Skip to main content

ADOS Drone Agent

The ADOS Drone Agent is the software that runs on the companion computer inside your drone. It sits between the flight controller (ArduPilot, PX4, Betaflight, or iNav) and everything else: your ground station, your phone, and the cloud. It is a hybrid system. The core long-running services (the supervisor, MAVLink router, HTTP front, cloud relay, logging daemon, and video pipeline) are native Rust binaries. Python handles AI and vision inference, hardware detection and first-boot bootstrap, the plugin runtime, and the setup and update webapp, all served behind the Rust front. It is licensed under GPLv3 and runs on ARM single-board computers from Raspberry Pi to Rockchip to NVIDIA Jetson.
ADOS setup webapp showing setup status and access URLs

What it does

The agent handles four core jobs:
  1. MAVLink proxy connects to your flight controller over serial UART, then forwards telemetry to WebSocket and TCP clients. Your GCS and the cloud all get MAVLink data without fighting for the serial port.
  2. Video pipeline detects your camera (CSI or USB), encodes H.264 via hardware or software, pushes it to a local RTSP server (MediaMTX), and optionally transmits over WFB-ng for long-range HD video.
  3. Cloud relay sends telemetry and receives commands through three layers (Convex HTTP polling, MQTT at 2 Hz, and WebRTC peer-to-peer signaling) so your drone is reachable from anywhere.
  4. Service supervisor manages all of the above as individual systemd services with automatic restart, resource limits, and health monitoring.

Key numbers

Architecture at a glance

The agent uses a multi-process architecture managed by systemd. Each service runs in its own process with its own cgroup resource limits. The supervisor service (ados-supervisor) is a Rust orchestrator that starts and monitors child services through systemctl. Most units now run a native Rust binary; a few stay Python by design (see the table below).
Services communicate through two Unix domain sockets:
  • /run/ados/mavlink.sock carries raw binary MAVLink frames (4-byte length prefix + payload)
  • /run/ados/state.sock carries JSON telemetry at 10 Hz

Hardware auto-detection

On first boot, the agent reads /proc/device-tree/model and matches it against 17 board profiles stored as YAML files. Each profile defines the SoC, available buses (UART, I2C, SPI, USB), video encoding capabilities, GPIO pins, and a default tier. The tier determines which features are enabled:
You do not need to configure the tier manually. The agent detects your board and picks the right tier automatically. You can override it in /etc/ados/config.yaml if needed.

Two profiles, one codebase

The same agent binary supports two profiles:
  • Drone profile (default): runs on the companion computer mounted on the drone. Manages FC, camera, video TX, and cloud telemetry.
  • Ground station profile: runs on a ground-side SBC (typically a Raspberry Pi 4B). Receives WFB-ng video, creates a WiFi hotspot, serves video to your laptop or phone via WebRTC, and optionally drives an HDMI display and OLED screen.
The profile is detected automatically at boot based on hardware fingerprinting. If the SBC has an OLED on I2C, four GPIO buttons, an RTL8812EU WiFi adapter, and no flight controller on UART, it picks the ground station profile.

Interfaces

You can interact with the agent in four ways:
  1. Setup webapp (port 8080): first-run identity, MAVLink, video, network, remote access, ground-station setup, logs, and advanced recovery.
  2. CLI (ados command): terminal status page, scriptable status, update, uninstall, plus subcommands for hardware, logs, network, plugins, profile, radio, and runtime mode. Works over SSH.
  3. REST API (port 8080): the native Rust front (ados-control) serves the REST API and reverse-proxies un-migrated routes to an internal FastAPI. Setup, telemetry, video, MAVLink, and command endpoints are available. Authenticated endpoints use X-ADOS-Key.
  4. ADOS Mission Control (GCS): the companion web app for maps, video, mission planning, hardware management, and fleet operations.

What’s next