ADR-005: ROS2 OSS Package Adoption Strategy

Status

Accepted

Context

Bennu currently has two custom ROS2 packages (bennu_camera, bennu_bringup) with minimal external dependencies beyond px4_msgs, rclpy, and sensor_msgs. As the project evolves through the platform-readiness phases, we need to decide which OSS packages to adopt versus building custom, and in what order.

We evaluated ~25 packages across 9 categories: PX4 integration, camera/image pipeline, Gazebo bridging, GPS/localization, recording/debugging, testing, diagnostics, autonomy frameworks, and reference projects.

Key constraints:

• Jazzy compatibility required — all adopted packages must have ROS2 Jazzy releases.
• Headless CI — packages must work without GPU or X11 in GitHub Actions.
• Python-first — Bennu nodes are Python; C++-only packages are deferred unless wrappable.
• PX4 uXRCE-DDS is the primary bridge — packages that assume MAVROS are not suitable.
• Event-triggered capture — Bennu captures images on PX4 camera trigger events, not continuous streaming.

Decision

Adopt OSS packages in four tiers, aligned with project phases:

Tier 1 — Immediate (supports current SITL and camera refactor work)

Package	Purpose
ros_gz_bridge	Bridge Gazebo Harmonic camera sensors into ROS2 for camera_backend:=rendered
ros_gz_image	GPU-accelerated image format conversion between Gazebo and ROS2
launch_testing_ros2	Launch-level integration tests for drone.launch.py and bennu_bringup
diagnostic_updater	Node health reporting (capture rate, GPS staleness, disk space, backend status)

These are installed via apt in Dockerfile.ros2:

ros-jazzy-ros-gz-bridge
ros-jazzy-ros-gz-image
ros-jazzy-launch-testing-ros2
ros-jazzy-diagnostic-updater

Tier 2 — Recording & Debugging (supports artifact-based test strategy)

Package	Purpose
rosbag2_storage_mcap	MCAP format rosbag recording for SITL runs, replay backend, and debugging
foxglove_bridge	WebSocket bridge for browser-based visualization without X11/GPU

Tier 3 — Camera Pipeline Maturity (supports real hardware path)

Package	Purpose
image_transport	Bandwidth-efficient image pub/sub with compression plugins
compressed_image_transport	JPEG/PNG compression for rosbag recording and WiFi preview
camera_info_manager	Loads and serves lens calibration YAML files alongside image topics
camera_calibration	Checkerboard-based intrinsic calibration for IMX477 + 6mm lens
image_pipeline	Image rectification using calibration data

Tier 4 — Evaluate in Phase 3+ (autonomy)

Package	Trigger
px4_ros2_msg_translation_node	When PX4 firmware and px4_msgs versions diverge
px4-ros2-interface-lib	When mission execution moves from MAVSDK into ROS2 nodes
robot_localization	When external sensors (optical flow, VIO) are added beyond PX4's EKF
Aerostack2	When scope grows to multi-vehicle operations
apriltag_detector	When landing-pad or calibration ground-truth testing is needed

Not Adopted

Package	Reason
px4_ros_com	Vestigial since PX4 v1.14+. uXRCE-DDS replaces it entirely.
v4l2_camera	Pi camera modules need libcamera, not V4L2. Raw Bayer output only via V4L2.
MAVROS	Bennu's architecture uses uXRCE-DDS as the primary PX4 bridge (ADR-004). MAVROS adds a MAVLink translation layer.
camera_ros	Designed for continuous streaming, not event-triggered capture. Monitor for future hybrid use (preview/calibration).
Nav2	Fundamentally a 2D ground-robot stack. 3D drone navigation is a stretch goal, not production-ready. PX4's built-in mission mode is more practical.

Consequences

Positive:

• Standard ROS2 interfaces improve interoperability with community tools (Foxglove, rosbag2, image_pipeline).
• Reduced custom code — ros_gz_bridge replaces custom Gazebo-to-ROS2 bridge code for the rendered backend.
• diagnostic_updater gives system health visibility at low integration cost (~15 lines per node).
• launch_testing_ros2 fills the test gap between unit tests and full SITL scenario tests.
• Tiered adoption prevents premature complexity while keeping a clear upgrade path.

Negative:

• Docker image size increases with each tier of apt packages.
• Version pinning overhead — must track Jazzy release compatibility across packages.
• ros_gz_image may not work in headless CI without GPU; may need to fall back to ros_gz_bridge with software rendering.

Neutral:

• No photogrammetry-specific ROS2 package exists. The photogrammetry pipeline remains post-processing (OpenDroneMap). ROS2's role is data acquisition only, which Bennu's architecture already handles correctly.
• MAVSDK remains the external mission runner for SITL tests. The PX4 ROS2 Interface Library is deferred until mission execution moves into ROS2 nodes.