The work carried out reports on a new machine learning-based software prototype for designing and optimising phased array antennas. The prototype can rapidly and accurately compute excitation coefficients for large-scale antenna arrays across diverse configurations and far-field requirements. The prototype relies on tailored neural networks integrating structural physics and domain knowledge using an encoder-decoder structure.

The SPOTTED project addresses the critical issue of marine debris, particularly plastic pollution, by enhancing detection and monitoring capabilities. Utilizing advanced spatio-temporal fusion (STF) techniques and Artificial Intelligence (AI) with high-resolution satellite imagery from Sentinel-2 (S2) and PlanetScope (PS), the project aims to refine S2 data and develop AI algorithms for identifying landfills and floating debris.

Background and justification:

A number of AR/VR projects have been implemented so far within ESA. The main focus has been on the application of AR/VR/MR to human and robotic operations as well as teleoperations/ maintenance tasks and lately on Space Safety use cases. With an increasing volume of information being represented on AR/VR/MR applications, especially for use cases involving visualisation of large datasets (e.g. scientific data) and Big Data, choosing the right approach is critical to meet performance limitations and offer an intriguing experience with greater fidelity without overwhelming the user.

In this study we explore the potential of Large Language Models (LLMs) to interact with Earth Observation (EO) data with some promising first results. We source a first testing set for the BIG Bench framework. With 20+ subtasks, we aim for a large variety of questions which are currently beyond the capabilities of LLM and other models. Further, we implement a methodology to test proprietary state of the art models where token distribution may not be available.

N/A

FITAI is a Machine Learning based technique that improves GNSS positioning engine accuracy by mapping signal strength and residuals into position errors. These inputs are fed into a neural network to learn and correct errors, offering a non-invasive solution that provides better location precision without relying on external sources like reference stations.