stera_recorder

An ARKit (iOS) / ARCore (Android) capture session for Flutter that writes an MCAP dataset — RGB video, depth, point cloud, mesh and IMU — plus the Dart API you drive it with.

This is a Flutter plugin: the Swift and Kotlin implementations ship here and register themselves through GeneratedPluginRegistrant. There is no manual wiring in your SceneDelegate / MainActivity.

Platform support

Platform	Minimum	Notes
iOS	14.0	ARKit. Depth and mesh need a LiDAR device.
Android	API 24	ARCore 1.44. Depth depends on device support.

A physical device is required — there is no AR session on the iOS Simulator or the Android emulator.

Install

flutter pub add stera_recorder

Setup

iOS — add the usage descriptions to ios/Runner/Info.plist. The camera key is always required; the microphone and speech keys are only needed if you leave voice commands enabled.

<key>NSCameraUsageDescription</key>
<string>Used to record AR sessions.</string>
<key>NSMicrophoneUsageDescription</key>
<string>Used for voice commands during recording.</string>
<key>NSSpeechRecognitionUsageDescription</key>
<string>Used to recognise recording voice commands.</string>

Android — nothing to add. The plugin's manifest merges in CAMERA, HIGH_SAMPLING_RATE_SENSORS, the optional android.hardware.camera.ar feature and the com.google.ar.core meta-data. Your app needs minSdk 24 or higher.

The plugin only ever checks permissions; requesting them is your app's job.

Usage

You supply persistence and permissions, then build your own UI on top of ArRecorderProvider — a ChangeNotifier holding the session state.

import "package:permission_handler/permission_handler.dart";
import "package:stera_recorder/stera_recorder.dart";

class AppRecorderPermissions implements RecorderPermissions {
  const AppRecorderPermissions();

  @override
  Future<bool> isCameraPermissionGranted() async =>
      (await Permission.camera.status).isGranted;

  @override
  Future<bool> openSettings() => openAppSettings();
}

final recorder = ArRecorderProvider(
  // Swap InMemoryRecorderPreferences for a RecorderPreferences backed by your
  // own key-value store to persist the recording settings across launches.
  preferences: InMemoryRecorderPreferences(),
  permissions: const AppRecorderPermissions(),
);

await recorder.initializeSession();
await recorder.startRecording();
// …
await recorder.stopRecording();
print(recorder.outputDirectory); // the MCAP dataset

Render the preview with Texture(textureId: recorder.textureId!) once recorder.textureId is non-null, and rebuild on the provider — see example/lib/main.dart for the whole screen in ~230 lines.

What you get

• ArRecorderProvider — the ChangeNotifier your UI binds to: session lifecycle, tracking state, preview texture, durations, and the recording settings forwarded from the config manager.
• ArRecordingConfigManager — the recording settings and their persistence through your RecorderPreferences.
• ArRecorderService — the method-channel / FFI facade, if you'd rather drive the native session yourself.
• RecordingConfig, RgbVideoResolution, ArRecorderState, ArTrackingState — the data types those APIs speak.

Output is a directory of MCAP files: the RGB track as encoded video, with depth, point cloud, mesh and IMU on their own channels, timestamped against the AR session clock.

Layout

lib/src/
  channels/    ar_recorder channel name + method enum (shipped native contract)
  service/     ArRecorderService — the method-channel / FFI facade
  ffi/         iOS interop: the ffigen output plus its typed wrapper
  data/        RecordingConfig and the recorder/tracking/resolution enums
  config/      RecorderPreferences + RecorderPermissions (host-supplied)
  managers/    ArRecordingConfigManager — the settings and their persistence
  providers/   ArRecorderProvider — the ChangeNotifier the UI binds to
  services/    audio cues, voice commands, battery monitoring
ios/           Swift Package (stera_recorder/Sources/) plus the podspec that
               compiles the same sources: session, frame pipeline, encoders,
               MCAP writers
android/       Kotlin: the same triad, plus the ARCore dependency
ffi/           ARRecorderInterop.h — ffigen input only, not compiled into the pod

RecorderPrefKeys holds the persisted setting keys. They are shipped state — never change a value, or every user's recording settings reset on upgrade.

Example

example/ is a one-screen app — permission, preview, record, stop:

cd example && flutter run

It needs a physical ARKit or ARCore device.

Native contract

The channel name (ar_recorder) and the method names in ArRecorderMethod must stay in lockstep with ArRecorderMethodChannelHandler.swift / .kt.

On iOS the hot paths (startRecording, stopRecording, getRecordingState, …) bypass the method channel and go through FFI. The generated bindings look classes up by their Objective-C runtime name, which is why ArRecorderInterop.swift pins them with explicit @objc(ARFRecordingConfig) style annotations — without those the names would carry the Swift module prefix and every FFI call would fail at class lookup.

Regenerate the bindings after changing ffi/ARRecorderInterop.h:

dart run tool/generate_ar_recorder_objc_bindings.dart

Known limitations

• Built-in Kotlin. android/build.gradle still applies the Kotlin Gradle Plugin itself. Moving to AGP's built-in Kotlin — which Flutter warns about today and will eventually enforce — needs AGP 9.

iOS builds under both Swift Package Manager and CocoaPods: ios/stera_recorder is a Swift package whose Objective-C ObjCTryBlock shim lives in its own target (SwiftPM can't mix the two in one module), and stera_recorder.podspec compiles the same sources for the CocoaPods path.

License

Apache 2.0 — see LICENSE.