Communications and robotics simulation in UAVs : A case study on aerial synthetic aperture antennas

(1)

Communications and Robotics Simulation in UAVs:

A Case Study on Aerial Synthetic Aperture Antenna

Miguel Calvo-Fullana, Alexander Pyattaev, Daniel Mox, Sergey Andreev, and Alejandro Ribeiro

Abstract—Driven by their versatility and autonomy, advanced unmanned aerial systems have indicated a substantial promise along the development of current communication networks, as well as in support of new communication platforms. The resulting integration of robotic and communication systems brings in- creased levels of complexity, which cannot be modeled accurately by conventional approaches. An example of this are aerial synthetic aperture antennas, in which an aerial robotic swarm coordinates their flight patterns and communication exchanges to produce a synthetic aperture. The accurate representation of such a system mandates the careful simulation of wireless communications, antenna radiation patterns, flight dynamics, and even environmental conditions. This article details an integrated communications and robotics simulation framework, which en- ables the assessment of these emerging systems. The benefits of this new approach are illustrated for the aerial synthetic aperture antenna example, but extend to a wide array of problems stemming from the union of communications and robotics. This novel modeling platform makes it possible to design and analyze joint flight and communication protocols for unmanned aerial systems. The obtained results provide valuable insights into the operation of swarm-based synthetic aperture antennas. Moreover, the methodology introduced in this work is flexible and can be readily applied to other instances of joint communication and robotic systems.

I. INTRODUCTION

Unmanned Aerial Vehicles (UAVs) have raised considerable interest in the research community. This attention has not only been motivated by the specific problems that arise when working with aerial autonomous platforms, but also due to a wide array of new challenges to which they offer potential solutions. Factors such as high autonomy and versatility of quadrotor platforms coupled with the steadily decreasing costs of off-the-shelf equipment have made UAVs a practical choice for a large variety of commercial applications. Examples range from agricultural monitoring to search and rescue missions, surveillance operations, and delivery of medical supplies to remote areas [1]. More importantly, when equipped with communication hardware, UAVs have shown promise both as an enhancer of communication networks as well as an enabler of new paradigms in communication technologies [2], [3].

UAV platforms are highly mobile and can be readily tailored to the requirements of the wireless network that they support.

A communication framework that seeks to utilize them effectively has to take these factors into account. Existing approaches include the use of UAVs as mobile relays, extension of network coverage, and dissemination and collection of data

Miguel Calvo-Fullana, Daniel Mox, and Alejandro Ribeiro are with the University of Pennsylvania; Alexander Pyattaev is with YL-Verkot Oy; Sergey Andreev is with Tampere University.

from sensor networks [4]. In all these scenarios, realistic mobility models that effectively represent UAV flight dynamics [5] are an important design consideration. Furthermore, due to tight coupling of network and robotic components, reliance on horizontal approaches to the design of such systems has severe limitations. To this end, co-design of robotic and networking modules is a promising alternative for developing practical UAV applications.

While methods favoring co-design of communications and robotics have been proposed in the past [6], [7], attaining high- fidelity simulations of such systems remains an open issue.

Robotic (or autonomous) systems are governed by perception- action loops (control loops), which in communication-enabled cases are often closed over a wireless channel. Hence, the operations of the robot and the wireless channel are coupled, thus necessitating a certain level of joint modeling. This is why accurate simulation of UAVs relies on robotic simulators, such as the Robot Operating System (ROS) [8], coupled with accurate physics simulators, such as Gazebo. However, the UAV systems are equipped with complex communication modules, which also need to be modeled accurately. Since the communication component plays a fundamental role in the control loop, it is essential to involve the communication simulation in a concurrent manner.

One of the application areas where UAVs have demonstrated practical promise is that of remote sensing. On the one hand, due to their flexibility and maneuverability, UAVs become exceptional platforms for radar sensors. However, on the other hand, the small size of UAV platforms also limits the size of the mounted antenna, thus restricting the power and the wavelength at which the system can operate. An example of a radar system requiring larger wavelengths is that of Ground Penetrating Radar (GPR). Typically implemented on space- based platforms or manned large-scale aircrafts [9], these mobile setups can safely survey hazardous or hard to access environments. While some approaches have been taken with regards to mounting GPR systems on single UAVs [10], their capabilities are still heavily limited by the antenna size. These size limitations motivate the use of synthetic aperture antennas, by employing a swarm of UAVs to generate an equivalent synthetic aperture of much larger wavelength [11]. However, radar imaging systems, when deployed with even a single platform, have been shown to be very sensitive to position and estimation errors [10], an issue that is further aggravated as the number of UAVs grows. All of these limitations point to a clear benefit in attempting a communications and robotics co- design to reduce the impact of position and estimation errors on the communication system.

(2)

This work provides an overview of the benefits brought by joint simulation of communication and robotic components in UAVs. Specifically, it considers the UAVs as enablers of new technologies, which is exemplified by the design of aerial synthetic aperture antennas. First, this article offers in Section II a review of synthetic aperture antennas and introduces specific issues of their multi-platform option. Section III outlines a novel communication-enhanced position controller. This new flight controller is based on a co-design approach, which exploits the Fine Time Measurement (FTM) ranging available in, e.g., IEEE 802.11 hardware. This allows the UAVs to obtain local distance estimates, which are fused with the local state estimates and the GPS positions of the UAVs to maintain a required formation more accurately. To evaluate this co- design approach, Section IV introduces a joint simulation setup supported on the ROS-NetSim framework [12]. This setup is then utilized to produce the numerical results in Section V, which are both rich in detail and illustrative of the benefits of the proposed approach. Final discussions and possible extensions to the current work are addressed in Section VI.

II. AERIALSYNTHETICAPERTUREANTENNAS

In applications, such as radar or other forms of remote sensing, achieving high antenna gain and directivity is crucial to successful system operation. Additionally, certain radars function at low frequencies, which imply large wavelengths and, consequently, bulky antennas. For example, GPR usually operates at around 10 MHz, which corresponds to the wavelengths of around 30 m. The above requirements (high antenna gain, sharp antenna directivity, and large wavelength) can easily be satisfied by having a broad antenna aperture, readily provided by a large antenna or, when prohibitive, a precisely constructed antenna array. While viable for static and permanent installations, large rigidly connected antenna arrays may become impractical for the applications that require a certain level of mobility.

Not limited to successful application of synthetic aperture antennas in large-scale single platforms, current interest has also concerned smaller-scale platforms including UAVs. Due to their small size, low cost, and high versatility, synthetic aperture antennas can also be mounted onto multiple aerial mobile platforms, thus forming an aerial synthetic aperture antenna. In a multi-platform setup, each measurement in the path is acquired by a set of platforms. Effectively, this means that a swarm of agents substitute a single platform while maintaining a specific formation.

However, aerial synthetic aperture antennas bring consider- ably more complexity than their single platform counterparts (specific details regarding the operation of UAV-based synthetic aperture antennas are available in, e.g., [13]). The swarm of agents have to maintain a tightly coordinated formation to obtain a precise synthetic aperture. Deviations from ideal positioning bring reductions both in terms of the directivity as well as the effective gain of the synthetic aperture. To maintain their formation, agents resort to flight control via a combination of on-board Inertial Measurement Unit (IMU), visual odometry (VO), and Global Positioning System (GPS)

equipment. Furthermore, they need to share their positions with other agents in the formation, typically over IEEE 802.11 equipment in practice. This tight coupling between positioning and communication features motivates the approaches based on a co-design of communications and robotics.

III. FLIGHTCONTROL ANDCOMMUNICATIONCO-DESIGN

Despite the benefits provided by multi-agent operation, there is a severe drawback of having to drastically mitigate the positioning errors when working with UAV-based synthetic apertures. The agents comprising a formation are often subject to environmental factors, such as wind gusts and turbulence, while at the same time positioning themselves via on-board equipment (e.g., IMU, VO, and GPS) that produces noisy state estimates. As a result, there may be errors in the antenna array alignment that would not occur if the platforms were fixed on the ground.

Deviations from the ideal position in the array formation make the resultant synthetic aperture less effective. In particular, the directivity of an antenna may be reduced, which yields decreased range at a fixed transmit power. More specifically, the antenna directivity is exponentially degraded with respect to the variance in the position errors of the array elements [11].

A. Fine Time Measurement Ranging

A feasible method to reduce the positioning errors is by following a co-designed approach: employ the communication capabilities of the agents in a formation to enhance their positioning control. A feasible way to achieve that is via FTM ranging. The FTM protocol was introduced with the IEEE 802.11mc standard and is a type of Round Trip Time (RTT) distance estimation. This ranging method is an im- provement over the more traditional Received Signal Strength Indicator (RSSI) based ranging, which utilizes signal power measurements to estimate the distance to a wireless transmitter.

Accordingly, RSSI-based approaches are more sensitive to factors like environmental attenuation and are more dramat- ically effected by multi-path propagation, interference, and even temperature. Consequently, FTM-based ranging provides a much improved distance estimation over the IEEE 802.11 communication equipment.

The thinking behind the FTM-based ranging is straight- forward. The time of flight between two signals transmitted by two FTM-enabled devices is used to estimate the distance between them. The exact protocol is as follows: an agent sends an FTM request to another agent, after which an acknowledgement is issued by the corresponding agent, thus initiating the FTM-ranging procedure. After the acknowledgement, a message and an acknowledgement are sent, which results in four timestamps (sent time of FTM message, receive time of FTM message, sent time of ACK message, receive time of ACK message). The round trip time of these messages is recorded and the distance is computed by taking into account the time it took them to travel at the speed of light. The accuracy of this system can be improved by repeating the procedure as often as necessary, thus resulting in bursts of messages that can then be averaged.

(3)

B. FTM-Augmented Position Control

All of the above leads to a possibility of enhancing the position control in the UAV swarm by the means of a communication technology, namely, FTM ranging. Under this operation, the agents execute a formation control algorithm to maintain the required array formation and synthesize the desired antenna. There are several key aspects of that algorithm.

First, the agents broadcast their GPS readings at a specified rate. They do so over the IEEE 802.11 wireless channel.

Further, each agent queries every other agent in the formation by using the FTM ranging – each agent obtaining a distance estimate to every other agent in the formation.

Then, at each agent, a fusion algorithm processes the on- board odometry estimates, the GPS positions of other agents in the formation, and the FTM distances with respect to every other agent. This information is then leveraged to anticipate the difference between the current position and the desired formation position. It updates the flight control loop to move in the preferred direction so as to reduce the error between the current and the intended positions.

IV. PROPOSEDSIMULATIONARCHITECTURE

To accurately model the aerial synthetic aperture antenna system described in the previous section, a sophisticated simulation platform is required. The system is composed of multiple agents, each of which is equipped with specific antennas that come together to generate a synthetic aperture. It is apparent that due to the integrated nature of communications and robotics in the co-designed system, simulation has to be approached as a joint endeavor. While one could naively decide to pursue the modeling in a separate manner, by capturing dynamics on the one side and communications on the other, this would result in inaccurate representation of the compound system.

The positioning is dependent on the communication and the other way around. Since co-design is inherent to the system, joint simulation is also a requirement. At its core, this is a well-known issue that plagues not only the UAV scenario considered in this work, but also other complex engineering systems [14]. Due to their integrated nature, these systems are composed of different technologies that end up relying on different simulators. However, the aggregate system behavior can only be represented accurately by utilizing a joint modeling.

This has been shown to be particularly important in the case of networked control systems [15], to which communication- enabled robotics belongs. There are several examples of the latter. A common case is tele-operated deployment, wherein a system is controlled over a wireless channel. More generally, systems comprising Perception-Action-Communication (PAC) loops are also subject to these simulation approaches. These are systems common in industrial applications, such a those forming part of what is commonly known as the Industrial Internet. Furthermore, multi-agent robotics, in which agents coordinate over a wireless channel, is also subject to these issues.

ROS

Network Coordinator

Physics Coordinator Network Simulator (WINTERsim)

Physics Simulator (Gazebo + PX4)

Antenna Simulator (NEC2)

Radiation Pattern Position Control

Communication Agent 1

Position Control Communication

Agent N

· · ·

SyncSyncSync ChannelChannelChannel

Agent Poses GPS Broadcast

FTM Request GPS Broadcast

FTM Request

Distances

FlightActuation

Flight Dynamics FlightActuation

Flight Dynamics GPS BroadcastFTM Request GPS BroadcastFTM Request

Distances

FlightActuation

Flight Dynamics FlightActuation

Flight Dynamics

Fig. 1. System architecture of co-designed position control integrated into ROS-NetSim simulation framework. Transmitted network messages are illustrated by red lines. Blue lines correspond to kinematic interactions. Message passing required for system operation is shown in green.

A. Developed ROS-NetSim System

To conduct an accurate modeling of the system at hand, the approach proposed in this work is built on top of ROS-NetSim [12]. This is an integrated simulation framework designed to enable the interfacing and concurrent operation of communication and robotic simulators. ROS-NetSim is based on the ROS [8], and its general architecture when integrated into the co-design approach outlined above is illustrated in Fig. 1.

Several components compose the ROS-NetSim framework, which can comprise (i) the external simulators, (ii) the internal coordinators, and (iii) the ROS nodes.

The external simulators, network, and physics exchange information with each other (via internal coordinators) through a series of coordination and message passing mechanisms.

Hence, these components are customizable to the user needs since a wide range of simulators is supported. To analyze the aerial synthetic aperture antenna case addressed by this work, it is required that the selected simulators implement certain functionalities needed to evaluate the scenario. First, to capture high-fidelity physics and flight dynamics, Gazebo is

(4)

used as the modeling tool. This is a widely supported physics simulator, which seamlessly integrates with the PX4 flight controller. This allows for modeling where the flight control is executed as Software-In-the-Loop (SITL), thus resulting in the simulation of on-board localization together with the actuation and environmental feedback to the UAV.

With respect to the modeling of communication capabilities, support for the IEEE 802.11 radio stack is required. Due to the ubiquity of this technology, several network simulators might be used. However, the co-design flight controller requires precise simulation of the FTM capabilities introduced in IEEE 802.11mc, which is not supported widely. To this end, we rely on the WINTERsim network simulator. The versatility of this modeling platform, together with its support of the IEEE 802.11mc standard, make it suitable for the scenario at hand. Finally, coupled with the networking aspect, the simulations make use of an electromagnetic modeler to obtain beamforming patterns. In this regard, we resort to Numeri- cal Electromagnetics Code 2 (NEC2), a popular and widely employed simulation platform for antenna design.

The inner operation of ROS-NetSim comprises two coordination blocks. The physics coordinator queries the physics simulator to extract geometric information from the environment that it then passes to the network simulator to generate the wireless channel. The network coordinator captures wireless traffic, which it then forwards to the network simulator for further processing. Finally, ROS nodes (that carry the application code) are launched in the simulation without any modification, thus interacting with the physics environment and having their network traffic captured explicitly.

B. Flight Dynamics and Control

The UAVs considered for the simulation are 3DR Iris quadrotor platforms, a popular commercial solution that is modeled in Gazebo with accurate measurements, weight, and aerodynamics. To attain high-fidelity modeling of flight dynamics, a full flight stack of the quadrotors is introduced in the modeling. Each of the UAVs runs an autopilot as SITL.

Specifically, the flight controller considered is PX4, an open source, industry standard autopilot, which ensures that the simulation is running the exact same processes as the UAVs would do in a physical platform.

On-board localization is modeled in high detail, with sensors like IMU, GPS, barometer, and magnetometer being modeled along with their corresponding noise profiles. On-board communication with the autopilot is handled via the MAVLink protocol. Positioning and orientation messages are exchanged between the node and the autopilot, which are then translated by PX4 into the corresponding actuation at each of the rotors of the UAV. Further, Gazebo simulates the resulting forces and physical feedback, by taking into account the environmental conditions like wind, which can be specified in the modeling platform.

C. Communication and FTM Ranging

With respect to the communication hardware, the UAVs are equipped with both an antenna to generate the synthetic

aperture and the IEEE 802.11 (Wi-Fi) communication equipment. Regarding the latter, IEEE 802.11ac-2016 Wi-Fi is the technology of choice, with each agent running it as a mesh station. The IP network is a flat address space with no routing, where all the UAVs know each other’s addresses, and no multi- hop communication takes place. This allows the neighboring UAVs to communicate the desired position changes and error vectors without over-complicating the networking logic.

To model the IEEE 802.11 interface, the capabilities of the WINTERsim system-level network simulator are used.

This allows to accurately capture the realistic conditions corresponding to a Wi-Fi system. The agents employ the Wi- Fi interface to communicate periodically by broadcasting their GPS positions encapsulated into UDP messages. Specifically, every 100 ms, each UAV sends a sequence of RTS-CTS, followed by a broadcast UDP packet that includes the distance measured to all the direct neighbors, current GPS position, as well as local GPS error estimate. In the case discussed in the previous section, with several agents sharing the same medium, effects like contention are essential. This is because high levels of contention are bound to have adverse effects on the position control of the agents. One needs to capture these effects, which is facilitated by the choice of the network simulator.

One of the properties of the outlined co-designed system is the use of Wi-Fi’s specialized functionality. In this case, FTM ranging functions are employed for accurate distance estimation between the UAVs. While the use of FTM may be questionable indoors, it produces sufficiently accurate measurements outdoors. The Line-of-Sight (LOS) conditions are typical for the UAV deployments, thus allowing for a feasible way to correct for the GPS errors. Furthermore, the capabilities of ROS-NetSim are used for this driver-level query operation by transferring FTM requests to the network simulator. This allows it to handle the actual MAC procedures in question, and return the results on a side channel (which in Linux systems is usually a netlink socket). Hence, specifics of FTM implementation are transparent to the ROS nodes and can be captured by the network modeler, as would be the case in real- world UAVs, where it may be handled by the firmware of the Wi-Fi chip.

D. Synthetic Aperture Antenna Modeling

Once the joint kinematic/networking simulation is per- formed, the resulting positions and orientations of the agents are provided to an electromagnetics modeler to obtain the eventual beamforming patterns of the synthetic aperture antenna. First, the antenna elements mounted on the UAVs need to be specified. Due to the dimensions of common quadrotor platforms, there is a considerable size restriction with regards to the mounted elements. Hence, compact solutions need to be employed.

In this sense, quadrotors mount a compact, single-loop me- ander antenna with a specific build configuration [13], having its radiation pattern almost equivalent to an electrically small dipole of5m of length. Then, with this antenna configuration, NEC2 is employed, which is a well-known antenna modeling

(5)

5

x z y

θφ

0.85λ 0.25λ

Fig. 2. Formation consisting of4×2agents. Formation acts as end-fire array;

hence, maximal beam gain is alongy-axis.

tool. Using NEC2 as a solver, the far-field gains of the system formation are computed as a function of azimuth and polar angles. For each instant of time, a separate instantaneous antenna pattern is generated. The sequence of patterns thus produced can then be processed to compute, e.g., probability density function of the beamforming gain, or other metrics of interest.

V. SÊLECTEDSÎMULATIONRÊSULTS

The previous sections described both the co-designed system and the necessary simulation architecture to accurately represent the UAV-based synthetic aperture antenna setup. In this section, simulations of the former by way of the latter are detailed. More specifically, the formation shown in Fig. 2 is modeled. This scenario consists of8 quadrotors in an open environment. The agents attempt to maintain the demonstrated 4 ×2 formation, with end-fire spacing of 0.25λ and front- fire spacing of 0.85λ. The simulation, as more thoroughly discussed in Section IV, consists of realistic flight dynamics (PX4) with environmental conditions (wind), state estimation (GPS and IMU), wireless communications (IEEE 802.11), and a mounted antenna to generate the synthetic aperture.

This discussion starts by considering the effect of wavelength on the performance of the system at hand. Wavelengths of5−20m are considered, which correspond to60−15MHz and cover the range typically used by ground penetrating radar and over-the-horizon radar. Two representative wind scenarios are evaluated: high wind of 5 N and low wind of 2 N. Both cases are illustrated in Fig. 3, where the resulting probability density function of the maximal gain of the synthetic aperture antenna is displayed. In general, very short wavelengths have difficulty achieving supergain array effects, which is clearly noticeable in the case ofλ= 5 m (has a near-flat probability density function). Further, the impact of wind is also apparent, as it clearly reduces the achievable maximal gain as well as flattens the distributions. If one would completely disregard the wind, the formation remained close to the ideal gain of 13.26dB. However, as wind increases, a noticeable reduction in gain occurs. This is because high winds have a clear effect on the angular deviation of the array.

Further, consider the resultant radiation pattern of the formation when exposed to high winds of 5 N. The wavelength is chosen to be λ= 10m, which corresponds to 30MHz, a typical ground penetrating radar frequency. The ideal (without wind and with perfect positioning) and the fully simulated radiation patterns of the synthesized antenna are demonstrated in Fig. 4. There are clear discrepancies between the ideal

Figure 4.16.Results for wavelength variation with2Nwind

54

Fig. 3. Resultant probability density function of maximal radiation pattern gain under high wind (5N, top) and low wind (2N, bottom) for different values of wavelength.

and the simulated cases. First, the main lobe loses part of its effective area and the overall gain is lower throughout the pattern. Second, due to imprecise positioning and clear impact of the wind, the pattern suffers from a visible deformation, thus losing its perfect symmetry.

Overall, Fig. 3 and 4 offer a comprehensive example of the type of information that can be collected by using an integrated modeling platform and a co-design approach like the one proposed in this work. The methodology based on detailed simulations advocated by this study can help make substantiated choices in the design of aerial antenna systems.

For example, Fig. 3 indicates that formations with λ= 5 m are much more sensitive to wind than formations withλ≥10 m. As a matter of fact, the effect of wind for formations with λ ≥ 10 m remains more or less similar. This indicates that for applications like ground penetrating radar, the formation is resilient to wind effects.

However, Fig. 4 also shows a region of reduced beamforming gain between φ = [180,225] and θ = [0,45], which for an ideal radiation pattern has adequate gain. This is caused by the UAV formation attempting to compensate for the wind;

it indicates that while the maximal gain of the main lobe is not heavily affected by the wind (as seen in Fig. 3), the overall shape of the formation is. Hence, one should be aware of this fact when designing the system, to not rely on those regions. While the results collected in this article have been narrowed down, the important choices, such as formation shape, antenna payload, communication protocols, flight hardware, and more, can be introduced into the system and be evaluated accurately. This provides an accurate way of making system-design decisions for the target UAV setup.

VI. DISCUSSION ANDCONCLUDINGREMARKS

This article discussed a novel framework for the accurate simulation of communications and robotics in unmanned aerial systems. The adopted approach was introduced through the lens of a study on aerial synthetic aperture antennas. In the considered scenario, a swarm of aerial quadrotors is equipped with radio communication capabilities, both for generating a synthetic aperture antenna and for communicating among each other. Due to the complexity and tightly coupled nature of the communication and robotic systems, a joint modeling

(6)

0 45 90 135 180 225 270 315 360 180

135 90 45 0

φ(^◦) θ(◦ )

−30 −24 −18 −12 −6 0 6 12 18 Gain (dB)

(a) Ideal radiation pattern.

0 45 90 135 180 225 270 315 360 180

135 90 45 0

φ(^◦) θ(◦ )

(b) Simulated radiation pattern.

Fig. 4. Resultant radiation patterns of a synthetic aperture antenna generated by4×2formation in Fig. 2. This scenario was modeled with high wind load of5N and wavelength ofλ= 10m. Horizontal and vertical axes correspond to horizontal and vertical angles in degrees, respectively (cf. axes in Fig. 2).

Main lobe is outlined in black.

method was advocated, which is being capable of extracting both accurate and rich information out of the aerial synthetic aperture system.

While the focus of this work was shaped by UAV platforms communicating over Wi-Fi, the co-design as well as the joint network and robotic simulation approach outlined here extends to many other use cases. The study of wireless autonomous systems, such as those envisioned in the Industrial Internet applications, connected autonomous driving vehicles, and others, can benefit from the proposed methodology. The framework put forward in this work can be maintained with simulators chosen to adapt to the use cases considered and the co-designed algorithms developed accordingly. Furthermore, while the emphasis was set on Wi-Fi communication, the emerging cellular networks like 5G and beyond are anticipated to see robotic systems as one of their major users. Therefore, to design and accurately evaluate these setups, approaches such as the one developed in this work are beneficial and, in some cases, necessary.

ACKNOWLEDGMENTS

This work was supported by ARL DCIST CRA W911NF- 17-2-0181, the Intel Science and Technology Center for Wire- less Autonomous Systems, the Academy of Finland (projects RADIANT and IDEA-MILL), the Jane and Aatos Erkkos Foundation (project STREAM), and the RAAS Connectivity RTF framework. We would like to acknowledge the con- tributions of Tanmay R. Godbole (formerly with Tampere University) to enable this research.

REFERENCES

[1] T. Tomic, K. Schmid, P. Lutz, A. Domel, M. Kassecker, E. Mair, I. L. Grixa, F. Ruess, M. Suppa, and D. Burschka, “Toward a fully autonomous uav: Research platform for indoor and outdoor urban search and rescue,” IEEE robotics & automation magazine, vol. 19, no. 3, pp. 46–56, 2012.

[2] M. Mozaffari, W. Saad, M. Bennis, Y.-H. Nam, and M. Debbah, “A tutorial on uavs for wireless networks: Applications, challenges, and open problems,” IEEE communications surveys & tutorials, vol. 21, no. 3, pp. 2334–2360, 2019.

[3] S. Hayat, E. Yanmaz, and R. Muzaffar, “Survey on unmanned aerial vehicle networks for civil applications: A communications viewpoint,”

IEEE Communications Surveys & Tutorials, vol. 18, no. 4, pp. 2624–

2661, 2016.

[4] Y. Zeng, R. Zhang, and T. J. Lim, “Wireless communications with unmanned aerial vehicles: Opportunities and challenges,” IEEE Com- munications Magazine, vol. 54, no. 5, pp. 36–42, 2016.

[5] D. Orfanus and E. P. de Freitas, “Comparison of uav-based reconnais- sance systems performance using realistic mobility models,” in2014 6th international congress on ultra modern telecommunications and control systems and workshops (ICUMT), pp. 248–253, IEEE, 2014.

[6] R. I. Bor-Yaliniz, A. El-Keyi, and H. Yanikomeroglu, “Efficient 3-d placement of an aerial base station in next generation cellular networks,”

in2016 IEEE international conference on communications (ICC), pp. 1–

5, IEEE, 2016.

[7] D. Mox, M. Calvo-Fullana, M. Gerasimenko, J. Fink, V. Kumar, and A. Ribeiro, “Mobile wireless network infrastructure on demand,” in2020 International Conference on Robotics and Automation (ICRA), IEEE, 2020.

[8] M. Quigley, K. Conley, B. Gerkey, J. Faust, T. Foote, J. Leibs, R. Wheeler, and A. Y. Ng, “Ros: an open-source robot operating system,”

inICRA workshop on open source software, vol. 3, p. 5, Kobe, Japan, 2009.

[9] D. J. Daniels, “Ground penetrating radar,” Encyclopedia of RF and Microwave Engineering, 2005.

[10] M. G. Fernández, Y. Á. López, A. A. Arboleya, B. G. Valdés, Y. R.

Vaqueiro, F. L.-H. Andr´es, and A. P. Garc´ıa, “Synthetic aperture radar imaging system for landmine detection using a ground penetrating radar on board a unmanned aerial vehicle,”IEEE Access, vol. 6, pp. 45100–

45112, 2018.

[11] S. H. Breheny, R. D’Andrea, and J. C. Miller, “Using airborne vehicle- based antenna arrays to improve communications with uav clusters,” in 42nd IEEE International Conference on Decision and Control (IEEE Cat. No. 03CH37475), vol. 4, pp. 4158–4162, IEEE, 2003.

[12] “ROS-NetSim.” https://github.com/alelab-upenn/ros-net-sim/, Accessed on October 8, 2020.

[13] T. R. Godbole, “Evaluation of coordinated drone swarm operation as a synthetic aperture antenna: A simulation-based study,” Master’s thesis, Tampere University, 2019.

[14] C. Gomes, C. Thule, D. Broman, P. G. Larsen, and H. Vangheluwe,

“Co-simulation: a survey,”ACM Computing Surveys (CSUR), vol. 51, no. 3, pp. 1–33, 2018.

[15] W. Li, X. Zhang, and H. Li, “Co-simulation platforms for co-design of networked control systems: An overview,”Control Engineering Practice, vol. 23, pp. 44–56, 2014.

Miguel Calvo-Fullana is a a postdoctoral researcher at the University of Pennsylvania. His research interests lie in the broad areas of learning and

(7)

optimization for autonomous systems. In particular, he is interested in multi- robot systems with an emphasis on wireless communication and network connectivity.

Alexander Pyattaevis the CTO of YL-Verkot Oy, consulting in the area of wireless communication system design. He has authored over 50 publications on the subjects of medium access control in wireless communications, simulation approaches for system design, heterogeneous networking, and mobile communication systems.

Daniel Mox is a PhD student in the Mechanical Engineering and Applied Mechanics department at the University of Pennsylvania. His research interests lie at the intersection of robotics and wireless systems and include multi-robot systems, mobile networks, and path planning.

Sergey Andreev is an associate professor of communications engineering and Academy Research Fellow at Tampere University, Finland. He (co- )authored more than 200 published research works on intelligent IoT, mobile communications, and heterogeneous networking.

Alejandro Ribeiro is currently Professor of Electrical and Systems Engi- neering at the University of Pennsylvania (Penn), Philadelphia. His research interests are in the applications of statistical signal processing to collaborative intelligent systems. His specific interests are in wireless autonomous networks, machine learning on network data and distributed collaborative learning.