Machine-learning-based open-loop control of an optofluidic deformable phase plate

We present a machine learning–based open-loop control method for deformable phase plates (DPPs), a class of electrostatically-actuated optofluidic wavefront modulators. Traditional control schemes assume a linear device response, which fails to capture nonlinear effects due to electrode gap variations and fluidic coupling, especially at high strokes. To address this, we train a gradient boosting regression model on a dataset of interferometrically measured Zernike-mode wavefronts. The model learns the inverse mapping from desired shapes to control signals, achieving sub-20 nm RMS wavefront error. This approach improves control fidelity and robustness, enabling more accurate adaptive optics correction using DPPs.