Abstract: The capabilities and agility of legged robots has evolved in leaps and bounds in recent years, with mobile robots now on the verge of becoming ubiquitous. In this talk, I will cover recent work by the Dynamic Robot Systems group at the University of Oxford in optimisation- and learning-based methods for locomotion, manipulation, and loco-manipulation. In particular, the talk will highlight recent advances in Model Predictive Control for achieving highly agile behaviours, how we can leverage generative models to uncover interpretable representations for locomotion control, and how we can make robotic systems operate more seamlessly in shared environments with people.

Bio: Wolfgang is a Robotics Research Associate at the Oxford Robotics Institute, University of Oxford focusing on optimal planning and control of high degree of freedom systems interacting with environments working with Dr. Ioannis Havoutis. He currently works in the context of the EU Horizon 2020 project Memory of Motion (MEMMO) and the Offshore Robotics for Certification of Assets Hub (ORCA). He completed his PhD in Robotics and Autonomous Systems supervised by Prof. Sethu Vijayakumar at the University of Edinburgh, where he was a core member of the humanoids research group working on the Edinburgh-NASA Valkyrie project. Previously, he completed a MSc by Research (Distinction) in Robotics and Autonomous Systems and a BEng (Honours, 1st Class) in Mechanical Engineering with Management.

Abstract: The Dynamic Legged Systems lab has been developing quadruped robots and their locomotion control for over 15 years. In this talk, I will first introduce important design upgrades of our 140kg hydraulic HyQReal robot. Next, I will show the most recent results of our project VINUM that focuses on the automation of winter pruning in vineyards with quadruped robots. Last, my talk will take you into the field of space exploration. I will present the current progress of our ESA-funded project ANT that investigates the control and navigation aspects of quadruped and hexapod robots during space exploration in hard-to-access areas like craters.

Bio: Claudio Semini is the head of the Dynamic Legged Systems (DLS) lab at Istituto Italiano di Tecnologia (IIT). He received an MSc degree from ETH Zurich in 2005 and spent 2 years in Tokyo for his research at Tokyo Tech and Toshiba. During his PhD (2010) and subsequent PostDoc at IIT, he developed the quadruped robot HyQ. He has published more than 100 publications in international journals and conferences. He is/was the coordinator/partner of several EU-, National and Industrial projects (HyQ-REAL, INAIL Teleop, Moog@IIT joint lab, ESA-ANT, etc). His research interests include the construction and control of heavy-duty legged robots for application in real-world operations, legged locomotion, hydraulic actuation.

Abstract: This is a tandem talk between Prof. Kim and Dr. Ding. Prof. Sangbae will present some of the recent work on locomotion for humanoid and quadruped robots, and Dr. Ding will present his recent work on MPC for the MIT Humanoid robot.

Bio: Prof. Sangbae Kim, is the director of the Biomimetic Robotics Laboratory and a Professor of Mechanical Engineering at MIT. His research focuses on the bio-inspired robot design by extracting principles from animals. Kim's achievements on bio-inspired robot development include the world's first directional adhesive inspired from gecko lizards, and a climbing robot, Stickybot, that utilizes the directional adhesives to climb smooth surfaces featured in TIME's best inventions in 2006. Recent achievement includes the development of the MIT Cheetah capable of stable outdoor running up to 13mph and autonomous jumping over an obstacles at an efficiency of animals. This achievement was covered by more than 300 media articles. He is a recipient of best paper award from International Conference on Robotics and Automation (2007), King-Sun Fu Memorial Transactions on Robotics (2008) and IEEE/ASME transactions on mechatronics (2016), DARPA Young Faculty Award (2013), NSF CAREER award (2014), and Ruth and Joel Spira Award for Distinguished Teaching (2015).

Dr. Yanran Ding received his Ph.D. from the Mechanical Science and Engineering Department, University of Illinois at Urbana-Champaign (UIUC) in 2021. He is currently a post-doc at the biomimetic robotics lab at MIT working with Prof. Sangbae Kim. His research focuses on optimization-based control for both quadrupedal and humanoid robots to achieve dynamic motions. He is one of the best student paper finalists in the IROS 2017. His work on SO(3) MPC is recognized as the best paper finalist in the model-based optimization for Robotics Technical Committee, 2021.

Abstract: In this tandem talk between Ruben Grandia and Takahiro Miki, two experts in MPC and RL working on the same machine, we will present some of our very recent works in MPC and RL for legged locomotion. We will elaborate and discuss what works (and what not).

Bio: Marco Hutter is Professor for robotics at ETH and co-founder of ANYbotics AG. Ruben Grandia and Takahiro Miki are PhD students at the same lab. Our passion is to work on new approaches to make legged robots able to move across very challenging terrain. For more information, check out www.rsl.ethz.ch

Abstract: This talk focuses on the recent advances in learning-based control for legged locomotion, featuring multi-skill locomotion and perceptual locomotion. The talk will briefly review the state of the art in legged locomotion first, primarily on the "deep reinforcement learning" for continuous robot control in recent years. Then I will elaborate on the ideas and principles of multi-expert architecture for diversifying and fusing expert skills, and show how new locomotion skills perform in fall recovery, trotting, target-following and any arbitrary transitions. Followed by a brief introduction of how this architecture can be extended for manipulation. Then, I will present the meta-learning approach, in which the robot can constantly updates the interaction model and applies Model Predictive Control for generating the optimal actions to maximize the reward. The meta-learning approach learns different latent vectors of each condition and achieves online model adaptation within 0.2s, and the SpotMicro robot can produce adaptive and robust locomotion under changing ground friction, external pushes, and different robot dynamics including motor failures and the whole leg amputation. Lastly, we will take a look how perceptual locomotion can be achieved using only sparse visual feedback from direct depth measurements and yet the Anymal robot can traverse over a range of common terrains in human-centered environments (steps, ramps, gaps, and stairs). The talk will be concluded with a short interactive open discussion about the future directions and how researchers from different disciplines can collaborate and explore new research areas.

Bio: Dr Zhibin (Alex) Li is an Associate Professor at the University College London and leads the Advanced Robotics Intelligence Laboratory. His work in 2019-2020 - "multi-expert learning of adaptive legged locomotion" - is the first implementation of a multi-expert learning architecture for adaptive quadrupedal locomotion on a real robot (December 2020 issue, Science Robotics). His research interest covers dynamic motion skills of locomotion and manipulation, optimization-based motion planning and control, and deep reinforcement learning for robot motor learning including locomotion, manipulation and grasping. He led the theme of "Shared and Autonomous Manipulation" in joint cross-hub demonstration of the UK National Robotics and Artificial Intelligence Hubs, and is working to develop novel optimization and learning based robot control to achieve a new level of robot autonomy and intelligence.

Abstract: Humanoid robots are versatile platforms that can be used to accomplish a variety of missions. We believe that they can be the most useful as avatars, i.e., deployed in remote and dangerous locations, teleoperated by a human. While high-level decisions and even gestures can be commanded by the human, the robot needs to be physically capable to execute a variety of tasks without prior planning, which requires generic whole-body controllers and resilience in face of potential damages, and to compensate for communication delays that may affect the performance of the teleoperation. In this talk, I will present our solutions to these problems, which is based on the use of machine learning techniques in combination with whole-body controllers. I will show several results obtained on the real robots iCub and Talos.

Bio: Serena Ivaldi is a tenured research scientist at Inria, leading the humanoid and human-robot interaction activities of the Team Larsen in Inria Nancy, France. She obtained her Ph.D. in Humanoid Technologies in 2011 at the Italian Institute of Technology. Prior to joining Inria, she was a post-doctoral researcher in UPMC in Paris, France, then at the University of Darmstadt, Germany. She has been PI of several collaborative projects, such as the EU projects CoDyCo (FP7), AnDy (H2020), EuRobin (HE), concerning the development of advanced anticipatory control and interaction skills for humanoid robots and exoskeletons. She was Program Chair of the conference IEEE/RAS Humanoids 2019, co-leader of the Humanoid Robotics Group of GDR Robotique (the French Robotics society), and one of the judges of the Xprize ANA Avatar competition on telexistance and teleoperation of robots. She is also serving as Editor in Chief of the International Journal of Social Robotics and associate Vice-President of IEEE/RAS Member Activies Board. She was recently awarded the Suzanne Zivi Prize for excellence in research and she was nominated by Robohub and Women in Robotics in the list of “50 Women in Robotics you need to know about” in 2021. Website link: https://members.loria.fr/SIvaldi/

Abstract: Maneuverability is paramount to survival for animals and will be equally important to legged robots if they are to ever leave the safety of the laboratory. The cheetah (Acinonyx jubatus) is the pinnacle of maneuverability and can rapidly accelerate and initiate turns at high-speed on unpredictable terrain. However, biomechanics researchers are still far from understanding the whole-body control of cheetah locomotion as conventional GPS/IMU collars cannot provide the required data. In this talk I will discuss my lab’s efforts towards the challenging problem of understanding the locomotion of the cheetah. We do this through the lens of robotics and employ novel techniques including multi-body modelling, feedback control, computer vision & deep learning, physical robots, and trajectory optimization. These results are also relevant to the design of future legged robotic systems which will perform time-critical tasks in unstructured world.

Bio: Amir Patel is an Associate Professor in the Departmental of Electrical Engineering at the University of Cape Town (UCT) and the Director of the African Robotics Unit (ARU). He completed his undergraduate degree in Mechatronics Engineering (2008) and MS in Electrical Engineering (2011) at UCT. Since 2012 he has been employed by UCT where he also completed his PhD in 2015. In 2018, he was appointed as a Visiting Research Scholar at Carnegie Mellon University (Robomechanics Lab with Aaron Johnson) and Johns Hopkins University (LIMBS Lab with Noah Cowan). His primary research focus is to understand maneuverability in Africa’s fastest animals with a view to building more agile robotic systems. He does this through the lens of a roboticist using techniques such as trajectory optimization, state estimation, feedback control and bio-inspired robotic experiments. He has been the recipient of awards including the Claude Leon Emerging Researcher Award (2012), the Oppenheimer Memorial Trust Fellowship (2017, 2022), the Future Leaders – African Independent Research (FLAIR) Fellowship (2021) and the Google Research Scholar Award (2021).

Abstract: The Boston Dynamics Atlas humanoid robot can run, jump, flip, and dance using its 28 hydraulically-actuated joints. In this talk, I will discuss how we design behaviors for Atlas using trajectory optimization and animation, how we map those plans onto the world using perception, and how we execute those behaviors using nonlinear model-predictive control. I will also show how our recent developments in model-predictive control have improved whole-body coordination and performance for Atlas.

Bio: Robin Deits has been excited about walking robots since his first Lego Mindstorms kit. He spent his PhD years in the Robot Locomotion Group at MIT, where he worked on footstep planning and whole-body control during the DARPA Robotics Challenge. He is now a principal engineer at Boston Dynamics, where he works on behavior development and control for the latest generation of Atlas humanoid robots.

Abstract:

Bio: Jonathan W. Hurst is a Professor of Robotics, co-founder of the Oregon State University Robotics Institute, and Chief Technology Officer and co-founder of Agility Robotics. He holds a B.S. in mechanical engineering and an M.S. and Ph.D. in robotics, all from Carnegie Mellon University. His university research focuses on understanding the fundamental science and engineering best practices for legged locomotion. Investigations range from numerical studies and analysis of animal data, to simulation studies of theoretical models, to designing, constructing, and experimenting with legged robots for walking and running, and more recently, using machine learning techniques merged with more traditional control to enable highly dynamic gaits. Agility Robotics is extending this research to commercial applications for robotic legged mobility, working towards a day when robots can go where people go, generate greater productivity across the economy, and improve quality of life for all.

Abstract: For decades the most useful robots have been found in highly structured spaces, following predetermined paths with sub-millimeter precision and on-off end effector activation. Recent advances in robotics is beginning to change this paradigm, offering potential for robots to be applied in far less structured settings including in operation around humans. The value of potential uses for such a robot are vast if the technological hurdles to create them can be overcome. In this talk we will discuss Apptronik's systems approach and progress in creating robots for the unstructured world that we live and work in every day.

Bio: Dr. Paine received his B.S., MS., and PhD. in the field of electrical engineering from the University of Texas at Austin. His research focused on high performance force-controlled robotic actuation technologies for unstructured applications. His academic contributions have seen adoption in over 1,000 peer reviewed citations, and numerous recreations all over the world. His academic accomplishments led to him being invited to collaborate on several high-profile robotics projects such as the Valkyrie Humanoid Robot at NASA-JSC, and robots developed by Meka Robotics. After graduating, he founded Apptronik to continue developing advanced robotic technology and to apply these technologies to commercial products. He currently serves as CTO and has led numerous robotics projects in the fields of exoskeletons, manipulators, and humanoids.

Abstract: Introduce some of the things that Unitree Robotics has done in the past year to promote the development of the high-performance quadruped robot industry. Including technology, products and so on. And introduce some of the valuable things we are doing right now.

Bio: Wang Xingxing is the founder, CEO&CTO of Unitree Robotics. During his master's degree (year 2013 to 2016), he independently completed the high-performance low-cost quadruped robot XDog, which is driven by the outer rotor brushless motor. Established Unitree Robotics at the end of 2016, continued to promote the global marketization of high-performance low-cost quadruped robots. Unitree Robotics has successively launched LaikaGo, AlienGo, A1, Go1 and B1 quadruped robot, along with Z1 dexterous robotic arm, he had led the development of the above, which are widely recognized by the market.