Optimal design of non-uniform diffraction gratings for uniform out-coupling brightness in waveguide combiners

Recent work in optical-see-through (OST) near-eye display shows significant process in expanding the field of view (FoV) and eyebox. However, improving the image quality of computer-generated holography (CGH) still poses a major challenge in providing an immersive experience of augmented reality (AR). The difficulty in CGH design lies in bridging the gap between the physical system and the idealized wave propagation model. In this work, we incorporate a camera into the full-color 3D CGH optimization of a waveguide with two dimensional eyebox expansion. This iterative optimization process is repeated across a dataset of 3D CGH patterns to train a neural network applicable to the optimization of previously unseen CGH patterns. The proposed method is shown to outperform conventional approaches in optimizing CGH patterns for two-dimensionally expanded waveguide holography.