clawft

Spatial intelligence 2026

Survey of world models and reconstruction for Urth, BVH, and Graph Views

This is the Fumadocs index for the 2026-09-21 survey. Canonical files live in the repo under docs/research/spatial-intelligence-2026/.

Scrollyteller: Open /urth-spatial

Stance

The world is not a video. Appearance may be generated. Metric shape lives in the BVH. Plans ride Graph Views and optional LeWM. Unobserved Urth stays unobserved.

Deep dives

IDFile in repoStance
Indexdocs/research/spatial-intelligence-2026/README.mdCandidate list
S1world-labs-marble-atlas.mdApply language (dual export)
S2 / S8interactive-video-world-models.mdPriors only
S3feedforward-reconstruction.mdApply pipeline adapters
S4scene-graphs-open-vocab.mdApply hierarchy
S5large-scale-3dgs-lod.mdAppearance LOD twin
S6vjepa-and-lewm.mdOptional visual index
Synthesisurth-applicability.mdDo / do not table
rUv two-wayruv-parallels-and-gaps.mdLearn from / contribute to rUv
StoryboardSTORYBOARD.mdScrolly beats

Queue (as of the 2026-09-21 survey, not re-checked)

Residual: live BVH publish, spatial RPC, F9 promote, Urth E1/E2 OSM, panopticon, DiskANN defer, ADR-087.

New 2026: VGGT/MASt3R before COLMAP, SpatialLM as W1 proposer, scene-graph hierarchy, Octree-GS appearance LOD, V-JEPA visual index_id, Marble dual-export copy.

See also Urth vision and LeWM.

On this page