Keynote Speakers
|
Heejin AhnKorea Advanced Institute of Science and Technology,Republic of Korea Safety guarantee in learning-based control August 22, 14:00-14:45 Detailed information |
Davide ScaramuzzaUniversity of Zurich, SwitzerlandLearning agile, vision-based drone flight: from simulation to reality August 23, 14:00-14:45 Detailed information |
|
Katja MombaurKarlsruhe Institute of Technology, Germany,University of Waterloo, Canada Motion intelligence for human-centred robots through model-based optimization and learning August 24, 14:00-14:45 Detailed information |
Behçet AçikmeşeUniversity of Washington, USAConvex Optimization Based Optimal Control August 24, 14:45-15:30 Detailed information |
Heejin Ahn
Date: August 22, 14:00-14:45
Title: Safety guarantee in learning-based control
Abstract: The advances of machine learning in perceiving environments enable control systems to perform challenging tasks previously difficult to handle. However, learning-based control systems also inherit the fragility issue of machine learning, thereby often failing to identify the environments correctly and violating safety constraints. In this talk, we will discuss our recent approaches to ensuring the safety of learning-based control systems from a stochastic control perspective. We will also provide an overview of other safe learning-based control approaches.
Biography: Heejin Ahn is an Assistant Professor at the School of Electrical Engineering, Korea Advanced Institute of Science & Technology (KAIST), South Korea. She received her S.M. and Ph.D. in Mechanical Engineering from the Massachusetts Institute of Technology (MIT), USA, in 2014 and 2018, respectively. She received her B.S. in Mechanical and Aerospace Engineering from Seoul National University (SNU), South Korea in 2012. Before joining KAIST in 2022, she worked as an Assistant Professor at the Department of Electrical and Computer Engineering at SNU, as a postdoctoral research fellow at the University of British Columbia, Canada, and as a visiting research scientist at Mitsubishi Electric Research Laboratories, USA. Her research interests include the design and analysis of multi-agent control and learning-based control with applications to intelligent transportation systems.
Davide Scaramuzza
Date: August 23, 14:00-14:45
Title: Learning agile, vision-based drone flight: from simulation to reality
Abstract: I will summarize our latest research in learning deep sensorimotor policies for agile vision-based quadrotor flight. Learning sensorimotor policies represents a holistic approach that is more resilient to noisy sensory observations and imperfect world models. However, training robust policies requires a large amount of data. I will show that simulation data is enough to train policies that transfer to the real world without fine-tuning. We achieve one-shot sim-to-real transfer through the appropriate abstraction of sensory observations and control commands. I will show that these learned policies enable autonomous quadrotors to fly faster and more robustly than before, using only onboard cameras and computation. Applications include acrobatics, high-speed navigation in the wild, and autonomous drone racing.
Biography: Davide Scaramuzza is a Professor of Robotics and Perception at the University of Zurich. He did his Ph.D. at ETH Zurich, a postdoc at the University of Pennsylvania, and was a visiting professor at Stanford University. His research focuses on autonomous, agile navigation of mini drones using standard and event-based cameras. He pioneered autonomous, vision-based navigation of drones, which inspired the navigation algorithm of the NASA Mars helicopter and many drone companies. In 2022, his team demonstrated that an AI-powered drone could outperform the world champions of drone racing, a result published in Nature and featured on the magazine's cover. His result marks the first time an AI defeated a human in the physical world. For his research contributions, he has won many awards, including an IEEE Technical Field Award, the IEEE Robotics and Automation Society Early Career Award, a European Research Council Consolidator Grant, and many paper awards, including the IROS 2023 Best Paper Award, the IEEE Robotics and Automation Letters and the IEEE Transactions on Robotics best paper awards. His results have been transferred to many products, from drones to automobiles, cameras, AR/VR headsets, and mobile devices. Davide counts several entrepreneurial achievements: In 2015, he co-founded Zurich-Eye, which became Facebook-Meta Zurich, and developed the world-leading virtual-reality headset Meta Quest. In 2020, he co-founded SUIND, which builds autonomous drones for precision agriculture. Davide has been consulting the United Nations on disaster response, the Fukushima Action Plan, disarmament, and AI for good. Many aspects of his research have been featured in the media, such as The New York Times, The Economist, and Forbes.
Katja Mombaur
Date: August 24, 14:00-14:45
Title: Motion intelligence for human-centred robots through model-based optimization and learning
Abstract: Human-centred robots are predicted to have a large societal impact in the future, e.g. in form of humanoid robots supporting people in dangerous or monotonous jobs. They also can take the form of wearable robots or physical assistive systems enhancing and restoring mobility and independence of seniors or patients with impairments. In order to take human-centred robots to this level, still a number of challenges have to be solved, since these robots have to enter in close physical interaction with humans in a safe and socially acceptable manner. For this, they require motion intelligence which makes them aware of the mechanics of their own motions and lets them predict the actions of humans. However, generating and controlling motions for whole-body humanoid robots or for wearable robots in close interaction with the human body is very challenging due to their nonlinear dynamics with many degrees of freedom, underactuation and inherent instability. In our research, we aim to gain a fundamental understanding of the biomechanics of human movement and human-human and human-robot interaction and to develop models at different levels of complexity. Model-based optimization and optimal control play and important role in motion analysis, prediction and control, and can be efficiently combined with model-free methods. We also explore potential benefits of model-based reinforcement learning for improving motions of human-centred robots. In this talk, I will present different examples of complex motions for human-centred robots which all require an effective coordination of the whole body and advanced stability control. I will show how our Reem-C full size humanoid robot learned to walk on different terrains, balance, ride the Segway and a skateboard, manipulate and carry objects with both hands, and interact with humans at close proximity during dancing. In addition, I will show some examples of mobility assistance robots like exoskeletons and robotic rollators and the effect of different types of controllers on physical assistance as well as user comfort.
Biography: 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 research activities at LAAS-CNRS in Toulouse and Seoul National University, as well as in the USA. She studied Aerospace Engineering at the University of Stuttgart and SupAéro and holds a PhD in Mathematics from Heidelberg University.
Katja's research focuses on understanding human movement by a combined approach of model-based optimization and experiments and using this knowledge to improve motions of humanoid robots and the interactions of humans with exoskeletons, prostheses and external physical devices. Her goal is to endow humanoid and wearable robots with motion intelligence that allow them to operate safely in a complex human world. The development of efficient algorithms for motion generation, control and learning is a core component of her research.
Behçet Açikmeşe
Date: August 24, 14:45-15:30
Title: Convex Optimization Based Optimal Control
Abstract: Many future aerospace engineering applications will require dramatic increases in our existing autonomous control capabilities. These include robotic sample return missions to planets, comets, and asteroids, formation ying spacecraft, swarms of autonomous spacecraft, unmanned aerial, ground, and underwater vehicles, and autonomous commercial robotic applications. A key control challenge for many autonomous systems is to achieve the performance goals safely with minimal resource use in the presence of mission constraints and uncertainties. In principle these problems can be formulated and solved as optimal control problems. The challenge is solving them reliably in real-time, while assuring: i) full utilization of the performance envelope for the autonomous system; ii) systematic verification of the control algorithms.
Our approach to solving these challenging control problems is optimization-based control where we formulate these control problems as optimization problems and then to exploit convex optimization theory and algorithms for their robust and numerically efficient solutions. This seminar introduces several real-world aerospace applications, where optimization-based control has provided dramatic performance improvements over the heritage technologies. For example, recent applications of reusable rockets have benefited from the ability to compute, in real time, numerical solution of the underlying optimal control problems. There are also other applications, which can benefit from advances in real-time optimal control, including autonomous aerial drones, spacecraft proximity operations for rendezvous, docking and servicing, autonomous multi-vehicle systems, hypersonic transportation, proximity operation near asteroids and comets to name few. This talk will point out some of the key control problems that real-time optimal control could address in these applications as well as the challenges we face as researchers in transitioning this enabling technology into practice.
Biography: Behçet Açikmeşe is a Professor of Aerospace Optimization and Control in the William E. Boeing Department of Aeronautics and Astronautics and an adjunct faculty member in the Department of Electrical Engineering at University of Washington, Seattle. He received his Ph.D. in Aerospace Engineering from Purdue University. He was a senior technologist at JPL and a lecturer at Caltech. At JPL, he developed control algorithms for planetary landing, spacecraft formation flying, and asteroid and comet sample return missions. He developed the " flyaway " control algorithms used successfully in NASA's Mars Science Laboratory and Mars 2020 missions during the landings of Curiosity and Perseverance rovers on Mars in 2012 and 2021. Dr. Açikmeşe invented a real-time convex optimization based planetary landing guidance algorithm (G-FOLD), which is the first demonstration of a real-time optimization algorithm on a reusable rocket. This novel optimization-based control algorithm proved to be a key development in aerospace guidance and control, especially in enabling advanced numerical optimization techniques for autonomous rockets. He is a recipient of the NSF CAREER Award, IEEE Technical Excellence in Aerospace Controls, numerous NASA Achievement awards for his contributions to NASA missions and technology development. He is an associate editor of IEEE Control System Magazine and AIAA Journal of Guidance, Control, and Dynamics. He is a fellow of IEEE and AIAA.