- Eiichi Yoshida (Tokyo University of Science, Japan)

“ Learning-based Understanding and Prediction of Human Motion for Symbiotic Robot Interaction ”

Human-symbiotic robotic behavior needs not only understanding and interpreting human behaviors, but also predicting their intentions. We have been addressing anthropomorphic whole-body motion understanding, basically model-based dynamic or musculo-skeletal analysis. While it is very useful, we have also realized the difficulty of understanding the underlying motion strategy and the synthesize anthropomorphic motions only by this approach. Recent advances on machine learning techniques can be one important key to address those challenges. In this talk, we introduce some recent research activities to leverage learning methodology, on contact detection and human-following robot. We first address contact detection and estimation from motions. Physical human-robot interaction (pHRI) needs to deal with various contacts. We started tacking this issue by variable autoencoder (VAE) to detect contacts and estimate their force at the same time from the motion input only. We introduce another study on a human-following mobile robot, predicting  motions friendly to both the human and robot by combining optimization and machine learning towards smooth interaction.

- Serena Ivaldi (INRIA, France)

“ Adaptation in human-robot collaboration ”

In this talk, I will present our experimental findings concerning the adaptation mechanisms that humans exhibit when physically interacting with a cobot during co-manipulation tasks. This knowledge is important to inform robotics controller that reason about the human ergonomics during collaboration. I will also show some software tools that we developed to assess and visualise ergonomics criteria online based on Digital Human Models.

- Tadej Petrič (Jožef Stefan Institute, Slovenia)

“ Enhancing human-robot collaboration through investigating human dyads ”

In this talk, I will focus on improving human-robot collaboration through investigating human dyads in workspaces. The interaction between humans and robots in shared workspaces presents numerous challenges that must be addressed to optimize their collaboration. In this context, studying human dyads in workspaces provides an effective means to understand and improve human-robot interaction. By exploring the dynamics of human dyads and the factors that influence their collaboration in shared workspaces, we can develop strategies to improve the effectiveness of human-robot collaboration. This talk will provide an overview of my current research on human dyads in workspaces and highlight the potential benefits of this approach for optimizing human-robot collaboration.

- Bram Vanderborght (Vrije Universiteit Brussel, Belgium)

“ Industry 5.0: a multidisciplinary research approach? ”

Human-robot collaboration has great potential to face societal challenges (as ageing population, need for better and healthier work), sustainability requirements and economic needs (small lot sizes, reshoring,…). In this talk we will focus how we achieve a human centered approach by combining expertise of not only engineering/AI fields, but also human and social sciences. As such we achieve a higher acceptance of the technology by the use. We provide examples on collaborative robots and exoskeletons for improved ergonomics and sustainable self healing soft grippers. At the VUB this work was performed in the Brussels Human Robotics Research Center, BruBotics, which is a joint initiative of 8 research groups of the Vrije Universiteit Brussel (VUB) sharing a common vision: improve our quality of life through Human centered Robotics.

- Dorsa Sadigh (Stanford University, USA)

“ Learning Representations for Human-Robot Collaboration ”

There have been significant advances in the field of robot learning in the past decade. However, many challenges still remain when considering how robot learning can advance interactive agents such as robots that collaborate with humans. In this talk, I will be discussing the role of learning representations for robots that interact with humans and robots that interactively learn from humans through a few different vignettes. I will first discuss how bounded rationality of humans guided us towards developing learned latent action spaces for shared autonomy. It turns out this “bounded rationality” is not a bug and a feature — i.e. we can develop extremely efficient coordination algorithms by learning latent representations of partner strategies and operating in this low dimensional space. I will then discuss how we can go about actively learning such representations capturing human preferences including our recent work on how large language models can help design human preference reward functions. Finally, I will end the talk with a discussion of the type of representations useful for learning a robotics foundation model and some preliminary results on a new model that leverages language supervision to shape representations.

- Xu Xu (North Carolina State University, USA)

“ Promote workers’ safety and health through ubiquitous sensing during human-robot collaborative assembly tasks ”

In recent years, the topic of ubiquitous sensing has gained significant attention due to the availability of low-cost portable sensors, such as cameras and inertial measurement units. Through the use of ubiquitous sensing, human-centric intelligent systems can collect signals from sensors and perform context-aware computing based on human physical and cognitive activities. This talk will explore various applications of ubiquitous sensing, with a particular focus on promoting safety and health during human-robot collaborative assembly tasks. Specifically, we will discuss three applications: collision avoidance between human workers and collaborative robots, finding an optimal point of operation for robots to minimize the risk of musculoskeletal disorders for workers, and examining the impact of robot factors (such as end-effector movement path) on workers’ mental stress.

- Eiichi Yoshida (Tokyo University of Science, Japan)

“ Title ”

TBA

- Dongheui Lee (Technische Universität Wien (TU Wien), Austria)

“ Decision Making in Physical Human Robot Joint Actions ”

In this talk, I will present our recent research in collaborative robots in joint assembly tasks. Especially I will discuss about human’s decision making in joint actions, considering human ergonomics, overall task performances. I will present human user study results in this context.

- Sylvain Calinon (Idiap Research Institute, Switzerland)

“ Learning from Demonstration for Collaborative Tasks with Physical Contacts ”

Collaborative tasks with physical contacts require robot controllers that can swiftly adapt to the ongoing situation. Efficient representations at the crossroad of control, planning and perception are required to facilitate this challenge. Learning from demonstration (LfD) can be exploited to build these representations, either by learning the hyperparameters or by learning the higher level organization. Our ongoing work explores several facets of these challenges. First, I will show that a probabilistic interpretation of optimal control can facilitate the links between learning and optimization. First, I will show that a cost function composed of a sum of quadratic error terms can be treated either as a linear quadratic regulator (LQR) problem from an optimal control perspective, or as a product of Gaussians (PoG) from an information fusion perspective. I will then show that this dual view can be extended to non-quadratic costs and non-linear systems, which can be used to extend the concept of movement primitives to control primitives, bringing a modular approach to (re)combine controllers in parallel and in series. I will finally show that the underlying dictionary of controllers can contain: 1) ergodic control behaviors to provide explorative controllers in which the areas to explore are learned from demonstration; and 2) impedance behaviors exploiting geometry (Riemannian manifold and geometric algebra) for object affordances modeling. The two can be exploited to reduce the number of human demonstrations required and to provide better generalization capability. I will showcase the proposed information fusion principle in a wide range of applications requiring shared control, including teleoperation, haptic guidance and physical assistance.

Dr Sylvain Calinon is a Senior Research Scientist at the Idiap Research Institute and a Lecturer at the Ecole Polytechnique Fédérale de Lausanne (EPFL). He heads the Robot Learning & Interaction group at Idiap, with expertise in human-robot collaboration, robot learning from demonstration and model-based optimization. The approaches developed in his group can be applied to a wide range of applications requiring manipulation skills, with robots that are either close to us (assistive and industrial robots), parts of us (prosthetics and exoskeletons), or far away from us (shared control and teleoperation).

- Luis Figueredo ( Technical University of Munich (TUM), Germany)

“ Planning for humans: Leveraging comfort-based manipulability for improved pHRC ”

Recent advances in robotics technologies are closing the gap between humans and robots. Nonetheless, robots are still rarely thought of being physically engaging with humans, and human-like physical human-robot collaboration (pHRC) is still one of the key open challenges in robotics research. Collaboration and teamwork are better achieved when members understand each other capabilities and preferences. When it comes to pHRC, that means robots need better reasoning of human’s physical capabilities, ergonomics, and sense of embodiment. Robots need to have intrinsic knowledge of human feasible postures, quantitative metrics about ergonomics and dexterity - that give way to human reactivity and predictability when meeting manipulation challenges and uncertainties - and finally, quantitative metrics of muscular requirements to achieve given tasks. In this talk, I will present recent advances in human-based manipulability metrics, efficient human embodied structures to analyze them, methods to transfer manipulability features to robotic structures and tools to ground robot decision-making capabilities based on such human comfort-based manipulability metrics.

- João Silveri ( Deutsches Zentrum für Luft- und Raumfahrt (DLR), Germany)

“ Exploiting prior task knowledge for assistance during learning and shared control ”

To make robot skill acquisition quick and effective, a promising route is to use a combination of prior knowledge and data-driven methods. Such prior knowledge can take various forms depending on the problem at hand, including constraints (e.g. ’hold a cup upright if it is not empty’) and object-centered behaviors (e.g. ‘provide more assistance near the task goal’). Interestingly, the specification and handling thereof has been addressed independently in different fields, such as control and machine learning, raising the question of how they can benefit each other when coming together in robotic systems. In this talk I will show how shared control formulations, which implement task constraints as virtual fixtures, can be leveraged to make reinforcement learning in the real world safe and efficient. In the other direction this talk will demonstrate how concepts from machine learning, namely probabilistic models, can be adapted to suit the needs of shared control problems allowing, for instance, for an adaptive handling of virtual fixtures.