IAWG turns a natural-language scene request into a physically stable USD scene. It is not a chat interface bolted onto an asset browser: scene intent, asset metadata, spatial solving, physics settling and deterministic composition are each a reusable, testable, deployable capability.
What it does
IAWG world generator walkthrough
Capabilities
| Area | What you get |
|---|---|
| Natural-language scene generation | A user request is parsed into a structured, executable scene intent |
| Asset matching | Semantic retrieval against an indexed asset library, with dimensional and category constraints |
| Spatial solving | Placement that respects support surfaces, clearance, reachability and task-specific constraints |
| Physics settling | Objects are simulated to rest, so nothing floats or interpenetrates when the scene loads |
| Deterministic composition | The same request and seed produce the same USD scene, which is what makes a generated scene usable as a test fixture |
Why determinism matters
A generated scene is only useful for evaluation if it can be regenerated exactly. IAWG treats the scene as a reproducible artefact: the request, the seed and the asset library version fully determine the output. That is what lets a failure case discovered in week one be replayed unchanged in week twenty, after the algorithm, the asset library and the engine have all moved on.
Where it fits
Embodied intelligence and manipulation teams that need hundreds of task-relevant environments rather than a handful of hand-built ones. Instead of an artist authoring each layout, the team describes the task distribution and the generator produces scenes that satisfy it, each one physically valid and ready to enter the data or evaluation pipeline.
Deliverables
- A scene intent schema matched to your task family
- Asset library indexing and retrieval against your existing content
- The generation service, producing USD scenes on request
- Determinism and physics-validity checks wired into your CI
Talk to us
Tell us the task family and the asset library you already have, and we will scope what generation coverage is achievable.