Robotic Grasping
Convert object poses into end-effector targets so a robot can choose grasps, avoid collisions, and place objects.
MVP Content
From geometric intuition to research papers
Explore five starting points: build intuition for 6D pose geometry, understand how leading methods recover pose, make sense of datasets and metrics, and navigate related research through a searchable knowledge library.
Knowledge System
Knowledge System
Start with rigid geometry, coordinate frames, camera models, and PnP, then connect them to data annotation, visual SLAM, robot calibration, and grasping.
Methods
Methods
Compare templates, keypoints, 2D–3D correspondences, dense matching, synthetic data, and foundation models through their inputs, outputs, and failure modes.
Datasets
Datasets & Evaluation
Compare LINEMOD, YCB-Video, T-LESS, and BOP, and understand ADD, ADD-S, VSD, MSSD, MSPD, and AR.
Directions
Research Map
Follow connections from 6D pose to visual localization, feature matching, open-vocabulary perception, 3D representations, robotic manipulation, VLA, and world models.
Paper Atlas
Paper Library
Find papers by research area, publication date, venue, CCF rank, dataset, metric, author affiliation, or code availability, then open their detailed reading notes.
Applications
Why object pose matters
Reliable estimates of an object's 3D pose in the camera frame connect perception to real-world grasping, assembly, inspection, and augmented reality.
AR & Digital Twins
Align virtual models with physical objects: stable 6D poses support overlays, measurement, and interaction.
Industrial Inspection
Use part poses to verify assembly, check orientation, and identify missing or incorrectly installed components.
Scene Understanding
6D poses describe spatial relationships beyond 2D bounding boxes, supporting planning, reasoning, and simulation.
Learning Path
Suggested learning path
Build a foundation in projection geometry, coordinate transforms, and evaluation metrics before diving into network architectures.
- 1. Camera models and coordinate frames
Understand how camera intrinsics and extrinsics connect pixels, camera coordinates, and object coordinates.
- 2. 2D–3D correspondences and PnP
Learn the classical geometric pipeline for recovering pose from keypoints or correspondences.
- 3. Deep networks and pose refinement
Compare the tradeoffs of direct regression, dense correspondences, RGB-D ICP, and pose refinement.
- 4. BOP-style evaluation
Use a consistent evaluation protocol to examine visibility, symmetry, and projection error alongside aggregate scores.