By Fernando Cladera and Mike Pehel
Every year the PX4 ecosystem sees milestone advancements in research. This research provides a foundation for the real-world applications that expand the capabilities of autonomous platforms. We wanted to share some exceptional papers we found to help inspire you and bring to light some of the possibilities in the world of PX4.
OmniNxt: A Fully Open-source and Compact Aerial Robot with Omnidirectional Visual Perception
Peize Liu et al, HKUST Aerial Robotics Group, The Hong Kong University of Science and Technology
It’s always exciting to see a full-fledged open source drone platform in a paper. Here, the researchers showcase a multi-fisheye camera setup utilizing a Jetson Orin NX capable of omni-directional perception. This low SWaP setup is capable of indoor navigation in cluttered indoor environments at speeds up to 1.0 m/s while avoiding obstacles.
The paper also claims the platform demonstrates superior VIO accuracy compared to limited-FoV stereo systems. Notably, the omnidirectional perception allows the drone to fly without yaw rotation for trajectory following, which improves both localization accuracy and energy efficiency. The project makes omnidirectional perception more accessible to the broader research community through complete open-source hardware and software and we’re excited to see what the team does next.
A Neural Network Mode for PX4 on Embedded Flight Controllers
Sindre M. Hegre et al, Autonomous Robots Lab, Norwegian University of Science and Technology
This paper presents a significant advancement by creating a neural flight controller for PX4. The key innovation is developing a custom module that enables TensorFlow Lite neural networks to run directly on embedded microcontrollers (like the STM32H743), completely replacing the traditional cascaded PID control system with an end-to-end neural controller that directly outputs motor commands.
The system as described by the paper demonstrates remarkable efficiency with inference, showing successful simulation-to-reality transfer while preserving PX4’s safety features. Flight controllers are extremely resource constrained compared to the devices neural networks (and other AIs) usually run on, so this progress is noteworthy, especially considering it was achieved without the need for a companion computer to offload processing.
Trajectory Planning and Control for Differentially Flat Fixed-Wing Aerial Systems
Luca Morando et al, The agile robotics and perception lab, New York University
Presented at the ICRA 25 Aerial Robotics Workshop and main conference track, this paper introduces a novel approach to fixed-wing navigation using differential flatness and Bernstein polynomials similarl to “minimum snap” method widely used in quadrotors. The implementation is dynamic and adapts to the curved trajectory constraints of the aircraft while adhering to the optimal path. This could potentially enable aggressive flight paths for fixed-wing platforms and enhanced maneuverability overall.
Beyond the practical results, the methodology behind the research can also lead to faster deployments. They trained their system in Gazebo and the results show a very close sim-to-real gap. This assures a smoother transition when moving out of the simulator and into the real world.
Dynamic End Effector Trajectory Tracking for Small-Scale Underwater Vehicle-Manipulator Systems (UVMS): Modeling, Control, and Experimental Validation
Niklas Trekel et al, Institute of Mechanics and Ocean Engineering, TU Hamburg
Sometimes it seems like the UAVs get all the attention, but PX4 also gets utilized in AUVs in maritime environments. This research showcases improved trajectory tracking for underwater vehicle-manipulator systems (UVMSs) on the BlueROV2 platform from our friends over at BlueRobotics. Even small disturbances can significantly impact the operation of manipulators or really any moving part of the AUV, so to see such buttery smooth motion from on the arm in this video is impressive. This represents important progress in making underwater manipulation more accessible through commercial, lightweight platforms rather than requiring expensive, large-scale research vessels.
Looking Forward
What makes this wave of research particularly exciting is its commitment to open-source accessibility, ensuring these cutting-edge capabilities reach the widest audience possible, and pushing the limits of what PX4 controllers can do.This is just a small slice of the great work being done, so feel free to share with us other interesting research or your own projects on social media or in the Dronecode Discord channel. If we get enough feedback, we’ll be sure to highlight ongoing research more often!
