Katja Mombaur
Karlsruhe Institute of Technology, Germany
Holistic models of humans, exoskeletons and their interactions
Abstract: Human models play an important role in human-robot interaction to allow the robot to identify the interacting human’s needs and to better predict their actions. In the case of exoskeletons, human-robot interactions are particularly challenging due to the permanent interaction with direct contact. In this case, it is important to consider holistic models of the combined human-exoskeleton system with subject-specific human models including mechanical, muscular and potentially neural components, exact models of the exoskeleton and its physical properties, and a focus on the interaction between the two including interaction forces and torques at all interfaces. In this talk, I will give an overview of different levels of detail of interaction modeling in humans and exoskeletons, which should be chosen based on the research questions asked. An important part of human-exoskeleton interaction is the adaptation process to the support provided by the exoskeleton resulting in a change of movement over time.
Bio: Katja Mombaur joined the Karlsruhe Institute of Technology in Germany in 2023 as Full Professor, Chair for Optimization & Biomechanics for Human-Centred Robotics and Director of the BioRobotics Lab. In addition, she holds an affiliation with the University Waterloo in Canada where she has been Full Professor and Canada Excellence Research Chair (CERC) for Human-Centred Robotics & Machine Intelligence since 2020. Prior to moving to Canada, she has been a Full Professor at Heidelberg University where she directed the Optimization, Robotics & Biomechanics Chair, as well as the Heidelberg Center for Motion Research. Her international experience includes two years as a visiting researcher at LAAS-CNRS in Toulouse and one year at Seoul National University. She studied Aerospace Engineering at the University of Stuttgart and SupAéro in Toulouse and holds a PhD in Mathematics from Heidelberg University.
Justin Carpentier
Inria Research Scientist
INRIA Paris, France
Recent progress and prospects in (differentiable) simulation for robotics and beyond
Abstract: Over the past few years, robotics simulators have largely improved in efficiency and scalability, enabling them to generate years of simulated data in a few hours. Yet, efficiently and accurately computing the simulation derivatives remains an open challenge, with potentially high gains on the convergence speed of reinforcement learning and trajectory optimization algorithms, especially for problems involving physical contact interactions. In this presentation, I will highlight the recent contributions on differentiable simulation developed in the WILLOW research group at Inria Paris, which have led to the development of a new fully differentiable simulator for robotics, called Simple, and shortly available at https://github.com/Simple-Robotics/Simple.
Bio: Justin Carpentier is a researcher at Inria and École Normale Supérieure, heading the Willow research team since 2023. He graduated from École Normale Supérieure Paris-Saclay in 2014 and received a Ph.D. in Robotics in 2017 from the University of Toulouse. He did his Ph.D. in the Gepetto team at LAAS-CNRS in Toulouse, working on the computational foundations of legged locomotion. In 2024, he received an ERC Starting Grant focusing on laying the algorithmic and computational foundations of Artificial Motion Intelligence. His research interests lie at the interface of optimization, machine learning, computer vision, simulation, and control for robotics, with applications ranging from agile locomotion to dexterous manipulation. He is also the leading developer and manager of widely used open-source robotics software, among them Pinocchio, ProxSuite, HPP-FCL, and Aligator
Nasser Rezzoug
Associate Professor
University of Poitiers, France
Upper-limb biomechanical capacity modeling for human/robot interaction
Abstract: One fundamental assumption of human centered robot control is the ability to measure or estimate the physical capacity of humans to continuously adapts the robot assistance level based on the real-time need of its human counterpart. Moreover, within the framework of computer-aided design of workstations, the knowledge of human force capacities enables us to evaluate to what extent a task is in adequacy with the capacities of the operators, to define ergonomic criteria of discomfort and to implement models of muscular fatigue. In this framework, the presentation will focus on the work carried out to propose biomechanical models of force capacity based on musculoskeletal models and convex polytopes, and the experimental campaigns used to validate them.
Bio: Nasser Rezzoug received his PhD thesis in 2000 at the university of Orsay. His research was focused on grasping force coordination during grasping and graspless manipulations. He, then, joined the LESP and HandiBio laboratory Toulon in 2002. His research interest focused on the study of the coordination of the upper-limb in healthy and patients suffering from spinal cord injury, learning based grasp synthesis as well as posture and gait analysis with wearable sensors. He also started to study human force capacities with joint-torque and musculoskeletal models. In 2020, he joined the INRIA AUCTUS team in Bordeaux to contribute to the biomechanical aspect of human robot collaboration and teleoperation. Since 2022, he is an associate professor at the Robioss team of the PPrime institute in Poitiers. His research interests concern human biomechanical capacities modeling and measure as well as the dynamics of human movement. He was the co-chair of the board of directors of the French speaking international society of biomechanics from 2020 to 2022 and is now co-chair of the scientific theme “movement and autonomy” of the French research group in robotics.
Ajay Seth
Associate Professor of Biomechanical Engineering
Delft University of Technology (TU Delft), Netherlands
Personalized musculoskeletal modeling for biomechanics-aware robotic rehabilitation and physical therapy of the human shoulder—a modeler’s perspective
Abstract: For musculoskeletal models to be useful in movement rehabilitation and physical therapy, they must help us understand the disorder/injury and elucidate better treatment options. But, it turns out that interpreting human movement—kinematics, muscle and joint forces—is already hard. Can we expect therapists to process and analyze model results? On the other hand, robots are good at navigating landscapes and reacting to perturbations. Therefore, we propose to make robots aware of the human/patient injury state based on a musculoskeletal model and to act and re-act accordingly during therapy—where it counts. The first step is creating a model that captures the outcome measures and/or injury risks during therapy. Then, using it to understand how key measures are affected by patient morphology and other properties? We employ and develop OpenSim Creator to create and personalize musculoskeletal models that enable us to quantify range-of-motion and rotator cuff muscle strains while performing physical therapy of the human shoulder.
Bio: Ajay Seth is an associate professor of Biomechanical Engineering at the Delft University of Technology (TU Delft), where he leads the Computational Biomechanics Lab. The lab’s mission is to develop computational models and algorithms that enable the acquisition, analysis and prediction of human and animal movement. He is interested in methods that quantify and explain the biological basis of human movement from the pathological (injury and stroke) to the exceptional (cycling and rowing). Before joining TU Delft, he was the architect of the modeling and simulation software, OpenSim, at Stanford University where he completed a Simbios distinguished postdoctoral fellowship in Bioengineering. Ajay received his PhD from the University of Texas at Austin in Biomedical engineering and predoctoral degrees in Systems Design Engineering from the University of Waterloo, Canada.
Shaoping Bai
Professor, Department of Materials and Production
Aalborg University, Denmark
Design modeling, motion sensing and control for high-performance human-robot interaction in wearable exoskeletons
Abstract: Wearable exoskeletons are being advanced rapidly for broad applications. An essential issue in exoskeleton development is the human-robot interaction, which requires in-depth study in biomechanical modeling, compliant actuation, human motion sensing and interaction control. This talk will provide a brief overview of wearable technology development at the Exoskeleton Lab, AAU, addressing these research challenges in human-exoskeleton interaction. Novel designs, biomechanical simulations and sensing methods will be presented, along with application examples.
Bio: Shaoping Bai is a full professor at Department of Materials and Production, Aalborg University (AAU), Denmark. His research interests include wearable sensors, medical and assistive robots, and exoskeletons. Prof. Bai leads several national and international research projects in exoskeletons, including EU AXO-SUIT and IFD Grand Solutions project EXO-AIDER, and Danish Independent Research Council project VIEXO, among others. He is a recipient of a number of awards including IEEE CIS-RAM 2017 Best Paper Award, IFToMM MEDER 2018 Best Application Paper Award and WearRAcon2018 Grand Prize of Innovation Challenges, the Best Student Paper in 2024 Inter. Workshop on Medical and Service Robots. Prof. Bai was an associate editor of ASME J. of Mechanisms and Robotics, IEEE Robotics and Automation Letters, and is currently an associate editor of Robotica and ASME Letters in Translations Robotics. He is the founder of BioX ApS, an AAU spin-off on wearable technologies. He is an elected member of IFToMM Executive Council and serves as a deputy chair of IFToMM Denmark.
Nicolas Perrin-Gilbert
CNRS Research Scientist
The Institute of Intelligent Systems and Robotics, France
Boosting sample efficiency to enable online reinforcement learning
Abstract: In recent years, reinforcement learning (RL) has shown great potential, positioning itself as a key technology for future robotic controllers. Currently, the most effective way to apply RL to real robots involves learning primarily in simulation, utilizing either high-fidelity simulators or domain randomization, followed by transfer learning to adapt the controller to the real-world system. However, this approach becomes impractical when a human operator constantly interacts with the system, as human behaviors are challenging to model accurately in simulation. To overcome this limitation, online learning through real-world interactions is preferable, but the high data demands of deep RL present a significant hurdle. Enhancing the sample efficiency of RL algorithms is therefore essential to enable effective online learning in human-robot interaction contexts. In this presentation, I will introduce two methods aimed at improving sample efficiency: the first is a divide-and-conquer strategy that leverages the sequential nature of robotic tasks, while the second is an off-policy RL algorithm for continuous control that cleanly separates actor and critic updates.
Bio: Nicolas Perrin-Gilbert obtained his Ph.D. in computer systems from Toulouse INP, France, in 2011. From 2008 to 2011, he was as a visiting PhD student at CNRS-AIST JRL in Tsukuba, Japan, where his research focused on locomotion planning and control using the HRP-2 robot. From 2011 to 2013, he held a postdoctoral position at the Italian Institute of Technology in Genoa, Italy, where he conducted research on balance control with the humanoid robot COMAN. Since 2013, he has been a permanent CNRS researcher affiliated with the Institute of Intelligent Systems and Robotics (ISIR) in Paris, France. His research interests include bipedal locomotion and walking aids, balance control, motion planning, reinforcement learning and machine learning in general.
