Between December 3-5, 2025, the European Space Agency (ESA) hosted a AI STAR Symposium - Artificial Intelligence Symposium on Theory, Application, and Research at European Space Operations Centre (ESOC), Darmstadt.
The conference was a great place to share recent space-oriented AI solutions. Bartosz presented a poster, entitled “Sparse Supervision for Traversability Estimation in Lunar Robotics”.
]]>
Between 25–30 October 2025 we co-organized the EASi workshop (Explainable AI in Space) and took part in the ECAI 2025 conference in Bologna. ECAI is the 28th European Conference on Artificial Intelligence, bringing together researchers, practitioners and innovators in AI through keynotes, paper sessions, workshops and tutorials.
The EASi workshop took place as part of ECAI, under the theme HYPERVIEW2. It was organised by Poznan University of Technology, KP Labs, the ESA Φ-Lab and Warsaw University of Technology. The workshop combines a challenge (HYPERVIEW2) and a call for papers, allowing participants to either compete in the challenge, publish a paper, or do both. The HYPERVIEW2 Challenge focused on advancing explainable AI methods for Earth observation through the analysis of multi and hyper hyperspectral satellite and airborne data. This second edition of the challenge encouraged participants to design models capable not only of achieving high accuracy in data interpretation but also of providing transparent and interpretable reasoning behind their outputs. More details can be found at workshop webpage - https://ai4eo.eu/portfolio/easi-workshop-hyperview2/
Explainable AI in space matters because AI is increasingly used for mission-critical tasks such as satellite operations, Earth observation, and environmental monitoring. Without transparency or interpretability, decisions made by AI systems might lack accountability or trustworthiness — especially in high-stakes or safety-critical scenarios.By promoting explainability, the workshop helps ensure that AI can be audited, understood by human operators, and aligned with ethical and safety requirements in space-related contexts.
]]>
Drone-based crowd monitoring is a key technology for applications in surveillance, public safety, and event management, primarily due to its dynamic, aerial perspective that surpasses the limitations of traditional ground-based systems. Recently, a new trend has emerged in tiny object localization and tracking, characterized by the use of point-oriented object sensing, which enables accurate monitoring of densely packed individuals in low-altitude aerial imagery. In this dissertation, advancements in this area are presented, including novel approaches for point-oriented object localization and a new solution for point-oriented object tracking. For localization and counting tasks, a series of enhancement mechanisms is introduced. These include the integration of motion-based features, the use of task-oriented synthetic data, and addressing the influence of varying image input resolutions in neural networks. A direct incorporation of drone altitude into the neural network architecture is also investigated, a new module that processes all pixels of high-resolution images without downscaling is proposed, and a novel loss function tailored to point-oriented localization is introduced. For object tracking and trajectory counting, an algorithm is proposed that enhances trajectory continuity and unique counting reliability in drone-based crowd monitoring, enabling the accurate tracking of individuals across video sequences. The approach extends the Simple Online and Real-time Tracking (SORT) framework by replacing the bounding-box assignment with a point-distance metric. It is further enhanced with three cost-effective techniques: camera motion compensation, altitude-aware assignment, and classification-based trajectory validation. Additionally, Deep Discriminative Correlation Filters (DDCF) are integrated, which reuse spatial feature maps from localization algorithms to improve computational efficiency and handle missed detections. To support this research, two new datasets, UP-COUNT and UP-COUNT-TRACK, are introduced, addressing challenges in modern drone imagery, including simultaneous camera and object motion, as well as changing flight altitudes. All proposed methods are quantitatively evaluated on both the publicly available DroneCrowd dataset and new datasets, demonstrating significant improvements in localization and tracking performance and achieving state-of-the-art results in drone-based people and trajectory counting. This dissertation makes substantial contributions to computer vision in aerial robotics, offering practical tools for rapid crowd size and movement estimation. These tools have been demonstrated to be applicable in real-world scenarios.
Between October 7-9, 2025, the European Space Agency (ESA) hosted the ASTRA 2025 symposium at Scheltema, Leiden (The Netherlands). The event brought together participants from academia, industry, and government agencies to discuss the latest advancements in AI and robotics for space exploration. The symposium featured keynote speeches, technical sessions, and panel discussions on a wide range of topics, including in-orbit servicing, planetary exploration, and autonomous systems.
During the conference, representatives from ESA and NASA discussed their approaches to AI and robotics in space exploration. Many speakers focused on the previous missions, such as the Mars rovers, how they developed their systems, how they maintained them, and what lessons they learned. The discussions also put a strong emphasis on the future direction of space exploration, including current and upcoming missions, such as the Sample Return mission and ESA Rosalind Franklin rover.
One of the highlights of the symposium was the special session dedicated to lunar rovers, which featured presentations from leading companies in the field, including Airbus, Space Applications Services, Thales Alenia Space Italia, Ispace Europe, and GMV. The session provided insights into the latest developments in lunar rover technology, including advancements in mobility, autonomy, and scientific instrumentation.
The networking was connected with the poster session, which showcased the latest research and development in AI and robotics for space exploration. The posters covered a wide range of topics, including machine learning, computer vision, and autonomous systems.
One of the posters was presented by our team member, Bartosz Ptak, who showcased his work on “Bird’s-Eye View Terrain Understanding for Lunar Robotics in Simulated Environments”. The poster received positive feedback from attendees, and Bartosz had the opportunity to discuss his research with other experts in the field.
The final day of the symposium included a tour of the European Space Research and Technology Centre (ESTEC) in Noordwijk, The Netherlands. ESTEC is the largest space research and technology center in Europe and serves as the technical heart of ESA. During the tour, attendees had the opportunity to see three of the center’s key facilities for robotics and AI research.
The 22nd Advanced Concepts for Intelligent Vision Systems (ACIVS) Conference was held in Tokyo, Japan, from July 28-30 at the University of Electro-Communications (UEC) Chōfu campus. This event, organized by the ACIVS Program Committee, focused on cutting-edge techniques for creating adaptive, intelligent, safe, and secure imaging systems. Attendees had a unique opportunity to present their research, exchange ideas, and network with peers. Beyond the keynotes, oral presentations, and poster sessions, the conference also offered additional activities, including visits to a temple, UEC research labs, and the UEC Telecommunication Museum. To learn more about the ACIVS conference series, visit their website acivs.org, whereas, for more information about this year’s edition select ACIVS 2025.
During the Computer Vision and Machine Learning session, Mateusz represented the PUT Vision laboratory, presenting his research on solar irradiance forecasting and adaptation to new sites and incoming data. His presentation, titled “On-Device Continual Adaptation for Reliable Solar Irradiance Forecasting”, was a collaborative effort with Marek Kraft and Alessandro Capotondi (from the University of Modena and Reggio Emilia). Mateusz highlighted how the inherent uncertainty of solar energy, mainly due to unpredictable weather, impedes its large-scale integration into power grids. He then introduced a continual adaptation strategy based on an incremental learning method that significantly improves forecasting accuracy to address this challenge, also bearing in mind edge computing limitations.
This work was supported by the ‘PhDBoost’ Program for doctoral students of the Doctoral School of Poznan University of Technology (in 2024) from the University’s subsidy financed from the funds of Ministry of Science and Higher Education.
In the same session, Dominik Belter, director of the Institute of Robotics and Machine Intelligence, presented his work. Entitled “Pointy – A Lightweight Transformer for Point Cloud Foundation Models”, the presentation focused on foundation models for point cloud data and was a collaborative effort of Konrad Szafer, Marek Kraft, and Dominik Belter.
]]>
In the mid of July 2025, 13-19, in Fiuggi, Italy, was the 21st International Summer School on Advanced Computer Architecture and Compilation for High-performance Embedded Systems (ACACES) organized by the High Performance, Edge And Cloud computing (HiPEAC) Network. Our PUT Vision Lab was represented by Mateusz, who was one of the 200 attendees from 100 institutions across 23 countries. Based on the proposal and registration form, he was again selected and awarded a grant covering the participation costs.
Within the week, Mateusz attended the following sessions:
On Wednesday, Marek presented our recent advancements in the Deepness plugin and its applications to a broad audience of approximately 450 QGIS experts. His presentation was followed by a networking session that generated many new and promising ideas for future model and plugin development. On Thursday, Przemysław and Marek served as experts on a discussion panel titled “AI in GIS: Will Artificial Intelligence Change the Industry?”
We are delighted to report that our Deepness plugin has gained significant traction since its presentation at last year’s conference! It has successfully encouraged GIS experts to integrate machine learning techniques with their expertise, exemplified by a presentation on the identification of asbestos roofs on aerial photographs using Deepness. Furthermore, one of the workshops at the conference specifically covered the usage of the Deepness plugin.
Deepness - an open-source plugin for the QGIS application, allowing the easy employment of neural network models on any raster layer representing a matrix of values or image data. Deep neural networks show a clear improvement in computer vision tasks, enabling the automatic performance of, among others, regression, segmentation and detection of objects in the images. The Deepness plugin supports model types that complete the abovementioned tasks, linking deep learning inference directly with the most popular geographic information system (GIS) application. Moreover, a model registry with ready-to-use models is provided, bringing the power of deep learning to users without machine learning expertise. This enables augmenting the familiar, established workflow with new functionalities.
Photo: Stowarzyszenie QGIS Polska
]]>
Between June 25th and 26th, 2025, we had the pleasure of attending the I Doctoral Workshop on Space Robotics in Málaga, Spain. The event was organised by the University of Málaga and the European Space Agency Academy. The workshop was a unique opportunity for PhD students and early-career researchers to present their work, exchange ideas, and network with peers in the field of space robotics. Heads of european space robotics laboratories were also present, providing valuable insights and feedback on the participants’ research.
We had the oportunity to represent our laboratory and present our research:
Dominik presented “Computer Vision Laboratory” - a talk about our research in the field of computer vision, including the latest advancements and applications in space robotics.
Bartosz’s presentation was titled “Space Robotics Meets Traversability Map Estimation With Bird’s Eye View Representation”, highlighting the potential of this approach.
Bartosz Ptak was awarded the Best Presentation Award for his talk on “Space Robotics Meets Traversability Map Estimation With Bird’s Eye View Representation”. It is a great honor to receive this recognition, and we are grateful for the opportunity to share our research with the community.
]]>
Tracking marine mammals is essential for gaining insights into their health, behaviour, and population dynamics. With the growing use of drones in ecological research, the demand for efficient, automated video analysis has increased. This paper introduces a novel tracking algorithm designed explicitly for online marine mammals tracking in UAV-recorded videos, addressing unique challenges such as complex seabed textures, water reflections, and irregular animal movements. By integrating particle filters with the widely used Simple Online and Real-time Tracking (SORT) algorithm, we tackle issues of missed detections in marine contexts, enhancing tracking robustness. We also present a new publicly available dataset for drone-based porpoise tracking and evaluate our SORT-PF against state-of-the-art tracking methods. The results demonstrate significant improvements, including a 9.3% increase in the HOTA metric, simultaneously keeping the lowest ID switches compared to baseline methods, highlighting the potential of the proposed SORT-PF for automated wildlife monitoring and other challenging tracking applications.
Check the dataset webpage at: https://putvision.github.io/UAVPorpoises/.
Tracking marine mammals is essential for gaining insights into their health, behaviour, and population dynamics. With the growing use of drones in ecological research, the demand for efficient, automated video analysis has increased. This paper introduces a novel tracking algorithm designed explicitly for online marine mammals tracking in UAV-recorded videos, addressing unique challenges such as complex seabed textures, water reflections, and irregular animal movements. By integrating particle filters with the widely used Simple Online and Real-time Tracking (SORT) algorithm, we tackle issues of missed detections in marine contexts, enhancing tracking robustness. We also present a new publicly available dataset for drone-based porpoise tracking and evaluate our SORT-PF against state-of-the-art tracking methods. The results demonstrate significant improvements, including a 9.3% increase in the HOTA metric, simultaneously keeping the lowest ID switches compared to baseline methods, highlighting the potential of the proposed SORT-PF for automated wildlife monitoring and other challenging tracking applications.
Link to the publication: https://doi.org/10.1016/j.neucom.2025.130503
This project has been supported by the Polish National Agency for Academic Exchange (NAWA) under the STER programme, Towards Internationalisation of Poznan University of Technology Doctoral School (2022–2024).
]]>