Beyond RGB-D: Perception of Glass, Mirrors, and See-Through Scenes for Robotics

This work presents a survey of robotic perception methods for transparent, reflective, and see-through environments. It reviews recent approaches in transparent object segmentation, depth completion, multimodal sensing, polarization-based geometry recovery, neural rendering, event-based vision, foundation-model-based depth estimation, uncertainty estimation, sim-to-real transfer, and embodied robotic intelligence. The paper emphasizes a robotics-oriented perspective by connecting sensing, scene understanding, uncertainty-aware reasoning, and task-level decision-making within a unified perception pipeline. It also discusses practical challenges such as SLAM failures, unreliable RGB-D sensing, computational constraints, benchmarking limitations, and real-time deployment issues in non-Lambertian environments. This upload is a preprint version intended for public dissemination and later journal submission. It has not yet undergone formal journal peer review.