1 Introduction

One of the significant research problems in UAV-swarm applications is to design and develop an intelligent control algorithm that help in coordinating with several multi-agents in run-time by utilising the set of instructions given to main leading agent. Such systems need to be structured very well enough so that it may easily compute the manoeuvres in need as well as facilitates coordination among the multi-agents effortlessly. Researchers have been seen while utilising dual loop control topologies along with external sensors and systems to ensure the accurate navigation in GPS-denied environments [1, 2]. Thus, the main research objective is to produce a swarm control algorithm without deploying any external sensor or system, that helps in tracking the different trajectories in GPS-denied environment without collisions. A simple swarm has been illustrated in Fig. 1 to show that it is necessary to acquire the simple run-time assessment of positions and velocities of multi-agents. This is the main reason that any unmanned aerial vehicle (UAV) in general or particularly CoDrones in our case, can track the trajectory in more stable way without any risk of drift or collisions. Researchers have concluded that an unmanned aerial vehicle, or drone, is susceptible to losing control if its control algorithm fails to compute or at least forecast its velocity and position. This is the main cause of concern while implementing swarm framework for autonomous indoor operations.

Fig. 1
figure 1

High-level indoor space swarm on resouce-limited CoDrones

Full size image

Deploying UAV swarms in complex indoor environments presents unique obstacles: real-time collision avoidance, limited onboard processing, unreliable GPS, and the constrained space introduced by motion capture systems. While earlier efforts heavily depend on external localisation systems, sophisticated onboard vision sensors, or computational offloading [3,4,5,6,7,8,9,10,11,12,13], these setups often introduce delays, increase payload weight, and compromise swarm autonomy. This research introduces a fully onboard, infrastructure-independent strategy by integrating RSSI-based relative localisation with PID and FOPID controllers for swarm coordination. This eliminates the need for external tracking or SLAM-based systems and enables MAVs to maintain formation and track complex trajectories indoors. This research manuscript consists of following main research contributions as mentioned below:

  • RSSI-Enabled FOPID Control Integration: This is the first work to couple RSSI-based relative localisation with FOPID control for swarm UAVs, improving tracking accuracy and autonomy over PID and even RSSI-PID strategies [21,22,23]. Unlike earlier RSSI studies, which are limited to detection or single-agent positioning [23], our approach exploits inter-agent RSSI values for dynamic multi-agent coordination.

  • Fully Onboard, Sensor-Minimal Implementation: Existing MAV platforms often require optical flow sensors [5], external vision systems [4], or SLAM processing [10, 11, 18]. Our method operates using only onboard sensors (e.g., RSSI and IMU), lowering hardware complexity, reducing latency, and improving resilience against GPS or vision system failure [3, 20].

  • Experimental Validation and Scalability Demonstration: We validate the proposed RSSI-FOPID system through both simulation and hardware trials, showing significant error reduction for square and helical trajectories. Unlike past systems requiring extensive computation [6, 10, 18], our design proves scalable and practical for 2–3 drone swarms without added infrastructure or external computing.

These contributions collectively represent a significant step forward in enabling lightweight, low-power, real-time swarm control in indoor environments-an essential requirement for scalable, autonomous MAV systems. This manuscript is organised into six sections: Section 1 contains the introductory context, while Section 2 presents a comprehensive literature assessment and the current state-of-the-art methodologies. Section 3 addresses the examination of the proposed methodology. The results and corresponding discussion can be found in section 4. Additionally, one may locate the hardware validation component within the same section. Section 5 serves as the conclusion, while Section 6 presents future recommendations and initiatives within the proposed study framework.

2 Literature review

This section shares the state-of-the-art approaches followed by researchers for implementing autonomous indoor swarm coordination using micro-air vehicles (MAVs). From this literature review one may be able to see that majority of the contributions involve GPS and motion capture systems for navigation and localisation and therefore these techniques faced numerous latency, precision and accuracy limitations. In addition to these limitations, delays in run-time feedback is the most frequent issue occurred in several research studies [3,4,5,6]. Moreover, onboard optical mouse sensors, albeit providing a degree of precision, necessitate oscillatory motion to precisely evaluate velocity, thus constraining manoeuvrability and control efficacy [5, 6]. Another hurdle occurs when omnidirectional cameras are affixed to small helicopters, which may encounter drift issues that undermine their capacity to sustain precise placement throughout flight [7]. The Delfly MAV, utilised for indoor exploration, features a stereo-camera for obstacle evasion. Nonetheless, it lacks a mechanism for velocity estimate, hence constraining its capacity for dynamic, autonomous navigation in real-time [8, 9]. Alternative research has suggested larger MAV swarms for navigation based on Simultaneous Localisation and Mapping (SLAM). This approach yields promising results; nevertheless, it requires substantial computational resources to handle the extensive data produced by numerous MAVs, which may pose a constraint in practical applications [10, 11]. Several studies have drawn inspiration from insect navigation for spatial exploration, incorporating onboard computing that emulates biological systems. These systems, while creative, remain constrained by the computational capabilities of the MAVs and their efficiency in processing environmental data [12]. GPS-based localisation systems, although efficient in outside settings, encounter considerable difficulties in indoor swarm coordination, especially regarding accuracy and the potential for signal loss. Indoor formation flights, exemplified by the Crazyflie MAV, frequently rely on external computers for real-time data processing, hence introducing added complexity and constraining the system’s autonomy. Relative localisation techniques utilising onboard cameras still necessitate additional external devices to precisely ascertain the positions of MAVs, hence further constraining the autonomy of these systems [16, 17]. Collaborative SLAM systems, commonly employed in swarm robots, have advanced in map integration and environmental exploration. Nevertheless, they remain significantly dependent on other computers for data processing and lack vital functionalities, such as obstacle avoidance, which are crucial for complete autonomy [18, 19]. The primary issue is to engineer lightweight MAVs that can attain robust, autonomous navigation and swarm coordination using only onboard processing. This represents a substantial advancement in the development of more efficient and autonomous systems for indoor exploration. Table 1 presents a review of these contributions and their limits in indoor space exploration.

Table 1 Limitations related to current MAVs for indoor swarming
Full size table

Above table shares a comprehensive summary related to the limitations of utilising micro-aerial vehicles (MAVs). In addition to this, one may find further details such as their processing techniques, navigation methods utilised, their constraints associated with other externally deployed systems and sensors. There are several research contributions where external optical flow for velocity estimation had been deployed and this led them to obstruct the run-time decision making while swarm operations [3]. Similarly, if one may read [4], an off-board processor for visual based navigation had been deployed and it resulted in delays in video processing and eventually led to run-time autonomous operations. In above table, one may get an idea of all state-of-the-art MAVs and the fact that all of them were in need of an external computing system such as custom quadrotors required on-board optical flow sensors for avoidance and for precise navigation in indoor environment. Moreover, by the deployment of external systems and sensors, these MAVs had experienced several issues such as the overall performance and flying dynamics had been changed [5, 6], experiencing drift issues and could not sustain stable formations [7] and face estimation issues as well [8, 9]. One may see Astec Pelican in previous table that uses onboard SLAM for multi-floor mapping but again in this case, it faces low latency issues. Moreover, GPS-enabled localisation used in Aquatic robots and Mikrokopters faced an issue of less robust swarming during the loss of GPS signals [10,11,12,13]. Similar sort of issues were witnessed in the research carried over Asctec Hummingbirds – disturbing the swarm coherence in indoor environment [14,15,16,17].

Thus, proposing RSSI-enabled fractional order proportional Integral Derivative (FOPID) control that will enhance the swarm coordination. By integrating the received signal strength indicator (RSSI) – one may retrieve the relative positioning and can eliminate the need for external systems or sensors to be deployed on MAV like GPS or off-board computation [18,19,20]. In this way, we will be able to lessen the latency, improve the autonomy and resilient positioning without depending on any external system or sensor [20]. From [21, 22], it has been proved that FOPID outperforms PID in several cases such as MAV-based trajectory tracking, swarm cooperation and adaptability to dynamic environment. This is because it provides flexible tuning for better response [22]. In this research we are further improving the performance of FOPID by enabling it with RSSI-based feedback system to minimize the computation demands and work in same way at least as other drones are working in table 1 but without the deployment of GPS or motion-capture cameras [3, 4, 23]. The main aim of this technique is to come up with a low-power extend swarm flights and give liberty to MAVs to swarm even in crowded environment like [22,23,24,25]. The overall results provided in this manuscript shares that the proposed technique is much reliable and adaptive kind of solution for indoor swarm operations. Please note that this study primarily compares the proposed RSSI-enabled FOPID control approach against traditional PID and RSSI-PID baselines, it is worth noting that other contemporary swarm navigation techniques-such as Visual-Inertial Odometry (VIO), LiDAR-SLAM, and UWB-based localization-offer high-precision mapping and positioning capabilities. However, such systems typically involve significant hardware complexity, external infrastructure, or onboard computational requirements, which may not be suitable for lightweight, cost-sensitive swarm platforms like CoDrone. Our current focus remains on evaluating control-level performance under minimalistic onboard sensing conditions. Future extensions of this work will include benchmarking against these advanced localisation frameworks to further validate the scalability, robustness, and performance of RSSI-based swarm coordination methods in broader scenarios.

A comprehensive review of state-of-the-art methodologies reveals that most existing swarm localisation and control approaches are heavily reliant on external infrastructure, such as GPS-based systems, motion capture setups, or Simultaneous Localisation and Mapping (SLAM) pipelines. These dependencies often introduce substantial limitations, including increased latency, elevated computational demands, and added hardware complexity. For example, vision-based navigation relying on off-board processors [4] or the use of onboard optical flow sensors [5, 6] tends to induce feedback delays, thereby reducing responsiveness and limiting scalability in real-time applications. While some research has attempted to mitigate these constraints by incorporating onboard sensing, many of these systems still depend on centralised computation or offloading for effective SLAM integration or relative localisation [10,11,12,13,14,15,16,17]. Moreover, several studies have reported performance degradation due to sensor drift, unstable formations, or the inability to operate effectively without external feedback mechanisms [7, 8, 15].

Therefore, Fractional Order PID (FOPID) controllers have shown superior control flexibility and adaptability over conventional PID methods in swarm navigation and trajectory tracking tasks [21, 22], but they have not been combined with RSSI-based relative localisation for coordinated multi-agent control [3,4,5]. Real-time RSSI feedback and FOPID control are combined in this study to solve these long-time problems. The proposed system uses onboard processing and no external localisation or communication infrastructure [12,13,14,15]. Previous research relied on GPS, external vision systems, or computationally intensive SLAM frameworks [7,8,9,10,11]. In Table 4, empirical evaluations show considerable velocity tracking error reductions [16,17,18], from 14.694 to 6.796 in three-drone square trajectory testing, without extra onboard sensing units. Thus, this work offers a computationally efficient, infrastructure-free, and empirically proven indoor swarm UAV control alternative. The proposed architecture improves localisation precision and control accuracy by using onboard RSSI measurements with FOPID control, making it scalable and lightweight for real-world autonomous swarm deployments. This explanation validates our clarification regarding the originality of our planned research work.

3 Proposed approach

The RSSI data are continuously monitored to detect any divergence from the expected range. Each drone in the swarm is identified when the Received Signal Strength Indicator (RSSI) of any Unmanned Aerial Vehicle (UAV) closely matches others within the group. Let \(\text {RSSI}_i\) denote the RSSI value of the i-th UAV. The RSSI-based distance estimation can be expressed as:

$$\begin{aligned} \text {RSSI} = -10n \log _{10}(d) + A \end{aligned}$$
(1)

where d is the distance between the transmitter and receiver, n is the path-loss exponent, and A is the RSSI value at a reference distance. In this study, the path-loss exponent n utilised in the RSSI-distance model is regarded as a constant, experimentally established by initial measurements performed in our indoor laboratory environment. This assumption streamlines the computing demands for real-time RSSI-based localisation and corresponds with the restricted processing capacities of the onboard devices utilised in CoDrones. Although it is recognised that n may fluctuate due to ambient variables such as wall materials, furniture arrangement, and multipath effects, sustaining a constant value enabled us to concentrate on assessing the fundamental control performance of the RSSI-enabled FOPID system. Furthermore, the testing setting was maintained uniformly across all trials to reduce the influence of environmental variability. Future iterations of this study may integrate adaptive calibration approaches or machine learning-based signal modelling to accommodate fluctuating path-loss circumstances for wider deployment in heterogeneous or dynamically changing indoor environments. Each drone uses RSSI to estimate distances to neighboring UAVs and adjusts its position to maintain the desired formation. The control system then minimizes the position error derived from RSSI by dynamically adjusting each drone’s velocity and heading. Consider three drones \(D_1\), \(D_2\), and \(D_3\). The desired RSSI values between each pair are denoted as \(\text {RSSI}_{D12}\), \(\text {RSSI}_{D13}\), and \(\text {RSSI}_{D23}\). The corresponding error for each pair is computed as:

$$\begin{aligned} e_{ij} = \text {RSSI}_{Dij} - \text {RSSI}_{ij}, \quad \text {where } i,j \in \{1,2,3\} \end{aligned}$$
(2)

Here, \(\text {RSSI}_{ij}\) denotes the received signal strength indicator measured between UAV i and UAV j, whereas \(\text {RSSI}_{Dij}\) is the desired signal strength representing the target inter-agent spacing. This distinction is important, as \(\text {RSSI}_{ij}\) reflects real-time, peer-to-peer signal measurements used for swarm formation control. The PID control input is then given by:

$$\begin{aligned} u(t) = K_p e_{ij}(t) + K_i \int _{0}^{t} e_{ij}(\tau )\, d\tau + K_d \frac{de_{ij}(t)}{dt} \end{aligned}$$
(3)

Similarly, the FOPID (Fractional Order PID) control input is defined as:

$$\begin{aligned} u_i(t) = K_p e_{ij}(t) + K_i \frac{1}{t^\lambda } \int _0^t e_{ij}(\tau )\, d\tau + K_d^{t^\mu } \frac{de_{ij}(t)}{dt} \end{aligned}$$
(4)

To ensure safe operation, an additional constraint is imposed to maintain a minimum RSSI threshold:

$$\begin{aligned} \text {RSSI}_{Dij}> \text {RSSI}_{\text {min}} \end{aligned}$$
(5)

The complete RSSI-enabled PID and FOPID swarm control algorithms are illustrated in Figures 2 and 3, respectively.

Fig. 2
figure 2

RSSI-enabled PID swarm control algorithm

Full size image
Fig. 3
figure 3

RSSI-enabled FOPID swarm control algorithm

Full size image

In Figures 2 and 3, the notation used for swarm dynamics is defined as follows: \(X_{\text {ref}}\) represents the global reference trajectory assigned to the swarm, while \(X_{l}\) denotes the current state (position and velocity) of the leader drone. The variable \(X_{cl}\) indicates the control state vector of the leader, generated by the RSSI-based error dynamics. Similarly, \(X_{f_{ij}}\) represents the state vectors of the follower drones \(D_j\), where each follower adjusts its trajectory based on the inter-agent RSSI error \(e_{ij}\) with respect to the leader or neighboring agents. These definitions support the overall swarm control mechanism driven by either PID or FOPID strategies, as visualized in Figures 2 and 3.

4 Results and discussion

This section presents the simulation results of the autonomous swarm control system, which employs PID, FOPID, RSSI-enabled PID, and RSSI-enabled FOPID methodologies for navigation in indoor environments. All simulations were conducted in MATLAB and Simulink. The scenario models a swarm of two unmanned aerial vehicles (UAVs) operating in a controlled indoor environment to monitor a predefined route. Each UAV is equipped with RSSI sensors to measure signal strength from neighbouring drones. The PID and FOPID controllers are triggered based on RSSI measurements to facilitate swarm identification and navigation, as illustrated in Figures 4 to 7. In the proposed control architecture, the agents operate under a velocity-controlled regime, where the reference trajectory is followed by continuously adjusting the drone’s velocity to match a time-dependent velocity profile. Therefore, the accuracy of path following is directly tied to how closely each drone’s actual velocity tracks the desired velocity. As such, velocity tracking error serves as a reliable and dynamic performance metric, with smaller deviations indicating more precise adherence to the planned path. This justifies the use of velocity plots in Figures 4 to 7 as a representative measure of swarm navigation performance.

Fig. 4
figure 4

Velocity tracking by a swarm of two drones using PID and RSSI-enabled PID

Full size image
Fig. 5
figure 5

Velocity error rates for two-drone swarm using PID and RSSI-enabled PID

Full size image

Figure 4 displays velocity profiles of two drones. The top plot shows results using standard PID control, where Drone 1 (red) and Drone 2 (blue) closely follow the desired trajectory (black dashed line). The bottom plot illustrates RSSI-enabled PID control, where Drone 1 (green) and Drone 2 (magenta) follow the desired trajectory more precisely, with reduced deviation. Figure 5 highlights velocity error rates. RSSI-enabled PID demonstrates consistently lower errors and faster convergence compared to PID alone. Figures 6 and 7 show similar evaluations for FOPID and RSSI-enabled FOPID. RSSI integration significantly reduces oscillations and tracking errors, particularly evident in dynamic transitions. The overall comparison is summarized in Figure 8 and Table 2.

Fig. 6
figure 6

Velocity tracking using FOPID and RSSI-enabled FOPID

Full size image
Fig. 7
figure 7

Velocity error rates using FOPID and RSSI-enabled FOPID

Full size image
Fig. 8
figure 8

Comparison of velocity tracking errors for PID, FOPID, RSSI-enabled PID, and FOPID

Full size image
Table 2 Velocity Error Rate Comparison using Exponential Decay Functions
Full size table

4.1 Performance Metrics Analysis

The performance of RSSI-enabled PID and FOPID controllers is further analyzed using metrics such as detection time, containment time, formation stability, and control effort. The simulation settings and results are summarized in Table 3.

Table 3 Performance Metrics for RSSI-enabled PID and FOPID Control Strategies
Full size table

4.2 Hardware validation with CoDrones

Based on promising simulation results, the RSSI-enabled FOPID controller was implemented on CoDrone platforms. Figure 9 shows the CoDrone units and the lab setup at the Advanced Unmanned Aerial Systems Lab, KFUPM. Hardware specifications such as sensors, frame, LED indicators, and micro-USB ports are shown in Figures 10. This experimental validation utilised the CoDrone platform created by Robolink Inc., featuring onboard BLE modules that measure RSSI data from peer drones at an average refresh rate of 10 Hz. The RSSI values were obtained through the drone’s firmware-level API and analysed in real time by a lightweight onboard CPU (ARM Cortex-M0, 64 MHz). The drones utilise Bluetooth Low Energy (BLE) for communication, and the Received Signal Strength Indicator (RSSI) readings were employed directly for estimating inter-agent distances within the control loop. The experiments were conducted in Lab 248 at the Advanced Unmanned Aerial Systems Laboratory, KFUPM. The test space is a confined indoor area measuring roughly 7.97 meters in length, 7.76 meters in width, and 5.52 meters in height, yielding a maximum diagonal distance of approximately 11 meters. The space was arranged with few dynamic impediments, yet featured typical sources of signal reflection, including desks, chairs, and wall-mounted equipment. These conditions were uniform throughout all trials to guarantee equity and reproducibility. Subsequent developments of this research will investigate higher-density and constantly evolving environments to enhance validation of robustness. The drones are programmed using Python to perform coordinated swarm operations. In Lab 248, both square and helical trajectory tracking were tested using two and three drones.

Fig. 9
figure 9

(a) CoDrone and remote; (b) Ground station setup at KFUPM

Full size image
Fig. 10
figure 10

(a) Top view and (b) bottom view of CoDrone

Full size image
Fig. 11
figure 11

Swarm of two CoDrones using RSSI-enabled FOPID

Full size image
Fig. 12
figure 12

Swarm of three CoDrones using RSSI-enabled FOPID

Full size image

Above is case 1 for performing square trajectory tracking for swarming two codrones and later three drones. The velocity tracking responses are shown in Figure 13. Whereas within this figure, one may see the error rates computed using exponential decay functions. In contrast, case 2 illustrates the execution of helical trajectory tracking for a swarm of two codrones, then expanding to three drones, as depicted in Figure 13. The velocity tracking reactions are illustrated in within the same figure and the error rates calculated utilising exponential decay functions. In Figure 11 and 12(a) to (f) shows the square trajectory tracking within the lab. Furthermore, Indoor environments inherently introduce signal fluctuations due to multipath effects caused by walls, furniture, and other reflective surfaces. To mitigate the impact of RSSI variability in such settings, a combination of strategies was employed. First, the path-loss exponent used in distance estimation was empirically calibrated under static lab conditions prior to testing. Second, RSSI values were filtered using a moving average technique across multiple samples (typically over 10 Hz sampling), thereby smoothing out short-term spikes caused by reflections. Lastly, the swarm was operated with sufficient inter-drone spacing to minimize signal overlap and interference. These precautions contributed to the observed formation stability and trajectory tracking performance shown in Figure 11, despite the challenges of indoor signal reflections.

Fig. 13
figure 13

Square trajectory: velocity responses using FOPID with and without RSSI

Full size image
Fig. 14
figure 14

Helical trajectory: velocity responses using FOPID with and without RSSI

Full size image
Fig. 15
figure 15

Velocity error rates for square trajectory using 2 and 3 drones

Full size image
Fig. 16
figure 16

Three CoDrone swarm in helical trajectory using FOPID and RSSI-FOPID

Full size image
Fig. 17
figure 17

Helical trajectory: velocity tracking and errors using FOPID vs RSSI-FOPID

Full size image
Fig. 18
figure 18

Velocity error rates for helical trajectory with 3 drones

Full size image

4.3 Comparative evaluation

Table 4 summarizes trajectory tracking performance based on error rates under different conditions. The error rate in Table 4 is computed as the mean deviation from the reference trajectory using exponential decay modeling, expressed in meters per second (m/s).

Table 4 Comparative Error Rates for FOPID vs RSSI-enabled FOPID
Full size table

The above table offers a comprehensive comparison of the performance of two control techniques-FOPID (Proportional-Integral-Derivative) and RSSI-enabled FOPID-utilized for trajectory tracking in a system comprising n CoDrones. The error rates for each technique are presented for two trajectory types: Square and Helical, as well as for varying numbers of CoDrones (2 and 3). In the Square trajectory with two CoDrones, the FOPID control strategy yields an error rate of 9.796, signifying a comparatively elevated performance error. Nonetheless, the implementation of the RSSI-enabled FOPID technique results in a substantial reduction of the error rate to 4.847, illustrating the efficacy of incorporating RSSI (Received Signal Strength Indicator) feedback in enhancing the system’s accuracy. The decrease in error rate indicates that RSSI feedback improves the control system’s capacity to adjust to changing external conditions, including fluctuations in signal strength. Discussing the tracking of square trajectory using three CoDrones one may see a significant rise in error rate which is 14.694 measured using Gaussian functions while implementing FOPID alone using 3 Codrones and it gets reduced to 6.796 while integrating RSSI with FOPID. This even gets much less in case of helical trajectory and yield an error rate of 5.7773 in the presence of three CoDrones by utilising RSS-enabled FOPID. Using alone FOPID again the error rate increases up to 5.9204 for helical trajectory tracking. This indicates that the integration of RSSI feedback mitigates specific challenges associated with larger drone swarms and specific trajectory perhaps by providing additional information on relative positioning and signal strength. The results indicate that RSSI-enabled FOPID control markedly diminishes trajectory errors across diverse swarm sizes and complexities, underscoring the essential function of real-time environmental feedback in improving control system efficacy for multi-agent systems. In contrast to pure FOPID control, which has scalability issues in bigger swarms due to heightened interaction dynamics, the RSSI-enabled method exhibits enhanced accuracy and adaptability. These findings highlight the significance of using sensor-based feedback, such as RSSI, to enhance accuracy, robustness, and scalability in autonomous drone swarm activities.

Despite the fact that the RSSI-enabled FOPID framework that has been described is intended to provide scale swarm coordination, the experimental validation that has been reported in this paper has purposefully been restricted to swarms consisting of about two and three CoDrones. This decision was influenced by a number of practical considerations, including the following: (i) the confined indoor lab environment posed collision risks for larger swarm configurations; (ii) during the trial period, there were only three CoDrone units available for testing; and (iii) precise tracking and RSSI-based localisation become increasingly difficult to benchmark accurately in cluttered physical spaces without the presence of external ground-truth systems. The underlying control method, on the other hand, continues to be lightweight and decentralised, which suggests that it is possible to achieve greater swarm sizes with adequate inter-agent spacing and communication planning. To assess performance metrics such as inter-drone interference, trajectory error propagation, and control resilience under more dynamic and scalable circumstances, future implementations will investigate swarm sizes that are more than three drones. These evaluations will take place in both simulated and hardware environments.

5 Conclusion

This research paper presents the performance evaluation of RSSI-enabled FOPID in contrast to conventional PID and RSSI-enabled PID control algorithms for number of drones to perform swarm in indoor facility. The main results demonstrate that with the integration of RSSI feedback the performance of RSSI-enabled FOPID outperforms other strategies in dynamic conditions by lowering the velocity tracking errors not only for square trajectory but also for helical trajectory tracking. At this stage, our research concludes that the proposed algorithm performs well up to the swarm of three drones but there is also a fact noticed in the results that the error rates may increase with an increase in bigger number of CoDrones performing swarm together. Moreover, another fact related to these error rates is that the values are much lower in case of Helical trajectory than those are in square trajectory. This leads us to another notion that the nature of the path followed by CoDrones plays a potential impact in the overall performance. If there is smooth path provided as reference trajectory to our CoDrones using RSSI-enabled FOPID algorithm, this results in less aggressive control adjustments. These are some of the important take-away points while implementing our proposed control algorithm. In simple words, RSSI-enabled FOPID control results in a most important swarm tracking control algorithm that does not only provide accurate results but fine resilience among the CoDrones as well which is itself a promising path forward for the development of multi-agent-based swarm navigation and collision avoidance.

6 Future recommendations and directions

Our research is looking over the integration of more intelligent control approaches to enable smooth and robust swarm behavior along with self-capability to adjust the fleet in response to ambient fluctuating situations. Thus, the study is still in progress to evaluate some of the machine learning algorithms to elevate the trajectory tracking performance. In addition to this, our focus is to go for more trials to confirm the robustness in dynamic hurdle-oriented environment where hurdles can come at anytime in between the swarm of drones. In this way, our suggest algorithm can gain more trust of researchers and field experts in facilitating the swarm of UAVs in GPS-denied indoor environment with enhanced efficiency and reliability.