91 lines
2.5 KiB
Markdown
91 lines
2.5 KiB
Markdown
# EmboFlow V1 Scope And MVP
|
|
|
|
## MVP Definition
|
|
|
|
The first release should prove that EmboFlow can turn raw embodied data assets into structured outputs through a visual workflow engine.
|
|
|
|
### MVP Success Path
|
|
|
|
1. A user signs into a workspace
|
|
2. The user creates a project
|
|
3. The user uploads or imports a raw asset
|
|
4. The platform probes the asset and generates a structure summary
|
|
5. The user previews the asset
|
|
6. The user composes a canvas workflow
|
|
7. The workflow executes asynchronously
|
|
8. The user reviews logs, outputs, and generated artifacts
|
|
9. The user exports a normalized dataset, delivery package, or training config
|
|
|
|
## In Scope For V1
|
|
|
|
- User login and workspace model
|
|
- Personal and team workspaces
|
|
- Project resource isolation
|
|
- Raw asset upload and import
|
|
- Object storage integration
|
|
- Asset probing and structure detection
|
|
- Raw asset preview
|
|
- Canvas workflow editor
|
|
- Built-in node library for ingest, transform, inspect, export
|
|
- Node configuration through schema-driven forms
|
|
- Code injection for processing nodes
|
|
- Workflow run orchestration
|
|
- Logs, status, retries, and artifact tracking
|
|
- Dataset conversion and delivery-package normalization
|
|
- Training config export
|
|
- Plugin registration skeleton
|
|
|
|
## Important Business Scenarios
|
|
|
|
### Embodied Dataset Conversion
|
|
|
|
- Import RLDS, LeRobot, HDF5, or Rosbag
|
|
- Map to canonical semantics
|
|
- Export to target dataset format
|
|
|
|
### Delivery Package Normalization
|
|
|
|
- Import customer-provided raw directory or archive
|
|
- Rename top-level folders
|
|
- Validate required file structure
|
|
- Validate metadata files
|
|
- Check video file quality and naming
|
|
- Export or upload normalized package
|
|
|
|
### Data Processing Workflow Authoring
|
|
|
|
- Drag nodes onto canvas
|
|
- Connect nodes into DAG
|
|
- Tune parameters
|
|
- Inject code into processing nodes
|
|
- Re-run pipeline with traceable history
|
|
|
|
## V1 Modules To Build Deeply
|
|
|
|
- Identity and workspace management
|
|
- Asset ingestion and probing
|
|
- Workflow editor and node model
|
|
- Execution engine
|
|
- Built-in dataset conversion nodes
|
|
- Built-in delivery normalization nodes
|
|
- Preview and inspection
|
|
- Artifact export
|
|
|
|
## V1 Modules To Keep Lightweight
|
|
|
|
- Annotation
|
|
- Collaboration
|
|
- Plugin lifecycle UX
|
|
- Advanced analytics
|
|
- Kubernetes and Volcano scheduling adapters
|
|
- Advanced multi-sensor synchronized visual analytics
|
|
|
|
## Explicit V1 Exclusions
|
|
|
|
- Platform-managed training execution
|
|
- Real-time multi-user canvas co-editing
|
|
- Full marketplace for third-party plugins
|
|
- Complex enterprise approval workflows
|
|
- Streaming data processing
|
|
- Large-scale distributed execution as a deployment requirement
|