integrating psychological perspectives with those offered by computer science. In the first part of the talk, I will focus on the analysis of human behaviour during conversations. This involves the detection of low-level behavioural cues, including eye contact and body movements, as well as more complex phenomena such as engagement and emotions. I will show how integrating prior knowledge of human interaction conventions can help build effective eye contact detection approaches without the need to collect manual annotations. Furthermore, I will present recent work on recognising emotion regulation strategies with the help of large language models fine-tuned on prompts incorporating multi- modal behaviour descriptions. The second part of my talk will focus on attention as a connecting link between humans and machines. On the one hand, I will describe how models of human attention allocation from cognitive psychology can be utilised to improve machine learning approaches. On the other hand, I will discuss ways in which models of human attention can enhance the seamlessness of human-machine interaction. Biography Dr. Philipp Müller is a researcher working at the intersection of machine learning, psychology, and human-machine interaction. He received a Bachelor's degree in Psychology as well as Bachelor's and Master's degrees in Computer Science from Saarland University. After staying as a Visiting Student Researcher at Stanford University, he conducted his PhD research at the Max Planck Institute for Informatics on "Sensing, Interpreting, and Anticipating Human Social Behaviour in the Real World." Subsequently, he worked at the University of Stuttgart and is now a Senior Researcher at the German Research Center for Artificial Intelligence (DFKI) in Saarbrücken. The goal of his current research is to lay the foundations for effective human-machine collaboration by advancing the capabilities to analyse multi-modal human behaviour. He has authored 40 peer-reviewed publications in computer science and psychology venues, including top conferences such as CHI, NeurIPS, ICCV, and ACM Multimedia. Since 2021, he has been an organiser of the annual ACM Multimedia Grand Challenge "MultiMediate" on automatic social behaviour analysis.
12:30-13:30	Lunch
13:30-14:00	Debkalpa Goswami Soft Robots for Hard Problems in Disease Modeling and Therapy Abstract and speaker’s biography >> Abstract In contrast to rigid-bodied robots built from metals, ceramics and hard plastics, soft robots are made of soft and/or compliant materials, making them inherently safe when working in close contact with humans. The growing field of soft robotics provides an ideal opportunity for the development of implantable devices and biomimetic simulation testbeds due to the constituent materials possessing mechanical properties comparable to that of biological tissue. Soft robotic devices are pushing the boundaries of robotics in accomplishing tasks that are out of the reach of traditional rigid body systems. In this talk, Dr. Goswami will present some of his recent work that leverages soft robotic technology to build both benchtop and in vivo models of cardiovascular disease. He will also discuss novel implantable soft robotic drug delivery devices with potential applications in the treatment of type 1 diabetes. He will explore how these platforms can be harnessed to simulate disease progression, enabling more accurate and personalized treatment strategies. Biography Debkalpa Goswami serves as the Director of Biomechanics at the Cleveland Clinic’s Cardiovascular Innovation Research Center and holds an academic appointment as an Assistant Professor at Case Western Reserve University, USA. Debkalpa received his Ph.D. from Purdue University and completed postdoctoral training at the Massachusetts Institute of Technology. He has held full-time research positions at ETH Zurich, Switzerland, and the University of Bremen, Germany, before starting as a faculty at Cleveland Clinic and Case Western Reserve University in early 2023. His research group combines soft robotics, 3D printing, biosensing tools, and computational modeling to build advanced physical and digital biomechanical models of disease.
14:00-14:30	Stephanie Woodman Toward General Shape-Changing Robots: Leveraging Stretchable Computation and Robot-Agnostic Shape Sensing to Close the Loop on Shape in Soft Robots Abstract and speaker’s biography >> Abstract Advancements in stretchable computation and robot-agnostic shape sensing are paving the way for robots that can autonomously adapt their form in real time. From multi-terrain surveillance to multifunctional caretaker robots, these systems will no longer be constrained by a single form, but instead will evolve dynamically to meet changing demands. Achieving such drastic shape transformations requires robots that are inherently deformable and capable of withstanding large strains. In this presentation, I will discuss my contributions to stretchable computation and robot-agnostic shape sensing, and how we can leverage these to achieve closed-loop shape control for soft robots. Biography Stephanie's research focuses on soft robotics and shape-changing robots. She earned her BS in mechanical engineering from Boston University, where she conducted research in both the Morphable Biorobotics Lab and the Mesoscale Soft Matter Lab. Now a NASA NSTGRO Fellow under Professor Rebecca Kramer-Bottiglio at Yale University, she has become an expert in soft computation and shape-sensing-both essential to her vision of general shape-changing robots (GSCRs) capable of real-time evolution.
14:30-15:00	Renate Sachse Computational Soft Robotics: Mechanics-Driven Design of Bio-Inspired, Intelligent Systems Abstract and speaker’s biography >> Abstract Soft robots, with their ability to adapt to imperfect environments and handle fragile objects, show great promise for applications ranging from medical assistance to delicate manipulation tasks. However, it remains challenging to design soft robots that can execute targeted maneuvers while accounting for their complex mechanical behavior. This talk presents a novel computational approach that combines insights from plant biomechanics with advanced methods in computational mechanics to design and optimize soft robotic systems. Through systematic analysis of biological motion principles and the development of tailored numerical methods, mechanical understanding drives the design process. The presentation introduces new computational frameworks for analyzing and optimizing complex deformations, considering both structural behavior and actuation strategies, and illustrates how mechanics-driven simulation methods can bridge the gap between biological inspiration and practical soft robot design. Biography Dr. Renate Sachse is currently a postdoctoral researcher at the Technical University of Munich and has completed a research stay at the Bertoldi Lab at Harvard University. She received her Ph.D. in Civil Engineering from the University of Stuttgart in 2020, combining computational mechanics with biomimetic design principles to focus on the development of soft robots and compliant mechanisms. Her research has been published in leading disciplinary journals in mechanics, such as the International Journal for Numerical Methods in Engineering, as well as high-ranking interdisciplinary journals, including PNAS and Advanced Science. This work has been recognized with several prestigious awards, including the Bertha Benz Prize 2022, the Dr. Wilhelmy-VDI Award, and the Klaus Tschira Boost Fund Fellowship. Beyond research, she is actively engaged in promoting young researchers and currently serves as the chairperson of the Young Investigators Committee of the European Community on Computational Methods in Applied Sciences (ECCOMAS).
15:00-15:30	Break
15:30-16:00	Ahalya Prabhakar Spectral-Based Methods for Efficient Human-Robot Collaboration Abstract and speaker’s biography >> Abstract A robot's success in the real-world is determined by effective reasoning of sensor, motion and task information in dynamic situations with limited data and time. This is made further challenging in collaborative scenarios where communication and understanding between robotic agents and human operators is crucial. I show that spectral-based methods are promising tools that push the boundaries towards effective and intuitive human-robot collaboration. I demonstrate how these methods can be used to develop a framework that allows human operators to communicate task and motion information through spatial representations that can directly be used in model predictive controllers, inherently accommodating dynamic constraints. Furthermore, these representations enable co-design of intuitive interfaces for human-robot communication. I illustrate these two aspects by applying it to human-swarm teaming, showing how they can be used to facilitate effective real-time human-swarm communication. Finally, I provide an outlook on future directions for exploring intuitive representations to facilitate multimodal human-robot interaction. Biography Ahalya Prabhakar is an Associate Research Scientist and Lecturer in the School of Engineering and Applied Sciences at Yale University. Previously she was a postdoctoral fellow at EPFL investigating multimodal sensor learning to enable safe, adaptive control for human-robot interaction. She received the PhD in Mechanical Engineering from Northwestern University, where she explored autonomous methods for enabling efficient human-robot communication and robot learning. Her research is focused on developing useful algorithmic interfaces for humans to intuitively interact with and control robotic systems.
16:00-16:30	Johannes Köhler High-Performance & Safe Control of Autonomous Systems Using Optimisation, Robustness, Adaptation Abstract and speaker’s biography >> Abstract This talk focuses on advanced control methods for autonomous operation in robotics. Specifically, online optimisation, robust designs, and online adaptation are used to enhance performance and ensure safety. High-level specification during autonomous operation can be naturally addressed using online re-planning with optimisation-based techniques. However, the resulting computational demand is often a bottleneck for embedded applications. This problem is addressed using tailored control architectures and machine learning techniques. Furthermore, an efficient method for predicting reachable sets is presented, which can ensure collision-free operation. Lastly, learning-based techniques are presented to enhance performance, ensure reliable operation in uncertain environments, and efficiently automate the control design. Efficiency and reliability of these methods are illustrated with robotic experiments. Biography Johannes Köhler is a postdoctoral researcher and lecturer at ETH Zurich, Switzerland. He received his doctoral degree in mechanical engineering in 2021 from the University of Stuttgart, Germany. During his academic career, he had research stays at the Autonomous Systems Laboratory at Stanford University, the Medical Devices and Technologies Group at the University of Auckland, and the School of Engineering and Applied Sciences at Harvard University. His research focuses on advanced control methods that leverage optimisation, robust design, and online learning to ensure high-performance and safe operation of autonomous robots. His research has been recognised with the EECI PhD Thesis Award for the best PhD thesis in Europe in the field of Systems & Control, the IEEE CSS George S. Axelby Outstanding Paper Award 2022 for the best paper published in the IEEE Transactions on Automatic Control, and the IFAC Journal of Process Control Paper Award 2023.
16:30-17:00	Carmelo Sferrazza The Path to Humanoid Intelligence Abstract and speaker’s biography >> Abstract Humanoid robots represent the ideal physical embodiment to assist us in the diversity of our daily tasks and human-centric environments. Driven by substantial hardware advancements, progress in artificial intelligence (AI), and a growing demand for adaptable automation, this vision appears increasingly feasible. Yet, to date, humanoid intelligence remains far from achieving its envisioned general-purpose capabilities. What sets it apart from other challenges in machine learning - and even within robotics? My research seeks to address this question and ultimately aims to bridge the gap between humanoid AI and human intelligence. In this talk, I outline two key strategies: (1) developing algorithms, systems, and architectures that leverage embodiment-aware priors and inductive biases to manage the high-dimensional complexity of humanoid robot learning, and (2) harnessing multi-sensory feedback from the environment - such as vision, touch, and audio - to enable a unified, versatile embodiment capable of effectively reasoning across a wide range of tasks. Biography Carmelo (Carlo) Sferrazza is a postdoctoral researcher at UC Berkeley, working with Prof. Pieter Abbeel. His research focuses on advancing humanoid robots’ intelligence by incorporating priors, inductive biases, and multi-sensory feedback. Carlo obtained his Ph.D. from ETH Zurich under the supervision of Prof. Raffaello D’Andrea. During his doctoral research, he worked on the design of vision-based, data-driven tactile sensors, and the applications of such sensors to robot control and dexterous manipulation. Carlo’s Ph.D. thesis was awarded the ETH Medal, and in 2022, he was selected as a Robotics Science and Systems Pioneer. He is also the recipient of an SNSF Postdoctoral Fellowship, the Best Paper Award at the 2020 IEEE International Conference on Soft Robotics, and the 2017 ETEL Award. He frequently shares his research with the general public, including presentations at events such as the WORLD.MINDS Annual Symposium and TEDxZurich.