中文版说明 →

Unreal Engine 5 project for procedural military simulation world generation driven by real-world GIS data.

Overview

Sphinx transforms real geographic data — OpenStreetMap exports, DEM elevation tiles, and ArcGIS Feature Services — into classified land-use polygons that drive UE5's Procedural Content Generation (PCG) framework to automatically populate terrain with buildings, vegetation, roads, and water features.

Key Capabilities

• Multi-source GIS ingestion — Local GeoJSON/DEM files or live ArcGIS REST queries
• Terrain analysis — DEM → slope/aspect → terrain classification → connected zone extraction
• Polygon derivation — Terrain zones cut by road/river/coastline vectors into land-use polygons
• Automatic classification — 9-rule system: Residential, Commercial, Industrial, Forest, Farmland, Water, Road, OpenSpace, Military
• PCG integration — Custom sampler node outputs points with density/type metadata for downstream spawners
• Offline → Online pipeline — Editor-time generation persisted as DataAsset, runtime PCG reads from asset

Architecture

Mode A/B/C: Single-Region Pipeline
───────────────────────────────────
GeoJSON (OSM) ──┐
                ├→ IGISDataProvider → PolygonDeriver → ULandUseMapDataAsset → PCG Sampler
ArcGIS REST   ──┘         │                │            (.uasset)                │
                          │                │                                     │
DEM tiles ────────→ QueryElevation   TerrainAnalyzer                     Attribute Filter
(GeoTIFF/PNG/RAW)         │                │                                     │
                          │          ┌─ With DEM → vector cut                   ├→ Building HISM
                          │          └─ No DEM  → vector polygons               ├→ Tree HISM
                          │                │                                     ├→ Road Spline
                          │          LandUseClassifier                           └→ Water Mesh
                          │                │
                          └──────→ TArray<FLandUsePolygon>

Mode D: Tiled Pipeline (Global Scale)
──────────────────────────────────────
Python Preprocessing (PBF/DEM/WorldCover)
        │
        ├→ tile_manifest.json
        ├→ tiles/tile_X_Y.geojson
        ├→ tiles/tile_X_Y_dem.png
        └→ tiles/tile_X_Y_landcover.json
                │
    TiledFileProvider (LRU cache, dedup)
                │
    TiledWorldBuilder (Editor batch tool)
                │
    Per-tile ULandUseMapDataAsset (spatial index)
                │
    TiledLandUseMapDataAsset (async streaming catalog)
                │
        PCG Sampler (tile-aware, LoadRadius)

Quick Start

Local File Mode

1. Export region data from overpass-turbo as GeoJSON
2. (Optional) Download DEM tiles from OpenTopography or SRTM
3. Place files under Content/GISData/YourRegion/
4. Drop AGISWorldBuilder into your level
5. Set DataSourceType = LocalFile, configure paths and origin coordinates
6. Generate In Editor → preview → Generate And Save Data Asset
7. Wire the DataAsset into your PCG graph via the GIS Land Use Sampler node

ArcGIS REST Mode

1. Set DataSourceType = ArcGISRest
2. Enter Feature Service URL and API key
3. Generate And Save Data Asset → same downstream PCG workflow

Tiled File Mode (Global Scale)

1. Prepare raw data: OSM PBF, DEM GeoTIFF, ESA WorldCover GeoTIFF
2. Run the Python preprocessing pipeline:

   python Tools/GISPreprocess/preprocess.py --input ./RawData --output Content/GISData/Region_01 --tile-size 1024

3. Set DataSourceType = TiledFile, point TileManifestPath to the generated tile_manifest.json
4. Use ATiledWorldBuilder in editor to batch-generate per-tile DataAssets with land cover fusion
5. Reference the UTiledLandUseMapDataAsset catalog in your PCG graph — tiles stream automatically

Plugin: GISProcedural

See Plugins/GISProcedural/README.md for detailed algorithm documentation.

Scaling Roadmap (Global Military Simulation)

All 4 phases are implemented. The plugin scales from ~10 km² demos to global coverage:

See PLAN.md for the original collaboration plan.

Development Workflow

1. Prepare GIS data
   overpass-turbo → GeoJSON export
   OpenTopography / SRTM → DEM tiles
   ESA WorldCover → LandCover GeoTIFF (optional, 10m resolution)

2. Run preprocessing pipeline (for tiled mode)
   python Tools/GISPreprocess/preprocess.py --input ./RawData --output Content/GISData/Region_01 --tile-size 1024

3. Editor: Generate & Preview
   Drop AGISWorldBuilder → Set DataSourceType → GenerateInEditor → visual check
   (or ATiledWorldBuilder for tiled batch generation)

4. Editor: Bake to DataAsset
   GenerateAndSaveDataAsset → ULandUseMapDataAsset persisted under /Game/GISData/
   (Tiled mode: TiledWorldBuilder generates per-tile DataAssets automatically)

5. PCG Graph
   GIS Land Use Sampler node → Attribute Filter → Building/Tree/Road spawners
   (Enable bEnableTiling + set LoadRadius for tile-aware sampling)

6. Runtime Streaming
   TiledLandUseMapDataAsset async-loads tile DataAssets on demand

Tech Stack

• Engine: Unreal Engine 5
• PCG: UE5 Procedural Content Generation Framework
• Data Formats: GeoJSON, GeoTIFF, PNG Heightmap (Mapbox Terrain RGB), RAW (.r16/.r32), SRTM HGT
• Online Services: ArcGIS REST Feature Services
• Preprocessing: Python 3.10+, GDAL/OGR, osmium-tool (Phase 1+)
• Projection: WGS84 ↔ UTM (built-in), Simplified Mercator for local regions
• Language: C++ (UE5 plugin), Python (external preprocessing)

License

Internal project — MilSim Team.