US What a companion computer does
A drone companion computer is the AI brain on a drone. It sits next to the flight controller and runs the heavy software the controller cannot: computer vision, path planning, object detection. The flight controller keeps the aircraft stable and executes commands; the companion computer decides what those commands should be, based on what its cameras and sensors see.
The term comes from the open autopilot projects ArduPilot and PX4, which call any onboard computer that talks to the autopilot a companion computer. It can be as simple as a single-board computer logging telemetry or as capable as an NVIDIA Jetson running several neural networks on live video.
Why drones need one
A flight controller runs on a microcontroller built for real-time stabilization, not for neural networks. To make a drone autonomous, so it can avoid obstacles, follow a target or map a site, you need a full computer running AI, and that is the companion computer. It connects to the flight controller over a serial UART, CAN or Ethernet link, usually with the MAVLink protocol, sending waypoints and velocity commands while reading telemetry back.
Keeping the two jobs apart is also a safety measure. If the companion computer crashes or reboots, the flight controller keeps flying and can hold position, return home or land under its own failsafe rules.
Flight controller vs companion computer
The two computers are designed for very different work, which is why one cannot simply replace the other.
| Role | Flight controller | Companion computer |
|---|---|---|
| Main job | Keep the aircraft stable and execute commands | Decide what the commands should be |
| Processor | Microcontroller, often an Arm Cortex-M | Arm SoC with GPU or NPU, such as NVIDIA Jetson |
| Software | Real-time autopilot: ArduPilot or PX4 | Linux with ROS 2, MAVSDK and AI models |
| Sensors | IMU, barometer, compass, GNSS | Cameras, LiDAR, radar, thermal payloads |
| Timing | Hard real time, hundreds of loops a second | Soft real time, frame by frame |
| Typical power | Around 1 to 2 W | Around 7 to 25 W for Jetson Orin Nano and NX |
| If it fails | The aircraft cannot fly | The flight controller falls back to a failsafe |
What a companion computer runs
Most companion computers run Linux. The software stack depends on the mission, but usually includes:
-
//
Vision. Object detection, tracking and obstacle avoidance from onboard cameras, often with AI models optimized for the GPU.
-
//
Navigation. Path planning and position estimation, often fused with GNSS, visual odometry or LiDAR.
-
//
Links to the flight controller. Commands and telemetry over UART, CAN or Ethernet via MAVLink, using libraries such as MAVSDK or MAVROS with ROS 2.
-
//
Payload control and data. Gimbal pointing, video encoding for the ground link and onboard recording.
Where companion computers are used
Any drone that has to make decisions in flight, rather than replay a route, carries a companion computer of some kind. Common applications include:
-
//
Inspection. Power lines, wind turbines, bridges and pipelines, where the drone keeps a set distance from the structure and flags defects as it flies.
-
//
Mapping and survey. Photogrammetry and LiDAR capture, with onboard processing to check coverage before landing.
-
//
Search and rescue. Thermal and color cameras with person detection to cover large areas quickly.
-
//
Agriculture. Crop scouting and targeted spraying guided by plant detection.
-
//
Defense and security. ISR missions that track targets and stream processed video to the ground station.
Choosing a companion computer
Every gram and watt on a drone costs flight time, so the choice is a trade-off between AI performance and size, weight and power, often called SWaP-C. Check these points:
-
//
AI performance. Enough GPU or NPU throughput for the models and the number of cameras you plan to run.
-
//
Camera inputs. MIPI CSI or GMSL for low-latency cameras, USB or Ethernet for others.
-
//
Flight controller interfaces. Spare UART, CAN and Ethernet ports on the carrier board, at the right logic levels.
-
//
Power and heat. A wide DC input that suits the drone battery, and a thermal design that works in the airflow under the frame.
-
//
Weight and mounting. A light carrier board or enclosure and vibration-isolated mounting.
On NVIDIA Jetson
NVIDIA Jetson Orin is the usual companion computer because it offers a high ratio of AI performance to weight and power, which is what a drone needs. Small Jetson Orin NX carrier boards keep the weight down for airframe integration, while sealed rugged Jetson computers suit larger UAVs and ground control vehicles. Our Airbus ISR drone case study shows a rugged Jetson AGX Orin in an airborne video pipeline.
We build drone companion computers on Jetson and integrate them with the flight controller. See our guide to NVIDIA Jetson for drones for the full picture.
Quick answers on companion computers
Drone companion computers: common questions
Does every drone need a companion computer?
No. A drone flown manually or on simple GPS waypoints only needs a flight controller. A companion computer is needed for vision, AI and autonomy beyond the autopilot.
Can a Raspberry Pi be a companion computer?
Yes, for light tasks. It handles telemetry, logging and simple vision. Real-time AI on several cameras usually needs a GPU module such as Jetson Orin.
What protocol does a companion computer use?
Usually MAVLink. ArduPilot and PX4 both speak MAVLink. PX4 can also share data with ROS 2 directly over its DDS bridge.
What happens if the companion computer fails?
The flight controller takes over. It detects the lost link and follows its failsafe: hold, return to launch or land.
Is a companion computer the same as a mission computer?
Closely related. In defense, mission computer is the usual name for the onboard computer that runs payloads and autonomy. On small drones the two terms often describe the same thing.
We supply Jetson companion computers with JetPack, drivers and the flight controller interface configured as standard, and validate each build before it ships. Tell us the airframe, payload and flight time, and we will recommend the module and carrier.