Research seminars

Abstract: In this talk, I will overview our work at Google Geo to get global maps available to a wide range of users on their phones, through the web, or through the Google Earth Engine APIs. I will discuss the high level journey from raw imagery to photogrammetric reconstruction. I will discuss algorithmic challenges involved in quantifying, measuring, and achieving high accuracy modeling of the surface of the Earth.

I will provide a brief introduction to frames of references, International Terrestrial Reference Frame (ITRF), measurement epoches, earth ellipsoid and geoid, tectonic shifts, earthquakes, subsidence (like CA central valley), and earth tides. And how each of these pose a challenge in estimating a fresh topographic map of the Earth’s surface.

Bio: Faraz Mirzaei, received in M.Sc. and Ph.D. in Computer Science at the University of Minnesota (USA). Previously, he obtained his degree at Shiraz University (Iran). From 2012 to 2015, he worked at Qualcomm Research Silicon Valley, and he is currently the Technical Lead for Photogrammetric Quality Metrics at Google, Mountain View. His interests include photogrammetry, mapping, map making, surveying, inertial navigation,  computer vision, and perspective geometry.

Abstract: For a researcher, publishing his/her results is one of the most important activities. The goal of these seminars is to get a deeper understanding of the academic publishing process. This first talk will address the following topics:

1. Why publish?

2. When publish?

3. Where publish? Journal and conference rankings. Impact Factor.

4. How are researchers evaluated? Quantity, Quality, Impact.

5. Useful tools: ISI web, Google Scholar, PoP, SCImago,...

This presentation will include practical examples of how to use the available tools to solve common questions such as how to find the most relevant journals and conferences, influential papers and "hot topics" in a research area, or how to find quality indicators of our publications and how to report them in a CV or an accreditation application.

Bio: Juan D. Tardós is professor on Systems Engineering and Automatic Control at the University of Zaragoza. His research area is perception and environment understanding in robotics. He is co-author of one book and more than 60 journal and conference papers on these topics. He has served for several conferences and journals, reviewing more than 200 papers. He has handled +80 papers as Associate Editor of the IEEE Transactions on Robotics, IROS and RSS, obtaining reviews and writing recommendations for their publication or rejection. This presentation reflects his own experience and opinions.

Abstract: Point clouds are one of the simplest data representations of 3D data. They are also particularly convenient because many real-world data acquisition techniques often result in point clouds, e.g., LiDAR scanning or photogrammetric applied to imagery. However, high-resolution and large-scale scans lead to several billion or even trillion points. Managing such massive data volumes is not only challenging for real-time tasks but also for out-of-core computations. I will briefly talk about the work I conducted during my PhD.  First, we will go through the rendering of huge point clouds using alternative rendering pipelines. Traditional pipelines designed for triangle meshes are often inefficient for point clouds, while GPGPU shaders can help us build custom pipelines. However, another problem arises from the memory limitations in the GPU, which ultimately leads to using Level of Detail systems. Building on the concepts applied in rendering, we will explore approaches to improve point cloud reconstruction. Although photogrammetry is the standard technique for reconstructing point clouds, it struggles with other nonfrequent data, such as thermal and multispectral imagery. By incorporating alternative pipelines that partially rely on photogrammetry, we can enhance both the efficiency and accuracy of point cloud reconstructions. Finally, I would like to talk about my work on LiDAR simulation to generate synthetic datasets that complement real-world data.

Bio: Alfonso López-Ruiz holds a PhD in Computer Science from the University of Jaén in 2023. He received a BSc and MSc in Computer Science from the same institution in 2019 and 2020. Recently, he was granted a Juan de la Cierva Postdoctoral Fellowship to conduct his research at the University of Zaragoza. His interests span GPGPU, parallel computing, real-time rendering, photogrammetry, image processing and geometric algorithms.

Abstract: Environmental perception plays a crucial role in the continuous improvement of Advanced Driver Assistance Systems (ADAS) and the development of autonomous driving technologies. In this talk, we will explore how Renault Group integrates innovative solutions in these two technological pillars to foster smarter and safer mobility.

We will delve into the distinct challenges and technological requirements for each application, highlighting how sensor selection, algorithm design, and processing unit choices vary depending on the specific objectives and operational contexts of the system. From the perception systems essential for autonomous vehicle decision-making to the validation methodologies ensuring the safety and robustness of ADAS in conventional vehicles, this session will provide an in-depth view of the cutting-edge technologies shaping the future of automotive perception.

Bio: Irene Cortés has a Degree in Electronic and Automation Engineering from Universidad de Zaragoza and worked for six years at the Carlos III University of Madrid. During her time there, she completed her Master's in Robotics and Automation and earned her Ph.D. in 2024 titled "Advanced Techniques for Autonomous Vehicle Perception on Real Platforms." She is the author of several scientific articles in journals and conferences, as well as industrial patents related to autonomous and connected vehicles. Since 2022, Irene has been working at Renault Group’s Software and Technology department, where she leads projects focused on environmental perception and data analysis for connected vehicles.

Abstract: Multi-robot systems emerge as a promising solution for tackling complex tasks that are beyond the capabilities of a single robot. Their inherent parallelism, robustness to individual failures, and ability to operate in large-scale environments make them particularly appealing for applications such as search and rescue, environmental monitoring, herding and agriculture, or warehouse automation. However, the coordination and control of multiple robots operating in a distributed infrastructure pose significant challenges. Specifically, this thesis aims to solve four key aspects of distributed multi-robot systems: effectiveness of coordination in highly nonlinear, volatile environments; fast and accurate reconstruction of collective information; scalability of the control policies in the number of robots; and power supply management. To address the four points, the thesis exploits three main tools: physical properties of networked systems, geometrical control techniques and distributed optimization and machine learning methods.

Bio: Eduardo is a PhD candidate in the the Robotics, Computer Vision and Artificial Intelligence group of the Universidad de Zaragoza, co-supervised by Eduardo Montijano and Carlos Sagüés. Eduardo research interests span topics related to multi-robot decision-making, networked system, control and learning (and, occasionally, power electronics). He obtained his Bachelor in Electronics and Automation (2019) and Master in Electronics (2020) at the Universidad de Zaragoza. He has been a visiting scholar at the Existential Robotics Laboratory at UCSD in 2022 and 2024. Currently, he holds a researcher position at the ProrokLab, at the University of Cambridge. Eduardo is a Fulbright Scholar and a DAAD AInet Fellow.

Bio:     Dr. Gonzalo Esteban-Costales is an Assistant Professor at the University of Zaragoza since 2024. He received his BSc in Computer Science in 2009, his MSc in Cybernetics in 2011, and his PhD in Production Engineering and Computation in 2020, all from University of León, Spain. His research interests include the application of haptic simulators for training and education, as well as cybersecurity. 

Abstract:     In the early stages of any research, it's common to feel overwhelmed by the sheer volume of information and unsure about how to identify the most relevant studies. Many researchers rely on basic methods such as searching through references or conducting simple database queries, but these approaches can be time-consuming, inefficient, and prone to bias. This talk will introduce Systematic Literature Reviews, a methodology originally developed in the health sciences, that provides a more structured and rigorous approach to identifying, evaluating, and synthesizing a wide range of relevant studies, data, and other forms of evidence. By following a clear, objective approach, systematic reviews ensure a more comprehensive and unbiased perspective, not only in health sciences but in any discipline. The session will highlight how adopting this methodology can improve the quality of research, streamline the search for relevant evidence, and offer a clearer path through the vast amount of available information.

Bio:     Adrián Jarabo is a research scientist at Meta Reality Labs Research, where he leads a small team focusing on real-time and inverse rendering of virtual humans. Before joining Meta in 2022, he was an Assistant Professor at Universidad de Zaragoza. He has published over 50 papers in topics including physics-based rendering, appearance modeling and computational imaging. He is a Eurographics Junior Fellow, and the recipient of the Eurographics Young Researcher Award and the Eurographics PhD Award in 2021 and 2017, respectively. 

Abstract:    In this talk we will describe our recent efforts on developing general, ray-traced volumetric representations for modeling complex volumetric materials in forward and inverse rendering applications, with applications to physics-based rendering and scene reconstruction. Volumetric heterogeneous materials are widespread in real-life in general, and humans in particular, with notable examples such as skin, cloth or hair. However, finding a good representation for such materials is still an open problem. During the talk, we will walk through the development of our ray-traced mixture-based approach for volumetric representations that provides a number of benefits with respect to the widely used voxel grids, discuss our choices, and show how our project grew from a small physics-based hair rendering project to a wider general framework for rendering and scene reconstruction.