The synchronization of indoor collaboration for a group of Unmanned Aerial Vehicle (UAV)s presents challenges due to instability of the Global Positioning System (GPS) signal, cluttered environments, and dynamic obstacles. This paper investigates the accuracy of the navigation of two UAVs during a synchronized flight using DJI Tello UAVs. Additionally, it studies the tracking problem by introducing delays into the system to provide a more accurate representation of a real-world control scenario. In addition to the theoretical analysis, this paper presents a set of experiments in which diferent synchronous flight control options are tested and compared under diferent conditions. A Proportional Integral Derivative Controller (PID) is considered, which is adapted to control the maintenance of a predetermined distance between two UAVs. Several simulations are conducted to evaluate the performance of the above approaches. The obtained results demonstrate that the proposed methods for evaluating positioning accuracy during the execution of synchronized actions by a UAV group are applicable to future indoor localization analysis scenarios.
The coordinated operation of robots is being utilized in automation projects in high-tech industries to leverage their collective capabilities. Coordinating the control of multiple devices presents significant technical challenges, particularly when using low-cost UAV models. For instance, the DJI Tello has Visual Positioning System (VPS) and comes with a Python library that enables synchronized flight of multiple UAVs. The objective of this work is to improve the quality of DJI Tello synchronization by selecting a control architecture and configuration that ensures
commands, improves flight stability, and prevents possible collisions. In general, synchronization is
fundamental to improve the positioning accuracy of UAVs flights, especially in indoor localization
scenarios, such as those related to logistic automation [
https://github.com/OlhaPohudina/SynchronizationDJITello/ (O. Pohudina)
CEUR
ceur-ws.org
the efectiveness of the control algorithms plays a crucial role in processing the data collected by these
sensors and adjusting the motor power in real time to correct the UAV’s position and orientation. As
well, the wireless communication between the control station and UAVs introduces delays in data
transmission, which can change the relevance of information related to position, speed, and direction
of UAV. If this information is delayed due to interference or channel variability, the response of local
UAV controllers may be inaccurate, jeopardizing its stability [
UAVs using a common Wireless-Fidelity (Wi-Fi) network; • leader-slave control: the slave UAV autonomously follows the leader UAV.
The following architectures are proposed to develop the three variants of synchronized flight, previously described. In the configurations, increasing levels of complexity and cooperation, both in network management and in coordination between UAVs, are considered. The leader UAV is visually identified by a QR code attached to its back.
In the first architecture, shown in Figure 1, each UAV operates in Access Point (AP) mode, creating its own independent Wi-Fi network. Each UAV is connected to a control station, like a computer. These two separate control stations run the same Python script in parallel to send flight commands. In this configuration, both UAVs receive the same sequence of instructions and follow the same flight plan. The synchronization of movements relies heavily on the simultaneous execution of scripts at the two control stations, without any explicit coordination mechanism between the two systems, relying solely on manual synchronization.
Communication between computers and UAVs is based on the protocol User Datagram Protocol (UDP): • commands are transmitted from the computer to the UAV via UDP port 8889; • telemetry data transmitted by the UAV is received downstream by the host computer via UDP port 8890.
The independent architecture can be used in a scenario with a limited number of UAVs and a suficient
number of operators. In case of good operator training, it is possible to achieve high maneuvering
accuracy and system responsiveness in the presence of interference and obstacles. The second
architecture, shown in Figure 2, uses a centralized approach. In this scenario, two UAVs operate in Station
(STA) mode and are connected to a Wi-Fi network, which is created by an external router configured as
an access point [
The ground control station is connected to the same network and sends UAV flight commands via UDP packets using port 8889. The results of the commands are sent by the UAV control system to ports 9010 and 9011 (one for each UAV). The ground station uses one Python script to transmit identical commands simultaneously. The centralized structure increases the scalability of the architecture. However, using a router as an access point can introduce additional delays and afect the frequency of receiving response data from UAV. In addition, in the presence of environmental disturbances (e.g., obstacles), UAVs may struggle to maintain synchronous flight.
The third architecture, shown in Figure 3, uses an approach based on the leader-slave model. In this scheme, the slave control program UAV recognizes the position of the QR code attached to the back of the leader UAV.
The leader operates under a script from an independent ground station, which sends flight commands via Wi-Fi, similar to previous flight organization methods. The slave UAV is connected to the second ground station, which receives a video stream on port 11111, recognizes the leader’s position using a QR code, computes corrective actions, and transmits control commands to maintain a fixed distance. This architecture is more adaptive and can prevent the risk of collisions.
All three considered architectures include the Vicon system, which provides UAV positioning tracking.
The experiments are carried out inside a secured area in our laboratory [
After evaluating various solutions for communication and control management in systems with multiple UAVs, the software implementation of synchronous modes is presented. For all methods, a control interface is implemented which assumes the following control options: take-of, landing, circle, square lfights, and emergency stop. The battery charge level is proposed to be displayed, which afects the operation of the control system: a charge level below 20% disallows take-of. The thread module is used to perform flight procedures in separate threads, preventing the graphical interface from blocking during the execution of instructions. The djitellopy library, and in particular the Tello class, provides a Python interface for direct interaction with the DJI Tello UAV over the Wi-Fi network, allowing you to send commands and receive sensor data. For the flight of the slave UAV, another interface is implemented, which allows you to adjust and, if necessary, change the control coeficients. The model for detecting the QR code of the slave UAV is divided into three main layers: • Layer 1: Pose Detection and Error Estimation. This layer is responsible for processing visual data to determine the position of the QR code and calculate errors relative to the desired pose. QR Controller receives a video stream from the front camera (CF - camera frame) and extracts the pose (position and orientation) of the QR code: {, , , } . Error Estimation computes errors { , , , } in relation to the target pose { , , , }. The values of the desired pose are the following: = 0.7, = 0, = 0, = 0. In this way, the slave UAV will be at a distance of 70 cm from the master UAV, at which the control errors are calculated as the diference between – Proportional term (P): – Integral term (I): – Derivative term (D): the desired pose and the detected one: = − , = − , = − , = − . • Layer 2: PID controller. Here, the errors obtained for each axis of motion are processed. Each controller calculates a specific control action that will be sent to the actuator.
the control signal () for each axis ( , , , ) as the sum of three PID controller contributions: () = + + .
Through extensive experimentation, it was possible to progressively calibrate the PID controller gains. The tuning was done dynamically by allowing real-time intervention in the parameters, allowing immediate observation of how changes afect the system’s response. The control action () for the diferent axes ( , , , ) is composed by the proportional, integrator and derivative components. For each recommended PID the gain values are also provided, determined experimentally to ensure stable and eficient operation of the control system.
= ⋅ , = ⋅ , = ⋅ , = ⋅ .
Recommended values: = 0.75, = 0.75, = 0.65, = 0.6.
0 0 () =
∫ ( ) , () = () = ∫ ( ) , () = ∫ ( ) .
Recommended values: = 0.0012, = 0.0012, = 0.002, = 0.01. ∫ ( ) , 0 0 () () ()
= ⋅ , = ⋅ , = ⋅ , = ⋅ Recommended values: = 0.4, = 0.4, = 0.4, = 0.4.
The recommended values ensure the stability of the slave UAV movements and the speed when correcting errors in the distance from the master UAV, both for translational movements (X, Y, Z ) and for rotation around the vertical axis (yaw angle ). • Layer 3: Drive and Engine Control. The control commands, produced by the PID controller, are transformed into low-level signals necessary to actuate UAV’s motors. This layer is not accessible for modification in DJI Tello
UAV, it is responsible for executing the motion along the trajectory. () .
The accuracy of executing the motion along the trajectory will be assessed by comparing the actual trajectory with the planned one in the next section.
Analyzing the trajectories of the two DJI Tello UAVs reveals how accurately they follow the desired path, highlighting any deviations and execution errors.
The planned flight path, indicated by the blue and red dotted lines in Figure 5, represents the expected results of the execution of the sent commands, while the actual flight path (blue and red) is displayed using Vicon motion capture data. The shape and continuity of the trajectories also allow us to assess the stability of the flight. In Figure 5, each column corresponds to the diferent experimental architecture (1independent control, 2- centralized, and 3- cooperative). The figures at the top show circular trajectories, while those at the bottom show square trajectories. The position error estimated in these experiments represents the distance, frame-by-frame, between the ideal planned trajectory and the one actually lfown by the UAVs. Thus, it represents a spatio-temporal error, as it relies on both the spatial distance between the two corresponding points and their accurate temporal relationship. During flight, it is evident that the error increases due to the accumulation of inaccuracies in the IMU sensors. These errors are partially compensated by the internal controller, whose calculations do not depend on the lfight starting point, but on the relative position at the start of the execution of the last DJI Tello library command.
The average values of the position error, calculated in the diferent experiments and deduced from the previous graphs, are summarized in Table 1. It should be noted that, in Experiment 3 (leader-follower), the position error still indicates the deviation from an ideal trajectory, but it is not directly comparable with the other cases. In this context, the value of the error also includes the dynamics of adaptation of the follower to the leader’s movements, influenced by the controller’s response and unexpected variations in the leader’s trajectory.
The results indicate that a good level of multi-UAV coordination can be achieved with low-cost platforms,
if they are supported by a specific architecture and appropriate compensation mechanisms. Although
hardware limitations cannot be eliminated, the combined use of computer vision and distributed control
techniques can overcome many operational limitations. A hybrid solution between centralized and
master-slave architectures will be proposed for further research, in which the master UAV and a group
of autonomous slaves use computer vision to maintain coordination. This approach will also combine
the advantages of scalability, direct control, and collision avoidance, making it suitable for more complex
operational scenarios. An external independent positioning system that can compensate for IMU errors
should also be added. The obtained results demonstrate that the proposed methods for evaluating
positioning accuracy during the execution of synchronized actions by a UAV group are applicable to
future indoor localization analysis scenarios. In particular, they can be used efectively in systems that
integrate visual data, IMU sensors, and external positioning measurement technologies [
Project co-funded by the European Union – Next Generation Eu - under the National Recovery and Resilience Plan (NRRP), Mission 4 Component 2 Investment 3.3 - Decree No. 117 (2nd March 2023) of Italian Ministry of University and Research - Concession Decree No. 2332 (22nd December 2023) of the Italian Ministry of University and Research, Project code D93D23000380003, within the Italian National Program PhD Programme in Autonomous Systems (DAuSy).
During the preparation of this work, the author(s) used X-GPT-4 and Gramby in order to: Grammar and spelling check. After using these tool(s)/service(s), the author(s) reviewed and edited the content as needed and take(s) full responsibility for the publication’s content.