Frontend Overview

What It Does

The Hephaes frontend is a local-use web application for managing robotics data assets and converting them into ML-ready formats. It runs alongside the backend on a user’s local machine and provides a UI for the full data pipeline workflow.

Users can:

• Manage local assets — register, upload, scan directories, tag, and browse MCAP files stored on disk
• Create conversion specs — inspect asset topics, draft conversion configurations, preview sample output, and save reusable configs
• Run conversions — convert MCAP files into TFRecord or Parquet output with a guided authoring workflow
• Monitor jobs — track conversion job progress, view results, and inspect output metadata
• Replay episodes — scrub through episode timelines with websocket-based realtime replay

Local-Only Design

The open source frontend is designed for local use only. It connects to a locally running backend server and operates on assets stored on the user’s filesystem. There is no authentication, multi-user support, or cloud deployment model in the current release.

Stack

Navigation

The app shell provides four primary navigation sections:

1. Dashboard — summary cards, trend views, and blockers with drill-through links
2. Inventory — asset listing with filters, upload, directory scan, bulk indexing, and tagging
3. Outputs — browse and inspect converted output datasets
4. Jobs — job history with status tracking and detail views

Additional routes are accessed contextually from within these sections:

• /assets/[assetId] — asset detail with metadata, episodes, and related jobs
• /convert — conversion authoring workspace (create new or use saved configs)
• /replay — episode timeline replay with playback controls

Future Direction

• Custom computation scripts as an option in conversion jobs
• Cloud-hosted deployment model beyond local use
• Enhanced visualization and replay capabilities