Research Map
Research Map
Discover the fields closely connected to object pose estimation, the problems they solve, and how their papers fit into the research library.
The Methods section focuses on approaches within 6D pose estimation. The research map organizes related work in visual localization, open-vocabulary segmentation, feature matching, VLA, and world models, keeping the central topic coherent while allowing the library to grow.
Each area links to its research-area filter in the library. Areas without papers are marked as pending; papers appear once the corresponding researchArea field is set in their metadata.
Core
Core focus
Object pose estimation remains the site's focus, starting with coordinate frames, correspondences, PnP, evaluation protocols, and major systems.
Geometry & Perception
Upstream perception & geometry
These areas supply perception frontends, interpretable correspondences, 3D geometry, and renderable representations for pose estimation.
Localization
Visual Localization / SLAM
Estimates camera poses within a scene or map, providing localization, mapping, and multiview geometry for object pose systems.
Connection to 6D pose
Shares camera poses, projection geometry, and map representations
A question to start with
How do camera poses, scene coordinates, and map matching affect object pose systems?
Matching
Feature Matching / Correspondences
Studies local features, dense matching, and 2D–3D / 3D–3D correspondences, underpinning PnP, registration, and render-and-compare methods.
Connection to 6D pose
Provides interpretable geometric correspondences
A question to start with
How can image points, object surface points, and 3D models be matched reliably?
Open-Vocab
Open-vocabulary Segmentation / Perception
Studies open-category object discovery, semantic segmentation, and mask generation, extending pose estimation beyond fixed object categories.
Connection to 6D pose
Provides perception frontends for open categories
A question to start with
How can a target object be found without a fixed set of categories?
3D Repr.
3D Representations / Reconstruction
Studies NeRF, 3DGS, TSDF, point clouds, and implicit representations to support CAD-free, sparse-view, and scene-level geometric modeling.
Connection to 6D pose
Supplies shape, appearance, and renderable geometry
A question to start with
Can reconstructed or generated 3D representations replace manually created CAD models?
Embodied Systems
Downstream robotics & temporal intelligence
These areas connect object states, language goals, action policies, and predictions of environmental behavior.
Robotics
Robot Perception & Manipulation
Studies how pose, masks, language instructions, and scene geometry support action planning for grasping, assembly, and manipulation.
Connection to 6D pose
Connects pose estimation to physical actions
A question to start with
How can pose estimates become executable, reliable robot actions?
VLA
VLA / Embodied AI
Studies vision-language-action models, robot foundation models, and multimodal policy learning, extending pose estimation into downstream tasks and high-level control.
Connection to 6D pose
Connects perception, language, and action policies
A question to start with
How can foundation models use geometric perception for long-horizon manipulation?
World Model
World Models
Studies video prediction, physical dynamics, latent dynamics, and planning, extending static recognition into temporal reasoning and models of interactive environments.
Connection to 6D pose
Supports prediction, planning, and physical consistency
A question to start with
How can models understand changes in object states under actions and over time?
How To Use
How to extend the map
Add papers beyond conventional 6D pose to the same research library. Use researchArea for the broad research area and methodTags for more specific method attributes.
For example, place open-vocabulary segmentation papers under Open-vocabulary Segmentation / Perception, with tags such as segmentation, foundation-model or promptable-mask.