Physical AI Tools
Physical AI workflows combine digital twin scene context, simulation-ready asset packages, operational constraints, and engineering review records. The AI tool surface helps agents and product workflows prepare layouts, import CAD and floor plans, run simulations, compare what-if scenarios, train surrogate models, and package results for downstream runtime use.
Use this page when an Agent workflow or backend integration needs a practical map of the tools behind scene preparation, simulation, optimization, and review handoff.
Tool layers
| Layer | Primary users | Access boundary | Output type |
|---|---|---|---|
/mcp/base/ MCP tools | Agent clients, copilots, workflow orchestrators | base.read, base.compute.run, and selected base.action.write scopes | Scene context, source data, simulation summaries, optimization results, reports, review records |
/ai/layout/* AI Engine endpoints | Layout optimizer, engineering studies, planning workflows | Product service authentication and layout permissions | Optimized layouts, evaluation KPIs, validation errors, sensitivity output, recommendations |
/ai/cad/* and /api/v1/model-assets/* | CAD, BIM, and model asset intake workflows | Product service authentication and asset permissions | Imported geometry, layout records, model asset versions, component geometry, twin bindings |
/ai/des/*, /ai/whatif/*, /ai/sim/*, and analysis engines | Simulation and scenario workflows | Product service authentication and simulation permissions | DES runs, what-if comparisons, EnergyPlus-oriented simulation, Monte Carlo, system dynamics, evolutionary optimization, surrogate output |
| SimRunner | Engineering automation and batch scene execution | Project runtime credentials and scene access | .digrec records, semantic scene JSON, run logs, scenario package artifacts |
Use MCP tools for governed Agent orchestration. Use AI Engine endpoints for controlled simulation, layout, analysis, and model workflows inside the product runtime.
MCP endpoint and scopes
| Endpoint | Scope | Use |
|---|---|---|
/mcp/base/ | base.read | Read scene records, documents, connector state, data quality, equipment status, and supporting evidence. |
/mcp/base/ | base.compute.run | Run approved simulation, optimization, spatial analysis, forecasting, report generation, and model workflows. |
/mcp/base/ | base.action.write | Store approved work-order or action records when the scenario review requires an operational follow-up. |
Most Physical AI Agent workflows start with base.read and base.compute.run. Add write scope only when approved review output must become an operational record.
MCP tool map
Scene and source-data intake
| Tool | Primary job | Typical question |
|---|---|---|
import_dxf | Import a DXF floor plan and recognize walls, doors, windows, and fences. | "Can this floor plan become a simulation layout?" |
import_data | Import external REST or CSV records for analysis. | "Load this line history as scenario input." |
list_connectors | List configured source connectors and current sync status. | "Which source systems can feed this scenario?" |
check_data_quality | Review completeness, accuracy, consistency, timeliness, and violations. | "Is this scenario input ready?" |
troubleshoot_connector | Diagnose connector errors and sync logs. | "Why did the process data stop updating?" |
get_equipment_documents | Retrieve manuals, drawings, SOPs, and maintenance records. | "Which equipment constraints apply to this scenario?" |
search_documents | Search ECM documents by keyword, type, or related entity. | "Find the process guide for this station." |
query_knowledge | Query equipment, failure modes, repair actions, rules, and schedules. | "Which operating constraints should be attached?" |
Simulation and scenario analysis
| Tool | Primary job | Typical question |
|---|---|---|
run_des | Run a discrete event simulation for process or queue modelling. | "What is the throughput for this process layout?" |
run_dag_simulation | Run a routed DES with paths, conditions, and Sankey-style flow output. | "Where does flow accumulate in this multi-path scenario?" |
run_abm | Run an agent-based crowd simulation. | "How do people move through this area?" |
run_simulation | Run a registered module simulation such as traffic, heating, or equipment lifecycle scenes. | "Compare this scenario with the baseline." |
cascade_simulation | Chain simulation engines across DES, ABM, and Monte Carlo. | "How does uncertainty affect this process plan?" |
run_montecarlo | Run stochastic risk or stress testing. | "What is the distribution of possible outcomes?" |
run_system_dynamics | Run stock-and-flow simulation for long-horizon behavior. | "How does capacity evolve under this policy?" |
run_doe | Run design of experiments and factor significance analysis. | "Which factors matter most?" |
simulate_logistics | Simulate AGV or forklift logistics on a facility layout. | "Can this warehouse flow support the planned workload?" |
Optimization and model acceleration
| Tool | Primary job | Typical question |
|---|---|---|
optimize_layout | Optimize facility layout using multi-objective search with DES evaluation. | "Which layout balances throughput and wait time?" |
run_optimization | Find parameter sets with multi-objective optimization. | "Which configuration gives the best tradeoff?" |
optimize_evolutionary | Run evolutionary multi-objective optimization. | "Explore design variables across a larger search space." |
optimize_bayesian | Tune black-box functions with Bayesian optimization. | "Find a strong candidate with fewer simulation runs." |
optimize_milp | Solve a mixed-integer linear programming problem. | "Find a feasible assignment plan under constraints." |
train_surrogate | Train a fast surrogate model from simulation or measurement data. | "Can we approximate this expensive simulation?" |
predict_surrogate | Run inference with a trained surrogate model. | "Estimate KPI output for this new candidate quickly." |
recommend_model | Recommend a model type for available data. | "Which model family fits this dataset?" |
automl_forecast | Select a forecasting model automatically. | "Forecast this process or load signal." |
conformal_predict | Produce distribution-free prediction intervals. | "What uncertainty range should we show?" |
explain_prediction | Explain a model prediction with SHAP-style evidence. | "Why did this scenario score change?" |
detect_drift | Detect data or concept drift. | "Has the scenario input changed from the training set?" |
estimate_causal_effect | Estimate treatment effects. | "What was the impact of this operating change?" |
find_optimal_policy | Find a policy using causal inference. | "Which policy should be reviewed for rollout?" |
Spatial review
| Tool | Primary job | Typical question |
|---|---|---|
analyze_spatial_anomaly | Detect unusual spatial sensor values. | "Where are abnormal readings clustered?" |
compare_zones | Compare statistics between zones or floors. | "Which zone has higher load or traffic?" |
recommend_sensor_placement | Recommend locations for additional sensors. | "Where is coverage weak?" |
find_path | Find a path between locations in a building. | "Can a worker or vehicle reach the station safely?" |
generate_report | Generate a simulation or scenario report. | "Create a scenario review package." |
AI Engine tool map
Layout and planning
| Endpoint | Purpose | Inputs to prepare | Output to review |
|---|---|---|---|
/ai/layout/optimize | Optimize a layout against objectives and constraints. | Baseline layout, movable elements, constraints, objective weights, run settings. | Candidate layouts, Pareto output, KPI tradeoffs, constraint notes. |
/ai/layout/evaluate | Evaluate one layout through a configured scene type. | Layout elements, scenario type, time horizon, replications. | KPI output, bottleneck or risk notes, run metadata. |
/ai/layout/validate | Validate layout bounds, overlap, route, and element constraints. | Layout geometry, rules, floor or zone bounds. | Validation issues and repair list. |
/ai/layout/templates | List reusable layout templates. | Scenario category or module context. | Template IDs and default parameters. |
/ai/layout/analyze-space | Analyze space utilization and spatial metrics. | Floor plan, zones, assets, occupancy or flow inputs. | Utilization, density, and spatial review summary. |
/ai/layout/sensitivity | Test response metrics against changed inputs. | Baseline layout, factor ranges, target KPIs. | Sensitivity results and factor ranking. |
/ai/layout/auto-arrange | Create an arranged candidate from supplied context. | Floor plan, assets, spacing rules, constraints. | Candidate arrangement and validation notes. |
/ai/layout/recommend | Recommend improvement ideas for a layout. | Evaluated layout, KPI output, constraints. | Suggested changes and expected impact. |
/ai/layout/compare | Compare candidate layouts. | Baseline and candidate layouts, shared KPI definitions. | Side-by-side metrics and review summary. |
/ai/layout/floor-plan/upload | Store floor plan material for layout workflows. | Floor plan file, units, origin, site context. | Upload record and parse status. |
CAD, DES, and simulation
| Endpoint group | Purpose | Inputs to prepare | Output to review |
|---|---|---|---|
/ai/cad/import, /ai/cad/import-dxf, /ai/cad/import-json | Import CAD, DXF, or JSON layouts. | Source file, units, coordinate assumptions, layer mapping, tenant context. | Imported layout, recognized entities, warnings, saved layout references. |
/ai/cad/scan-dxf and /ai/cad/import-dxf-selective | Inspect DXF content and selectively import chosen layers or objects. | DXF file, layer selection, object filters. | Scan result, selected import output, skipped object list. |
/ai/cad/import-json-async, /ai/cad/import-dxf-async, /ai/cad/jobs/{job_id} | Run and track async CAD import jobs. | Source file, import profile, job metadata. | Job state, parsed output, errors, and saved records. |
/ai/cad/simulate | Run a simulation from CAD-derived layout. | Imported layout, process assumptions, scenario parameters. | Simulation result and KPI summary. |
/ai/des/run, /ai/des/batch-run, /ai/des/runs, /ai/des/runs/{run_id}, /ai/des/runs/{run_id}/export | Run and retrieve DES scenarios. | Scene type, entities, resources, routes, service times, run settings. | Throughput, wait time, bottlenecks, utilization, exported result. |
/ai/des/entity-types, /ai/des/scene-types, /ai/des/dag-scenes, /ai/des/templates, /ai/des/custom-scenes | Discover and configure DES scene definitions. | Scenario category, template requirements, custom scene payload. | Available scene types, templates, and custom scene records. |
/ai/sim/extract-kpis, /ai/sim/build-idf, /ai/sim/run | Prepare and run EnergyPlus-oriented building simulation. | Geometry, zones, materials, schedules, weather, energy assumptions. | KPI schema, IDF payload, run status, simulation outputs. |
/ai/whatif/run, /ai/whatif/engines, /ai/whatif/batch, /ai/whatif/submit, /ai/whatif/batch/submit, /ai/whatif/job/{job_id} | Run sync or async what-if comparisons. | Baseline, changed parameters, engine choice, batch settings. | Scenario comparison, job status, output metrics, cancellation state when requested. |
Analysis engines
| Endpoint | Purpose | Inputs to prepare | Output to review |
|---|---|---|---|
/ai/montecarlo/run | Run stochastic simulation over uncertain variables. | Input distributions, run count, target metrics, random seed. | Output distribution, percentiles, probability of threshold crossing. |
/ai/montecarlo/risk | Summarize risk from stochastic results. | Monte Carlo output, risk thresholds, business context. | Risk categories and review notes. |
/ai/sd/simulate | Run system dynamics simulation. | Stocks, flows, parameters, horizon, timestep. | Time-series output and state trajectories. |
/ai/sd/policy | Compare policy changes in a system dynamics model. | Baseline model, policy changes, KPI definitions. | Policy comparison and scenario notes. |
/ai/evolve/optimize | Run evolutionary optimization over design variables. | Variables, bounds, objectives, constraints, population settings. | Candidate set, objective scores, constraint results. |
/ai/evolve/layout | Run evolutionary layout search. | Layout candidates, constraints, objective functions. | Layout candidates and tradeoff summary. |
/ai/surrogate/train | Train a surrogate model for faster scenario evaluation. | Sampled simulation runs, input features, target KPIs. | Model ID, training metrics, validation summary. |
/ai/surrogate/predict | Estimate KPI output with a trained surrogate model. | Model ID, candidate inputs, feature mapping. | Predicted KPIs, confidence indicators, model reference. |
/ai/surrogate/models | List trained surrogate models. | Optional project or scenario context. | Model catalog and metadata. |
Recommended tool sequences
Simulation-ready asset package
import_dxf or CAD import endpoint
-> check_data_quality for source records
-> search_documents and get_equipment_documents for constraints
-> layout validation and asset-readiness review
-> generate_report with scene, asset, and assumption references
Use this sequence when a project needs a reusable package with geometry, semantics, physical assumptions, operating constraints, and reviewer notes.
Layout and process study
/ai/layout/validate
-> /ai/layout/evaluate for baseline
-> /ai/layout/optimize or optimize_layout
-> /ai/layout/compare
-> reviewed engineering summary
Use this sequence when comparing layout candidates, line balance, routing, or process options.
What-if and uncertainty review
run_des or /ai/des/run
-> /ai/whatif/run for changed scenarios
-> run_montecarlo or /ai/montecarlo/run for uncertainty
-> generate_report for review
Use this sequence when the team needs a structured comparison of baseline, candidate, and uncertain outcomes.
Surrogate-assisted iteration
DES, EnergyPlus-oriented, or external simulation samples
-> /ai/surrogate/train
-> /ai/surrogate/predict for fast candidate screening
-> selected high-value scenarios return to the original simulation engine
Use this sequence when a slower simulation must support many design iterations. Keep the surrogate model tied to the sampled dataset, feature schema, and validation metrics.
Runtime handoff
scene and model asset IDs
-> simulation-ready asset review
-> SimRunner or target runtime adapter
-> semantic scene JSON, run output, and validation notes
-> approved handoff package
Use this sequence for Omniverse, Isaac, PhysX, Newton, robotics middleware, or other project runtime handoff.
Data requirements
Physical AI AI tools are strongest when the scenario package includes:
- tenant, scene, model asset, model asset version, and component geometry IDs;
- source file reference, checksum, source format, coordinate system, units, origin, floor, and zone references;
- semantic labels, material hints, collision assumptions, articulations, constraints, motion ranges, and safety zones;
- process route, task sequence, service time, demand profile, operating limits, and exception rules;
- sensor, inspection, work-order, production, meter, or field-observation records prepared through DFS;
- simulation engine, runtime adapter, version, seed, run duration, timestep, and changed parameters;
- output references such as KPIs,
.digrec, semantic scene JSON, rendered output, and reviewer notes; - validation owner and reuse target such as process planning, robot training, engineering review, or downstream application handoff.
Use Create an AI Agent-Ready Dataset and DFS Mapping Fields before workflows that depend on governed scenario input.
Output review checklist
- Scene, model asset, model asset version, and component geometry references are visible.
- Coordinate system, units, scale, and floor or zone mapping are explicit.
- Runtime, engine, version, run duration, random seed, and changed parameters are recorded.
- Simulation assumptions, simplifications, and missing physical parameters are visible.
- Layout output includes constraint status and a baseline comparison.
- Surrogate output includes model ID, feature schema, training dataset, and validation metrics.
- Robot-training or external-runtime output is reviewed by the engineering owner before reuse.
- Field validation notes and corrections are captured for later scene and model improvement.