The coordinated operation of multiple Unmanned Aerial Vehicles (UAVs), often referred to as a drone formation, has emerged as a critical area of research, offering significant advantages over single-vehicle operations. Inspired by the efficiency and robustness of flocking birds, drone formation enables enhanced mission capabilities in areas such as environmental monitoring, search and rescue, precision agriculture, and aerial surveying. The benefits of a well-controlled drone formation include improved task efficiency through parallel operation, increased system robustness through redundancy, and potentially reduced energy consumption through aerodynamic drafting effects.
Despite substantial theoretical advancements in multi-agent coordination and drone formation control algorithms, a significant gap persists between simulation-based research and reliable, real-world deployment. Many proposed control strategies are validated solely in simplified numerical simulations, which often fail to capture the full spectrum of real-world complexities. These include hardware limitations, communication delays and dropouts, sensor noise, aerodynamic interactions, and unpredictable environmental disturbances. This disconnect underscores the urgent need for integrated experimentation frameworks that can bridge this gap, allowing for the rapid transition of drone formation theories from concept to field-ready application.
This article presents the design and validation of a comprehensive rapid experimentation system for drone formation flight. The core philosophy is to create a seamless pipeline from high-fidelity simulation to outdoor flight testing. The system comprises two synergistic subsystems: an X-Plane-based Hardware-in-the-Loop (HIL) simulation environment and a physical flight test platform utilizing fixed-wing UAVs. Crucially, both subsystems share identical core hardware (autopilots) and software (ground control station), enabling the direct transfer and validation of control algorithms. This integrated approach allows researchers to identify and rectify issues in the controlled HIL environment, drastically reducing the time, cost, and risk associated with outdoor flight tests for drone formation, thereby accelerating the development cycle.
Design of the Rapid Experimentation System
The efficacy of the rapid experimentation system stems from its mirrored architecture. The HIL simulation subsystem emulates the physical world with high fidelity, while the flight test subsystem executes algorithms in the real environment. Their parallel design ensures continuity and rapid iteration.
Hardware-in-the-Loop (HIL) Simulation Subsystem
The HIL subsystem provides a risk-free, repeatable, and high-fidelity environment for developing and initially validating drone formation control algorithms. It replaces the physical airframe, environment, and sensors with software models while keeping the actual flight computer (autopilot) in the loop.
1. Flight Simulation Software (X-Plane): The subsystem is built around X-Plane, a professional-grade flight simulator renowned for its high-fidelity physics engine. X-Plane utilizes blade element theory to model aerodynamic forces, providing highly realistic flight dynamics for various fixed-wing and rotary-wing aircraft. For drone formation research, multiple instances of X-Plane can be run simultaneously, each modeling the complete six-degree-of-freedom dynamics, kinematics, and sensor output (simulated GPS, IMU data) for an individual UAV within the formation.
2. Networked Autopilot Hardware: The heart of both subsystems is a custom-designed, network-centric autopilot. It is built around a dual-processor architecture featuring two ARM Cortex-M4 cores. One processor is dedicated to core flight control tasks (stabilization, navigation, actuation), while the second is reserved for higher-level functions essential for a drone formation, such as path planning, cooperative control law execution, and inter-agent communication management. The autopilot is equipped with abundant interfaces, including multiple serial ports, SPI, CAN bus, and—most importantly for formation flight—dual Ethernet ports. This extensive connectivity supports complex sensor suites and, crucially, enables direct peer-to-peer communication between UAVs via a network, which is fundamental for decentralized drone formation control strategies.
3. Unified Ground Control Station (GCS): A single, network-aware GCS software is used across both subsystems. It provides real-time monitoring of multiple UAVs simultaneously, displaying attitude, position, velocity, and health status. The GCS allows for in-flight parameter tuning, mission planning (including editing complex paths for the entire drone formation), and comprehensive data logging and analysis. Its ability to communicate via UDP with multiple networked autopilots is key to managing a scalable drone formation.
4. System Integration: The HIL system integrates these components through a standard network switch. Each autopilot, programmed with the candidate drone formation control algorithm, connects to the network. It sends actuator commands (simulated servo and throttle signals) via UDP to its dedicated instance of X-Plane. X-Plane computes the new vehicle state and returns simulated sensor data (attitude, position, velocity) back to the autopilot via UDP, closing the control loop. The GCS also connects to this network, listening to the state broadcasts from all autopilots to display the entire drone formation. This setup creates a realistic testbed where the actual flight hardware executes control laws in a simulated world. The components and their correspondence to the real flight system are summarized below:
| HIL Simulation Component | Real Flight System Component |
|---|---|
| X-Plane Software Instance | Physical UAV, Aerodynamic Environment, Sensors |
| Network Switch | Wireless Data Radio Network |
| Networked Autopilot | Networked Autopilot |
| Ground Control Station Software | Ground Control Station Software |
Physical Flight Test Subsystem
Upon successful validation in HIL simulation, the drone formation control algorithm, embedded in the very same autopilot hardware, transitions directly to the physical flight test subsystem. This subsystem replaces the simulated components with their real-world counterparts.
1. UAV Platform: The testbed utilizes “Skywalker”-type fixed-wing electric UAVs. These platforms are cost-effective, modular, and large enough to carry the necessary avionics payload. Their flight characteristics are well-understood, making them suitable for testing collaborative behaviors in a drone formation. Key parameters are listed in the table below.
| Parameter | Value |
|---|---|
| Wingspan | 1680 mm |
| Length | 1180 mm |
| Maximum Take-off Weight | 1800 g |
| Cruise Speed | 15 – 20 m/s |
2. Avionics and Sensing: A commercial-off-the-shelf integrated navigation module provides the essential state estimation. It combines a MEMS-based IMU (accelerometers, gyroscopes, magnetometers), a GPS receiver, and a barometric sensor. Using sensor fusion algorithms, it outputs real-time estimates of position, velocity, and attitude—critical data for both individual flight control and maintaining the drone formation geometry.
3. Communication Network: A robust, low-latency communication backbone is vital for any multi-UAV system. The subsystem employs Microhard nVIP 2400 radio modules operating in the 2.4 GHz band. Configured in an ad-hoc wireless network, these radios provide Ethernet-like connectivity between all UAVs and the ground station. This architecture enables direct UAV-to-UAV communication, allowing for decentralized information sharing within the drone formation without relying on a ground station relay, which is essential for scalable and fault-tolerant control.
4. Integrated Flight Control System: The autopilot from the HIL tests is integrated with the physical UAV. It connects to the navigation module via a serial port, to the wireless radio via Ethernet, and to the servos and electronic speed controller (ESC) via its internal interfaces. This creates a self-contained flight control system for each UAV in the drone formation.
System Test and Drone Formation Validation
To demonstrate the effectiveness of the integrated experimentation framework, a classic leader-follower drone formation control strategy was implemented and tested through the complete pipeline.
Control Architecture for Drone Formation
Individual UAV Control: Each UAV’s low-level flight control uses an Active Disturbance Rejection Control (ADRC) scheme for the inner attitude loops. ADRC is chosen for its robustness to model uncertainties and external disturbances common in lightweight UAVs. The attitude controller generates roll ($\phi_d$) and pitch ($\theta_d$) commands. These are tracked by the ADRC loops to produce actuator outputs ($\delta_a$, $\delta_e$, $\delta_r$, $\delta_{th}$). Outer loops for trajectory following are implemented using Proportional-Integral-Derivative (PID) controllers. They regulate the vehicle’s ground track. The overall control law for an individual vehicle can be summarized as:
$$ \boldsymbol{\tau}_{inner} = \text{ADRC}(\boldsymbol{\eta}_{d}, \boldsymbol{\eta}, \dot{\boldsymbol{\eta}}) $$
$$ \boldsymbol{\eta}_{d} = \text{PID}_{outer}(\chi_d, \gamma_d, V_d, \boldsymbol{\nu}, \mathbf{p}) $$
where $\boldsymbol{\tau}_{inner}$ represents the actuator commands, $\boldsymbol{\eta}$ and $\dot{\boldsymbol{\eta}}$ are the attitude and angular rate, $\boldsymbol{\nu}$ is the velocity vector, and $\mathbf{p}$ is the position. $\chi_d$, $\gamma_d$, and $V_d$ are the desired heading, flight path angle, and airspeed, respectively.
Leader-Follower Formation Control: A decentralized leader-follower strategy governs the drone formation. The leader UAV follows a predefined mission path. Each follower UAV is tasked with maintaining a prescribed relative position with respect to the leader. Let $\mathbf{p}_L = [x_L, y_L, h_L]^T$ and $\mathbf{p}_F^i = [x_F^i, y_F^i, h_F^i]^T$ denote the north-east-down (NED) positions of the leader and the i-th follower, respectively. The desired relative position is defined in a leader-body frame: longitudinal offset $f_c^i$, lateral offset $l_c^i$, and vertical offset $h_c^i$. The position error for follower $i$ is calculated as:
$$ \mathbf{e}_f^i = \begin{bmatrix} l_c^i \\ f_c^i \\ h_c^i \end{bmatrix} – \mathbf{R}_{L2N}^T (\mathbf{p}_F^i – \mathbf{p}_L) $$
where $\mathbf{R}_{L2N}$ is the rotation matrix from the leader’s body frame to the NED frame. A PID controller then acts on this error to generate the velocity and/or heading commands for the follower’s outer-loop controller:
$$ \begin{bmatrix} \chi_d^i \\ \gamma_d^i \\ V_d^i \end{bmatrix} = \text{PID}_{formation}(\mathbf{e}_f^i) $$
This structure allows the entire drone formation to move as a cohesive unit, with the followers automatically adjusting their flight to maintain the set geometry relative to the moving leader.
HIL Simulation Results for Drone Formation
The complete drone formation control system was first implemented and tested in the HIL environment. A scenario with three UAVs was executed, where the leader followed a rectangular path, and two followers maintained specific offsets. The simulation confirmed the stability and performance of the control laws. The GCS successfully displayed and logged data from all three simulated UAVs in real-time, verifying the multi-vehicle monitoring capability. The HIL tests allowed for extensive tuning of control gains and validation of the inter-UAV communication logic in a safe environment, proving the concept of the networked drone formation before any physical flight.
Physical Flight Test Results for Drone Formation
The autopilots, with the control code validated in HIL, were directly installed into two “Skywalker” UAVs for field testing. The objective was to achieve and maintain a simple leader-follower horizontal drone formation with a longitudinal separation of 50 meters and a lateral separation of 0 meters. The flight tests successfully demonstrated autonomous drone formation flight. The ground control station provided real-time tracking of both UAVs. Analysis of the flight data shows the follower UAV actively regulating its position relative to the leader. While sensor noise, wind disturbances, and communication latency introduced tracking errors not present in simulation, the fundamental drone formation behavior was achieved stably. The key result is that the transition from HIL to flight test required only minor fine-tuning of a few control parameters (e.g., PID gains for the formation controller to account for real-world dynamics and latency), rather than a lengthy redesign process. This highlights the primary benefit of the rapid experimentation system: it drastically reduces the iteration time for deploying and testing advanced drone formation algorithms outdoors.
The evolution of the lateral ($e_l$) and longitudinal ($e_f$) tracking errors during a representative segment of the flight can be modeled to show the convergence tendency:
$$ e_f(t) \approx A_f e^{-\lambda_f t} \cos(\omega_f t + \psi_f) + b_f $$
$$ e_l(t) \approx A_l e^{-\lambda_l t} \cos(\omega_l t + \psi_l) + b_l $$
where $A$, $\lambda$, $\omega$, $\psi$, and $b$ are parameters related to the controller response, platform dynamics, and steady-state bias. The data indicates $\lambda_l > \lambda_f$, suggesting faster convergence in the lateral channel for this specific platform and controller tuning.
Conclusion
This article presented the design and validation of a holistic rapid experimentation framework tailored for advancing drone formation control research. By integrating a high-fidelity X-Plane-based Hardware-in-the-Loop simulation subsystem with a network-centric physical flight test subsystem, the framework effectively bridges the gap between theoretical algorithm development and practical field deployment. The use of identical autopilot hardware and ground control software across both subsystems is the cornerstone of this approach, enabling a “code once, test everywhere” workflow.
The system directly addresses critical challenges in multi-UAV experimentation, including realistic dynamics simulation, robust inter-vehicle communication, and centralized multi-vehicle monitoring. The validation using a leader-follower drone formation control strategy demonstrated that algorithms exhaustively tested and debugged in the HIL environment can be transferred directly to physical UAVs, with flight readiness achieved after only minimal parameter adjustment. This capability significantly reduces the time, cost, and risk associated with outdoor flight trials, accelerating the research and development cycle for complex drone formation behaviors.
Future work will leverage this flexible framework to investigate more advanced, distributed, and self-organizing drone formation control algorithms, fault-tolerant formation reconfiguration, and collaborative sensing missions, further pushing the boundaries of autonomous multi-agent aerial systems.