Simulink Run Real Time Application

Running Simulink Real-Time Applications

Own the run lifecycle for Simulink Real-Time — build the model, install/load the application on the target computer, start and stop it, set its stop time, drive root inports and Playback blocks via tg.Stimulation — plus the first-pass runtime workflows around parameter tuning, parameter sets, recording, file logs, and live streaming.

Release compatibility: Designed and validated for MATLAB R2026a and later. Not recommended for earlier releases: APIs, system target files, and workflows differ.

When to Use

• Building a model with slbuild; installing or loading an .mldatx (install, load)
• Starting (start, with or without options), stopping (stop), or setting stop time (setStopTime)
• Reading run state (isLoaded, isRunning)
• Driving root inports, Playback blocks, or parameter schedules via tg.Stimulation, including reloading data from a timeseries, Dataset, parquet file, or struct
• Reading or writing tunable parameters (getparam, setparam) or one-time signal values (getsignal) on a loaded or running application
• Saving, loading, importing, exporting, or selecting parameter sets
• Recording, importing file logs, exporting SDI runs, or choosing between file logging and live streaming
• Inspecting or removing installed applications (getInstalledApplications, removeApplication, removeAllApplications)

When NOT to Use

If the task crosses one of these lines, STOP and load the named skill first before writing any code, then come back.

• Model configuration — system target file, fixed-step solver, Simulink Real-Time options → simulink-configure-real-time-model. Never patch model configuration here.
• Target definitions and target ops — IP, credentials, RTOS update, connect(tg) troubleshooting, status(tg), reboot(tg), setStartupApp → simulink-configure-real-time-target.
• Signal capture — slrealtime.Instrument, addForFileLog, addForSDI, addForMATLAB → simulink-instrument-real-time-application.
• Model structure edits — adding File Log blocks, logging badges, root inports, Playback blocks, or tunable variables to make signals loggable, streamable, or tunable → the Simulink model-editing workflow or simulink-configure-real-time-model.

Prerequisites

• MATLAB R2026a or newer
• Simulink Real-Time installed (ver lists "Simulink Real-Time")
• A relevant support package installed for the target hardware:
  • Speedgoat hardware → Speedgoat Add-On
  • Linux x86-64 devices → Embedded Coder Support Package for Real-Time Linux (requires Embedded Coder)
• The Simulink model is configured for Simulink Real-Time (see simulink-configure-real-time-model)
• A target computer is configured and reachable on the network (see simulink-configure-real-time-target)

Workflow

Use only the documented forms in references/api-reference.md (every call this skill makes, the tg.Stimulation methods, and the parameter-stimulation setup step) and references/start-options.md (the start named parameters). Dot-form and function-form are equivalent (load(tg, app) = tg.load(app)). start takes the target only — the loaded application is what runs — and AutoImportFileLog on stop is name-value.

1. Confirm Preconditions

Connect to the target and confirm the model is configured for Simulink Real-Time before the first build. Below, tg is the target object and app the .mldatx application name.

matlabCopy code
model = "myModelName";
app   = model;

tg = slrealtime; % Default Target
connect(tg);
assert(isConnected(tg), "Failed to connect to %s", tg.TargetSettings.name);

The model must be resolvable on the MATLAB path before slbuild, but do not require bdIsLoaded(model) for every runtime operation. If the target computer is not configured, route to simulink-configure-real-time-target. Do not patch model or target settings here.

Preflight before the first slbuild — always verify the model is configured. A model's name or slrt_ prefix is not evidence of it.

matlabCopy code
% Model must target Simulink Real-Time. speedgoat.tlc | speedgoat_aarch64.tlc | slrtlinux_x64.tlc
load_system(model);
stf = get_param(model, "SystemTargetFile");
disp(stf);

If stf is not a Simulink Real-Time STF (e.g. grt.tlc, ert.tlc, or empty), STOP and load simulink-configure-real-time-model.

There is no host C compiler to check for — the Speedgoat Add-On ships the cross-compiler slbuild uses. A compiler or toolchain error from slbuild means model configuration or a missing support package. See references/guardrails.md.

2. Build the Model

Build the real-time application from the configured model.

matlabCopy code
slbuild(model);

Produces <model>.mldatx in the build folder, destroying any previous one. Ask first if a current artifact is in use on a running target computer.

On failure, classify before routing: model configuration, install/toolchain, path/environment, custom code/library, or unsupported block. See references/build-and-load-pipeline.md.

3. Inspect Target State

Before any destructive op, read what is on the target computer.

matlabCopy code
disp(isLoaded(tg, app));
disp(isRunning(tg, app));
disp(getInstalledApplications(tg));

If something is already running, surface that to the user before calling load or start — those calls will refuse to clobber a running app.

4. Install or Load the Application

load installs the .mldatx and loads it into the runtime. Use install for upload-only; 'force' re-uploads an artifact the target already has.

matlabCopy code
load(tg, app);
install(tg, app);
install(tg, app, 'force');

Constraints:

• The app's SystemTargetFile must match the target's — load fails with a clear message. Surface it and route to simulink-configure-real-time-model.
• If an app is running, load refuses to overwrite it; stop it first.
• If an app is loaded but not running, load stops it first. Surface this before invoking.

See references/build-and-load-pipeline.md for the full sequencing.

5. Set Stop Time When Needed

Skip this when the model/start settings already cover the intended duration. Use these only for a runtime override.

matlabCopy code
setStopTime(tg, 30);                    % requires loaded; fires StopTimeChanged
start(tg, "StopTime", 30);              % per-run override

setStopTime validates nonnegative scalar and requires the app to be loaded.

6. Start the Application

Run the loaded application on the target computer.

matlabCopy code
start(tg);

start takes the target only — the loaded application is implied — plus 8 optional name-value pairs. See references/start-options.md for the full table.

Common forms:

matlabCopy code
start(tg, "StopTime", 30, "LogLevel", "debug");
start(tg, "StopTime", 30, "StartStimulation", "off");   % finite run, no auto-stimulation
start(tg, "FileLogMaxRuns", 5, "AutoImportFileLog", false);

start auto-connects if the target is not connected, but prefer an explicit connect(tg) so the user sees it happen.

Ask First before a non-default LogLevel or AutoImportFileLog=false — see Guardrails.

7. Verify Run State

Confirm the application is actually running before reporting success.

matlabCopy code
assert(isRunning(tg, app), "Application failed to start.");

Read state back after every change. For Stimulation runs, also check tg.Stimulation.getStatus("all").

8. Drive Stimulation When Needed

Stimulation is optional — use it for scheduled root-inport data, Playback block data, or time-varying parameter schedules. Not for a one-time parameter change; use setparam.

Every Stimulation method requires the target connected and the application loaded. For selector forms, reloadData shapes, accepted sources, and parameter-entry registration, see references/stimulation.md.

9. Reload Stimulation Data When Needed

Reload only to replace the scheduled source for an inport, Playback block, or parameter schedule. The affected entry must be stopped first. For parameters, reloadData is also the registration step that makes the parameter stimulable.

10. Monitor, Tune, and Log During Runs

Choose the data path by intent: getparam/setparam for one-off tunable reads and writes; getsignal for one-off signal reads; parameter sets for saving or selecting groups of values between runs; Stimulation for time-varying inputs or schedules; streaming for live host-side monitoring; File Log blocks or file-log instruments for high-rate or long-duration post-run data; startRecording/stopRecording for separate recording intervals and separate imported runs.

matlabCopy code
value = getparam(tg, "myModel/Gain", "Gain");
setparam(tg, "myModel/Gain", "Gain", 3.5);
signalValue = getsignal(tg, "myModel/OutBlock", 1);

saveParamSet(tg, "baseline");
loadParamSet(tg, "baseline");

startRecording(tg);
stopRecording(tg);
tg.FileLog.import(app);

See references/monitoring-tuning-logging.md for parameter sets, recording controls, and file-log import/export; references/signal-access-and-streaming.md for getsignal, model-reference paths, streaming overload, and decimation. Route instrument setup to simulink-instrument-real-time-application.

11. Stop the Application When Needed

Skip this when the application should run to its configured stop time. Call stop to stop early, end an indefinite run, or control file-log import.

matlabCopy code
stop(tg);                              % AutoImportFileLog defaults to true
stop(tg, AutoImportFileLog=false);     % skip file-log import

stop stops the application and unloads it — afterwards isLoaded(tg, app) is false, so a second run needs load(tg, app) before start(tg). It is safe to call when the app is not running. The runtime also deletes the target's currParamSet file. With AutoImportFileLog=false, logs stay on the target and FileLogMaxRuns=1 lets the next run overwrite them.

12. Manage Installed Applications

List what is installed on the target computer, and remove applications only on request.

matlabCopy code
apps = getInstalledApplications(tg);
disp(apps);

removeApplication(tg, "myApp");
removeAllApplications(tg);

Ask First — these are irreversible without a rebuild and re-install.

Constraints:

• Cannot remove a running app — stop first.
• Cannot remove the configured startup app without first calling clearStartupApp(tg). That belongs to simulink-configure-real-time-target.

Verification

matlabCopy code
disp(isConnected(tg));
disp(isLoaded(tg, app));
disp(isRunning(tg, app));
disp(getInstalledApplications(tg));
disp(tg.Stimulation.getStatus("all"));

After a full run cycle, expect: isConnected=true, isLoaded(tg, app)=true between load and stop, isRunning(tg, app)=true between start and stop, and both isRunning(tg, app)=false and isLoaded(tg, app)=false after stop — stop unloads the application, so a subsequent run needs a fresh load(tg, app) first.

Guardrails

references/guardrails.md carries the reasoning behind every rule below, plus the full table of invented forms with the correct replacement for each.

Always:

• Read SystemTargetFile before slbuild — never infer configuration from the model name or an slrt_ prefix. Route to simulink-configure-real-time-model if it is not a Simulink Real-Time STF.
• Verify isConnected(tg) before any install, load, start, stop, or Stimulation call.
• Read isLoaded(tg, app) and isRunning(tg, app) before mutating run state.
• Let load validate STF compatibility; surface its error message verbatim.
• Stop the affected stimulation entry (Stimulation.stop(selector) or stop("all")) before reloadData — required even on a freshly loaded app.
• Check a reloadData source against the app's recorded root-inport metadata before the call — mismatched datatypes or dimensions are rejected at best, misinterpreted at worst.
• For parameter stimulation, call reloadData before start; that call is what registers the entry. Without it, start is a silent no-op. Use setparam for one-time tuning.
• Use Time64[us] (duration) timestamps in parquet stimulation files.
• Confirm parameters are tunable before promising getparam/setparam/parameter sets; route tunability fixes to simulink-configure-real-time-model.
• Classify the data need (live monitoring, post-run integrity, deployed-app access, outside-MATLAB access) before choosing streaming, recording, file logging, or routing to simulink-instrument-real-time-application.

Ask first:

• slbuild when a current .mldatx is loaded or running — the rebuild invalidates it.
• start with non-default StopTime or LogLevel — surface the value and intent.
• stop(tg, AutoImportFileLog=false) — logs stay on the target and the next run can overwrite them.
• removeApplication / removeAllApplications — irreversible without rebuild + re-install.
• setparam that changes plant/controller behavior on hardware — surface block path, parameter, value.
• loadParamSet / setDefaultParamSet — one call can change many tunable values.

Never:

• Invent function names, named parameters, or selector forms absent from references/api-reference.md and the other reference files. If a form is not shown, do not use it. The forms agents invent most often — refuse each and use the right-hand column:

  Invented	Use instead
  tg.run(app) / tg.execute(app) / tg.deploy(app)	load(tg, app) then start(tg)
  start(tg, app, ...)	start(tg, ...) — the loaded app is implied
  load(tg, app, "SkipInstall", true)	load(tg, app) — load has no name-value pairs
  tg.unload() / tg.kill()	stop(tg)
  pause(tg) / tg.pause() / tg.suspend()	tg.Stimulation.pause(selector), or stop the app — the target has no pause
  stop(tg, app) / stop(tg, "myApp")	stop(tg)
  stop(tg, false) (shown in the published doc)	stop(tg, AutoImportFileLog=false) — the runtime rejects the positional form
  "Duration" / "Timeout" / "MaxRuntime" on start	"StopTime" — only the 8 documented parameters exist
  tg.getApps() / tg.listInstalled()	getInstalledApplications(tg)
  tg.Stimulation.startAll()	tg.Stimulation.start("all")
  Stimulation.swapData/loadData/replaceData	Stimulation.reloadData(selector, source)
  Stimulation.reset/clear/restore	load(tg, app) again — defaults live in the .mldatx

The remaining Never rules — the run-state ordering traps, the Stimulation sequencing traps, the inlined-literal trap, the recording-continuity claim, and editing build artifacts or model settings from here — are listed with the failure mode behind each in references/guardrails.md. Read that file before any lifecycle, Stimulation, or logging call.

Task Routing

Copyright 2026 The MathWorks, Inc.