Onboard software for the underwater robot (ROV) built by Triton Robotics, a student team from Seattle that competes in the MATE ROV Competition (Ranger class). This software ran the team's 2026 vehicle, which placed 2nd overall at the 2026 MATE ROV World Championship.
The vehicle's software is split across three repositories:
| Repository | Runs on | Role |
|---|---|---|
| TritonOS (this repo) | Raspberry Pi on the ROV | Control loop, thruster output, sensor telemetry, camera streams |
| TritonPilot | Pilot computer | Controller input, live video and telemetry, logging |
| TritonAnalysis | Separate analysis laptop | Stereo measurement, 3D reconstruction, mission tools |
Written and maintained by Thomas Gust, Triton's software and electrical lead.
TritonOS is the onboard runtime for Triton's ROV. It runs on the vehicle computer, listens for pilot commands from TritonPilot, drives the vehicle's PWM outputs, publishes sensor telemetry, manages camera streams, and exposes a small management RPC surface for configuration and calibration tasks.
TritonOS is intentionally separate from mission analysis code. During competition, mission-specific detection and scoring workflows should run from the TritonAnalysis repository on a separate computer. TritonOS should stay focused on safe vehicle operation, telemetry, and hardware control.
main_rov.py starts the onboard services:
- Pilot receiver and control loop on
tcp://0.0.0.0:6000 - Sensor telemetry publisher on
tcp://0.0.0.0:6001 - Video stream RPC server on
tcp://0.0.0.0:5555 - Management RPC server on
tcp://0.0.0.0:5556 - Optional network diagnostics server on port
7700
The most important runtime flow is:
TritonPilot controller input
|
v
PilotFrame JSON over ZeroMQ
|
v
PilotReceiver -> ControlService -> Mixer/Autopilot -> ThrustWriter
|
v
Navigator/PCA9685 PWM outputs
Sensor and video flows are separate so telemetry, control, and camera problems can be diagnosed independently.
main_rov.py ROV service supervisor and normal entry point
rov_config.py Operator-tunable runtime configuration
control/ Pilot intake, arming, control loop, hold controllers, RPC
motion/ PWM backends, thruster writer, channel mapping
sensors/ Hardware drivers, telemetry wrappers, derived processors
video/ GStreamer stream manager and video RPC server
schema/ Shared pilot-control wire schema
utils/ Shared config, Navigator import, reference, ZMQ helpers
tools/ Operator diagnostics and bench/field utilities
bin/ Install, update, tether, and debug shell scripts
tests/ Hardware-free unit tests
docs/ Setup, networking, architecture, and subsystem docs
- Documentation Index
- Setup Guide
- Network Guide
- Operations Guide
- Architecture Overview
- Subsystem Reference
- Configuration Guide
- Testing And Troubleshooting
From a development machine:
python -m venv .venv
. .venv/bin/activate
python -m pip install -r requirements.txt
python -m pytestThe test suite is designed to run without physical ROV hardware. Hardware-only
checks live in tools/ and should be run deliberately on the vehicle.
Initial provisioning on the Raspberry Pi:
sudo bash bin/install_configure.sh --project-dir /home/TritonOSNormal code update on the Pi:
sudo bash bin/update_code.sh --dir /home/TritonOSCheck the service:
sudo systemctl status tritonos-rov.service
sudo journalctl -u tritonos-rov.service -fSee Setup Guide for the full install path, expected hardware dependencies, and recovery options.
The normal tethered layout is:
- Pilot computer:
192.168.1.1 - ROV Ethernet:
192.168.1.4 - Pilot commands: ROV port
6000 - Sensor stream: ROV port
6001 - Video RPC: ROV port
5555 - Management RPC: ROV port
5556 - Network diagnostics: ROV port
7700
See Network Guide for tether setup, internet sharing, route configuration, video routing, and troubleshooting.
TritonOS starts disarmed. The control loop only commands non-neutral thruster
outputs after an arm command, fresh pilot input, and configured arming checks.
Hardware scripts in tools/ can bypass pieces of the normal control stack, so
use them with props removed or the vehicle secured.
Before any water test, run:
python -m tools.rov_preflight
python -m tools.print_channel_mapThen use the operations checklist in Operations Guide.