Research seminars

Abstract: In this presentation, we describe how Qualcomm is accelerating the future of extended reality (XR) with our Snapdragon XR technologies. Our best-in-class XR solutions include processors, software and perception technologies, reference designs, and developer tools that help create a new future of unlimited potential for both enterprises and consumers. In particular, we describe the XR Labs Europe activities in developing state-of-the-art perception technologies, hardware and software optimized, and delivered to the ecosystem via the Snapdragon Spaces platform. The XR perception features include head and hand tracking, image & object recognition and tracking, 3D reconstruction and scene understanding, and visual localization and mapping, all utilizing the latest computer vision and deep learning techniques.

Abstract: Electron microscopes (EM) can now provide the nanometer resolution that is needed to image synapses, and therefore connections, while Light Microscopes (LM) see at the micrometer resolution required to model the 3D structure of the dendritic network. Since both the arborescence and the connections are integral parts of the brain's wiring diagram, combining these two modalities is critically important.

In this talk, I will therefore present our approach to building the dendritic arborescence from LM images, to segmenting intra-neuronal structures from EM images, and to registering the resulting models. I will also discuss our recent work on building neural representations using functional data. 

Bio: Pascal Fua joined EPFL (Swiss Federal Institute of Technology) in 1996, where he is a Professor in the School of Computer and Communication Science and head of the Computer Vision Lab. Before that, he worked at SRI International and at INRIA Sophia-Antipolis as a Computer Scientist. His research interests include shape modeling and motion recovery from images, analysis of microscopy images, and Augmented Reality. He has (co)authored over 300 publications in refereed journals and conferences. He has received several ERC grants and has co-founded three spinoff companies. 

Abstract: For autonomous robots and automated driving, the ability to robustly perceive environments and execute their actions is the ultimate goal. The biggest challenge is that no sensors and actuators are perfect, which means that robots and cars must be able to properly deal with the resulting uncertainty. In this talk, I will introduce the probabilistic approach to robotics, which provides a rigorous statistical methodology for dealing with state estimation problems. In addition, I will discuss how this approach can be extended using state-of-the-art machine learning technologies to deal with complex and changing real-world environments.

Bio: Wolfram Burgard is a professor for computer science at the University of Nuremberg and head of the research lab for Autonomous Intelligent Systems and former head of the research lab for Autonomous Mobile Systems at the University of Freiburg. Over the past years he and his group have developed a series of innovative probabilistic techniques for robot navigation and control. They cover different aspects such as localization, map-building, SLAM, path-planning, exploration, and several other aspects. He has published over 350 papers and articles in robotics and artificial intelligence conferences and journals and several books. He has also obtained numerous awards for his research contributions. In 2008, he became a fellow of the European Coordinating Committee for Artificial Intelligence and he obtained the 2009 Gottfried Wilhelm Leibniz Prize, the most prestigious German research prize.

Abstract: This presentation deals with two relevant topics in aerial robotics that are converging: Aerial Manipulation and Bioinspired Aerial Robots. I will present recent results in aerial robotics manipulation, mainly obtained in the H2020 AERIAL-CORE project that I am coordinating. They will include not only aerial manipulation while flying, but also while holding with one arm and manipulating with other and also while perching. I will present recent results in my ERC Advanced Grant GRIFFIN dealing with bioinspired (flapping-wing) aerial robots. I will present new more efficient prototypes and the first fully autonomous indoor and outdoor flapping wing robots with on-board perception for guidance and obstacle detection and avoidance. I will also present the fist fully autonomous perching of these flapping-wing robots. Finally, I will point to the future, presenting preliminary results about manipulation with flapping-wing robots.

Bio: Anibal Ollero (Fellow, IEEE) is currently a Full Professor and the Head of the GRVC Robotics Laboratory, University of Seville, and the Scientific Advisor of the Center for Aerospace Technologies (CATEC), Seville. Since November 2018, he has been running the GRIFFIN ERC-Advanced Grant with the objective of developing a new generation of aerial robots that will be able to glide, flap the wings, perch, and manipulate by maintaining the equilibrium. He has transferred technology to 20 companies and has been awarded with 23 international research and innovation awards. He has also been elected one of the three European innovators of the year and IEEE Fellow "for contributions to the development and deployment of aerial robots."

Abstract: The appearance of an object is more than shaded visualization of its 3D shape. The appearance of every macroscopic surface position we observe is a result of light scattering in the microgeometry of the object. This presentation is on the influence of the microgeometry on the scattered light reaching an observer. We can feasibly solve Maxwell’s equations in a small microgeometry of say three microns cubed, but full digital representation of the microgeometry and calculation of light scattering for an object measured in centimeters is infeasible. There is currently no standard for full digital representation of the appearance of a physical object. In addition, there is a significant lack of validation of the actual photorealism of computer graphics rendering techniques. With a focus on the role of the microgeometry, I will in this presentation discuss the challenges in specification and predictive simulation of the appearance of physical objects.

Bio: Jeppe Revall Frisvad is an associate professor at the Technical University of Denmark (DTU). He has more than 15 years of experience in material appearance modeling and rendering and has, in a handful of research projects, served as leader of work packages focusing on analysis and synthesis of product appearance. As a highlight, his work includes the first directional dipole model for subsurface scattering, and his research on material appearance includes methods for both computation and photographic measurement of the optical properties of materials.

Abstract: Faithfully reproducing real-world objects on a 3D printer is a challenging endeavor. A large number of available materials and freedom in material deposition make efficient exploration of the printing space difficult. Furthermore, current 3D printers can perfectly capture only a small amount of objects from the real world which makes high-quality reproductions challenging. Interestingly, many of the applications for 3D printing are explored by humans either using our sense of touch, sight, or hearing. These senses have inborn limitations given by biological constraints: eyes have limited capability to distinguish high-frequency information; fingers feel applied forces in a non-linear fashion.

In this talk, I will introduce you to the concept of perception-aware fabrication that aims to exploit these limitations to create more efficient fabrication techniques which offer equal or higher quality as perceived by a human observer. A core element of perception-aware fabrication is a perceptual space which models the response of the human sensory system to external stimuli. I will show you how to derive such spaces and how to apply them in the context of computational fabrication. Finally, I will show you how we can leverage perceptual insights to design more efficient numerical simulations. I will demonstrate this general concept in the context of two applications: manufacturing objects with prescribed compliance properties, and designing customized digital styli that mimic the behavior of traditional drawing tools. Last but not least, I will present a technique for an efficient design of surfaces with prescribed reflectance behavior.

Bio: Michal Piovarci is a postdoc at Institute of Science and Technology Austria in the Computer Graphcis and Digital Fabrication group led by Bernd Bickel. He obtained his Ph.D. at USI Lugano under the supervision of Piotr Didyk in 2020. His thesis entitled Perception-Aware Computational Fabrication recieved the prestegious Eurographics PhD Award. Michals research interests are computer graphics, computational fabrication, haptic reproduction, appearance reproduction, and perception. 

Abstract: Many manufacturing tasks will benefit from Humans and cobots helping each other. For these collaborations to be effective and safe, robots need to model, predict and include human's intents in their decision making processes. In this talk, we'll show how to formulate human reaching intent prediction as an approximate bayesian computation problem using our sampling and ABC libraries. We will describe two key performance innovations to allow interactive rate predictions and show the importance of intent prediction through experimental results of human-robot collaboration on a shared task. Finally we will show our ongoing work on applications of human-robot collaboration for memory testing tasks and discuss some of our current research lines.

Bio: Dr. Javier Felip Leon is a Research Scientist at Intel Labs since 2016. Javier completed his PhD on Computer Science at Universitat Jaume I (Spain) in early 2016. His research focused on Robotic manipulation under uncertainty (some robot videos here: https://sites.google.com/site/javierfelipphdthesis/videos). Currently his research focuses on human-robot collaboration, analysis-by-synthesis, applied probabilistic methods and sampling methods. Javier has authored more than 20 peer-reviewed paper publications in top-tier conferences and journals such as ICRA, IROS, HUMANOIDS and RAS and more than 20 granted patents.

Abstract: We move towards an era of smart cities, where autonomous vehicles will provide on-demand transportation while making our streets safer and mobile robots will coexist with humans. The motion plan of mobile robots and autonomous vehicles must, therefore, account for the interaction with other agents and consider that they are, as well, decision-making entities. For example, when humans drive a car, they are fully aware of their environment and how other drivers and pedestrians may react to their future actions.

Towards this objective, I will discuss several methods for motion planning and multi-robot coordination that a) account for the inherent uncertainty of dynamic environments and b) leverage constrained optimization, game theory and reinforcement learning to achieve interactive behaviors with safety guarantees. The methods are of broad applicability, including autonomous vehicles and aerial vehicles.

Bio: Dr. Javier Alonso-Mora is an Associate Professor at the Cognitive Robotics department of the Delft University of Technology, where he leads the Autonomous Multi-robots Laboratory. He was a Postdoctoral Associate at the Computer Science and Artificial Intelligence Lab (CSAIL) of the Massachusetts Institute of Technology (MIT) and he received his Ph.D. degree in robotics from ETH Zurich, in the Autonomous Systems Lab in partnership with Disney Research Zurich. His main research interest is in navigation, motion planning and control of autonomous mobile robots, with a special emphasis on multi-robot systems, on-demand transportation and robots that interact with other robots and humans in dynamic and uncertain environments. He is the recipient of multiple prizes and grants, including an ERC Starting Grant (2021).

Abstract: Do you imagine visualizing from your mobile what is inside your fridge and automatically add to your shopping list what is missing? In BSH, we are developing the home appliances of the future using machine learning techniques.

This process involves several complex and different steps, which are fundamental for the success of the project, and go beyond just implementing an ML algorithm.

In this talk, we will explain the lifecycle of a machine-learning project in production, we will deepen into all different stages and roles you need to succeed.

Bio: Alejandro Rituerto is Industrial Engineer and got his PhD in Computer Vision and Robotics at the University of Zaragoza. He has more than 10 years of experience in applying Computer Vision in many different fields, like human assistance navigation, Augmented Reality or defect detection in production processes. Now he is part of the AI team at BSH that works in developing smart functionalities for home appliances based on AI.

Héctor Martínez holds PhD in Augmented Reality and has 15 years of experience in research and production projects. Since he joined BSH Home Appliances, he has worked on several projects launched to the market, including cutting edge AI projects.

Abstract: Accurately rendering virtual humans is a major endeavor, involving ever-evolving appearance capture systems, coupled with high-quality material models, sophisticated offline rendering techniques, and a significant portion of manual intervention. This, despite over 20 years of breakthroughs in rendering realistic humans and creatures in VFX, animation and videogames, closing the gap between real and digital footages, significant challenges still exists, specialty when targeting VFX-quality for personalized humans in tight-constrained real-time applications. In this talk, we will discuss the existing state of the art and current challenges, and the steps my team is doing towards high-quality virtual humans rendered in real time.

Bio: Adrián Jarabo is a Research Science at Meta Reality Labs Research, where he works in topics such as off-line and real-time rendering, a light transport modeling, and appearance modeling and capture, with focus on digital humans. Before that, he was an Assistant Professor at Universidad de Zaragoza, where he got his PhD in 2015. He is a Eurographics Junior Fellow, and the recipient of the EG PhD Award 2017 and the EG Young Researcher Award 2021.

Bio: Valentin is a Research Scientist at Adobe Research in the London lab, working on virtual material creation and editing. He previously worked in the Realistic Graphics and Imaging group of Imperial College London hosted by Abhijeet Ghosh. He obtained his PhD from Inria, in the GraphDeco research group, under the supervision of Adrien Bousseau and George Drettakis. During his PhD he spent 2 months under the supervision of Frédo Durand, at MIT CSAIL. His Thesis received the French Computer Graphics Thesis award 2020 and UCA Academic Excellence Thesis Award 2020. His research cover material and shape (appearance) acquisition, creation, editing and representation, leveraging deep learning methods.

Abstract: In the last 20 years, virtual 3D environments have become an integral parts of many industries. Of course, the video game and cinema industries are the most prominent examples of this, but cultural heritage, architecture, design, medicine, all benefit from virtual environment rendering technologies in different ways. These renderings are computed by simulating light propagation in a scene with various geometries and materials.  In this talk, I will focus on how materials define the light behaviour at the intersection with a surface, and most importantly how they are created nowadays. I will then discuss recent research and progress on material authoring and representation, before concluding with open challenges to further help artists reach their goals faster.

Abstract: This talk targets people identification in video based on the way they walk (i.e. gait). While classical methods typically derive gait signatures from sequences of binary silhouettes, in this talk we present the use of deep neural networks for learning high-level descriptors from low-level motion features (i.e. optical flow components) and body pose. First of all, I will overview how the non-silhouette-based gait recognition has evolved in recent years within our group and collaborators. Then, I will focus on recent approaches proposed for addressing different challenges wrt the gait recognition problem, as transfer learning and multimodal approaches with missing modalities, including the use of body pose for gait recognition.

Bio: Manuel J. Marín Jiménez, received the BSc, MSc and PhD degrees from the University of Granada, Spain. He has worked, as a visiting student, at the Computer Vision Center of Barcelona (Spain), Vislab-ISR/IST of Lisboa (Portugal) and the Visual Geometry Group of Oxford (UK); and, as visiting researcher, at INRIA-Grenoble (Perception and THOTH teams) and the Human Sensing lab (CMU).  He is the coauthor of more than 70 technical papers at international venues and serves as a reviewer of top computer vision and pattern recognition journals. Currently, he works as Associate Professor (tenure position) at the University of Cordoba (Spain).  In addition, he is member of the Governing Board of AERFAI (Spanish Association of Pattern Recognition and Image Analysis). His research interests include object detection, human-centric video understanding (including biometrics), visual SLAM and machine learning.

Abstract: High-accuracy per-pixel depth is vital for computational photography, so smartphones now have multimodal camera systems with time-of-flight (ToF) depth sensors and multiple color cameras. In this talk, I provide the principles of 3D imaging applicable to mobile phone cameras, including stereo imaging, structured light, and time-of-flight imaging. However, producing accurate high-resolution depth is still challenging due to the low resolution and limited active illumination power of ToF sensors. Fusing RGB stereo and ToF information is a promising direction to overcome these issues, but a key problem remains: to provide high-quality 2D RGB images, the main color sensor's lens is optically stabilized, resulting in an unknown pose for the floating lens that breaks the geometric relationships between the multimodal image sensors. Leveraging ToF depth estimates and a wide-angle RGB camera, we design an automatic calibration technique based on dense 2D/3D matching that can estimate camera extrinsic, intrinsic, and distortion parameters of a stabilized main RGB sensor from a single snapshot. This lets us fuse stereo and ToF cues via a correlation volume. For fusion, we apply deep learning via a real-world training dataset with depth supervision estimated by a neural reconstruction method. For evaluation, we acquire a test dataset using a commercial high-power depth camera and show that our approach achieves higher accuracy than existing baselines.

Bio: Min H. Kim is the KAIST ICT-Endowed Chair Professor of Computer Science at the KAIST School of Computing, where he directs the Visual Computing Lab (VCLAB). Prior to joining KAIST, he was a postdoctoral researcher at Yale University. He holds a Ph.D. in Computer Science from University College London (UCL). He has received numerous awards, including the SIGGRAPH Technical Paper Award. His main research areas are computational imaging, computational photography, 3D imaging, BRDF acquisition, and 3D reconstruction. He has served as Technical Paper Chair for Eurographics 2022, Course Chair for SIGGRAPH Asia 2022, and on many computer graphics and computer vision program committees, including SIGGRAPH, CVPR, ICCV, and Eurographics. He has served as an Associate Editor for top CS journals, including ACM Transactions on Graphics (TOG) and IEEE Transactions on Visualization and Computer Graphics (TVCG). 

Bio: Prof. Samuel Morillas has a BSc in Computer Science from Unvieridad de Granada and a PhD in Applied Mathematics from Universitat Politècnica de València (UPV) in Spain. From 2019 he is a full professor of the Applied Math department at UPV, where he is a member of the Instituto of Pure and Applied Math.

His undergraduate teaching includes fundamental algebra and calculus for first-year students and Continuous modeling for the 3rd year students of the Data Science degree. His graduate teaching at UPV concerns fuzzy logic and applications. He has also been an invited professor in a series of Universities: University of Granada (Spain), Rochester Institute of Technology (NY, USA), Chonbuk National University (South Korea), University at Buffalo (NY, USA), Colorado University at Boulder (Co, USA), Harvard University (Massachusetts, USA), University of Edinburgh (UK) and Metropolia University of Applied Sciences (Finland), where he has taught different courses and seminars about fuzzy logic, image processing, and image appearance models.

His research has been focused on fuzzy logic and fuzzy metrics both in the theoretical field and, mostly, in their applications. He has developed different solutions based on fuzzy logic and fuzzy metrics for problems in different fields including image processing, color science, and pavement engineering. He has published more than 50 papers in scientific journals and presented more than 70 contributions to scientific conferences.

Abstract: The use of robots is becoming increasingly common in various fields, and the way we communicate with them is an important aspect to consider. Gestures can be a natural and intuitive way to communicate with robots, as they allow for nonverbal communication that can convey meaning quickly and effectively. In recent years, there has been significant research in the field of human-robot interaction (HRI) focused on developing gesture-based interfaces for communicating with robots. These interfaces aim to improve the efficiency, ease of use, and effectiveness of robot communication. This talk will explore¡ the current state of research on gesture-based communication with robots, highlighting the advantages and challenges of using gestures in HRI, and outlining the potential future directions for this field. We will present the first steps of our research, which aims to develop a gesture-based interface for communicating with a humanoid robot. We are  exploring the use of machine learning algorithms to train the robot to recognize and respond to these gestures. This research has the potential to significantly improve the usability and user experience of humanoid robots in a variety of applications, from healthcare and education to entertainment and personal assistance 

Bio: Anaís Garrell Zulueta is a Lecturer Professor at the Polytechnic University of Catalonia (UPC) and a Robotics Researcher at the Robotics Institute (CSIC-UPC, Barcelona). In 2013, she earned her European doctorate and was awarded the second prize for the best Spanish robotics thesis. During her PhD, she spent a year at the Institute of Robotics at Carnegie Mellon University, USA. She was a finalist for the best paper award at the 2009 IROS conference. Recently, she was the first runner-up of the 2022 CEA Female Talent Award in Automation for her outstanding international scientific career. Her research interests include robot cooperation, Human-Robot Interaction, People Motion Behavior and Robots Companion.