Top 5 Companion Computers for UAVs

Updated September 2026. This article first ran in 2024 and named the Jetson TX2 and Raspberry Pi 4; both have since been superseded, and the line-up below reflects what developers are actually designing in today.

What a companion computer does

A companion computer is the onboard Linux computer that works alongside a drone's flight controller. The flight controller keeps the aircraft stable and executes commands; the companion computer gives it perception. It ingests image sensor and video data in real time and runs the computer vision and AI workloads that make a drone autonomous: visual inertial odometry (VIO) for GPS-denied navigation, SLAM, path planning, obstacle avoidance and object detection. Add a radio and the same computer encodes video and streams it beyond the workshop walls.

Traditionally the flight controller has been a separate board wired to the companion computer over a serial link. That works, but every extra board costs weight, volume, power and an integration effort. A companion computer with the flight controller already on it removes all four, which is why the two ModalAI boards below are a different category from the three general-purpose computers that follow them.

1. ModalAI VOXL 2

VOXL 2 is a Blue UAS Framework autopilot that fuses a PX4 or ArduPilot flight controller and an AI companion computer into one 16 g board. It is built around the Qualcomm QRB5165: eight cores up to 3.091 GHz, 8 GB LPDDR5, 128 GB flash, an Adreno 650 GPU and 15 TOPS of NPU, with the flight stack running on a dedicated sensors DSP so it never competes with vision workloads for CPU time. Six MIPI image sensors connect directly, video encodes at up to 8K30 in H.264 or H.265, and 4G/LTE, 5G, Wi-Fi and Microhard add-ons stack straight onto the board-to-board connector.

VOXL 2 is NDAA '20 Section 848 compliant and assembled in the USA, which is why it appears inside so many Blue UAS aircraft. Developers work in Ubuntu with ROS 2, OpenCV, Docker and TensorFlow Lite, and the VOXL SDK ships VIO, SLAM, obstacle avoidance and object detection ready to run. Documentation is open.

VOXL 2 starts at $1,399.99, board only.

2. ModalAI VOXL 2 Mini

VOXL 2 Mini is the same QRB5165 stack on an 11 g, 42 mm x 42 mm board with the 30.5 mm mounting pattern that FPV and sub-500 g airframes already use. It runs the same VOXL SDK and the same PX4 or ArduPilot flight stack on the sensors DSP, takes four MIPI image sensors, and adds connectivity through its USB 3 port. NDAA '20 Section 848 compliant, assembled in the USA and on the Blue UAS Framework list, it is the compute inside Stinger Vision FPV, a Blue UAS Cleared Select aircraft.

Choose VOXL 2 when you need six cameras or the board-to-board add-on ecosystem; choose VOXL 2 Mini when every gram and millimetre counts. VOXL 2 Mini starts at $1,349.99.

3. NVIDIA Jetson Orin Nano Super Developer Kit

The Jetson Orin Nano 8GB is NVIDIA's entry-level module: a 1024-core Ampere GPU with 32 Tensor Cores, a 6-core Arm Cortex-A78AE CPU and 8 GB of LPDDR5, rated at 67 sparse INT8 TOPS in its Super configuration across a 7 W to 25 W power range. The 69.6 mm x 45 mm module needs a carrier board for I/O, and the developer kit provides one along with camera and USB connectors.

For drone work the trade-off is the same as it was with the TX2 it replaced: the Orin Nano has no flight controller, no IMU or barometer and no cellular or mesh radio path of its own, so an autonomous aircraft built on it carries a separate autopilot, a carrier board and the cabling between them. The CUDA ecosystem is the reason to accept that weight, and JetPack supports TensorFlow and PyTorch directly.

4. NVIDIA Jetson Orin NX

The Jetson Orin NX 16GB shares the Orin Nano's footprint but keeps the same 1024-core Ampere GPU with 32 Tensor Cores but steps up to an 8-core Cortex-A78AE CPU and 16 GB of LPDDR5, rated at 157 sparse INT8 TOPS in Super mode. Its power range runs from 10 W to 40 W, which puts it above what most small airframes can dissipate and squarely in the territory of larger inspection and mapping platforms where a heavier compute payload is acceptable.

Like the Orin Nano, it is a module rather than a flight-ready system: budget for a carrier, a separate flight controller and a thermal solution. NVIDIA's larger AGX Orin, at 275 TOPS and 100 mm x 87 mm, sits beyond the size and power envelope of most drones and is better thought of as a ground-vehicle part.

5. Raspberry Pi 5

The Raspberry Pi 5 remains the introductory companion computer. Its Broadcom BCM2712 has four Arm Cortex-A76 cores at 2.4 GHz and a VideoCore VII GPU, memory runs from 2 GB to 16 GB of LPDDR4X, and two 4-lane MIPI transceivers accept cameras directly. It decodes 4Kp60 HEVC and runs Raspberry Pi OS with a large hobbyist and university following, and it is the cheapest way to get MAVLink, MAVROS or a simple vision pipeline running next to a PX4 or ArduPilot flight controller.

What it lacks is an NPU, a hardware video encoder and any flight hardware, so on-board neural networks run on the CPU and the drone still needs a separate autopilot, IMU and barometer. It suits teaching, low-compute telemetry and prototyping rather than real-time autonomy.

At a glance

Figures are for the compute module or board alone. Developer kits and carrier boards add weight and size on the NVIDIA and Raspberry Pi parts.

How to choose

If the aircraft has to fly without GPS, stream encrypted video and satisfy NDAA or Blue UAS requirements, start with VOXL 2 or VOXL 2 Mini: the flight controller, sensors and autonomy software are already on the board. If your team's models are built on CUDA and the airframe can carry a carrier board and a separate autopilot, a Jetson Orin is the natural fit. If you are learning, teaching or prototyping a low-compute payload, the Raspberry Pi 5 gets you flying for the least money. Whichever you pick, weigh the whole stack, not the module, because the boards a companion computer needs around it are where the grams hide.

See how VOXL 2 fits into a complete aircraft on the VOXL platform page, or browse development kits.