| 40 | || '''08.05.2023''' || '''Jalo Nousiainen[[BR]](LUT-Universität, Lappeenranta, Finnland)''' || '''Model-based reinforcement learning and inverse problems in extreme adaptive optics control[[BR]][[BR]]'''The control of eXtreme Adaptive Optics (XAO) systems is crucial for the direct imaging of potentially habitable exoplanets on ground-based telescopes. However, current XAO control laws leave strong residuals, particularly at small angular separations from host stars where most habitable exoplanets are located. To address this issue, our recent work has focused on two approaches: Model-based Reinforcement Learning (MBRL) and spatio-temporal Gaussian process (ST-GP) regression.[[BR]]MBRL is a data-driven approach that learns control strategies from system feedback and promises to effectively manage factors that can hamper XAO performance, such as temporal delay, calibration errors, photon noise, and optical gains. I will present recent results from the GHOST test bench at ESO and discuss our future goals.[[BR]]ST-GP regression, however, allows for the theoretical examination of predictive control strategies. Factors that affect predictive controllers' performance include the wavefront sensor type, measurement noise level, AO system geometry (aliasing, actuator spacing), and atmospheric conditions (e.g., seeing, wind speed). Through ST-GP regression, we can explore the theoretical limits of predictive control under different conditions and geometries.[[BR]]Overall, our work aims to advance XAO control methods to enable high-contrast imaging of potentially habitable exoplanets using ground-based telescopes.[[BR]][[BR]]Presentation: English[[BR]]Slides: English[[BR]]Questions: German, English || |