Several recent reviews and survey studies have comprehensively explored a wide range of topics related to UAV autopilots and flight controllers, including foundational hardware and software architectures, open-source ecosystems, and the role of simulators in testing and validation.

In terms of UAV components and system architecture, the work in [11] presents a state-of-the-art survey of multirotor drones covering drone types, construction methods, and the underlying technologies. Similarly, reference [15] provides an interdisciplinary overview of UAV hardware and software architecture and their applications, while references [16,17] specifically focus on flight control systems.

A number of studies address guidance, navigation, and control in UAV systems. For instance, reference [18] discusses techniques for autopilot integration in small-unmanned aircraft systems (UAS), reference [19] emphasizes obstacle avoidance strategies, and reference [20] reviews estimation techniques across different flight phases. The reliability of UAV computing platforms is examined in [21], which analyzes fault and failure modes in mission-critical applications.

The role of open-source ecosystems and operating systems is another key consideration in UAV autopilot development. For example, reference [2] surveys open-source drone platforms, including hardware, software, and simulation environments, while reference [22] reviews open-source autopilots used in drones and compares common platforms. Further, reference [10] provides an overview of open-source UAV autopilot platforms, and reference [23] discusses these platforms from the perspective of real-time and operating systems, offering insights into their strengths and limitations.

Finally, simulation and co-simulation methodologies are recognized as critical tools for enabling safe and cost-effective UAV development. The work in [24] reviews current UAV simulators, while reference [25] advances this area by presenting step-by-step implementations of real-time co-simulations between Simulink and flight simulators such as X-Plane and Flight Gear, demonstrating model validation and controller testing in both SITL and HITL configurations.

All these overview and survey papers provide a nuanced understanding of the UAV landscape, highlighting advancements in flight controller architecture design, the influence of open-source ecosystems, and the essential role of simulators in testing and validation.

Despite the extensive survey literature on UAV autopilots, existing reviews show significant gaps in their treatment of the Pixhawk/PX4 ecosystem. A primary limitation is the fragmented focus of prior work, which typically addresses specific aspects, such as software architecture, guidance and control algorithms, or open-source development, in isolation, rather than evaluating the synergistic potential of the integrated Pixhawk/PX4 platform. Furthermore, there is an evident absence of synthesis concerning the practical integration of this ecosystem with complementary development tools and emerging technologies.

To address these gaps, the primary objective of this paper is to analyze the multifaceted capabilities enabled by the Pixhawk/PX4 ecosystem for developing advanced UAV applications. Specifically, this survey examines scientific literature to aggregate evidence on: (1) the integration of companion computers for onboard high-level processing; (2) the integration with ROS middleware; (3) the combination with MATLAB/Simulink for advanced modeling and control design; (4) the deployment of SITL and HITL environments for testing and validation; (5) the incorporation of cutting-edge technologies, including machine learning (ML) and artificial intelligence (AI) models as well as cybersecurity protocols. In addition to these technical dimensions, this survey also investigates the diverse contexts in which the Pixhawk/PX4 platform has been employed in educational context.

By mapping these convergent areas, this survey aims to provide researchers and educators with a holistic reference for leveraging the Pixhawk/PX4 ecosystem in the development of autonomous systems.

The remainder of this survey is organized as follows. Section 2 outlines the materials and methods guiding this review. Section 3 provides a technical overview of the Pixhawk/PX4 platform, detailing its key hardware and software features. In Section 4, we analyze the scientific literature on utilizing the Pixhawk/PX4 ecosystem for UAV applications. Section 5 shifts focus on discussing its role in educational context. Section 6 synthesizes the main findings and highlights critical issues. Finally, Section 7 summarizes the main conclusions.

The Pixhawk hardware usually comes as a single board autopilot consisting of two main units: the flight management unit (PX-FMU) and the input-output board (PX4-IO). The FMU units are based on 32-bit ARM microcontrollers running the PX4 flight controller, and feature backup failsafe 32-bit co-processors to enable recovery from a primary system failure. All platforms provide several PWM outputs for servos control and other uses and come with some internal sensors including a three-axis gyroscope, a three-axis accelerometer, a three-axis magnetometer, a barometric pressure sensor, a global navigation satellite system (GNSS), a voltage sensor for battery under voltage detection, a built-in safety switch, etc. In addition, it provides sockets for direct connection with external sensors and hardware such as GPS and compass, a buzzer, and an I2C bus for connecting to four external sensors. Finally, it offers connectivity with peripherals through UART and CAN buses. The flight controller (FC) can be operated through a radio control unit or remotely via a ground control station using the MAVLink protocol [27]. Fig. 2 illustrates the interface architecture of the Pixhawk and its associated peripherals.

Pixhawk autopilots come in various versions, differing in hardware specifications such as processor speed, PWM channel count, and IMU redundancy, providing flexibility for diverse applications. For example, the high-end Pixhawk 6X features an STM32H753 processor, triple redundant IMUs, and dual barometers, making it ideal for demanding use cases. Meanwhile, older models like the Pixhawk-1 and Pixhawk-2 have been discontinued. In addition to official Pixhawk units, several third-party autopilots are derived from the platform, offering enhanced features or cost-effective alternatives. Examples include MindRacer, Cube Orange, and Kakute. Advanced configurations are also available, such as integrated boards combining a Pixhawk 5X/6X with a Raspberry Pi CM4 [28] or a Pixhawk 6X paired with an NVIDIA Jetson Orin in a single unit [29]. Additionally, Auterion SkyNode, launched in 2021, is an enterprise-grade Pixhawk/PX4 designed for professional and industrial drones [30]. It supports optional companion computers like the Raspberry Pi CM4 or i.MX 8M, enabling edge computing, cloud integration, and enhanced security features. Table 1 shows some versions developed according to Pixhawk standards, and Fig. 3 illustrates the appearance of select Pixhawk models.

The progression of Pixhawk hardware, as summarized in table-1, has significantly enhanced its suitability for research and education. The shift from STM32F4 to more powerful STM32H7 processors, along with increased clock speeds, enables the execution of computationally intensive algorithms, such as advanced filtering, real-time sensor fusion, and lightweight machine learning, directly on the autopilot, reducing reliance on external companion computers. The introduction of features like triple redundant IMUs and dual barometers improve fault tolerance and safety, which is critical for experimental reliability and student projects. Expanded connectivity (e.g., more PWM outputs, CAN bus, and Ethernet) supports the integration of diverse sensors, custom payloads, and companion computers, facilitating scalable and interdisciplinary UAV platforms. This ongoing hardware evolution, coupled with strong standardization, has made Pixhawk a reproducible, accessible, and professional-grade tool that lowers the barrier to advanced experimentation while supporting innovation in both academic and applied research contexts.

PX4 is the open-source flight control software that operates on Pixhawk hardware. It runs on NuttX [31], a POSIX-compliant real-time operating system (RTOS) that enables multi-threading and advanced scheduling. The software is written in C/C++ and consists of two primary components: the flight control stack and the middleware, which handles internal communications [26].

The flight control stack is the core of PX4, responsible for navigation and control algorithms used for various drones, including multirotor, fixed wing, and vertical takeoff and landing (VTOL) airframes. It supports various flight modes, such as manual, altitude hold, and position hold. At its core, the stack manages body position, velocity, and attitude (yaw, pitch, roll) using independent PID controllers. Additionally, PX4 also employs an extended Kalman filter (EKF) to fuse data from IMUs, GPS, compasses, air speed sensors, and barometer, estimating position. The EKF algorithm serves two key functions: estimating unmeasurable states like position and velocity and refining angular orientation by filtering sensor noise.

The middleware layer acts as a bridge between the flight control stack and hardware, facilitating communication with external devices like GCS and companion computers. It consists of three main components. The micro–Object Request Broker (uORB) [32], a messaging system enabling real-time data exchange between modules using a publish-subscribe pattern and relies on shared memory. The second consists of parameter subsystem, a data table for storing and retrieving system parameters such as vehicle configurations, PID values, and mission waypoints. PX4 provides users with C/C++ callback functions to listen to changes in uORB topics associated with parameter up-dates. The third component is a hardware drivers’ layer, managing system initialization, configuration, and low-level interactions like data acquisition and actuator control. This middleware allows developers to focus on high-level control algorithms without delving into hardware specifics. Finally, PX4 supports custom module integration, enabling advanced users to extend functionality by embedding C/C++ modules into the firmware. Fig. 4 illustrates the PX4 software architecture.

A Ground Control Station (GCS) is a computer or mobile device that enables remote monitoring and control of an unmanned aerial vehicle through wireless serial communication. GCS allows operators to plan and execute autonomous missions, receive real-time telemetry data, issue flight commands, and in some implementations, process collected data and imagery. Within the Pixhawk/PX4 ecosystem, QGroundControl [33] serves as the primary reference implementation for GCS software. However, the platform maintains compatibility with alternative solutions including Mission Planner [34], MAVProxy [35], among others.

MAVLink (Micro Air Vehicle Link) [27] is a lightweight communication protocol for exchanging messages between unmanned vehicles and GCSs. Originally developed by Lorenz Meier at ETH Zurich, its latest version, MAVLink 2.0, was released in 2017. The protocol enables bidirectional communication, allowing UAVs to transmit telemetry data (e.g., system state, location, velocity, altitude) to the GCS while receiving commands (e.g., takeoff, landing, mission execution) in return. MAVLink employs a compact binary packet format, with a messages size ranging from 12 to 280 bytes for signed messages. Each packet includes 2 bytes for checksum data integrity and requiring acknowledgment, and it includes 13 bytes signature to ensure tamper-proof communication. Messages are serialized into binary for transmission and deserialized upon receipt, optimizing efficiency in low-bandwidth environments.

MAVLink packets can be transmitted over various communication links, including sub-GHz frequencies (433, 868, 915 MHz), Wi-Fi, serial telemetry, or TCP/IP networks. Pixhawk 5X and later models feature Ethernet support, enabling MAVLink streaming via UDP or TCP over IP networks. A key capability of MAVLink is its routing feature, which facilitates communication between networked systems (e.g., vehicles, GCSs, antenna trackers) and onboard components (e.g., autopilot, cameras, gimbals), each assigned a unique ID number. Tools like MAVLink Inspector allow users to monitor traffic, inspect messages, and visualize real-time data through dynamic charts.

The MAVLink protocol supports a diverse range of units and systems, with PX4 flight controllers serving as the central node for vehicle control and telemetry. The autopilot is often augmented by an onboard companion computer (e.g., Raspberry Pi, NVIDIA Jetson) that handles high-level tasks such as computer vision, path planning, and obstacle avoidance. On the ground, Ground Control Stations (GCS) like QGroundControl or Mission Planner provide the primary human interface for monitoring and command. The protocol also extends to onboard payloads and peripherals. For camera control specifically, MAVLink includes the dedicated Camera Protocol [36], a specialized subset that enables real-time remote adjustment of zoom, focus, shutter speed, aperture, and image or video capture. Cameras must support MAVLink either natively or through a compatible companion computer. Beyond onboard components, the protocol facilitates interoperability with external ecosystems such as ROS/ROS2 and MATlab/Simulink through dedicated bridge implementations. For testing and validation, SITL simulations and HITL rigs communicate via MAVLink to validate system behavior before deployment. Finally, custom applications built with MAVLink SDKs (e.g., DroneKit, MAVSDK) further demonstrate the protocol’s versatility, enabling automated and programmable interaction with unmanned systems across a wide range of use cases. Fig. 5 illustrates the interoperability of PX4 Flight Controller through MAVLink protocol.

The Pixhawk/PX4 system supports integration with companion computers (CCs), such as Raspberry Pi, NVIDIA Jetson, or Intel NUC devices, to offload computationally demanding tasks, such as real-time image processing, autonomous navigation, and custom flight mode development. These CCs can be deployed in three configurations: onboard, swappable, or offboard.

In the onboard configuration, the CCs interface with the Pixhawk via serial/USB over MAVLink. Such configuration can be found in works such as in [40], which combines a Raspberry Pi for low-level tasks and an Intel NUC for high-level mission planning. The work [41] proposes a multi-UAV architecture leveraging the PX4 autopilot, where Raspberry Pi and Jetson companion computers are integrated to facilitate task distribution and optimized communication, thereby supporting swarm autonomy. Further, the works [42,43] combines NVIDIA Jetson TX1 and Nano for stereo vision and depth map processing for obstacle avoidance, [44–46] that utilize Raspberry Pi and NVIDIA Jetson for image processing and visual servoing, and [47,48] that utilize Odroid M1 and Intel i7 NUC CCs deploying dynamic flight maneuvers and predictive control.

The swappable configuration employs integrated solutions such as the Pixhawk CM4 module (combining a Pixhawk 5X/6X with a Raspberry Pi CM4) [28] or the Pixhawk 6X paired with an NVIDIA Jetson Orin in a unified board design [29]. This approach has been used for Time-of-Flight (ToF) positioning [49], implementation of a model predictive control [50,51] demonstrating its ability to handle computationally intensive tasks within compact form factors.

The offboard configuration leverages external computing power without increasing UAV weight, utilizing Wi-Fi or 5G for high-bandwidth communication between the autopilot and CC. In this setup, the CC serves as a bridge between the UAV and the GCS, managing data transmission, telemetry, remote commands, and high-level autonomy algorithms like mission planning or swarm coordination. For instance, reference [52] demonstrates an indoor localization system using a VICON motion capture system with reflective markers, where an offboard companion computer communicates with the UAV’s Pixhawk via MAVLink. Similarly, reference [53] enables aggressive UAV maneuvers through predefined waypoints, validated indoors with an OptiTrack system and offboard CC for real-time pose estimation and correction. Further, reference [54] extends this concept to multi-UAV cooperative transport of cable-suspended loads, employing Odroid-XU4 CCs (one per UAV) with centralized trajectory planning via WiFi, tested on three PX4-based UAVs.

DroneKit [36] and MAVSDK [37] are powerful APIs that enable advanced UAV control and mission management by interfacing with PX4 autopilots. DroneKit is a Python-based library that simplifies interaction with PX4 via the MAVLink protocol, allowing developers to monitor and control drone operations programmatically. It provides high-level functionalities such as real-time telemetry access, direct vehicle control, and seamless integration with other Python libraries for enhanced data processing. DroneKit can be executed on both GCS and CC. Built in C++, MAVSDK supports multiple programming languages (C++, Python, Swift, Rust, Java), though only its C++ binding allows simultaneous multi-UAV connections; other language wrappers are restricted to single-vehicle operation. A key feature is its offboard mode permitting programmatic control of UAVs by sending MAVLink commands (e.g., position, velocity, or attitude setpoints) from GCS or CC. While DroneKit continues to be utilized in legacy development projects, it is critical to note that it is no longer maintained, which poses a risk for long-term reproducibility and educational sustainability. In contrast, MAVSDK, released in 2018, is the actively maintained and most up-to-date high-level API for PX4, serving as the primary recommended interface for MAVLink-based drone control.

In this context, several articles demonstrate the use of DroneKit for UAVs control and coordination. For example, the papers [55–57] employ DroneKit to implement guided flight modes, multi-UAV cooperation, and obstacle avoidance, with commands transmitted from CC to Pixhawk flight controllers. Innovative approaches in using DroneKit to translate hand gestures [58] and voice commands [59] into UAV maneuvers, offering an alternative to traditional GCS or joystick inputs.

On the other hand, MAVSDK is also adopted for advanced UAV operations and swarm coordination. The work [60] uses MAVSDK’s offboard mode to execute precision tasks like target tracking, obstacle avoidance, and surveillance through direct command and waypoint transmission, while reference [61] demonstrates the API capability for precision landing in dynamic environments by updating flight paths based on visual feedback. Swarm-specific applications are explored in [62–64], where MAVSDK enables MAVLink-based inter-vehicle communication for real-time monitoring, navigation, and collaborative mission execution.

While DroneKit offers accessible Python-based scripting for basic UAV automation, MAVSDK provides a robust and future-proof solution for advanced control and swarm applications.

Pixhawk/PX4 ecosystem leverages ROS/ROS2 middleware [38], an open-source framework for robotics programming that provides a comprehensive suite of tools and utilities for UAV development, enabling rapid prototyping, simulation, and deployment. Its modular architecture and community support make it ideal for building autonomous UAV systems. Early integration relies on ROS and the MAVROS package, which acts as a bridge translating MAVLink messages into ROS topics and services. However, the current pathway for new developments is ROS2, which connects to PX4 natively via the Micro XRCE-DDS middleware. This shift is not merely nominal; ROS2 offers critical advantages for UAV applications, including a true real-time capable architecture, enhanced security features, and more reliable quality-of-service (QoS) controls for networked and multi-vehicle systems.

ROS2 is typically installed on a companion computer and interfaces with PX4 by reading or writing to internal uORB topics [32]. Beyond core communication features (topics, services, actions) and sensor integration (cameras, IMUs, GPS, LIDAR), ROS2 provides specialized packages for critical UAV functions, such as state estimation (e.g., vision inertial odometry/VIO, extended Kalman filters/EKF for sensor fusion, SLAM for navigation in unknown environments), perception functions (e.g., point cloud libraries/PCL for LIDAR and depth camera data processing), and visualization and debugging (e.g., RViz for real-time monitoring, Rosbag for data logging).

In this regard, the integration of ROS/ROS2 with the PX4 ecosystem has been widely explored in recent research. Studies such as [65–67] demonstrate the use of vision inertial odometry (VIO) and extended Kalman filters (EKF) for robust state estimation in GPS-denied or indoor environments. Simulation platforms leveraging PX4-ROS-Gazebo environments are another key focus, with enabling customizable testing of multi-robot collaboration, SLAM, and navigation algorithms [68–70], while reference [71] provides a comparative analysis of PX4-ROS-AirSim simulations versus real-world experiments. Further applications include autonomous navigation [72,73], marker-based precision landing [74,75], fault-tolerant control is explored in [76], developing a ROS-based con-troller for rotor failure, and [77,78] focus on point cloud data generation and laser-based localization, respectively, using ROS utilities like the PCL and SLAM. Collectively, these works showcase the versatility of the PX4-ROS ecosystem in advancing UAV capabilities across simulation and real-world deployment.

Furthermore, the ROS enables integration of third-party libraries through standardized packaging. By wrapping specialized algorithms like LIO-SAM (Lidar Inertial Odometry via Smoothing and Mapping), RTAB-Map (Real-Time Appearance-Based Mapping), VINS-Fusion (Visual-Inertial SLAM), and Kimera (real-time metric-semantic localization and mapping) into ROS packages, developers can leverage advanced capabilities while maintaining interoperability with existing ROS nodes.

Numerous contributions have adopted this approach, demonstrating their effectiveness across various challenging scenarios. For example, reference [79] combines ORB-SLAM3 for visual-inertial SLAM in GPS-denied environments with VINS-Fusion for robust visual-inertial odometry in dynamic conditions. Ref. [80] presents the application of LIO-SAM for precise localization using LiDAR-inertial odometry in challenging outdoor scenarios. Similarly, reference [81] leverages RTAB-Map for large-scale environmental mapping using RGB-D and LiDAR data, while reference [82] uses Kimera, a library for semantic mapping, which enables UAVs to interpret and interact with their environment at a higher level. These implementations collectively showcase how ROS-integrated perception algorithms can significantly enhance UAV autonomy across diverse operational conditions.

The ROS ecosystem’s flexibility in integrating cutting-edge algorithms significantly enhances the Pixhawk/PX4 platform’s capabilities, enabling seamless incorporation of state-of-the-art technologies for advanced UAV applications.

MATLAB offers UAV developers two complementary UAV toolboxes tailored for different development needs. The UAV Toolbox provides platform-agnostic tools for simulating and designing UAV algorithms, including mission planning, sensor modeling (e.g., cameras, LiDAR, IMUs), and environment simulation in both 2.5D and 3D. This toolbox is particularly suited for theoretical research and early-stage validation. The UAV Support Package for Pixhawk/PX4 [13] adds hardware-specific functions, such as direct Pixhawk interfacing, MAVLink communication, and automatic C++ code generation optimized for PX4. Together, these toolboxes streamline the entire development workflow, from algorithm design and simulation to real-world deployment, significantly accelerating prototyping and testing. This integrated solution makes MATLAB a valuable tool for UAV researchers and engineers working with PX4 autopilots.

In this regard, several contributions demonstrate the capabilities of MATLAB’s UAV Toolbox for PX4-based development. Works such as [12,83] utilize MATLAB/Simulink for multicopter system development, providing step-by-step guidance that serves as valuable educational and research resources. Further contributions [84–86] focus on system identification and PID controller tuning, leveraging MATLAB’s automatic C/C++ code generation to deploy optimized control algorithms directly on Pixhawk autopilots. For advanced control applications, papers [87–89] implement and validate techniques like sliding mode control and nonlinear backstepping control in MATLAB/Simulink before hardware deployment, demonstrating their effectiveness for challenging scenarios such as aerial slung load transportation. The work [90–92] demonstrate the use of model-based design and automatic code generation to deploy customized control algorithms on Pixhawk and companion computers like the Jetson Xavier. Finally, [93,94] address path planning and glider control, respectively, with simulations conducted in MATLAB/Simulink and real-world validation on Pixhawk-based systems. These works underscore the synergy between MATLAB/Simulink and PX4 for developing, testing, and deploying advanced UAV control systems.

SITL and HITL simulation tools [39] are indispensable for safely testing and validating UAV software and control algorithms in a controlled environment, bridging the gap between design and real-world deployment.

In SITL simulations, the PX4 flight controller runs on a computer and controls a simulated vehicle in a virtual environment, which may operate on the same machine or a separate one. Communication between the simulated vehicle and a ground control station (GCS) is carried out via MAVLink messages, mirroring real-world interactions. In HITL simulations, the PX4 firmware runs on actual Pixhawk flight controller hardware, controlling a simulated vehicle that provides sensor feedback to the PX4 system. This approach allows developers to test most of the flight code on real hardware while maintaining a safe testing environment.

PX4 supports several open-source simulators for SITL and HITL simulations, with the most prominent being Gazebo [95], jMavSim [96], and AirSim [97]. Gazebo is the most widely used simulator in the Pixhawk/PX4 ecosystem. It enables complex 3D environment simulations, supports multi-vehicle operations, and integrates sensor models (e.g., cameras, LiDARs, sonars). Additionally, it offers ROS compatibility and cloud connectivity. AirSim, developed by Microsoft, simulates both aerial and ground vehicles in diverse environments. It features realistic visuals, dynamic weather conditions, and high-fidelity physics. jMavSim, created by the Pixhawk team, is lightweight and easy to configure for basic UAV simulations. However, it lacks support for additional sensors beyond its default set and cannot simulate multiple vehicles simultaneously. All these simulators communicate with PX4 via MAVLink messages, exchanging sensor data from the simulation and actuator commands from the flight controller. In Gazebo, this interaction is further optimized through direct PX4 uORB topic integration.

Concerning SITL simulations, often integrated with Gazebo and ROS, have been extensively used in UAV research to validate control algorithms, path planning, and mission-specific applications. Works such as [88,98,99], leverage PX4-SITL to test advanced control methods like quasi-static state feedback (QSF), nonlinear backstepping, and disturbance observer-based control, while [100–102] focus on path planning for obstacle avoidance, ground target tracking, and automated power line inspection. Multi-UAV coordination, including swarm behaviors and collaborative payload transport, are explored in [103–105], with PX4-SITL serving as the validation platform. Further works [71,106,107] integrate MATLAB/Simulink, AirSim, and ROS with PX4-SITL to simulate and compare multi-drone scenarios against real-world experiments, while [108,109] employ PX4-SITL to evaluate vision-based algorithms and sensor fusion. Practical applications, such as optimized landing strategies and Cartesian mode implementation, are demonstrated in [110,111], highlighting PX4-SITL’s versatility as a robust tool for developing, testing, and deploying advanced UAV systems.

On the HITL front, numerous contributions demonstrate the extensive use of PX4-Gazebo integrations for realistic UAV testing. For example, works [112–114] utilize Gazebo to simulate UAV dynamics and sensor models while implementing flight control algorithms on Pixhawk hardware. Advanced control techniques are investigated in [115], which validate incremental nonlinear dynamic inversion and robust control methods through Pixhawk-based HITL simulations. Vision-based applications are ad-dressed in [116,117], testing autonomous navigation and inspection systems via PX4-ROS-Gazebo integration. The versatility of HITL platforms is further shown in [118–120], which employ MATLAB/Simulink, X-Plane, and Flightmare Renderer. Finally, [121,122] leverage AirSim and custom physics engines for object detection, path planning, and agile control validation.

The open and modular architecture of the Pixhawk/PX4 autopilot system, supported by its robust development ecosystem, as well as while it compatibility with GPU-enabled companion computers allows the implementation of AI/ML solutions to enhance UAV capabilities, including AI-driven sensor fusion for improved perception by combining LiDAR, IMU, and camera data, deep learning-based semantic perception for, and reinforcement learning optimized for navigation in complex environments. It also supports swarm intelligence for collaborative multi-agent operations. By leveraging simulation tools like Gazebo, AirSim, and jMAVSim, developers can train and validate AI models using synthetic data. This section explores the contexts and key use cases that demonstrate the integration of these emerging technologies.

In fact, numerous initiatives have successfully integrated AI-based solutions to enhance UAV capabilities. For instance, reference [123] creates a synthetic visual dataset using PX4-SITL and AirSim to train ML-algorithms for autonomous personal aerial vehicles, while references [124,125] leverages the PX4-ROS-Gazebo ecosystem to generate synthetic data for deep learning-based pose estimation and explainable path planning, respectively. Similarly, references [126–128] validate AI-driven path planning and swarm models using PX4-ROS-Gazebo simulations. Real-world implementations often rely on edge computing platforms, such as NVIDIA Jetson Nano and Jetson TX2 [129], for executing trained models in tasks like image processing and obstacle avoidance. Fault-tolerance control is also a key research area. In this front, reference [48] proposes an ANN-based fault-tolerant control system to replace faulty sensors, reference [130] introduces a real-time propeller fault-detection method using SVM classifiers, and reference [131] addresses GPS faults through machine learning models integrated with EKF estimators.

Environmental perception has been advanced using YOLO for fire detection and geo-localization [132] and FPGA-accelerated powerline perception combining mmWave radar and RGB cameras [133]. Further applications include DNN-based real-time flood tracking [134] and Mask R-CNN-assisted litter detection [135]. Additionally, natural language processing (NLP) is also integrated into UAV control, with [128,136] employing DNNs and GPT-3.5-based architectures to translate voice commands into executable tasks within PX4-ROS-Gazebo frameworks.

Researchers are leveraging a range of advanced solutions to address UAV security challenges. Initial efforts involved several ad-hoc approaches, including encryption techniques such as using a Raspberry Pi for encrypted communication links [137] and implementing the ChaCha20 algorithm through the MAVSec framework [138,139]. Other proposed methods include software rejuvenation, which employs intermittent system restarts to mitigate runtime attacks [140]. Machine learning techniques, particularly LSTM-based neural networks, have also been explored for detecting denial of service (DoS) attacks [141,142]. GPS-spoofing attacks, where hijackers use devices to deceive GPS receivers, have also garnered attention. In this sense, work in [143] describes a method to detect whether it has been hijacked using only GPS measurements. Beyond these, more sophisticated solutions continue to emerge in response to evolving threats. For instance, reference [144] introduces an intelligent intrusion detection system (IDS) that utilizes principal component analysis (PCA) and one-class classifiers to detect attacks, leveraging flight logs for training the models. Similarly, reference [145] proposes a deep learning-based pilot authentication scheme, DP-Authentication, which dynamically verifies pilot legitimacy using flight data collected from the PX4 flight stack. Further, the work [146] proposes a fusion multi-tier DNN-based collaborative intrusion detection and prevention system (CIDPS) for UAV-enabled 6G networks to neutralize attacks, which are implemented on actual PX4-based UAVs. These contributions underscore the potential of AI techniques to address evolving cybersecurity challenges in UAV systems.

All these studies demonstrate the PX4 ecosystem’s effectiveness in supporting AI integration across various UAV applications.