Max Robotics — Global Service Robot Directory + US Certification

About

Andrew Spek is a researcher at the forefront of 3D scene understanding for autonomous robotics, with a focus on deep learning and computer vision. His major contributions center on enabling robots to perceive complex 3D structures from single RGB images, a critical capability for fully autonomous navigation and manipulation. Spek pioneered a novel deep learning framework that jointly predicts depth, surface normals, and surface curvature—a trifecta of geometric information that provides a richer understanding of a scene than depth alone. His most-cited work, "Joint prediction of depths, normals and surface curvature from RGB images using CNNs" (2017), has accumulated 31 citations and established a new paradigm for multi-task geometric learning. He further advanced the field by developing a real-time model for joint semantic segmentation and depth estimation, addressing the practical hurdles of deploying deep networks on robotic hardware with limited computational resources. Spek also contributed a fast, robust method for computing principal curvatures from noisy range data, a foundational technique for object segmentation and robotic grasping. His work directly bridges the gap between high-level scene interpretation and low-level geometric reasoning, making him a key figure in the push toward more perceptive and autonomous robots.