Shuyan Hu, Xin Yuan, Wei Ni, Xin Wang, Abbas Jamalipour
©SHUTTERSTOCK/DMITRY KALINOVSKY
The applications of unmanned aerial vehicle (UAV)-enabled visual monitoring span the areas of public security, nature resilience, and disaster rescue. Covertness can play an indispensable role in applications demanding UAVs to be unnoticeable by targets, e.g., tailing and interception and police surveillance. This article discusses the types and technical challenges of visual camouflage for UAV-based surveillance. A particular interest is given to an agile disguising method, which adopts both distance keeping and elevation changing and confuses the target by constantly changing its relative position in the target’s view. The path design of the UAV monitor is nonstraightforward under this disguising approach due to nonconvex disguise objectives, UAV propulsion power, and control dynamics. A new control framework is presented to plan and refine the trajectory of the UAV monitor online. The framework employs model predictive control (MPC) to decompose the control decisions between slots, mitigating the impact of the inaccurate prediction of the target’s path and allowing the planned trajectory to be refined online. Simulations validate the merits of the new framework over the benchmark approach with no camouflage and demonstrate the different performances of fixed-wing and rotary-wing UAVs on a covert video surveillance mission.
With their eminent maneuverability, quick deployment, and expansive coverage, UAVs are increasingly employed in bushfire monitoring, nature conservation, disaster recovery, and emergency communications [1], [2]. UAVs equipped with onboard cameras or other sensing and perception systems (e.g., lidar, radar, etc.) have also been utilized to tail and monitor moving targets, as shown in Figure 1. The applications of UAV monitors span the areas of police surveillance, wildlife observation, and firefighting.
Figure 1 A schematic of covert video surveillance by joint distance keeping and altitude changing where a fixed-wing UAV monitor is tracking a moving vehicle.
In some of these applications, e.g., public safety and security, the stealth of the UAVs is critical, or the targets can notice the UAVs and execute countermeasures. A UAV performing stealthy monitoring must keep its target within its view while preserving its spatial separation from the target for stealth. To this end, the trajectory of the UAVs has to be carefully designed. Albeit solar power has been increasingly utilized to power UAVs for sustainability, the trajectory planning of the UAVs becomes more challenging due to new energy causality constraints arising.
Control-based approaches, such as receding horizon control, geometric control, dynamic programming, and Lyapunov theory, have been applied to UAV trajectory design and monitoring [3], [4], [5]. Deep learning and deep reinforcement learning have been applied to UAV-enabled target tracking to distinguish the target from backgrounds, address aspect ratio changes in images, and make coarse-to-fine tracking policies [6]. However, no study has considered visual camouflage or addressed nonconvexity in trajectory planning. Moreover, the trajectory needs to be planned online since there is no destination, and it may need to change frequently.
This article discusses the different types, application scenarios, and potential technical challenges of visual camouflage for UAV-based surveillance. Existing techniques that achieve visual camouflage are summarized and compared, such as distance keeping, heading regulation, obstruction, and relative stationarity (see Table 1). The focus of this article is on a new disguising method that adopts both distance keeping and elevation changing to confuse the target. Different from traditional disguising methods, such as line-of-sight (LoS) obstruction and relative stationarity, the new method can work effectively even when the target is equipped with radar or RGB-D camera and is applicable to both rotary-wing and fixed-wing UAVs. In this sense, the new method is less restrictive.
Table 1 A comparison study of existing techniques on visual camouflage for UAV-based disguised surveillance.
Trajectory planning is nontrivial under the new disguising method because of a nonconvex objective and constraints. By leveraging MPC, a new framework is presented to plan and refine the trajectory of the UAV online. The framework decouples the control decisions between slots and adopts a control strategy of multislot trajectory planning and per-slot trajectory control, hence mitigating the impact of inaccurate predictions and suboptimal multislot trajectory planning.
The contributions of this article are summarized as follows:
Simulations validate the merits of the new framework over its alternatives without camouflage and demonstrate the performances of fixed-wing and rotary-wing UAVs on a covert video surveillance mission.
The rest of this article is organized as follows. The “Motivations of Visual Camouflage” section summarizes the motivations of covert UAV surveillance, followed by popular techniques to achieve visual camouflage. The “Challenges of Visual Camouflage” section discusses the challenges faced by visual camouflage implementation during UAV surveillance. The “Online Trajectory Planning for Energy-Efficient Covert Video Surveillance” section provides a case study of energy-efficient covert UAV surveillance, followed by simulation results in the “Performance Evaluation” section. The “Future Research Opportunities” section discusses future directions. Finally, the “Conclusion” section concludes the article.
Covertness can play a critical role in applications demanding UAVs to be unnoticeable by targets, e.g., tail and interception, wildlife tracking, and electronic combat. UAV-based covert video surveillance is promising for the following reasons:
The state-of-the-art methodologies of achieving visual camouflage for the UAV monitors are elucidated as follows.
Figure 2 (a) and (b) Schematics of visual camouflage by (a) relatively fixed location and (b) virtual reference point.
In a nutshell, the visual camouflage of video surveillance is typically accomplished by dynamically planning the UAV’s flight path based on and adapting to the target’s trajectory or a reference point to be inconspicuous to the target. This article focuses on visual camouflage based on joint distance keeping and altitude changing for its flexibility, which can apply to both fixed-wing and rotary-wing UAVs.
The visual camouflage of UAVs leveraging joint distance keeping and altitude changing imposes new challenges arising from UAV dynamics and energy consumption/supply.
The additional maneuvers of a UAV for disguising can consume more power, leading to new challenges concerning the power consumption and energy supply of the UAV. In particular, the maneuvers consume more power than flying in a straight line or performing simple turns since more complex maneuvers require the UAV’s engines to work harder. This results in increased power consumption and subsequent requirements for additional energy supply.
With different aerodynamics and control mechanisms, rotary-wing and fixed-wing UAVs follow distinct dynamic models.
A fixed-wing UAV has to maintain a minimum horizontal velocity to remain airborne. It adjusts its velocity incessantly with no sudden or sharp movements because of its maneuverability and aerodynamics. The fixed-wing UAV cannot hover, and its power consumption is captured by a single propulsion power model that couples the UAV’s horizontal and vertical movements. The factors contributing to the energy consumption of a fixed-wing UAV are the horizontal and vertical velocities and accelerations [12, Eq. 5], weight, wing surface, and span of the UAV [8].
In contrast, a rotary-wing UAV can fly much more freely. The rotary-wing UAV has different propulsion and thrust power models for level and vertical flight (ascending or descending), respectively. The factors contributing to the energy consumption of a rotary-wing UAV are the horizontal and vertical velocities, with separately controlled propulsion power for level flight and thrust power for ascending or descending [2, Eq. 24], fixed blade power, the rotating speed of the blade tip, the incurred power, and the mean rotor-incurred velocity when the UAV hovers [2].
Popular methods of visual camouflage require knowledge of the target’s locations at each time slot. A monitor can predict the target’s trajectory based on historical observations, typically using Kalman filtering. The trajectory may also be estimated by techniques based on data mining, long short-term memory, and sliding window polynomial fitting. The prediction accuracy relies on road conditions and the target’s reaction. It can have a strong impact on the effectiveness of the surveillance and the control of the monitor.
It is also important to balance surveillance and camouflage. When disguising itself, a UAV usually needs to stay away from its target. Trees, buildings, and thick clouds can help the camouflage but block the UAV’s sight. A camera with a higher resolution can capture objects at a greater distance yet result in a larger video file, which calls for a UAV with a higher storage capability and data transmission rate.
Traditionally, the power supply of a UAV comes from its onboard battery, a laser beam, or a tethered power line [2]. UAV batteries typically last only half an hour, which means frequent visits to home bases for recharging or short mission durations. A laser beam can transmit up to hundreds of watts [13], posing safety risks to the target and others. The UAV would be restrained to a small area if tethered.
Solar energy has been increasingly exploited in UAVs [14]. The amount of energy harvested depends on weather conditions and the time of day. Joint power management and trajectory design are essential for UAVs on disguised surveillance as a new energy causality constraint arises to prevent a UAV from consuming more energy than it has harvested at any moment. The additional maneuvers of the UAV for disguising consume more power, making the constraint more stringent.
In this article, we develop a new online trajectory-planning framework for a solar-powered UAV monitor on a disguised monitoring and tracking task.
In our considered scenario, a UAV with a panoramic camera is deployed to monitor a suspicious target, e.g., a driving vehicle, as depicted in Figure 1. The UAV monitor has a solar panel installed on its top for solar energy harvesting. Monitoring begins immediately after the target is discovered and continues until the tailing task is completed. The surveillance mission lasts T seconds. The mission duration can be equally separated into M time slots with ${\delta}$ seconds per slot ${(}{T} = {\delta}{M}{).}$
Unlike many existing trajectory-planning studies, the UAV monitor does not have a specific destination. The target may take any reasonable motions, including linear, curved, and randomly changing movements. The UAV monitor needs to be flexible and adaptable, able to accurately track the target regardless of the nature of the target motions. This makes it suitable for a variety of applications where the movement of the target is unknown or cannot be accurately predicted for a long time horizon.
The UAV camouflages its monitoring intention by tailoring its path and heading slot by slot. Hence, it seems to be far away and moving in all directions in the target’s view. An important application of this disguised surveillance is monitoring and tracking suspicious targets (for law enforcement) which makes it more difficult for the target to notice the UAV and predict its movements. It is also more difficult for the target to use any fixed points as a reference, such as landmarks or other objects in the environment, to track and locate the monitor, further increasing the uncertainty and confusion for the target.
It is critical for the UAV monitor to meticulously plan its trajectory to enhance its operation duration and energy sustainability (i.e., maintaining an energy reserve of no less than ${\eta}{E}_{0}$ with ${\eta}\in{(}{0},{1}{)}$ at any time instant; here, ${E}_{0}$ is the battery capacity, and ${\eta}$ is the ratio of the energy reserve). However, it is hard to accurately forecast the target’s path. The prediction can be increasingly inaccurate further into the future. Consequently, a planned trajectory would be increasingly unreliable, especially when a disguise is required.
A novel framework of online trajectory planning arises from combining optimization techniques, i.e., SCA, and control theory, i.e., MPC (Figure 3).
Figure 3 The flowchart of the MPC-based online trajectory planning where, at any slot ${\tau},$ the monitor forecasts the target’s path over the next N slots, e.g., by using Kalman filtering. Following MPC, the monitor can design its path from its current position for the next N slots. Only the first step of the N-step path is executed at slot ${(}{\tau} + {1}{)}{.}$
One reason for using this integrated framework is that it is impossible to achieve global or local optimal solutions as this would require global information (i.e., the future movements of the target and future energy harvesting and consumption of the UAV); incur prohibitive complexity arising from significant optimization variables (i.e., all waypoints, accelerations, and velocities); and, above all, violate causality. Using a control framework, it is possible to decouple control decisions over time, plan the UAV trajectory online with only causal information, and reduce the estimation error.
Another reason is that pure control strategies are unsuitable for this situation because the visual disguise requires constant and irregular changes in the UAV’s relative position to the target. Specifically, at any time slot ${\tau},$ the multiobjective trajectory planning for the upcoming N slots (i.e., slots ${\tau} + {1},\ldots,{\tau} + {N}{)}$ is nonconvex (due to nonconvex objectives concerning both disguise metric maximization and flight energy consumption minimization and nonconvex constraints concerning energy harvesting and aerodynamics) and cannot be solved by standard solvers, such as the interior point method. Nevertheless, the problem can be iteratively convexified using SCA. A convex upper bound for the energy usage of the UAV can be minimized, as opposed to minimizing the energy usage directly. Likewise, the nonconvex aerodynamics constraints can be convexified. The resultant convex problem can be readily solved by standard tools, such as CVX, with a polynomial computational complexity of ${\mathcal{O}}{(}{N}{}^{3.5}{)}$ [8].
The new framework is suitable for both fixed-wing and rotary-wing UAVs.
The key difference between the two types of UAVs is that the trajectory of a rotary-wing UAV can be optimized directly. By contrast, the trajectory of a fixed-wing UAV has to be optimized together with its velocity and acceleration. By combining the SCA and MPC, a quantifiable (local) optimal trajectory is obtained for a foreseeable future period and can be refined on the go.
This section offers MATLAB simulation results to validate the benefits of the new solar-powered UAV-based disguised visual surveillance and tracking method. The UAV weighs 4 kg. It has a lithium battery with a voltage of 22 V and a capacity of ${1.6}\,\times\,{10}^{4}\,{\text{mAh}}$ [15]. In other words, the battery can store up to ${22}\,\times\,{1.6}\,\times\,{10}^{4}\,\times\,{3.6} = {1.27}\,\times\,{10}^{6}\,{\text{J}}$ of energy. The UAV battery initially has ${E}_{0} = {40}\,{\text{kJ}}$ of energy, and the minimum energy reserve at any moment is set to ${\eta}{E}_{0} = {16}\,{\text{kJ}}$ with ${\eta} = {0.4.}$ Here, we use the same initial battery level (or capacity) for a fair comparison between the fixed-wing and rotary-wing UAVs. The minimum and maximum UAV speeds are 4 m/s and 30 m/s, respectively. The maximum acceleration and pitch angle of the fixed-wing UAV are ${4}{\text{m}/\text{s}}^{2}$ and ${\pi}{/}{3},$ respectively. The initial altitudes of the target and monitor are 100 m and 120 m, respectively. The mission duration is ${T} = {300}\,{\text{s}}{.}$ The rest of the simulation parameters concerning the UAV, such as the fixed blade power, incurred power, and mean rotor-incurred velocity, are consistent with those in [8].
The computational overhead and frame rate depend on the computational complexity, CPU frequency, and CPU core performance. At each time slot, the computational overhead of the new method is dominated by that of the interior point method used to plan the trajectory of the UAV monitor for the next N slots, which is ${\mathcal{O}}{(\text{N}}^{3.5}{),}$ i.e., about $(3\text{N}{)}^{3.5}$ floating point operations (FLOPs) for a rotary-wing UAV since the 3D position needs to be optimized per slot or $(9\text{N}{)}^{3.5}$ FLOPs for a fixed-wing UAV as the 3D position, velocity, and acceleration need to be optimized per slot. Here, we set ${N} = {30}$ and ${\tau} = {0.2}$ s/frame, or in other words, the frame rate is five frames per second. We also assume that the CPU frequency is 3.3 GHz, which can readily support the computational overhead, i.e., ${90}^{3.5} = {6.9}\times{10}^{6}$ FLOPs for a rotary-wing UAV or ${270}^{3.5} = {3.2}\,\times\,{10}^{8}$ FLOPs for a fixed-wing UAV, within a frame.
No existing study has considered the energy sustainability of disguised UAV video surveillance in the literature; i.e., the UAV power has never been optimized in the studies of disguised UAV surveillance. To this end, no existing work is directly comparable to the new approach described in this article. With due diligence, we come up with three effective alternatives to the new approach, including the counterparts of the approach without disguising (labeled “ND” in the figures) and without energy harvesting and a scheme minimizing the traveling distance of the UAV monitor (labeled “DST” in the figures).
Figure 4 compares the fixed-wing and rotary-wing UAV monitors under a random target trajectory (random accelerations along the x- and y-axes) when the proposed MPC-based online algorithm (labeled as “Prop.” in the figures) and the baseline schemes (ND and DST) are performed. The weighting coefficients of visual disguise are set to ${(}{0.4},{0.6}{)}{.}$ The corresponding 3D trajectories are demonstrated in Figure 5.
Figure 4 The UAV type and optimization scheme. (a)–(c) The 3D trajectories and objective values of the fixed-wing and rotary-wing UAV monitors by the online scheme, given a random target trajectory, where Objective refers to the obtained value of the objective function, Energy refers to the total energy consumption of the UAV, and Disguise refers to the total value of the disguising metric. Prop.: proposed MPC-based online algorithm.
Figure 5 The 3D trajectories of the fixed-wing and rotary-wing UAV monitors by the online schemes.
Compared to the ND and DST approaches, the proposed MPC-based online algorithm strikes an elegant balance between distance preservation and elevation adjustment, resulting in effective visual camouflage. It is observed that with a less constrained mobility model, the rotary-wing monitor takes more and sharper level twists, preserves a wider level distance to (and behind) the target, demonstrates stronger randomness and better visual camouflage on the $(\text{x},\text{y})$-plane, and does not vary its elevation as frequently as compared to the fixed-wing monitor. While the rotary-wing UAV is more flexible in control and movement, the fixed-wing UAV can carry a larger solar panel and is more energy efficient. Nevertheless, both types of UAV monitors demonstrate a good visual disguise.
Figure 6 shows the change in the battery level [in watt hours (Wh)] of the UAV monitor over time during the surveillance task, with and without solar energy harnessing (labeled as “With EH” and “Without EH,” respectively), averaging more than 20 realizations of the random target trajectory along the x-axis. The battery level is reasonably stable and maintains the required energy reserve throughout the operation under the proposed scheme thanks to its energy-harvesting capability, as shown in Figure 6. Specifically, at the beginning of its mission, the UAV consumes more power than it can harvest, causing the battery level to decrease even when the UAV is equipped with energy-harvesting capability. As the UAV continues its flight, it begins to harvest more energy than it consumes, allowing its battery to be recharged and causing the battery level to increase. In contrast, without disguising or energy harvesting, the UAV monitor would either lose its capability of disguising (as shown in Figure 4) or quickly deplete its battery (as shown in Figure 6).
Figure 6 The battery level of the fixed-wing and rotary-wing UAV monitors with and without solar energy harnessing (EH).
Existing studies have been on disguised video surveillance by a single UAV. However, single-UAV surveillance is prone to visual occlusion, exposure, and fragility, posing challenges in the design of UAV surveillance. This section discusses the techniques that can potentially address these challenges.
In an urban environment, UAVs may experience visual occlusion due to trees, buildings, and junctions. Road sensors with fixed locations, broader FOVs, and advanced hardware may help address this issue. Observation failures due to occlusion can be significantly reduced since targets are unlikely to be obscured in all sensors’ FOVs simultaneously. To achieve disguised video surveillance, it would be helpful to investigate how to dynamically coordinate many sensors to collaborate with a UAV, e.g., sending their sensing results to the UAV. Other techniques for tackling visual occlusion are still desired, especially for regions without (sufficient) ground sensors.
The formation and coordination of multiple UAV monitors, e.g., a UAV swarm, can improve concealed video surveillance. But it is hard to collect all the UAV data and perform a centralized global optimization as the UAVs may have different owners, are dispatched for different purposes, and are unwilling to share the data. Multi-UAV surveillance can be potentially coordinated by distributed machine learning techniques, such as federated learning (FL) and multiagent learning. FL can further incorporate differential privacy and homomorphic encryption to preserve privacy. The design of learning models, loss functions, and privacy-preserving techniques for multi-UAV surveillance is an interesting topic to investigate.
This article presented a new trajectory-planning framework for UAV-based disguised tracking. While keeping its distance, the UAV monitor constantly changes its elevation and relative position in the target’s view. The framework utilizes MPC to plan and refine the trajectory of the UAV monitor online, effectively improving the tolerance of the trajectory to the inaccurate prediction of the target and suboptimal short-term trajectories planned per slot. Computer simulations confirmed the improvement of the framework over methods with no camouflage in disguise and energy sustainability and its support of both fixed-wing and rotary-wing UAVs on covert video surveillance.
Work in this article was supported by the National Natural Science Foundation of China, under Grants 62231010, 62071126, and 62101135, and the Innovation Program of Shanghai Municipal Science and Technology Commission, under Grants 20JC1416400 and 21XD1400300.
Shuyan Hu (syhu14@fudan.edu.cn) is currently a postdoctoral research fellow with the School of Information Science and Technology, Fudan University, Shanghai 200433, China. She received her B.E. degree in electrical engineering from Tongji University, Shanghai, China, in 2014, and her Ph.D. degree in electronic science and technology from Fudan University, Shanghai, China, in 2019. Her research interests include machine learning and convex optimizations and their applications to unmanned aerial vehicle networks and intelligent systems.
Xin Yuan (xin.yuan@data61.csiro.au) is currently a research scientist at the Commonwealth Scientific and Industrial Research Organization, Marsfield, NSW 2122, Australia. She received her B.E. degree from the Taiyuan University of Technology, Shanxi, China, in 2013, and her dual Ph.D. degree from the Beijing University of Posts and Telecommunications, Beijing, China, and the University of Technology Sydney, Sydney, Australia, in 2019 and 2020, respectively. Her research interests include machine learning and optimization and their applications to unmanned aerial vehicle networks and intelligent systems.
Wei Ni (wei.ni@data61.csiro.au) is currently a principal research scientist at the Commonwealth Scientific and Industrial Research Organization, Marsfield, NSW 2122, Australia. He is also a conjoint professor at the University of New South Wales, Kensington, NSW 2052, Australia, an adjunct professor at the University of Technology Sydney, Broadway, NSW 2007, Australia, and an honorary professor at Macquarie University, Sydney, NSW 2000, Australia. He received his B.E. and Ph.D. degrees in electronic engineering from Fudan University, Shanghai, China. His research interests include machine learning, online learning, stochastic optimization, and their applications to system efficiency and integrity.
Xin Wang (xwang11@fudan.edu.cn) is currently a Distinguished Professor and the chair of the Department of Communication Science and Engineering, Fudan University, Shanghai 200433, China. He received his B.Sc. and M.Sc. degrees from Fudan University, Shanghai, China, in 1997 and 2000, respectively, and his Ph.D. degree from Auburn University, Auburn, AL, USA, in 2004, all in electrical engineering. His research interests include stochastic network optimization, energy-efficient communications, cross-layer design, and signal processing for communications.
Abbas Jamalipour (a.jamalipour@ieee.org) is currently a professor of ubiquitous mobile networking at the University of Sydney, Sydney. NSW 2006, and the editor-in-chief of IEEE Transactions on Vehicular Technology. He received his Ph.D. degree in electrical engineering from Nagoya University. He was the president of the IEEE Vehicular Technology Society in 2020–2021. He is a fellow of the Institute of Electrical, Information, and Communication Engineers and the Institution of Engineers Australia, and he is an IEEE Distinguished Speaker.
[1] A. V. Savkin and H. Huang, “Bioinspired bearing only motion camouflage UAV guidance for covert video surveillance of a moving target,” IEEE Syst. J., vol. 15, no. 4, pp. 5379–5382, Dec. 2021, doi: 10.1109/JSYST.2020.3028577.
[2] S. Hu, Q. Wu, and X. Wang, “Energy management and trajectory optimization for UAV-enabled legitimate monitoring systems,” IEEE Trans. Wireless Commun., vol. 20, no. 1, pp. 142–155, Jan. 2021, doi: 10.1109/TWC.2020.3023816.
[3] H. Huang, A. V. Savkin, and W. Ni, “Online UAV trajectory planning for covert video surveillance of mobile targets,” IEEE Trans. Autom. Sci. Eng., vol. 19, no. 2, pp. 735–746, Apr. 2022, doi: 10.1109/TASE.2021.3062810.
[4] G. C. Calafiore and L. Fagiano, “Robust model predictive control via scenario optimization,” IEEE Trans. Autom. Control, vol. 58, no. 1, pp. 219–224, Jan. 2013, doi: 10.1109/TAC.2012.2203054.
[5] S. Wang, F. Jiang, B. Zhang, R. Ma, and Q. Hao, “Development of UAV-based target tracking and recognition systems,” IEEE Trans. Intell. Transp. Syst., vol. 21, no. 8, pp. 3409–3422, Aug. 2020, doi: 10.1109/TITS.2019.2927838.
[6] W. Zhang, K. Song, X. Rong, and Y. Li, “Coarse-to-fine UAV target tracking with deep reinforcement learning,” IEEE Trans. Autom. Sci. Eng., vol. 16, no. 4, pp. 1522–1530, Oct. 2019, doi: 10.1109/TASE.2018.2877499.
[7] S. Hu, W. Ni, X. Wang, A. Jamalipour, and D. Ta, “Joint optimization of trajectory, propulsion, and thrust powers for covert UAV-on-UAV video tracking and surveillance,” IEEE Trans. Inf. Forensics Security, vol. 16, pp. 1959–1972, Jan. 2021, doi: 10.1109/TIFS.2020.3047758.
[8] S. Hu, W. Ni, X. Wang, and A. Jamalipour, “Disguised tailing and video surveillance with solar-powered fixed-wing unmanned aerial vehicle,” IEEE Trans. Veh. Technol., vol. 71, no. 5, pp. 5507–5518, May 2022, doi: 10.1109/TVT.2022.3157705.
[9] R. Strydom and M. V. Srinivasan, “UAS stealth: Target pursuit at constant distance using a bio-inspired motion camouflage guidance law,” Bioinspiration Biomimetics, vol. 12, no. 5, Sep. 2017, Art. no. 055002, doi: 10.1088/1748-3190/aa7d65.
[10] I. Rano, “Direct collocation for two dimensional motion camouflage with non-holonomic, velocity and acceleration constraints,” in Proc. IEEE Int. Conf. Robot. Biomimetics (ROBIO), Shenzhen, China, Dec. 2013, pp. 109–114, doi: 10.1109/ROBIO.2013.6739444.
[11] J. Kim, “Three-dimensional discrete-time controller to intercept a targeted UAV using a capture net towed by multiple aerial robots,” IET Radar Sonar Navig., vol. 13, no. 5, pp. 682–688, Jan. 2019, doi: 10.1049/iet-rsn.2018.5408.
[12] Y. Zeng and R. Zhang, “Energy-efficient UAV communication with trajectory optimization,” IEEE Trans. Wireless Commun., vol. 16, no. 6, pp. 3747–3760, Jun. 2017, doi: 10.1109/TWC.2017.2688328.
[13] M. C. Achtelik, J. Stumpf, D. Gurdan, and K. M. Doth, “Design of a flexible high performance quadcopter platform breaking the MAV endurance record with laser power beaming,” in Proc. IEEE Int. Conf. Intell. Robots Syst., 2011, pp. 5166–5172, doi: 10.1109/IROS.2011.6048336.
[14] S. Schopferer, M. Brizon, C. Liersch, and S. Froese, “Evaluating the energy balance of high altitude platforms at early design stages,” in Proc. Int. Conf. Unmanned Aircraft Syst., Arlington, VA, USA, Jun. 2016, pp. 170–177, doi: 10.1109/ICUAS.2016.7502577.
[15] N. K. Ure, G. Chowdhary, T. Toksoz, J. P. How, M. A. Vavrina, and J. Vian, “An automated battery management system to enable persistent missions with multiple aerial vehicles,” IEEE/ASME Trans. Mechatronics, vol. 20, no. 1, pp. 275–286, Feb. 2015, doi: 10.1109/TMECH.2013.2294805.
Digital Object Identifier 10.1109/MVT.2023.3263329