Get Started
Research questionHow can warehouse robots estimate 9D box pose from one RGB-D view under clutter and occlusion without per-object CAD models?Warehouse manipulation requires both a box’s 6D pose and its dimensions, but stacked boxes can hide surfaces and provide little visual texture. Box symmetries also make pose ambiguous, while maintaining CAD models for changing inventories is costly.
Computer Vision
Robotics
Latest papersRecent research connected to this question, newest first.AnyBox: Efficient Zero-Shot 9DoF Pose Estimation of Boxes for Robotic ManipulationThe described system targets box pose and dimension estimation from a single RGB-D observation using a canonical box template rather than instance-specific CAD. Evidence comes from public benchmarks, an in-house warehouse dataset, and a cluttered box-shelving manipulation task; broader generalization beyond these settings is not established.research paper · Sep 3, 2026
Related questions
How can we predict local 3D occupancy from a single underwater RGB image despite unreliable visual geometry?How can robots recognize known objects without labeled examples or CAD when their 2D appearance is ambiguous?How can zero-shot 6D pose front-ends detect partially visible objects while rejecting similar distractors?How can multi-robot teams estimate 3D positions and orientations from body-frame bearings without knowing orientations?
Home
Topics
Search
Library