USV (VRX)
Repo: pantelis/usv-vrx-2022 Environment: ros.dev.gpu Status: External repo, awaiting platform integration Upstream: osrf/vrx (VRX 2023 competition) Epic: AURA-698 Design doc: Marine World Model for USV Navigation
Overview
Unmanned Surface Vehicle (USV) reference application targeting the Virtual RobotX competition. ROS 2 + Gazebo Harmonic simulation of the WAM-V platform on the Sydney Regatta course, with controllers for the standard VRX tasks (station keeping, wayfinding, perception, navigation). Serves as the maritime counterpart to TurtleBot Maze (ground) and AR4 Physical-AI (manipulation) on the user plane.
Architecture
- Simulator: Gazebo Harmonic via upstream
vrxandros_gz, WAM-V URDF with thruster + sensor plugins, wind/wave/current environmental forcing - Controller stack: ROS 2 nodes in
my_usv_components/— inverse-kinematics thruster allocation, station-keeping PID, wayfinding goal tracker - Sensing: GPS + IMU + camera + LiDAR fusion for localization and perception (per VRX 2023 task descriptions)
- References: SINGABOAT-VRX motion control, artificial potential field path-keeping for USVs
VRX tasks covered
| Task | Capability | Status |
|---|---|---|
| Station Keeping | Hold pose under wind/wave forcing via sensor fusion | Implemented |
| Wayfinding | Follow a sequence of geo waypoints | Implemented |
| Perception / Navigation | Buoy classification, channel following | Planned |
Platform services
- User plane: Gazebo + ROS 2 stack runs on
ros.dev.gpuas Ray workers; long-running sim sessions as Ray actors - Data plane: ROS bag recordings and Rerun (
.rrd) traces of sim runs stored inwarehouse/, surfaced via R2 for the Rerun web viewer - Control plane: ClusterAgent monitors
ros.dev.gpu; NotebookAgent submits training/eval jobs for perception models against bag-replay datasets
Integration plan
- Move repo under
aegean-ai/org and add toapplicationstable in the index - Devcontainer alignment with
ros.dev.gpu(ROS 2 Jazzy + Gazebo Harmonic) — current repo targets an earlier ROS 2 + Gazebo combo - Wire bag recording → lakehouse landing tier; Rerun visualization for trajectory + thruster commands
- Add VRX perception task (buoy detection) as a Counter-UAS-style training pipeline on
torch.dev.gpu