Physical AI
Physical AI in FactVerse starts from an operational digital twin that machines and engineering systems can use: model assets, component geometry, source data, layouts, simulation engines, scenario records, and reviewed operating evidence. The module helps teams prepare industrial scenes for layout planning, process validation, robotics context, equipment training, and simulation-driven engineering review.
Use this guide when a factory, facility, construction, logistics, or engineering team needs to turn CAD, BIM, digital twin, and operating data into simulation-ready scenario packages.
Current service boundary
Physical AI is implemented across several runtime surfaces:
| Surface | Current role |
|---|---|
| FactVerse frontend | Provides layout optimization, layout comparison, BIM import, spatial import, sensitivity analysis, simulation playback, simulation comparison, digital twin views, platform asset center, and simulation result pages. |
| Core backend | Provides unified twin identity, model assets, model asset versions, component geometry, twin-model bindings, BIM import replay jobs, building-energy simulation run status, KPI retrieval, layout persistence, and TrafficOps simulation lifecycle APIs. |
| AI Engine | Provides layout optimization, layout evaluation, layout validation, floor plan upload, CAD and DXF import, DES simulation, what-if analysis, Monte Carlo simulation, system dynamics, surrogate modeling, evolutionary optimization, and EnergyPlus-oriented building simulation endpoints. |
| SimRunner | Runs FactVerse cloud or local digital twin scenes from the command line, applies property overrides, records .digrec simulation output, and exports AI-readable Semantic.Scene JSON for downstream review. |
| DFS | Prepares source data, connector mappings, data quality checks, governed datasets, and operational constraints used by simulation and validation workflows. |
| External runtime adapters | Project integrations can hand prepared scenes and asset metadata to Omniverse, Isaac, PhysX, Newton, robotics stacks, or other approved rendering and physics runtimes when the project runtime is configured. |
This boundary makes FactVerse responsible for scene preparation, data governance, asset identity, scenario packaging, and review evidence. Runtime-specific physics fidelity, robot controller behavior, and field safety acceptance are handled by the project engineering workflow that owns the target runtime.
What users can do today
The current application exposes these user workflows:
| Workflow | What users can review or perform |
|---|---|
| Digital twin context | Review digital twin views, platform asset inventory, model assets, model asset versions, component geometry, and twin-model bindings. |
| BIM and CAD intake | Replay BIM and IFC import results into native model, resource, twin, component, and external binding records; import CAD or DXF-derived layouts into simulation scenarios. |
| Layout optimization | Optimize, evaluate, validate, save, compare, auto-arrange, and recommend layout variants; upload floor plans and run sensitivity analysis. |
| DES simulation | List scene types and templates, create custom scenes, run single or batch DES simulations, inspect recent runs, and export run results. |
| What-if analysis | Compare baseline and changed scenarios across registered engines such as DES, ABM, Monte Carlo, system dynamics, network analysis, evolutionary optimization, Bayesian optimization, DOE, and module-specific optimization. |
| Building simulation | Extract KPIs, build IDF input, queue simulation runs, review run status, and retrieve KPI payloads for EnergyPlus-oriented workflows. |
| Scenario recording | Run FactVerse scenes through SimRunner, apply property overrides, create .digrec outputs, and export semantic scene JSON for AI-assisted review. |
| AI-assisted preparation | Use surrogate models, evolutionary search, Monte Carlo risk analysis, system dynamics, and report generation to turn simulation results into review evidence. |
Before you start
Prepare the scene, data, and ownership model before using the workflow:
| Requirement | Notes |
|---|---|
| Tenant context | Model assets, twins, simulation runs, and replay jobs should resolve under the intended tenant. |
| Permissions | Layout editing, simulation run access, reports, platform asset access, model asset management, and BIM replay use separate platform scopes. |
| Asset identity | Equipment, spaces, model assets, component geometry, source files, and digital twin objects need stable IDs before simulation records can be reused. |
| Coordinate system | CAD, BIM, layout, and digital twin objects should use consistent units, origin, orientation, and floor or zone references. |
| Source data | Process limits, production rules, task sequences, meter readings, inspection evidence, and field observations should be prepared through DFS or another governed integration path. |
| Scenario owner | Assign an engineering owner for assumptions, runtime selection, result review, and field validation. |
| Runtime adapter | Confirm which runtime consumes the scenario package: FactVerse SimRunner, EnergyPlus-oriented building simulation, DES, Omniverse, Isaac, PhysX, Newton, robotics middleware, or another approved environment. |
For data preparation, start with Getting Started with DFS, Create an AI Agent-Ready Dataset, and DFS Mapping Fields.
Open Physical AI workflows
Open the FactVerse application and use these current entry points:
| View | Route |
|---|---|
| Digital twin | /digital-twin |
| Platform asset center | /platform/asset-center |
| Layout optimizer | /layout-optimizer |
| Layout compare | /layout-compare |
| BIM import | /bim-import |
| Simulation run result | /sim/runs/:id |
| Spatial import | /spatial/import |
| Sensitivity analysis | /sensitivity |
| Simulation playback | /playback |
| Simulation comparison | /comparison |
| Report generator | /reports |
| DOE lab | /doe |
| TrafficOps simulation | /modules/trafficops/sim |
| FMS simulation | /fms/sim |
Tenant module availability controls which industry simulation pages appear in navigation.
Build simulation-ready asset packages
Start with the platform asset layer:
- Register the model asset with source format, source file reference, metadata, and optional BigSpace context.
- Add a model asset version when geometry, source data, or component extraction changes.
- Register component geometry for addressable parts that need separate binding, analysis, or simulation behavior.
- Bind model assets, versions, and component geometry to the corresponding unified twin objects.
- Replay BIM or IFC import output into native ResourceRef, ModelAsset, ComponentGeometry, ExternalBinding, and twin foundation records when a parsed import needs to become platform data.
- Keep coordinate system, units, source version, checksum, and component references visible in the scenario package.
For Physical AI work, treat the asset package as the bridge between visual digital twin content and simulation or robotics consumption. A reusable package should include model identity, version, geometry, semantic metadata, coordinate assumptions, physical parameters, and validation notes.
Optimize layout and process flow
Use layout optimization when a team needs to compare physical arrangement, routing, capacity, or process options:
| Action | Current support |
|---|---|
| Optimize layout | /ai/layout/optimize runs objective-based layout optimization with movable elements, adjustable capacities, simulation time, population size, generation count, and replication settings. |
| Evaluate layout | /ai/layout/evaluate runs a single layout through the configured simulation scene type and returns KPI output. |
| Validate constraints | /ai/layout/validate checks layout bounds, overlap, route, and element constraints. |
| Analyze space | /ai/layout/analyze-space returns utilization and spatial review data. |
| Save and reuse | /ai/layout/save, /ai/layout/saved, and /api/v1/layouts support saved and persisted layouts. |
| Sensitivity | /ai/layout/sensitivity tests response metrics against changed layout inputs. |
| Auto-arrange | /ai/layout/auto-arrange creates an arranged candidate from the provided layout context. |
| Recommend changes | /ai/layout/recommend returns improvement suggestions from the evaluated layout. |
| Compare layouts | /ai/layout/compare compares candidate layouts. |
| Floor plan upload | /ai/layout/floor-plan/upload stores floor plan material for later layout work. |
A layout study should preserve the baseline, every candidate, objective weights, changed parameters, run duration, and reviewed KPIs. Use layout output as an engineering decision input together with safety zones, operator movement, equipment access, and field constraints.
Run simulation and what-if studies
Physical AI studies can combine several simulation and analysis engines:
| Engine area | Current support |
|---|---|
| DES | /ai/des/run, /ai/des/batch-run, scene templates, custom scenes, recent runs, run details, and run export. |
| CAD import | /ai/cad/import, /ai/cad/simulate, /ai/cad/import-dxf, /ai/cad/scan-dxf, /ai/cad/import-json, async import jobs, and saved layout records. |
| Building simulation | /ai/sim/extract-kpis, /ai/sim/build-idf, /ai/sim/run, plus backend /api/v1/sim/runs status and KPI endpoints. |
| Unified what-if | /ai/whatif/run, /ai/whatif/engines, /ai/whatif/batch, async submit, job status, and cancel endpoints. |
| Monte Carlo | /ai/montecarlo/run and /ai/montecarlo/risk for stochastic output distribution and risk quantification. |
| System dynamics | /ai/sd/simulate and /ai/sd/policy for stock-and-flow projection and policy comparison. |
| Evolutionary optimization | /ai/evolve/optimize and /ai/evolve/layout for search over design variables or layout candidates. |
| Surrogate modeling | /ai/surrogate/train, /ai/surrogate/predict, model listing, and model deletion for fast approximation of expensive simulations. |
| TrafficOps simulation | /api/v1/trafficops/simulations/runs and /api/v1/trafficops/simulations/runs/{runId}/result for traffic and checkpoint-oriented runs. |
Choose the engine that matches the question being reviewed: DES for flow and queue behavior, EnergyPlus-oriented simulation for building-energy inputs, Monte Carlo for uncertainty, system dynamics for long-horizon capacity, and surrogate modeling for faster exploration after enough sampled runs exist.
Use SimRunner for scene execution
SimRunner is the batch execution path for FactVerse digital twin scenes:
| Capability | Current support |
|---|---|
| Scene source | Pull a cloud scene by path or run a local scene JSON in offline mode. |
| Configuration | Read a JSON config file and apply repeated --override values with dot-path keys. |
| Batch execution | Login, prepare preset, apply property overrides, run the scene, and record output from a command-line workflow. |
| Record output | Write .digrec simulation output for replay or downstream inspection. |
| Dry run | Validate config, scene retrieval, and override application before starting the simulation run. |
| Semantic export | Export Semantic.Scene JSON after scene retrieval for AI-assisted scene understanding and scenario planning. |
| Protocol output | Print the embedded semantic scene protocol for downstream AI consumers. |
| Runtime platform | Windows x64 and Linux x64 runtime paths are supported in the current runner package. |
Use SimRunner when an engineering team needs repeatable scene execution, scheduled batch studies, regression-style simulation checks, or AI-readable scene export. The output should be stored with the scenario ID, scene source, overrides, run duration, runner version, output file path, and review owner.
Hand off to external physics and robotics runtimes
When a project uses Omniverse, Isaac, PhysX, Newton, robotics middleware, or another physics runtime, prepare a handoff package from FactVerse:
| Package item | Purpose |
|---|---|
| Scene identity | Tenant, scene path or scene ID, model asset IDs, model asset version IDs, and source file references. |
| Geometry and semantics | Component geometry, coordinate system, units, material hints, articulation hints, collision assumptions, and semantic labels. |
| Operating constraints | Task sequence, process rules, safety zones, speed limits, load limits, environment conditions, and validation references. |
| Runtime parameters | Target runtime, adapter version, physics settings, random seed, run duration, and output format. |
| Review evidence | Simulation results, KPI tables, .digrec references, semantic scene JSON, screenshots or render outputs, and reviewer notes. |
FactVerse supplies stable identity, governed data, scenario structure, and review records. The runtime adapter supplies engine-specific conversion, simulation execution, and controller integration for the target project.
API surface
Physical AI uses several API groups:
| Group | Example endpoints |
|---|---|
| Model assets | /api/v1/model-assets, /api/v1/model-assets/{modelAssetId}/versions, /api/v1/model-assets/{modelAssetId}/components, /api/v1/model-assets/{modelAssetId}/components/{componentGeometryId}/geometry, /api/v1/twin-model-bindings |
| BIM replay | /api/v1/platform/bim-imports:replay, /api/v1/platform/bim-imports/replay-jobs, /api/v1/platform/bim-imports/replay-jobs/{jobId}, /api/v1/platform/bim-imports/replay-jobs/{jobId}:retry |
| Building simulation runs | /api/v1/sim/runs, /api/v1/sim/runs/{id}, /api/v1/sim/runs/{id}/kpis |
| TrafficOps simulation | /api/v1/trafficops/simulations, /api/v1/trafficops/simulations/runs, /api/v1/trafficops/simulations/runs/{runId}, /api/v1/trafficops/simulations/runs/{runId}/result |
| Layout AI | /ai/layout/optimize, /ai/layout/evaluate, /ai/layout/validate, /ai/layout/templates, /ai/layout/analyze-space, /ai/layout/sensitivity, /ai/layout/auto-arrange, /ai/layout/recommend, /ai/layout/compare |
| CAD and layout import | /ai/cad/import, /ai/cad/simulate, /ai/cad/import-dxf, /ai/cad/scan-dxf, /ai/cad/import-dxf-selective, /ai/cad/import-json, /ai/cad/import-json-async, /ai/cad/import-dxf-async, /ai/cad/jobs/{job_id} |
| DES | /ai/des/entity-types, /ai/des/scene-types, /ai/des/dag-scenes, /ai/des/templates, /ai/des/custom-scenes, /ai/des/batch-run, /ai/des/runs, /ai/des/runs/{run_id}, /ai/des/runs/{run_id}/export, /ai/des/run |
| Building AI simulation | /ai/sim/extract-kpis, /ai/sim/build-idf, /ai/sim/run |
| What-if | /ai/whatif/run, /ai/whatif/engines, /ai/whatif/batch, /ai/whatif/submit, /ai/whatif/batch/submit, /ai/whatif/job/{job_id}, /ai/whatif/job/{job_id}/cancel |
| Analysis engines | /ai/montecarlo/run, /ai/montecarlo/risk, /ai/sd/simulate, /ai/sd/policy, /ai/evolve/optimize, /ai/evolve/layout, /ai/surrogate/train, /ai/surrogate/predict, /ai/surrogate/models |
Validation checklist
Before using Physical AI output in an engineering review, robot-training workflow, process planning session, or field trial, confirm:
- tenant context, permission scopes, and scenario owner are clear;
- scene ID, model asset ID, model asset version ID, and component geometry references are recorded;
- coordinate system, units, scale, and floor or zone references match the target site;
- physics parameters, collision assumptions, articulation hints, and material metadata are reviewed;
- source data, process limits, task sequence, and field observations are linked to governed records;
- simulation engine, runtime adapter, run duration, random seed, and changed parameters are captured;
- outputs such as KPIs,
.digrec, semantic scene JSON, and review notes are stored with the scenario package; - field validation and safety acceptance are handled by the responsible project engineering owner.