The pursuit of autonomous drone formation flight represents a significant frontier in robotics and intelligent control systems. The ability for multiple Unmanned Aerial Vehicles (UAVs) to operate in a coordinated manner unlocks capabilities far beyond those of a single agent. A cohesive drone formation can distribute sensor payloads for enhanced situational awareness, perform complex multi-angle imaging, execute synchronized maneuvers, and increase overall mission robustness. While theoretical research into cooperative control algorithms is prolific, a persistent challenge lies in bridging the gap between simulation and reliable, real-world flight experimentation. Many proposed methods for drone formation control remain confined to software simulations due to the practical difficulties of implementing robust inter-vehicle communication and reliable ground control infrastructure on resource-constrained platforms. This work addresses this critical gap by presenting a complete, pragmatic framework for achieving autonomous drone formation flight. We detail the design and validation of a leader-follower cooperative control algorithm, the development of a novel double-serial port ground control station (GCS) to overcome communication limitations, and the presentation of results from both ground vehicle and actual flight tests using our self-developed mini-UAV platform.
The core of any drone formation system is the cooperative control algorithm that dictates how each member of the group behaves in relation to the others. For our initial implementation targeting a two-UAV system, we adopted the well-structured leader-follower paradigm. This mode designates one UAV as the ‘leader’, tasked with following a pre-defined mission path, and the other as the ‘follower’, responsible for maintaining a specific geometric relationship with the leader. This approach simplifies the control problem by decomposing the overall formation task into independent trajectory tracking (for the leader) and relative position keeping (for the follower). Our control architecture is based on a dual-loop design. The inner-loop controllers for attitude, altitude, and velocity stabilization were previously developed and tested extensively on individual drones. The focus here is on the outer-loop drone formation controller, which generates guidance commands for the follower UAV based on the relative state of the pair.
We model the three-dimensional drone formation geometry by decomposing it into separate horizontal plane and vertical plane control problems. Let the leader and follower UAVs be denoted by subscripts \(l\) and \(f\) respectively. Their positions in a local North-East-Down coordinate frame attached to the leader’s desired path are \((x_l, y_l, z_l)\) and \((x_f, y_f, z_f)\). The control objective is for the follower to maintain a desired relative position \((d_x, d_y, d_z)\) with respect to the leader, which often translates to maintaining a specific horizontal distance \(d_c\) and vertical separation \(h_c\). The relative position errors are therefore defined as:
$$e_x = x_f – x_l – d_x, \quad e_y = y_f – y_l – d_y, \quad e_z = z_f – z_l – d_z$$
The kinematics of the relative motion can be described by the following equations, assuming small heading errors and constant leader velocity as a first-order approximation for controller design:
$$\dot{e_x} = V_f \cos(\psi_e) + \dot{\psi}_l e_y – V_l$$
$$\dot{e_y} = V_f \sin(\psi_e) – \dot{\psi}_l e_x$$
$$\dot{e_z} = V_{fz} – V_{lz}$$
Here, \(V_l, V_f\) are the horizontal speeds, \(V_{lz}, V_{fz}\) are the vertical speeds (climb rates), and \(\psi_e = \psi_l – \psi_f\) is the heading error between the two UAVs. All these state variables (position, velocity, heading, etc.) are provided in real-time by the onboard navigation systems of both drones and transmitted to the ground station.
To maintain the drone formation, we designed a Proportional-Integral (PI) controller for the follower. This controller processes the measured relative position errors and outputs velocity, heading, and altitude commands for the follower’s inner-loop autopilot. The control laws are formulated as follows:
$$V_{f}^{cmd} = K_{P,x} e_x + K_{I,x} \int_{0}^{t} e_x \, d\tau$$
$$\psi_{f}^{cmd} = K_{P,y} e_y + K_{I,y} \int_{0}^{t} e_y \, d\tau$$
$$h_{f}^{cmd} = K_{P,z} e_z + K_{I,z} \int_{0}^{t} e_z \, d\tau$$
In these equations, \(V_{f}^{cmd}\), \(\psi_{f}^{cmd}\), and \(h_{f}^{cmd}\) are the commanded airspeed, heading, and altitude for the follower UAV. The \(K_P\) and \(K_I\) gains are tuned for stable and responsive performance. This structure allows for precise tracking of the relative setpoint, effectively keeping the drones in the desired formation geometry. All computation for this drone formation algorithm is performed in real-time on the ground control station.
| Control Channel | Error Input | Command Output | Primary Gain |
|---|---|---|---|
| Horizontal Speed | Longitudinal Error (\(e_x\)) | Follower Airspeed (\(V_{f}^{cmd}\)) | \(K_{P,x}, K_{I,x}\) |
| Horizontal Heading | Lateral Error (\(e_y\)) | Follower Heading (\(\psi_{f}^{cmd}\)) | \(K_{P,y}, K_{I,y}\) |
| Vertical Altitude | Altitude Error (\(e_z\)) | Follower Altitude (\(h_{f}^{cmd}\)) | \(K_{P,z}, K_{I,z}\) |
A critical bottleneck for practical multi-UAV experiments is communication. Small UAVs have strict weight and power constraints, making onboard, air-to-air communication links challenging to implement reliably. To circumvent this, we developed a centralized ground control station with a novel double-serial port architecture. This GCS acts as the communication hub and the ‘brain’ for the drone formation. It receives telemetry data (at 5 Hz) from both the leader and follower UAVs via two separate digital radio links, each connected to a distinct serial port (e.g., COM1 and COM2). It runs the formation control algorithm using the synchronized data, calculates the follower’s commands, and then transmits those commands back to the follower UAV through its dedicated link. The leader simply follows its pre-planned mission.
The software implementation of the dual data streams required careful handling. We utilized a multi-threaded serial communication class. Upon receiving a character, each port triggers an event. The system identifies the port of origin and appends the character to a dedicated buffer. When a complete data packet (86 bytes containing 37 channels of navigation and attitude data) is received, it is unpacked and validated using a 16-bit Cyclic Redundancy Check (CRC) to ensure data integrity. Perhaps the most crucial software challenge was data synchronization. For the formation control algorithm to be accurate, the state estimates for both drones must correspond to the same moment in time. We achieved this by time-stamping incoming data packets using the Coordinated Universal Time (UTC) from each drone’s GPS. The control algorithm only processes pairs of data packets from the leader and follower that possess identical or very closely matched UTC timestamps, ensuring synchronous processing with an error margin of less than one second, which is sufficient for our drone formation purposes.
| Component | Function | Specification/Detail |
|---|---|---|
| Serial Ports | Data I/O for UAVs | Two virtual COM ports (e.g., COM3, COM6) |
| Data Rate | Telemetry Update Frequency | 5 Hz per UAV |
| Data Packet | Transmitted UAV State | 37 parameters (Position, Attitude, Velocity, etc.) |
| Synchronization | Time Alignment | UTC timestamp matching from GPS |
| Error Checking | Data Integrity | 16-bit CRC validation |
Prior to flight tests, we conducted extensive ground validation using a vehicle-based test platform. Two complete autonomous navigation systems were installed in a ground vehicle. Driving along a ~65 km route over 2684 seconds, we simulated flight conditions. The GCS successfully established and maintained stable dual communication links, received synchronized data from both systems, and logged all parameters. The data, as shown in the trajectory and heading plots, demonstrated excellent consistency between the two onboard systems, confirming the reliability of the hardware and the basic data handling software before committing to flight.
The ultimate validation occurred during field flight experiments. The tests utilized two identical, self-developed mini-UAVs with a wingspan of 3.5m, a mass of 9 kg, and a cruise speed of 25-35 m/s. After manual take-off and climb, both UAVs were switched to fully autonomous mode, with UAV #2 designated as the leader and UAV #1 as the follower. The mission consisted of two sequential drone formation phases. In Phase 1, the leader flew a circular path at 220m altitude with a 350m radius. The follower’s objective was to maintain a circular path at a lower altitude of 150m with the same 350m radius, effectively flying a concentric circle below the leader. The GCS computed the relative distance and issued speed, heading, and altitude commands to the follower in real-time.
Phase 2 involved a more complex maneuver: drawing the drones closer vertically while altering their flight radii. The leader maintained 220m altitude and a 350m radius, while the follower was commanded to fly at 210m altitude with a tighter 250m radius. Throughout the 705-second autonomous flight, the GCS interface operated flawlessly, displaying real-time data from both ports. The formation control algorithm effectively managed the relative geometry. Analysis of the flight data indicates that the controller successfully maintained the horizontal distance with an error typically within 30 meters, and the vertical separation error was within 10 meters. The distance between the two UAVs over time during the second phase confirms the stability of the drone formation.
| Flight Phase | Leader Parameters (Alt, Radius) | Follower Command (Alt, Radius) | Performance Metrics |
|---|---|---|---|
| Phase 1 | 220m, 350m | 150m, 350m | Concentric circle tracking established. |
| Phase 2 | 220m, 350m | 210m, 250m | Mean distance error ~20m; Altitude error < 10m. |
This work successfully transitions a theoretical drone formation control concept into a demonstrated real-world capability. We presented a holistic solution encompassing a leader-follower control law designed for practical implementation, a novel double-serial port ground control station that solves the critical data synchronization and communication problem for small UAVs, and a validation process through ground and flight tests. The results prove that our cooperative algorithm is effective and that the ground control system reliably manages the dual data links and real-time command generation necessary for autonomous drone formation flight. The achieved performance, with horizontal tracking errors around 30 meters and vertical errors under 10 meters, is a strong foundation. Future work will focus on improving controller robustness against wind disturbances, optimizing gain scheduling for different flight regimes, reducing altitude error during turns, and scaling the architecture to manage formations with more than two drones. This pragmatic approach provides a clear pathway for advancing from simulated drone formation algorithms to reliable, airborne cooperative systems.