Get Started
Home
Topics
Search
Library
Research questionHow can robots recognize known objects without labeled examples or CAD when their 2D appearance is ambiguous?Robots may encounter visually similar, low-texture parts whose images do not distinguish their identities. The difficulty is greater when no labeled training set or CAD model is available, leaving the system to rely on information gathered from the objects themselves.
AI
Computer Vision
Image & Video Processing
Robotics
Latest papersRecent research connected to this question, newest first.Where Appearance Fails, Geometry Recognizes: A CAD-Free 3D Shape Prior That Complements Vision Foundation ModelsThe evidence concerns short object-centric scans represented using RGB-D-derived geometry or 3D Gaussian Splatting and combined with frozen DINOv2 image features. Results cover the HOPE and T-LESS recognition datasets, including partial occlusion and varying lighting; the study does not address the BOP pose benchmark.research paper · Sep 3, 2026
Related questions
How can warehouse robots estimate 9D box pose from one RGB-D view under clutter and occlusion without per-object CAD models?How can robots recognize social gestures in real time despite occlusion, viewpoint changes, and noisy keypoints?How can humanoid robots anticipate future footholds when current visual observations are insufficient?How can dual-arm robots recover occluded object surfaces from calibrated, synchronized multi-view observations?