Autonomous Harbor Patrol

A modular C++/ROS 2 autonomy stack for an unmanned surface vehicle (USV), built on VRX (Virtual RobotX) — the Gazebo-based maritime robotics simulator developed by Open Robotics and the Naval Postgraduate School. The vehicle receives a set of waypoints, fuses GPS/IMU into a pose estimate, avoids obstacles detected by lidar, and enters a safe-stop state if any sensor stream goes stale.

Why VRX instead of a generic ground-robot sim

USVs have different dynamics than wheeled robots: no fixed wheelbase, thrust differential instead of steering, wave/current drift, and (eventually) COLREGs-style right-of-way behavior. VRX simulates a real WAM-V catamaran platform with wave dynamics and twin thrusters, so the control problem this project solves is the same shape as the one a real USV autonomy stack has to solve.

Architecture

                     ┌────────────────────┐
   GPS / IMU  ─────► │ state_estimator     │──► /vehicle_state
                     └────────────────────┘
   GPS/IMU/Lidar ───►┌────────────────────┐
   (raw topics)      │ health_monitor      │──► /sensor_health
                     └────────────────────┘
   waypoints  ───────►┌────────────────────┐
   (action goal)      │ mission_planner     │──► /target_waypoint
                     └────────────────────┘
   /target_waypoint ─►┌────────────────────┐
   /vehicle_state     │ obstacle_avoidance  │──► /desired_heading
   /scan (lidar) ────►│                     │──► /desired_speed_scale
                     └────────────────────┘
   /desired_heading ─►┌────────────────────┐
   /desired_speed ───►│ vehicle_controller  │──► /wamv/thrusters/{left,right}/thrust
   /vehicle_state ───►│                     │
   /sensor_health ───►│                     │
                     └────────────────────┘

Six independent nodes, each doing one job, communicating only through topics/actions — no node reaches into another's internals. vehicle_controller is the single point that actually enforces a safe stop (zero thrust) when sensor_health reports a fault; mission_planner also aborts the mission so the action client gets an honest result.

Prerequisites (your machine, not this sandbox)

• Ubuntu 24.04
• ROS 2 Jazzy
• Gazebo Harmonic
• VRX (vrx_gz package) — see the VRX wiki for install/build instructions

Build

# from the workspace root (this repo)
colcon build --symlink-install
source install/setup.bash

Run

Terminal 1 — bring up VRX itself:

ros2 launch vrx_gz competition.launch.py world:=sydney_regatta

Terminal 2 — bring up the autonomy stack:

source install/setup.bash
ros2 launch simulation/launch/harbor_patrol.launch.py

Terminal 3 — send a mission:

ros2 action send_goal /navigate harbor_patrol_msgs/action/Navigate \
  "{waypoints: [{x: 20.0, y: 0.0, z: 0.0}, {x: 20.0, y: 20.0, z: 0.0}], \
    acceptance_radius_m: 2.0}" --feedback

Testing

Pure decision logic (obstacle avoidance, thrust allocation) lives in header-only functions with no ROS dependencies specifically so they can be unit tested without Gazebo:

colcon test --packages-select obstacle_avoidance vehicle_controller
colcon test-result --verbose

Covered: clear-route behavior, full-stop distance threshold, steer-away direction on either side, speed scaling near the slow-distance threshold, malformed/empty scan handling, heading-error wraparound at ±π, and thrust clamping under large gains.

Suggested regression scenarios (build these into VRX worlds)

• Clear route
• Single obstacle, narrow passage
• GPS dropout mid-mission
• Lidar dropout mid-mission
• Unreachable waypoint (should abort cleanly, not hang)

Track mission completion time, final position error, and safe-stop count across runs after any control-logic change.

Status

In progress — see commit history. state_estimator uses a simple equirectangular GPS projection and raw IMU yaw, no Kalman filter (intentional for v1). Thrust allocation is a proportional controller, not model-predictive. Both are natural next steps once the base loop is proven reliable in sim.

Project layout

autonomous-harbor-patrol/
├── src/
│   ├── harbor_patrol_msgs/    # custom msgs + Navigate.action
│   ├── mission_planner/       # action server, waypoint sequencing
│   ├── obstacle_avoidance/    # lidar -> heading/speed commands (+ unit tests)
│   ├── vehicle_controller/    # heading/speed -> thruster commands (+ unit tests)
│   ├── health_monitor/        # sensor staleness -> safe-stop
│   └── state_estimator/       # GPS/IMU -> fused VehicleState
├── simulation/launch/         # top-level launch file
├── .github/workflows/ci.yml   # colcon build + test on push/PR
└── README.md