Actuality from the project

Three people (hired in 2024) spent 1.5 months (April–May) at the Subaru Telescope (Hawaii) thanks to European funding, collaborating with the SCExAO instrument teams on: image denoising using autoencoders, new approaches to wavefront reconstruction and control based on neural networks, and the integration of these developments into the real-time software stack. Deliverable 1.1 (specifications document) was finalized, as was a contribution to the integration report for the STREAMS computing platform. The project produced four contributions (two presentations and two posters) to the 8th AO4ELT Conference (Chile, Oct. 27–31), the flagship event of 2025 for the Adaptive Optics community. The project is also actively participating in the design, development, testing, and validation of the SAXO+ system upgrade on the SPHERE instrument, leading the work package dedicated to the real-time computer and working in close collaboration with the other work packages.

SCExAO: https://www.naoj.org/Projects/SCEXAO/scexaoWEB/000home.web/indexm.html

STREAMS: https://streams.pages.obspm.fr/streams/

SPHERE+: https://sites.lesia.obspm.fr/sphereplus/

Focus

Continue multi-scale hardware and software co-design to optimize and plan the execution of new OA data processing pipelines, combining AI with more traditional task flows.

Implement and evaluate an intelligent data interface based on FPGA boards with sufficient capacity to simultaneously collect large volumes of data, filter and compress raw data, and distribute it with low latency within a heterogeneous architecture.

Integrate, test, and validate a demonstrator for the SAXO+ module, taking into account all the specific requirements of the SPHERE+ instrument upgrade.

Develop a prototype to demonstrate the scale-up for PCS on the ELT.

Relation with others WP

not applicable

Relation with industries

not applicable

Publication links to the project

• hal-05519474 – “A CNN encoder for modal phase reconstruction in Adaptive Optics systems”, Pierre Vermot, Damien Gratadour
• hal-05520431 – “Convolutional neural networks for fast myopic deconvolution of AO observations”, Pierre Vermot
• hal-05529869 – “Upgrading SPHERE with the second stage AO system SAXO+: Hardware interface integration”, P Fontanillas, F Ferreira, A Sevin, D Gratadour
• hal-05627730 – Exploring Neural Network Approaches for Pyramid Wavefront Sensor Denoising in Adaptive Optics, Jordan Raffard, Damien Gratadour