Owli-AI Spatial Mapping

Independent Android research app for the pre-development of a local live 3D environment map for Owli-AI Assist. The initial reference device is the Samsung Galaxy S23+.

Current scope

This bootstrap repository deliberately implements only:

• an Android app that runs on devices and ordinary CI AVDs,
• ARCore capability diagnostics in AR Optional mode,
• camera permission diagnostics,
• a pure Kotlin mapping-core module,
• deterministic geometry and depth-projection tests,
• CI, AVD smoke tests, agent guidance, and downloadable WebAgent repo packs.

It does not yet start an ARCore session, read camera frames, build a point cloud, estimate a floor plane, detect obstacles, or claim that a path is clear.

Safety invariant

Unknown, unavailable, stale, or low-confidence sensor information is never interpreted or presented as free space.

Modules

• app/ — Android/Compose diagnostic shell and future ARCore adapter.
• mapping-core/ — platform-independent geometry, depth projection, and later rolling-map algorithms. It must not import Android, AndroidX, or ARCore classes.

First build

Requirements:

• JDK 17
• Android SDK Platform API 36.1 for the compile SDK
• targetSdk remains API 36 and is not the compile SDK package requirement
• internet access on the first run to obtain the Gradle distribution and dependencies

PowerShell:

.\gradlew.bat :mapping-core:test :app:testDebugUnitTest :app:lintDebug :app:assembleDebug

Linux/macOS/WSL:

chmod +x gradlew scripts/*.sh
./gradlew :mapping-core:test :app:testDebugUnitTest :app:lintDebug :app:assembleDebug

The repository includes the verified Gradle 8.13 wrapper JAR. The existing bootstrap script remains the trusted recovery path when the JAR is absent and verifies the pinned SHA-256 checksum before use. See gradle/wrapper/README.md.

AVD smoke test

./gradlew pixel2Api35DebugAndroidTest \
  -Pandroid.testoptions.manageddevices.emulator.gpu=swiftshader_indirect

The AVD verifies the safe unsupported/diagnostic path. ARCore Depth is hardware-dependent and must later be validated on the S23+.

Start here

• Humans: docs/ROADMAP.md, then docs/ARCHITECTURE.md.
• Repository assistants: CHATGPT.md, then AGENTS.md and .ai/repo-map.md.
• Coordinate work: read docs/COORDINATE_SYSTEMS.md before changing any math.
• Test data: read docs/DATASET_POLICY.md before adding recordings.

Planned workspace role

After the bootstrap is green on GitHub, this repository can be added to KreutzM/Owli-AI-Assist-WS as the independent spatial/ submodule. The workspace remains coordination-only; this repository remains independently buildable.

Status

Research prototype. No production release, navigation assurance, medical-device claim, or replacement for established mobility aids.