Learned pi metalens for wide-field-of-view near-infrared imaging

Wide—field-of-view (FoV) near-infrared (NIR) imaging is vital for compact augmented reality and robotic systems. Conventional wide-FoV optics rely on bulky compound refractive optics, limiting miniaturization. Metalenses composed of sub-wavelength meta-atoms offer a compact alternative. However, achieving high transmission and complete 2 π phase coverage simultaneously over wide angle-of-incidence (AOI) remains challenging, leading to blur and non-uniform efficiency across the field. Here, we present a learned π metalens based on an AOI-aware design framework that achieves uniform-quality 100° FoV NIR imaging. Rather than enforcing full 2 π phase control, we leverage meta-atoms that maintain high-transmission across wide angular ranges, which inherently exhibit restricted phase coverage. To design such a phase-limited metalens, we introduce a differentiable phase-limited metalens representation that incorporates the AOI-aware vectorial response of meta-atoms. This effectively learns a metalens using phase-limited high-transmission meta-atoms, yielding improvements in focusing efficiency and image sharpness across a wide FoV. Our learned π metalens demonstrates a 14% improvement in modulation transfer function at the largest AOI and a 2.4x enhancement in focusing-efficiency uniformity compared to conventional quadratic 2 π metalenses. Our results suggest that, when AOI-dependent responses are properly modeled, complete 2 π phase coverage is not strictly required for high-performance wide-FoV metalenses, opening new opportunities for compact NIR imaging systems in robotics, human-computer interaction, and eye-tracking in extended reality systems.