As we look back on 2025, we are pleased to highlight a year marked by strong scientific progress, deepened collaboration, and increasing convergence across the ASAP consortium. Building on the foundations laid in earlier phases, the project has advanced from concept development toward integrated solutions, with each Work Package (WP) contributing essential expertise. From coordinated in-person workshops and assemblies to refined algorithms, improved simulation environments, and continued progress toward on-board intelligence and FPGA-based acceleration, 2025 has been a year of consolidation and forward momentum.
In this year’s ASAP annual newsletter, we share key highlights from across the Work Packages, illustrating the collective effort, technical achievements, and shared vision that continue to drive the project toward its long-term objectives.
Work Package 1 (WP1) led the consortium through a pivotal year of face-to-face collaboration and scientific convergence. In 2025, Work Package 1 orchestrated three major events: the Smart Orbits Workshop, the second General Assembly, and a topical panel at European Space Weather.
The 2025 Smart Orbits Workshop
To address telemetry bottlenecks for future missions, WP1 organized the 2025 Smart Orbits workshop, a convening point for physicists, engineers, and data scientists. During sessions curated by the WP1 team, speakers such as Enrico Camporeale (Queen Mary University of London) and Vicki Toy-Edens (Johns Hopkins University) explored the design of "grey-box" models and the need to define algorithms that meet strict spacecraft power constraints. The workshop also bridged the gap between software and hardware; Pedro Antunes demonstrated that neural inference is feasible on FPGAs within a few Watts of power envelope. The workshop presentations can be found on Youtube, and a summary of the key points are available here: Spart Orbits.
The General Assembly in Leuven
On October 6th, 2025, WP1 facilitated the project's most significant management milestone to date: the second ASAP General Assembly (GA), hosted by KU Leuven in Belgium, bringing together 16 participants: 12 on-site and the remainder joining virtually, marking the consortium's first major face-to-face gathering. The meeting solidified the roadmap for the final phase of the project. The event involved the participation of the External Expert Industrial Advisory Board (EEAB). Three distinguished experts joined the proceedings: Jorge Amaya (Space Weather Modelling Coordinator, ESA), Stefan Strålsjö (Head of Engineering Sweden), and Daniele Gregori (Chief Scientific Officer, E4 Computer Engineering SpA). WP1 has successfully managed collaboration between WP2 (AI development) and WP3/4 (Technical implementation), ensuring strategic access to AI algorithms for the hardware engineers.
ESWW Topical Discussion
WP1’s influence extended beyond the consortium to the broader scientific community at the European Space Weather Week (ESWW) in Umeå. Organized and moderated by WP1 members George Miloshevich, Panagiotis Gonidakis, Ekaterina Dineva, and Stefaan Poedts, the panel titled "Scientific Outlooks for the Analysis of Space Weather Data in the Age of AI" addressed the most pressing issues in the field. The discussion, featuring experts such as Matthew West (ESA), highlighted the Vigil mission as a case study for future onboard autonomy and, in ESA's interest, for onboard AI applications. The organizers guided the debate toward the necessity of reproducible workflows and well-curated datasets as the bedrock of operations. A summary of the panel can be found here: AI Meets Space Weather.
Work Package 2 (WP2) focuses on the selection, development, and optimisation of machine learning (ML) algorithms for space science applications. Its role within the project is to identify the most effective approaches for autonomous data analysis and on-board processing, and to adapt and refine these algorithms for use with both simulated and real mission data, in close collaboration with the other work packages.
During the past year, WP2 activities focused on the comparison and optimisation of machine learning algorithms, carried out in close collaboration with WP5. The performance of different ML approaches was systematically evaluated using both virtual spacecraft synthetic data generated within WP5 and real observational data from the Solar Orbiter and MMS missions. This comparative analysis allowed WP2 to assess the robustness, reliability, and transferability of the algorithms across simulated and real mission scenarios, supporting the selection of the most suitable methods for further development and application within the project.
Significant progress was also achieved in the optimisation of ML techniques for the analysis of particle distribution functions. These developments improved the capability to separate the alpha particle population and to quantitatively assess the complexity of particle distributions, supporting more accurate real-time plasma characterisation. In parallel, advances were made in solar image analysis, including refinements to Variational Autoencoder (VAE) architectures and the implementation of a lightweight version of the YOLO algorithm applied to SDO data, enabling efficient feature extraction and event detection.
WP2 activities were also actively disseminated within the scientific community. The WP2 leader participated in the Smart Orbits Workshop, presenting future perspectives on space observations and artificial intelligence/machine learning. In addition, key results obtained within WP2 were presented at the European Geophysical Union (EGU) General Assembly 2025, further strengthening the visibility and impact of the project.
In the ASAP project, Work Package 3 (WP3) is responsible for accelerating key scientific applications based on artificial intelligence (AI-- developed in other Work Packages) using Field-Programmable Gate Arrays (FPGAs). This includes adhering to the performance- and energy-constraints that our space requires while at the same time reaching execution times suitable for the use-case.
One of the key results has been a solid survey of the available FPGA technologies (now and in the past) that have been used to accelerate AI space. Furthermore, this survey identified future opportunities and trends in the field that are used to develop hardware in the WP3 package.
Another key result is initial FPGA accelerators of four ASAP use-cases (two on remote-sensing and two in-situ). These use cases were vastly different and required different hardware implementations, subject to their neural network architecture. We implemented several accelerators using the Xilinx Vitis AI and High-Level Synthesis (HLS) methodologies, and quantified how much faster the accelerators are compared to executing them on modern ARM Cortex multicore processors, demonstrating the benefit of the approach. An early version of this work where presented at the AI for Space Application workshop: https://youtu.be/XBdxdZa4oYE?si=r2bU3sUmzYVfOV17
Currently, a more specialized accelerator is being developed for the use case. The accelerator aspires to be even more energy-efficient than the baseline accelerators (outlined above) without sacrificing performance for the target four use-cases, as well as investigating having built-in fault detection
Within the ASAP project, WP4 plays a central role in bridging algorithm development and hardware realization by enabling the deployment and evaluation of machine learning models on space-grade FPGA platforms. Building on the ML models defined in WP2 and the FPGA prototypes developed in WP3, WP4 provides the software framework and validation environment needed to assess how these solutions perform under realistic mission constraints.
In 2025, Work Package 4 (WP4) advanced the development of the ASAP hardware–software testbed, focusing on enabling and validating AI inference on space-grade hardware. The work concentrated on two main areas: the design of a flexible software interface layer and the definition of a validation framework for machine learning algorithms under representative in-flight scenarios.
Led by IngeniArs, with contributions from CNRS, KTH, and KU Leuven, WP4 progressed on the testbed software interface based on an application-independent soft-GPU. This virtual layer connects high-level AI frameworks (e.g. TensorFlow, PyTorch, ONNX) to radiation-tolerant FPGA platforms. Initial analyses using the GPU@SAT development kit show that all ASAP neural network use cases can be inferred, with ongoing efforts focused on porting selected models and optimizing them through quantization.
In parallel, CNRS led the definition of the testbed validation plan, in collaboration with IngeniArs, KTH, KU Leuven, and INAF. The plan targets functional and performance testing of selected ML algorithms across near-Earth and deep-space mission scenarios, based on requirements defined in WP2. Discussions on test scenarios, algorithm selection, and test bench design are ongoing, with close coordination across WP2, WP3, and the wider consortium.
WP4 will next focus on executing selected models on hardware, finalizing the test plan, deploying the test bench, and conducting validation tests leading to the test report in 2026.
Within the ASAP project, Work Package 5 makes significant contributions to the development of a virtual spacecraft that supports intelligent, autonomous space missions using machine learning techniques.
One key accomplishment is the creation of a virtual flight environment in which a spacecraft (or a set of satellites, as in multi-spacecraft missions) can realistically fly through numerically simulated plasmas. A spacecraft operating in a virtual environment can follow various trajectories and retrieve virtual measurements of electromagnetic fields and plasma properties, analogous to what a real spacecraft detects in space.
The project has also produced virtual instruments that mimic the measurement of plasma particles by sensors on satellites. A plasma analyzer reproduces the distribution in terms of energy and direction. This produces virtual data that can be easily compared with data acquired by missions such as Cluster, MMS, and Solar Orbiter. This diagnostic tool can work under a broad range of simulated space conditions.
Furthermore, WP5 has a simulation tool that follows the trajectory of energetic particles in a varying electromagnetic environment. This enables the simulation of how particles gain energy and interact with regions such as shocks and turbulent regions. Simulating a spacecraft flyby in a virtual plasma environment has reproduced phenomena already detected in situ.
Finally, there have been new approaches to preparing particle data in a helpful format for machine-learning techniques, aiming to classify regions in space automatically. Particle distribution is analyzed using fundamental modes decomposed by a recently upgraded and optimized Hermite transform algorithm. Neural networks can be trained on these transforms to identify relevant phenomena autonomously. This could eventually lead to missions with an improved data filtering system and optimized data filtering, storage, and transmission of space probe data.
Collectively, these outcomes enable a flexible and practical digital twin of a space mission and will facilitate the development of machine learning technologies to ensure that future spacecrafts can operate more efficiently, autonomously, and reliably.
In 2025, WP6 built on the solid dissemination foundations laid in the previous year. The focus this year was on sharing ASAP’s work more broadly, encouraging discussion, and highlighting the project’s progress across space and AI research.
One of the main highlights was the organisation of the “AI for Space Applications Exploring” workshop, which brought together researchers and practitioners to explore how AI can support future space missions. WP6 also continued to grow the project’s YouTube channel, adding recordings from both the AI in Space workshop and the Smart Orbits Workshop organised by WP1.
Throughout the year, WP6 maintained an active presence on LinkedIn, sharing project updates, event highlights, and consortium achievements. This steady engagement has helped expand the project’s reach and keep the community informed. In addition, the project event page captures the workshops, conferences, and meetings attended by ASAP team members, offering more detailed insights and personal perspectives from across the consortium.
Together, these activities reflect WP6’s ongoing role in ensuring that ASAP’s work remains visible, accessible, and connected to the broader community as the project continues to grow.
As we wrap up this year’s newsletter, we’re proud of what the ASAP consortium has achieved in 2025 — from deepening scientific progress and cross-WP collaboration to strengthening community engagement through workshops, shared content, and ongoing dialogue. Each contribution has helped move us closer to our collective vision of enabling intelligent, autonomous space missions.
Looking ahead to 2026, we’re excited to build on this momentum, continue sharing insights, and strengthen the connections that make ASAP’s work meaningful both within the project and across the broader space and AI communities. Thank you to everyone — partners, collaborators, and followers alike — for your continued support and engagement. Here’s to another inspiring year of discovery and impact.
Warm regards,
The ASAP Project Team
Vi använder cookies för att analysera webbplatstrafik och optimera din webbplatsupplevelse. Genom att acceptera vår användning av cookies kommer dina data att aggregeras med alla andra användardata.