Reconfigurable metasurface enabled by ferroelectric superdomain for AI-driven shape completion

Metasurfaces offer a promising compact platform for miniaturizing optical systems and enabling advanced functions such as geometric reconstruction of target objects. However, efficiently extracting multi-dimensional optical information within the targets acquired by metasurfaces and subsequently reconstructing the geometry of partially obscured objects has remained challenging. Here, we propose an infrared metasurface that consists of integrating a continuous graphene with a periodic ferroelectric domain array, which enables in - situ tunable relative transmittance spanning from negative to positive values for all-optical computing tasks. The designed metadevices exhibit a tunable spectral response ranging from 8 to 17 μm across diverse ferroelectric domain patterns by reconfiguring the ferroelectric domains. We also show that the reduced transmittance can be modulated from −6% to 10% under low gate voltages. By encoding the reduced optical transmission of metadevices with a shape completion network, we further propose robust three-dimensional shape completion from sparse and incomplete point clouds by passive thermal imaging, achieving a high-resolution geometric model through the recovery of occluded structural details. Our work merges reconfigurable metasurface design with AI-driven reconstruction, potentially opening pathways for intelligent vision systems in autonomous navigation and precise three-dimensional sensing.