Physical AI Workflow Guide

Use this guide to build a Physical AI workflow with FactVerse AI Agent, digital twin scenes, simulation-ready asset packages, and simulation services. The workflow helps engineering teams prepare scenes, run approved simulation or rendering tasks, review assumptions, and reuse validated results for process planning, robot training, and virtual operation checks.

Prerequisites

Requirement	Details
Scene context	Designer scene, factory or process layout, equipment placement, asset metadata, and version history.
Simulation-ready assets	Geometry, material, collision, motion, articulation, and behavior metadata required by the target simulation workflow.
Operational context	Production constraints, operating envelope, safety zones, task sequence, source data, and real-world observations.
Simulation runtime	Enabled FactVerse simulation services, SimRunner, Omniverse, PhysX, Newton, Isaac, or other approved simulation and rendering services for the project environment.
Validation owner	Engineering owner who can accept assumptions, compare simulation output with field evidence, and approve reuse.

Endpoint and scopes

Use the current base and module endpoints for Physical AI workflows. The base endpoint provides shared scene, asset, compute, and action-draft access; module endpoints add operational signals that constrain the scenario.

Endpoint	Scope	Use
/mcp/base/	base.read	Read scene records, asset metadata, documents, validation notes, and operational context.
/mcp/base/	base.compute.run	Run approved analysis, simulation preparation, or result summarization tasks.
Module endpoints	Module read scopes	Bring in domain data such as facility operations, predictive maintenance, or industry-specific signals when enabled.
/mcp/base/	base.action.write	Store approved scenario records, review notes, or task drafts after human approval.

Workflow steps

1. Define the physical task: layout review, process planning, robot training context, packaging flow, or virtual operation check.
2. Prepare the scene: confirm scene version, model asset IDs, model asset version IDs, component geometry, coordinate system, geometry scale, and equipment placement.
3. Check asset readiness: verify collision shapes, materials, articulations, constraints, motion ranges, and required metadata.
4. Bind operational data: attach process limits, task sequence, safety zones, historical observations, source data, and validation references.
5. Run approved compute: use the enabled FactVerse simulation, SimRunner, or project runtime to generate trajectories, interactions, measurements, KPIs, or scene outputs.
6. Review assumptions: record model simplifications, expected error range, missing parameters, and differences from field behavior.
7. Reuse validated context: export approved results to engineering review, robot training, process planning, or downstream applications.

Product surfaces

Surface	Use in a Physical AI workflow
Physical AI module	Code-backed reference for model assets, BIM replay, layout optimization, simulation engines, SimRunner, and runtime handoff.
DFS mappings	Map units, ranges, topology, physics classification, source IDs, and operating constraints.
DFS Pro datasets	Prepare governed scenario input data and reusable records for review.
Workflow run record	Store scenario ID, assumptions, output references, validation notes, and reuse target.

DFS setup for operational constraints

Physical AI workflows often need current operating constraints, process limits, environment readings, and historical observations alongside the Designer scene.

Data need	DFS workflow
Source signals and process data	DFS Lite Connectors
Units, ranges, topology, and physics classification	Mapping Fields
Governed scenario input data	DFS Pro Datasets
Repeatable data preparation	Governance Studio
Agent-readable scenario evidence	Create an AI Agent-Ready Dataset

For tool selection, endpoint groups, and recommended Agent sequences, see Physical AI Tools.

Recommended output structure

Section	Content
Scenario package	Scene version, asset versions, task goal, operating assumptions, and simulation backend.
Asset readiness	Assets ready for simulation, assets needing repair, and missing metadata.
Simulation result	Measurements, interactions, constraints, generated trajectories, or rendered scene output.
Validation notes	Field comparison, accepted assumptions, limitations, and required follow-up tests.
Reuse path	Engineering review, robot training dataset, process planning package, or application handoff.

Engineering guidance

• Keep simulation assumptions visible to reviewers and downstream users.
• Compare generated results with field observations before expanding usage.
• Use faster physics and rendering iteration to explore options, then reserve high-fidelity validation for the cases that need it.
• Treat robot-training outputs as versioned engineering assets with scenario ID, source scene, asset versions, and validation notes.
• Record corrections from field tests so the scene, asset metadata, and workflow prompts improve over time.

Common failure modes

Symptom	Likely cause	Response
Simulation behaves unrealistically	Missing collision, mass, articulation, friction, or constraint data	Return an asset-readiness repair list before running further analysis.
Robot-training context transfers poorly	Scenario assumptions differ from field conditions	Capture the gap, update the scene or asset metadata, and rerun the scenario.
Result reuse is blocked	Scene version, asset version, or backend parameters are missing	Require a scenario package before accepting the result.
Compute action is blocked	Scope or approval policy is missing	Keep the request as a review task until the project owner approves execution.

Validation checklist

• The workflow references scene version, asset version, and simulation backend.
• Each compute result includes assumptions, limitations, and source references.
• Asset-readiness issues are reported before the workflow treats a result as reusable.
• Approved results are versioned for engineering review, robot training, or process-planning reuse.