Artificial Intelligence for Adaptive Optics – AI4AO

The project relies on an innovative co-design approach combining optical hardware and artificial intelligence algorithms. It revisits the three core components of adaptive optics: wavefront sensing, control, and global system management.

WP1 focuses on developing new wavefront sensors through joint optimization of optical design and neural network-based reconstruction. This approach overcomes traditional trade-offs between sensitivity, linearity, and robustness, with applications in astronomy (ultimate sensitivity) and space surveillance (strong turbulence conditions).

WP2 addresses predictive control. Reinforcement learning (RL) approaches are explored to replace or complement classical controllers (e.g., LQG), enabling anticipation of turbulence evolution and reduction of temporal errors. Comparative studies between model-based and model-free RL approaches, as well as multi-sensor data fusion, will be conducted.

WP3 is dedicated to experimental validation. The developed solutions will be tested on AO benches (LOOPS, PICOLO) and validated in real conditions on-sky (PAPYRUS, FEELINGS). Real-time implementation, robustness, and operational performance will be key aspects of this phase.

AI4AO will demonstrate the feasibility of integrating artificial intelligence at multiple levels of adaptive optics systems.

Expected results include:

new wavefront sensor concepts optimized through AI–optics co-design, improving sensitivity and robustness,
predictive control strategies based on reinforcement learning, outperforming classical approaches in dynamic conditions,
experimental validation both on laboratory benches and on-sky,
real-time compatible implementations meeting operational constraints.

These results will address key limitations of current AO systems, particularly nonlinear effects and temporal delays, while improving robustness under extreme turbulence conditions.

AI4AO paves the way for a new generation of autonomous and intelligent adaptive optics systems. In the short term, the project will enable a maturation phase toward operational deployment.

In the medium term, developments can be integrated into major infrastructures such as the ELT for astronomy or space surveillance platforms like PROVIDENCE. Applications also extend to optical communications, a rapidly growing strategic domain.

In the long term, the integration of AI into AO systems could transform instrument design, enabling simpler, more robust, and self-adaptive architectures. The project will contribute to strengthening European sovereignty in space observation and advanced photonic technologies.

To come

With the rapid expansion of human activities in space, whether it be manned missions, communication satellites, constellations for Earth observation, or New Space initiatives, the management of the space domain is becoming increasingly complex. This increased complexity necessitates enhanced space surveillance to ensure the security of orbital infrastructures. Optics, and more specifically adaptive optics (AO), offer considerable potential for monitoring space objects with high resolution.

Adaptive optics is a key technology for correcting the effects of atmospheric turbulence, thereby enabling detailed imaging from ground-based telescopes. Today, AO is integrated into all major ground-based telescopes. With the construction of the Extremely Large Telescope (ELT), AO has become indispensable. Equipped with ultra-high-performance AO, the ELT opens up fascinating prospects, such as detecting Earth-like planets and potentially analyzing their atmospheres.

AO also plays a strategic role in space surveillance, particularly in observing satellites and securing optical communications between the ground and satellites. By improving angular resolution, AO allows for the distinction of details such as solar panels, antennas, or variations in satellite shapes. It also facilitates the detection of changes, such as mechanical degradations or malfunctions, and enhances astrometric precision for determining orbits. These capabilities are essential for monitoring abnormal behaviors and identifying potentially hostile satellites. The PROVIDENCE platform, developed by ONERA, aims to address these challenges by providing high angular resolution observation, classification, identification, characterization, and surveillance capabilities for low Earth orbit.

By combining advancements in adaptive optics and machine learning, the AI4AO project aims to push the current limits of high angular resolution imaging, with direct applications in astronomy, optical communications, and space surveillance. AI4AO aspires to transform AO systems, making them more performant, robust, and autonomous. Developed over 36 months, AI4AO positions itself as a structuring project for future generations of high angular resolution instruments. The advancements made will directly benefit projects like the ELT and PROVIDENCE, offering major innovation prospects for astronomy, optical telecommunications, and space observation. The project also paves the way for future technological maturation, aiming for large-scale operational deployment.

The consortium, composed of the Laboratoire d'Astrophysique de Marseille, ONERA, and the University of Durham, brings together complementary expertise in photonics, AI, and the integration of real-time embedded solutions.