This guide contains Amazon affiliate links. If you buy through them, DroneCoreGuide earns a small commission at no extra cost to you.
A companion computer is the second brain on an autonomous drone. It runs everything the flight controller shouldn’t: computer vision, mission logic, mapping and networking. Choosing one means balancing processing power against weight, power draw and heat.
What a companion computer does
- Reads one or more cameras.
- Runs AI models for detection, tracking or segmentation.
- Sends commands to the flight controller over MAVLink.
- Streams video or telemetry to the ground.
- Logs data for later analysis.
The main options
| Board | Strengths | Trade-offs |
|---|---|---|
| Raspberry Pi 5 | Cheap, huge community, easy Linux setup, good for light AI | Modest AI speed without an accelerator |
| Raspberry Pi + AI accelerator (Hailo, Coral) | Big boost to AI speed at low power | Extra cost, model conversion required |
| NVIDIA Jetson Orin Nano | Strong GPU for real-time AI, CUDA and TensorRT | Higher cost, more heat, more power |
| Jetson Orin NX | Even more AI performance | Heavier and pricier |
| Small x86 boards | Run full desktop software | Power hungry, heavy |
Shop Raspberry Pi 5 · Shop Jetson Orin Nano · Shop Pi AI accelerators
How to choose
- Learning MAVLink and simple vision: Raspberry Pi 5.
- Real-time object detection on a budget: Raspberry Pi 5 with an AI accelerator.
- Multiple cameras, larger models or depth processing: Jetson Orin Nano or NX.
Power
Companion computers need clean, stable power. Use a dedicated 5 V regulator (BEC) rated well above the board’s peak draw: a Raspberry Pi 5 can draw around 5 A under heavy load with peripherals, and Jetson boards often take a higher input voltage. Never power a companion computer from the flight controller’s small 5 V pad. A voltage dip can reboot it mid-flight. Shop 5 V regulators
Weight and mounting
- Add the computer, camera, cooling and regulator weight to your thrust calculation.
- Mount on vibration-damped standoffs.
- Keep it away from GPS and compass, since USB 3 ports and Wi-Fi can interfere with GPS.
Cooling
AI workloads make these boards hot. Use a heatsink and fan, and make sure there’s airflow even when the drone hovers. A throttled processor runs your model slower exactly when you need it.
Software stack
- OS: Raspberry Pi OS or Ubuntu; JetPack on Jetson.
- MAVLink libraries: pymavlink, MAVSDK or ROS 2 with MAVROS.
- Vision: OpenCV plus a detection framework.
- Optional: ROS 2 for larger robotics projects.
Next step: wire your Pi to the flight controller.
Power budget example
Suppose your build includes:
| Component | Typical draw |
|---|---|
| Raspberry Pi 5 under AI load | up to ~5 A at 5 V (about 25 W peak with peripherals) |
| Camera module | under 0.5 A |
| AI accelerator | a few watts |
| Fan | under 0.2 A |
That’s roughly 25–30 W at peak. On a 4S battery (about 15 V average), it adds around 2 A of draw. On a GPS drone hovering at 15 A, that’s a noticeable but manageable share, and it means slightly shorter flights. Size your regulator for the peak with margin, and include this load in flight-time estimates.
Wireless links to the ground
- Wi-Fi: easy for bench testing and short range, but range is limited and it can interfere with GPS and 2.4 GHz radios.
- Telemetry radios: long-range, low-bandwidth MAVLink to the ground station.
- Cellular modems and long-range IP radios: used on advanced platforms for video and data.
Keep the companion computer’s radios away from the GPS antenna, and choose frequencies that don’t clash with your RC link.
Boot and safety design
- Make the companion computer start its software automatically at boot.
- Have it send a heartbeat to the flight controller; configure the autopilot to react if the heartbeat stops.
- Never let the companion computer arm the drone without explicit confirmation from the pilot.
- Log everything: commands sent, detections and errors. Logs make debugging possible.
Choosing cameras
| Camera type | Use |
|---|---|
| CSI camera (Pi Camera) | Low latency, easy on Raspberry Pi |
| USB camera | Simple, flexible, higher latency |
| Global shutter camera | Less motion distortion for fast movement |
| Depth camera | Obstacle sensing and 3D mapping |
| Thermal camera | Inspection and search tasks |
Frequently Asked Questions
What is a companion computer on a drone?
An onboard computer, separate from the flight controller, that runs high-level tasks such as computer vision, AI and mission logic.
Is a Raspberry Pi powerful enough for drone AI?
For learning and lightweight detection, yes. For faster real-time AI, add an accelerator or move to a Jetson.
How do I power a companion computer on a drone?
Use a separate, high-current 5 V regulator connected to the battery. Don’t rely on the flight controller’s small 5 V output.
Raspberry Pi or Jetson for an AI drone?
Raspberry Pi is cheaper and easier for learning. Jetson is much faster for real-time AI but costs more and uses more power.