How jansen-liang/graphworld works
GraphWorld is a platform for building interactive building-scale scene assets, running embodied-agent simulations, and evaluating task performance through web and runtime adapters. The principal workflow is scene preparation, agent or human action, world-state evolution, and replay/metrics output. The graph preserves the API, scene/run services, generation, runtime, and evaluation branches. Evidence is limited to the README and the supplied source excerpts and index; unsampled connections are omitted or marked unverified.
Main entry points: run_service.py Run service · graph.py World graph · App.tsx Web application · scene_service.py Scene service
Components 24
Web client
- Web application
frontend/web/src/app/App.tsx - Scene tools
frontend/web/src/features/scene-builder/SceneBuilderPage.tsx - Run and replay UI
frontend/web/src/features/run-monitor/RunDetailPage.tsx
Platform API
- HTTP API
backend/app/main.py - Scene service
backend/app/services/scene_service.py - Scene importer
backend/app/runtime/scene_importer.py - Run service
backend/app/services/run_service.py - Live simulation
backend/app/services/simulation_service.py - Persistent records
backend/app/db/models
World model
- Scene generation
backend/generation - Nodes and state
backend/core/node.py - Relations
backend/core/edge.py - Action model
backend/core/action.py
Simulation runtime
- World graph
backend/runtime/world/graph.py - Action resolution
backend/runtime/action_engine.py - Action execution
backend/runtime/action_executor.py - Input adapter
backend/runtime/input_adapter.py - Scene preparation
backend/runtime/scene_preparation.py - Runtime adapters
backend/adapter
Agent and evaluation
- Agent decision
backend/runtime/agent/decision.py - Agent planning
backend/runtime/agent/planning.py - Task evaluation
backend/runtime/eval/matrix_evaluator.py - Replay and metrics
backend/experiments/episode_records.py
Other components
- Researcher
Connections 26
- Researcher → Web application: uses
- Web application → Scene tools: routes to
frontend/web/src/app/App.tsx - Web application → Run and replay UI: routes to
frontend/web/src/app/App.tsx - Scene service → Scene importer: imports
backend/app/services/scene_service.py - Run service → Persistent records: uses records
backend/app/services/run_service.py - Run service → Scene service: uses repository
backend/app/services/run_service.py - Run service → Agent decision: chooses actions
backend/app/services/run_service.py - Run service → Input adapter: uses
backend/app/services/run_service.py - Scene service → Scene preparation: prepares scenes
backend/app/services/scene_service.py - Live simulation → Input adapter: resolves input
backend/app/services/simulation_service.py - Live simulation → Action execution: executes actions
backend/app/services/simulation_service.py - Scene importer → Nodes and state: imports schema
backend/app/runtime/scene_importer.py - Scene generation → Nodes and state: creates assets
backend/generation/symbolic_scene_generator.py - Scene generation → Relations: defines relations
backend/generation/ontology/model.py - Input adapter → Action resolution: resolves input
- Action resolution → Action model: binds actions
backend/runtime/action_engine.py - Action execution → Action resolution: delegates resolution
backend/runtime/action_executor.py - Action execution → World graph: mutates world
backend/runtime/action_executor.py - World graph → Nodes and state: stores nodes
backend/runtime/world/graph.py - World graph → Relations: stores relations
backend/runtime/world/graph.py - Agent decision → Agent planning: uses planning
backend/runtime/agent/decision.py - Run service → Task evaluation: evaluates runs
backend/app/services/run_service.py - Replay and metrics → Task evaluation: records evaluation
backend/experiments/episode_records.py - World graph → Runtime adapters: exports snapshots
backend/runtime/world/graph.py - Runtime adapters → Web application: renders world
- Run service → Replay and metrics: emits replay and metrics