Depth cameras produce missing or corrupted point clouds on transparent and specular warehouse items, causing robotic pick failure rates above 35% for those SKUs

Commercial structured-light and time-of-flight depth cameras (Intel RealSense, Microsoft Azure Kinect) fail to return valid depth data for transparent plastic packaging, glass bottles, shrink-wrapped products, and highly reflective metallic items because infrared light passes through or specularly reflects off these surfaces instead of scattering back to the sensor. This produces holes, phantom surfaces, or background bleed-through in the 3D point cloud, making grasp-pose estimation unreliable. Why it matters: transparent and reflective items account for a significant fraction of e-commerce SKUs (bottles, blister packs, glossy electronics packaging), so robotic picking systems must fall back to manual human intervention for these items, so fulfillment centers cannot achieve full automation and must maintain parallel manual pick stations, so the labor savings promised by pick-and-place robots are capped at roughly 65% of total SKU coverage (Amazon Sparrow's reported ceiling), so warehouse operators cannot close the business case for fully automated piece-picking lines, so the $15B+ warehouse automation market grows slower than projected. The structural root cause is that all mainstream commercial depth sensors rely on infrared light reflection, which physically cannot produce returns from materials whose refractive index allows IR transmission or whose surface geometry creates specular (mirror-like) reflection away from the sensor, and alternative sensing modalities (polarization cameras, plenoptic sensors, tactile probing) remain too slow, too expensive, or too immature for production-speed warehouse picking.