Best Autonomy Software Stacks for Drones
For most drone builders, the best autonomy software stack starts with an open-source flight stack, PX4 or ArduPilot, with perception added on a companion computer through ROS 2, NVIDIA Isaac ROS or ModalAI’s VOXL SDK. Teams that want a supported commercial layer on top can look at AuterionOS, while defense programs that need mission-level AI pilots can look at Shield AI Hivemind or Palladyne AI.
A drone autonomy stack has layers. The flight stack keeps the aircraft stable and follows commands. The perception layer estimates position and sees obstacles. The mission layer decides what to do next. No single product covers all three equally well, which is why most real systems combine two or three of the options below.
How we chose
Quick comparison
| Stack | HQ / steward | Best for | Notable |
|---|---|---|---|
| PX4 Autopilot | Dronecode (Linux Foundation) | Commercial flight control with a permissive license | v1.17 stable, BSD-3 license |
| ArduPilot | ArduPilot project | Broad vehicle support and a large community | Copter 4.7.1 stable, GPLv3 |
| AuterionOS | Arlington, VA and Munich | A supported OS across a fleet | Runs on Skynode and Jetson Orin |
| ModalAI VOXL SDK | San Diego, CA | GPS-denied flight on small drones | Open-source VIO built on OpenVINS |
| NVIDIA Isaac ROS | Santa Clara, CA | GPU-accelerated perception on Jetson | Isaac ROS 4.0 supports Jetson Thor |
| Nav2 (ROS 2) | Open Navigation / ROS community | Ground robots and mixed fleets | Behavior-tree navigation |
| Shield AI Hivemind | San Diego, CA | AI pilots for defense aircraft | Flown on MQ-20 Avenger and BQM-177A |
| Palladyne AI | Salt Lake City, UT | Collaborative autonomy and swarming | SwarmOS runs on SWaP-limited compute |
| Skydio Autonomy | San Mateo, CA | Buyers who want autonomy in a finished drone | Runs only on Skydio aircraft |
PX4 Autopilot
PX4 is an open-source flight control system for uncrewed vehicles, hosted under the Dronecode Foundation. Its BSD-3 license lets companies ship modified code without publishing it, which is why many commercial drones use it. PX4 docs list v1.17 as the current stable release, with v1.18 in beta. Version 1.16 added bidirectional DShot, a rebuilt rover architecture, a ROS 2 message translation node and log encryption. PX4 runs on Pixhawk-standard hardware and on companion boards such as ModalAI’s VOXL 2, and Dronecode holds its trademarks so no single company controls the roadmap.
ArduPilot
ArduPilot supports multicopters, helicopters, planes, rovers, boats and submarines from one code base. It is licensed under GPLv3, so changes you distribute must be shared. Copter 4.7.1 is the current stable release on the ArduPilot firmware server. ArduPilot has deep support for sensors, including scanning lidar such as the LightWare SF45/B and optical flow sensors like ARK Flow, and it runs on most Pixhawk-class controllers. Its documentation and forums are extensive, which helps small teams solve integration problems without a vendor contract.
AuterionOS
Auterion describes AuterionOS as a vendor-independent operating system for autonomous robots. It ships on Auterion’s Skynode X and Skynode S and is included free with select partner hardware, and a Jetson Orin version targets heavier AI work. Operators can deploy apps from Auterion or third-party developers, and Auterion positions the OS as a way to manage fleets of interconnected robots rather than single aircraft. In July 2026, Auterion and SkyFall announced 50,000 Shrike strike drones for Ukraine using Skynode S and Auterion’s terminal guidance.
ModalAI VOXL SDK
VOXL SDK is the open-source software stack that ships on every ModalAI VOXL platform, including VOXL 2 and VOXL 2 Mini. It supplies camera pipelines, visual-inertial odometry, SLAM, obstacle avoidance and object recognition, with source code on GitLab. ModalAI recommends its voxl-open-vins-server, built on the OpenVINS research platform, as the current VIO. The SDK hands estimates to PX4 or ArduPilot running on the same board, and it integrates with ROS and OpenCV. The older single-camera voxl-qvio-server is no longer being developed.
NVIDIA Isaac ROS
Isaac ROS is NVIDIA’s open-source collection of accelerated ROS 2 packages. Version 4.0 supports JetPack 7.0 and Jetson AGX Thor. Key packages include Isaac ROS Visual SLAM, built on NVIDIA’s cuVSLAM, and Isaac ROS Nvblox, which builds 3D maps and cost maps for navigation in real time. Version 4.0 also updated Visual SLAM to cuVSLAM 11. For drones, Isaac ROS runs on a Jetson companion computer and can feed position estimates and obstacle data to the flight stack over ROS 2.
Nav2
Nav2 is the professionally supported successor to the ROS Navigation Stack. It uses behavior trees to coordinate planners, controllers and recovery behaviors, and it works with most robot shapes, drive types and sensor setups, indoors or out. Nav2 is available for ROS 2 Humble, Jazzy, Kilted and Rolling, with pre-built Docker images. Open Navigation LLC, led by Nav2 creator Steve Macenski, leads its maintenance. It is built for ground and surface robots, so drone teams mostly use it on ground vehicles in mixed fleets.
Shield AI Hivemind
Shield AI’s Hivemind is an autonomy stack for military aircraft, sold as Hivemind Enterprise with tools including EdgeOS, Pilot, Commander and Forge. Shield AI offers a software development kit so partners and integrators can build their own behaviors on top. Hivemind has flown General Atomics’ MQ-20 Avenger and, in 2025 Navy tests, the Kratos BQM-177A target drone. It also powers Shield AI’s own V-BAT. Hivemind is aimed at defense customers, so expect a licensing and program engagement rather than a download.
Palladyne AI
Palladyne AI sells Palladyne Pilot, platform-agnostic AI software for closed-loop detection, tracking and control on tactical UAVs, and SwarmOS, a decentralized swarming platform. In 2026 it ran SwarmOS on SWaP-limited drone compute during a U.S. Army exercise and announced work with Orkid Tech to bring both products to fixed-wing UAVs. It also integrated SwarmOS with Draganfly drone components. SwarmOS lets drones share sensor data and coordinate without a central controller, and Palladyne Pilot fuses vision, lidar, radar and acoustic inputs for persistent tracking.
Skydio Autonomy
Skydio Autonomy is Skydio’s built-in AI flight system for obstacle avoidance, 360° awareness and GPS-denied flight. It runs on Skydio’s own drones, including the X10, X10D, R10 and F10. Skydio says the system draws on nearly a decade of flight data. Skydio does not publish a license for other manufacturers’ aircraft. It offers APIs and ICDs for integrating third-party systems with its drones instead, so it belongs on this list as a benchmark and a buy option, not as a stack you can install.
Common questions
Should I choose PX4 or ArduPilot?
Both are mature and widely used. PX4’s BSD-3 license suits companies that want to keep code changes private. ArduPilot’s GPLv3 license requires sharing distributed changes, but its community and vehicle support are very broad. Many teams pick based on the hardware they use and the engineers they can hire.
What is the difference between a flight stack and an autonomy stack?
A flight stack such as PX4 or ArduPilot handles stabilization, navigation to waypoints and failsafes. An autonomy stack adds perception and decisions: estimating position without GPS, mapping obstacles and choosing actions. Most drones run the flight stack on a flight controller and the autonomy layer on a companion computer.
Can I run Skydio’s autonomy on my own drone?
Not based on Skydio’s public information. Skydio Autonomy runs on Skydio aircraft, and Skydio’s integration options are APIs and ICDs for connecting other systems to its drones. If you build your own airframe, look at PX4 or ArduPilot with VOXL SDK, Isaac ROS or a licensed stack such as AuterionOS.
Which stack works best without GPS?
Visual-inertial odometry is the usual answer. ModalAI’s VOXL SDK and NVIDIA’s Isaac ROS Visual SLAM both provide it, and Skydio Autonomy includes GPS-denied flight on Skydio drones. Results depend heavily on cameras, lighting and texture, so test in your real environment.
Putting the stack on hardware
Software only performs as well as the compute, cameras and flight controller it runs on. Aerora designs flight and motion control, including ADRC control and embedded Linux BSPs, as part of its engineering work. For partners who build that hardware, see the best edge AI integration partners for autonomous drones and the best robotic autonomy hardware partners.
Sources (32)
- https://px4.io/px4-autopilot-release-v1-16-what-you-need-to-know/
- https://docs.px4.io/main/en/releases/
- https://github.com/PX4/PX4-Autopilot
- https://github.com/ArduPilot/ardupilot
- https://firmware.ardupilot.org/Copter/
- https://ardupilot.org/copter/docs/common-lightware-sf45b.html
- https://en.wikipedia.org/wiki/ArduCopter
- https://auterion.com/product/auterion-os/
- https://auterion.com/product/auterion-os/auterionos-nvidia-jetson-beta/
- https://auterion.com/auterion-and-skyfall-to-ship-50000-shrike-strike-drones-to-ukraines-front-lines/
- https://docs.modalai.com/voxl-sdk/
- https://docs.modalai.com/flying-with-vio/
- https://docs.modalai.com/voxl-open-vins-server/
- https://nvidia-isaac-ros.github.io/releases/index.html
- https://github.com/NVIDIA-ISAAC-ROS/isaac_ros_visual_slam
- https://nvidia-isaac-ros.github.io/repositories_and_packages/isaac_ros_nvblox/index.html
- https://docs.nav2.org/
- https://github.com/ros-navigation/navigation2
- https://steve.macenski.com/about/
- https://shield.ai/hivemind-edgeos-a-game-changer-for-autonomous-robotics/
- https://www.therobotreport.com/shield-ai-raises-250m-scale-hivemind-enterprise-autonomy-developer-platform/
- https://shield.ai/autonomy-for-the-world-mq-20-avenger/
- https://shield.ai/shield-ai-demonstrates-hivemind-on-bqm-177a-supporting-navy-manned-unmanned-teaming-goals/
- https://en.wikipedia.org/wiki/Shield_AI
- https://militaryembedded.com/unmanned/isr/v-bat-drones-and-hivemind-software-to-be-supplied-to-indian-army-by-shield-ai
- https://www.palladyneai.com/products/ai-software/palladyne-pilot-ai-drones/
- https://investor.palladyneai.com/news-releases/news-release-details/palladyne-ai-deploys-true-autonomy-drone-swarming-software
- https://www.therobotreport.com/palladyne-ai-draganfly-reach-integration-milestone-autonomous-drone-swarms/
- https://www.palladyneai.com/press-releases/palladyne-ai-and-orkid-tech-partner-to-extend-palladyne-pilot-and-swarmos-autonomy-platform-to-fixed-wing-uavs/
- https://www.skydio.com/skydio-autonomy
- https://www.skydio.com/developer-tools
- https://en.wikipedia.org/wiki/Skydio
Ready to build what’s next?
Whether you’re developing autonomous aircraft, robotics, intelligent sensing systems or connected platforms, Aerora can help bring your vision to life.
Submit a design inquiry