Skip to content

ADR-007: Platform Readiness — Phases 0–2

Status

Accepted (implemented 2026-03-08 through 2026-03-17)

Context

Bennu began as a photogrammetry prototype with basic camera capture and EXIF geotagging. To feed an independent geospatial analysis platform, the drone needed to produce versioned, signed, schema-validated mission bundles — not just a folder of geotagged JPEGs.

The platform's only coupling point with Bennu is the mission bundle contract. Everything downstream (ingestion, STAC catalog, ML inference) depends on bundles conforming to that contract.

Decision

Implement platform readiness in three phases, each gated by passing tests.

Phase 0: Foundation & Governance

  • Contract schema (contract/v1/): JSON Schema defining manifest.json structure and images.csv 18-column spec. The schema is the single source of truth for what constitutes a valid bundle.
  • DroneIdentity (bennu_core): Dataclass encoding hardware config (flight controller, PX4 version, GPS, sensors). Every manifest references this.
  • Ed25519 signing (bennu_dataset/signer.py): Manifest is canonicalized (deterministic JSON), signed with PyNaCl. Platform rejects unsigned bundles.
  • CI pipeline (.github/workflows/ci.yml): ruff + pytest on every PR. Python 3.12 on Ubuntu 24.04.
  • License: Apache-2.0 across all packages.

Phase 1: Data Quality Baseline

  • Image quality scoring (bennu_camera/quality.py): Laplacian variance for blur, histogram analysis for exposure. Produces 0.0–1.0 score with categorical flags.
  • 18-column ImageMetadata (bennu_camera/geotag.py): Full per-image metadata including RTK fix type, GSD, quality score, ambient light. Replaces the original basic lat/lon/alt geotag.
  • Mission manifest generator (bennu_mission/mission_manifest.py): Aggregates identity + images + quality into a contract-v1-compliant manifest.json and images.csv.
  • Bundle packager (bennu_dataset/packager.py): Assembles the physical directory structure, computes SHA-256 checksums for integrity chain.
  • E2E integration test (tests/integration/test_bundle_e2e.py): Full pipeline from identity → capture → score → manifest → package → sign → validate.

Phase 2: Sensor Configuration & Calibration

  • YAML sensor configs (bennu_camera/sensor_config.py): Three profiles — survey (RGB+NIR+ambient), inspection (RGB+thermal), mapping (RGB only). Config drives which capture nodes launch.
  • Calibration data (bennu_camera/calibration.py): BH1750 ambient light readings captured alongside images when sensor config includes ambient light sensor. Written as calibration.csv in the bundle.

Consequences

  • Any drone (not just Bennu) can produce valid bundles if it honors the contract schema.
  • The integrity chain (checksums → manifest → signature) provides tamper-evidence without requiring network connectivity during flight.
  • Quality scoring happens onboard in real-time, enabling future abort-on-quality-degradation.
  • Sensor configs make the system extensible to new sensor combinations without code changes.
  • The E2E test is the project's most important regression gate — if it passes, the drone produces valid bundles.
  • Design doc: docs/plans/2026-03-08-drone-platform-readiness-design.md
  • Phase 3 (Survey Intelligence): in progress — grid planner, terrain following, coverage analysis