Focal plane wavefront control with model-based reinforcement learning

Focal plane wavefront control with model-based reinforcement learning

Focal Plane Wavefront Control with Model-Based Reinforcement Learning Breakthrough in Exoplanet Imaging

Section 1 – What happened?

Researchers at a leading institution have made a groundbreaking discovery in the field of exoplanet imaging, developing a new model-based reinforcement learning algorithm called Policy Optimization for NCPAs (PO4NCPA). This innovative approach enables the automatic detection and correction of both dynamic and static non-common-path aberrations (NCPA) in high-contrast imaging instruments on extremely large telescopes. The PO4NCPA algorithm interprets the focal-plane image as input data and determines phase corrections that optimize both non-coronagraphic and post-coronagraphic point spread functions (PSFs) without prior system knowledge.

Section 2 – Background & Context

The direct imaging of potentially habitable exoplanets is a prime science case for high-contrast imaging instruments on extremely large telescopes. However, the observation of exoplanets orbiting close to their host stars is limited by fast-moving atmospheric speckles and quasi-static NCPA. Conventional NCPA correction methods often rely on mechanical mirror probes, which compromise performance during operation. The development of a machine-learning-based NCPA control method is crucial for advancing exoplanet imaging and potentially leading to groundbreaking discoveries in the field of astrobiology.

Section 3 – Impact on Swiss SMEs & Finance

While the breakthrough in exoplanet imaging may not have an immediate impact on Swiss SMEs and finance, it highlights the potential of machine learning and artificial intelligence in solving complex problems in various fields. The development of PO4NCPA demonstrates the power of model-based reinforcement learning in optimizing performance in high-contrast imaging instruments. This innovation may inspire Swiss companies to explore the application of machine learning in their own industries, potentially leading to new business opportunities and revenue streams.

Section 4 – What to Watch

The PO4NCPA algorithm's sub-millisecond inference times and performance make it suitable for real-time low-order correction of atmospheric turbulence beyond high-contrast imaging. Researchers and scientists will continue to explore the applications of PO4NCPA in exoplanet imaging and other fields, potentially leading to new breakthroughs and discoveries. Additionally, the development of PO4NCPA may inspire the creation of new startups and companies focused on applying machine learning and AI in various industries.

Source

Original Article: Focal plane wavefront control with model-based reinforcement learning

Published: April 1, 2026

Author: Jalo Nousiainen

Disclaimer: This article is for informational purposes only and does not constitute financial advice. Consult a licensed financial advisor before making investment decisions.