Context of the AI4AO Project

Ground-based observation of the Universe and of Earth’s orbital environment is undergoing a profound transformation. Modern astronomy now relies almost entirely on adaptive optics (AO) to overcome the blurring effects of atmospheric turbulence, which otherwise limit even the largest telescopes to the resolution of a 20 cm amateur instrument. AO has enabled major breakthroughs—from the study of the Milky Way’s central black hole to the direct imaging of exoplanets—and has become indispensable for all current and future large observatories, including the upcoming Extremely Large Telescope (ELT)

At the same time, AO is becoming a critical technology for space surveillance and tracking (SST/SSA), where ground-based optical systems must track rapidly moving satellites and debris with high precision. With the exponential increase of LEO constellations and the growing congestion of orbital space, Europe faces an urgent need for autonomous, high-resolution monitoring capabilities to protect its strategic space assets.

Although AO has matured significantly over the last three decades, it now faces new and severe challenges. Next-generation astronomical instruments require nanometric wavefront control, an order of magnitude beyond current systems, to detect faint exoplanets hidden in the glare of their host stars. In the space-surveillance domain, AO systems must operate at low elevation angles, through extreme turbulence and strong scintillation, while tracking objects that move rapidly across the sky. These scenarios expose intrinsic limitations of existing AO architectures: their wavefront sensors become highly non-linear, their measurements degrade in strong turbulence, and their control loops are too slow to cope with rapidly evolving atmospheric conditions. As a result, the performance of even the most advanced AO instruments is still far from the theoretical diffraction limit.

At the same time, artificial intelligence (AI)—and in particular deep learning and reinforcement learning—has opened new possibilities for sensing, prediction, and control. Early results worldwide show that AI can reconstruct wavefronts from complex WFS signals, infer non-linear dynamics, and learn predictive control policies that outperform classical algorithms. Yet these efforts remain partial and largely simulation-based. What is missing is a comprehensive, physically grounded framework that integrates AI directly into the three pillars of AO: wavefront sensing, real-time control, and global system management. This is precisely the ambition of AI4AO: to rethink AO from end to end by embedding intelligence within both the optical hardware and the computational pipeline.

AI4AO brings together leading expertise from the Laboratoire d’Astrophysique de Marseille, ONERA, and the University of Durham, building on decades of joint developments in AO for astronomy, laser communication, and SSA. The project leverages cutting-edge experimental platforms—LOOPS, PICOLO, PAPYRUS, FEELINGS—to move beyond simulations and validate AI-driven concepts directly on sky, a decisive step toward operational deployment on the ELT and on next-generation SSA facilities such as PROVIDENCE

By combining optical co-design, advanced neural networks, predictive modelling, and real-time hardware optimisation, AI4AO aims to deliver a new generation of adaptive optics: more sensitive, more robust, and fully autonomous, capable of unlocking the scientific potential of future telescopes and strengthening Europe’s capacity to monitor and secure its space environment.

Page updated

Google Sites

Report abuse