What was on board.
Platform
- WAM-V catamaran hull
- Custom engine pods at the rear bulkheads
- Sensors and UAV dock on the payload tray, electronics shelved beneath
Propulsion System
- ePropulsion Navy 3.0 Evo (×2)
- Differential drive, parallel along the keel
- RS485 battery and propeller status monitoring
Power System
- ePropulsion E60 lithium (×2)
- Hexsoon power distribution board
- DC-DC rails to 24 V, 12 V and 7 V
- Solid-state relays on the 48 V feed
Onboard Computer
- NVIDIA Jetson Orin NX
- Seeed reComputer J4012 carrier
Control System
- Hexsoon CubeOrangePlus
- u-blox M9N backup GPS on the Cube
- PID on yaw rate and velocity
- L1 waypoint following, manual joystick override
Perception Sensors
- Stereolabs ZED 2i stereo camera
- YOLO object detection
Navigation System
- Hemisphere R632 GNSS receiver, RTK to 4 cm
- Hemisphere A25 GNSS antenna
- 0.01° heading accuracy
Acoustics
- Teledyne RESON TC4013 hydrophone
- Signal frequency and emitter presence
Wireless Communications
- Ubiquiti airMAX Rocket Prism 5AC (USV)
- Ubiquiti Rocket 5AC (GCS)
- Ubiquiti airMAX Omni antenna (USV)
- Ubiquiti airMAX Sector antenna (GCS)
- Hexsoon Herelink telemetry and control
- MAVLink protocol
Manipulators & Payload
- Flywheel racquetball launcher with drum reload
- UAV landing dock, four stepper-driven clamping rods
- UAV 1.0
Software Architecture
- ROS 1 Noetic on Ubuntu 20.04 LTS
- YOLO over ZED 2i point clouds
- Dubins path gates, A* obstacle avoidance
- Single-frame status GUI
Others
- 4-channel 3.3 V relay module for the signal tower
- Bud PTR-28483 connectors, sealed enclosures
What it had to solve.
The RobotX 2024 Autonomy Challenge runs as one continuous autonomous run, and a task is only finished if the vehicle can carry what it learned into the next one. These are the problems that shaped USV 1.0.
Finding the gate by sound
The course marks the correct entry gate with a submerged acoustic beacon rather than with colour, so no camera can answer it. A Teledyne RESON TC4013 hydrophone reports signal frequency and the presence of an emitter, which is the only sensor on the boat that can pick the gate. It was also the team's lowest-ranked subsystem by priority, and the test plan takes it as far as deriving a location in the lab.
Detection that commits only when sure
A ZED 2i stereo camera feeds a YOLO detector, and object positions come from the depth point cloud rather than from bearing alone. Detections must hold across a confidence threshold in frame count before the planner is allowed to act, because one false positive on open water commits the whole vehicle to the wrong buoy.
One reading, four decisions
The light tower shows three colours for one second each. The first sets which way to run the path and which creature to circle, the second sets the docking bay, the third sets the tin the aircraft collects. Perception here is not a per-task problem. Misread the sequence once and four later tasks execute correctly against the wrong target.
Planning around obstacles without a LiDAR
Obstacle buoys sit at random along the path, and USV 1.0 had no LiDAR to find them with. Avoidance runs as A* over a graph whose edges are weighted by a repulsion field around each detected obstacle. Docking fits a plane across the three bay panels from stereo data and approaches on waypoints perpendicular to it.
Propulsion sized for control, not speed
The 2022 boat's Minn Kota thrusters struggled to move it. Two ePropulsion Navy 3.0 Evo units, 60 kg of thrust each, drive differentially along the keel from custom engine pods at the rear bulkheads. The pods also stabilise the hull in pitch, which is what keeps the forward camera on the horizon instead of swinging between sky and water.
A flight deck on a moving boat
The aircraft has to launch from the vehicle and land back on it, at sea. The landing dock clamps the aircraft's legs with four linearly actuated stepper-driven rods, so the UAV is held through transit and sea state rather than resting on a pad that is never level.