A SIMULATION SETUP FOR COMMUNICATION HARDWARE INTHE LOOP EXPERIMENTS

Markus Sauer and Florian ZeigerZentrum fur Telematik e.V., Allesgrundweg 12, Gerbrunn, Germany

{markus.sauer,florian.zeiger}@telematik-zentrum.de

Klaus SchillingDepartment of Computer Science VII, University of Wurzburg, Am Hubland, Wurzburg, Germany

schi@informatik.uni-wuerzburg.de

Keywords: Networked mobile robots, Communication, Simulation.

Abstract: Simulations are a very powerful tool in robotics to design and verify new algorithms before doing time-consuming tests with real hardware. Nowadays, a lot of very realistic simulation environments are available tosimulate robot kinematics and dynamics and any type of multi-robot systems in a virtual physical environment.Unfortunately, the communication in these simulations is often only considered in a very simplified matter,although the characteristics of a real communication link are very complex and might have a strong influenceon the performance of a multi-robot algorithm. This contribution proposes a setup to perform communicationhardware in the loop tests with the 3D simulation environment USARSim. For this setup any communicationdevice which can be connected to a PC architecture like WLAN,UMTS or Bluetooth can be used. A coop-erative collision avoidance algorithm is presented as an example which is realized with this setup, while realhardware is used for the communication link between the robots. Finally, the limitations are presented.

1 INTRODUCTION

The progress in the area of telecommunication tech-nology together with the demand of networked mo-bile robot systems to assist humans in many differentareas (e.g. disaster management, security and surveil-lance, or search and rescue applications) forces thedevelopment of multi robot systems which incorpo-rate several autonomy functions like formation driv-ing and obstacle avoidance. Hereby, due to the re-quired flexibility and dynamic communication topol-ogy, distributed control algorithms are very desirable.For the development of these mechanisms to controland coordinate swarms of mobile systems or multirobot systems capable simulation or emulation en-vironments are a useful and necessary tool for effi-cient development and analysis. But the use of simu-lation environments for networked mobile robot sys-tems also implies some consideration with respect tosignificance and validity of the simulation. On theone hand the complete dynamics and kinematics ofeach system must be modeled appropriately. On theother hand, also the available communication link in-between the robots must be represented in a suitablemanner. With respect to the simulation of the dynam-

ics and kinematics of mobile robots in multi robot sys-tems several simulation environment were developedin the recent years. Two well-known examples of themany available simulators are Player/Stage (Gerkeyet al., 2003) and USARSim (Carpin et al., 2007).Player is a robot device server to realize multi-robotor sensor-network systems. Stage can be used to-gether with Player and can simulate large populationsof robots in a 2D environment. USARSim is based onthe famous Unreal Tournament 2004 game engine. Itis a general purpose 3D - multi-robot simulator whichprovides basic physical properties of the robot andthe simulated environment which closely match thereal implementation of the robots and the real envi-ronment. In addition, it is also possible to simulatecamera images from cameras inside the simulation.Compared to Player/Stage it is only a simulation with-out a device server and controller concept like Player.Figure 1 shows a typical environment simulated withUSARSim for the virtual RoboCup Rescue league.

With respect to the simulation of the communica-tion link also many approaches and even products areavailable to be integrated. Of course, the importanceof these simulations of communication link technolo-gies is not only limited to the area of multi robot sys-

312

Figure 1: Typical environment simulated for the virtualRoboCup Rescue with USARSim.

tems. In the area of network testing and evaluation ofwireless network systems (Doshi et al., 2007) oftenthe QualNet network simulator is used for the setupof real-time emulations. This simulator is also usedin (Xu et al., 2003) for simulations regarding qualityof service provisioning in wireless ad-hoc networks,as well as in (Bagrodia et al., 2006), where a sys-tems simulation environment for future combat sys-tems is presented. In the area of networked hapticvirtual environments (Sankaranarayanan et al., 2007)used NIST-Net to create realistic Internet-like charac-teristics in a laboratory setting. NIST-Net (cf. (Car-son and Santay, 2003)) is a tool to facilitate testingand experimentation with network code through emu-lation which can model communication performancecharacteristics like packet delay, jitter, bandwidth lim-itations, congestion, and packet loss.

Of course, there exist other powerful simulationtools like NS2 or OPNET. All these simulation en-vironments are very mighty tools which have focusedon the simulation of the characteristics of the commu-nication channel. Unfortunately, they are often verycomplex and time-consuming to operate and most ofthem cannot be easily integrated with the simulationenvironments for mobile robot dynamics and kine-matics mentioned before. It is also known that thesimulation tool itself influences the outcome of a sim-ulation (Liu and Kaiser, 2005). In addition, you needto test the algorithm anyway later with real commu-nication hardware. Currently, in the area of simula-tion of networked robot systems and robot swarmsthe simulation of the communication interface is of-ten represented in a very abstract or simplified way.Nevertheless, several publications for networked con-trol systems turned out the importance of the knowl-edge about the communication characteristics and itsinfluence on the implemented control algorithms. In(Lopez et al., 2006), experiments of closed-loop net-worked control systems are evaluated focusing specif-ically on the performance and time delays effectsfor different compensation actions. In (Wei et al.,

2001) stability of networked control systems is in-vestigated for different network-scheduling protocols.Also methods for compensating network-induced de-lay are presented together with experimental resultsfor networked control systems with packet loss on thecommunication link. (Walsh et al., 2002) provided ananalytical proof of global exponential stability for anovel control network protocol and commonly usedstatically scheduled access methods. There, the fo-cus is set on communication constraints which areimposed by the network and the performance of theproposed protocol and the statically scheduled pro-tocols are examined in simulations. As above men-tioned, the behavior of the communication channel isvery important for the analysis and implementationof coordination and distributed control algorithms fornetworked robotic systems and may influence the be-havior of the complete system. Thus, this work pro-poses an approach how real communication hardwarecan easily be included into hardware simulation en-vironments - in this case USARSim. The communi-cation hardware is used as in real world applicationsbut nevertheless directly integrated to the algorithmsto be analyzed. The environment consisting in a mapand the dynamics and the kinematics of the physicalentities (mobile robot clients) is provided by the US-ARSim server. This modular design allows flexibleextensions in terms of replacing the simulated robothardware by real mobile robot hardware which accel-erates the development duration of multi robot sys-tems. As the proposed system integrates real wire-less communication hardware and standard protocolstacks directly into the simulation an intensive anal-ysis of implemented coordination and control algo-rithms for robot teams under consideration of the ef-fects of real wireless communication is possible.

The remainder of this work is structured as fol-lows. First, the hardware in the loop setup whichintegrates the real communication stack in the sim-ulation is introduced. Then an implementation of acooperative collision avoidance algorithm as exampleapplication is presented. Afterwards, the areas andthe constraints of the proposed setup are investigated.

2 HARDWARE IN-THE-LOOPSETUP

The objective of the simulation system design is theuse of real communication hardware while simulatingmultiple robots with USARSim. Therefore, the pre-sented system can be divided into three main parts: alocal area network segment, the clients, and a wirelesscommunication segment (cf. Figure 2).

A SIMULATION SETUP FOR COMMUNICATION HARDWARE IN THE LOOP EXPERIMENTS

313

Figure 2: Setup of Hardware in-the-loop Simulation Components.

2.1 Hardware Setup

The local area network segment uses standard Ether-net communication to provide connectivity betweenthe USARSim server and all clients which supportshigh bandwidth communication with low delays. Thisconnection is used for the exchange of drive com-mands and sensor data between the USARSim simu-lation environment and the different clients. This seg-ment represents the indirect communication betweeneach client and its environment, and as it is realizedvia the Ethernet segment, the direct communicationbetween each client over the wireless link is not inter-fered. Communication between the clients is not im-plemented via this link. The clients are equipped witha mini PC architecture with 1200MHz, 1GB RAM,a 8GB compact flash card as hard disk, and DebianEtch as operating system provides a platform to exe-cute the programs and algorithms for navigation andcooperation tasks which should be investigated andanalyzed. This mini PC represents the computingpower of a single robot. The LAN segment is onlyused by each client to retrieve environment data fromthe USARSim server. Communication between theclients is only realized via the wireless communica-tion segment. The wireless segment is based on IEEE802.11 wireless LAN and represents the communi-cation hardware which is directly integrated into thesimulation setup. This communication link is exclu-sively used for the communication between all clientsi.e. robots and human operators. As a standard oper-ating system is used the correspondingprotocol stacksare available and also routing mechanisms for wire-less ad-hoc networks like OLSR, DSR, or AODV caneasily be used.

2.2 Software Components

For each of the hardware components described inSection 2 also dedicated software components areexisting. On the simulation layer a USARSim serveris running which provides an environment model,the physical behavior of the clients, and sensordata for clients. As only the Ethernet segment isused for the communication between the USARSimserver and the clients, the inter client communicationvia the wireless segment is not affected. On theclients no specific installation for the USARSimsimulation and for maintaining the connection to theUSARSim server is required. The communication tothe USARSim server, and respectively the simulatedrobots are realized with simple string messages overTCP-socket (Carpin et al., 2007). Each client isrunning on one of the described mini PC. Basicallyhere, the distributed control algorithms can beimplemented. Furthermore, the operating systemis also maintaining the communication link to theUSARSim server for sensor data acquisition andsending commands. The client PCs are also equippedwith WLAN PCMCIA cards supporting the IEEE802.11 b/g standard. The wireless communicationis exclusively used for inter-client-communicationwhich represents one of the key issues of the pro-posed architecture. In the presented setup all standardprotocol versions which are available for the clientoperating system (e.g. Debian Linux) can be used.As the wireless communication link is exclusivelyused for inter client communication, the real protocolstacks and real physical behavior of the link allowsfor meaningful hardware in-the-loop simulations.Thus, the navigation, coordination and cooperation

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

314

algorithms which should be analyzed are exchangingdata via communication links with a realistic behav-ior - including external disturbances.

2.3 The Simulation System Design

This system setup is designed as simulation envi-ronment for network control systems and scenariosof robots or robot formation driving with real IEEE802.11 wireless LAN communication. In this work,one client represents the formation leading robot andthe other three clients are robots which should keepa certain formation. The leading robot sends its po-sition data with a frequency of 10Hz to the otherteam members via the wireless link. A communica-tion from the team members back to the leader is notpresent. The communication between the USARSimserver and each client uses the standard USARSim in-terface based on TCP connections. All robots run thesame distributed cooperative collision avoidance al-gorithm while moving to their respective goal points.

3 EXAMPLE: COOPERATIVECOLLISION AVOIDANCE

Typically, distributed control algorithms for roboticnetworks (Bullo et al., 2008) often assume a cer-tain simplified model of the communication channel.Here, a setup is proposed to test these control algo-rithms with a real communication stack. As applica-tion example for this contribution, a cooperative col-lision avoidance control algorithms based on the con-cepts of (Stipanovic et al., 2007) is used.

In the example scenario, a group ofn mobilerobots should move through an environment withoutcolliding with objects in the environment or with eachother. There is no central instance coordinating themovement of the robots. In the shown simulation mo-bile robots with differential drive are used. They areequipped with a simulated laser range finder for ob-stacle detection. The laser range finder has a fieldof view of 180 degree and is mounted to the frontof the robots. The leader robotai=1 drives a rhom-bus in this environment and continuously sends itspose to the other followingn−1 robots. The robotsai∈{2..n} receive this pose over the wireless communi-cation segment and set their own new goal pose rela-tive to the pose of the received leader pose. Thus, theformation shown in Fig.3 is established in equilibriumof the controller. Any other logical communicationtopology can be realized with this kind of setup e.g.robot one can only communicate with robot two and

robot three and four can only communicate with robottwo. This is especially interesting for the investigationof the system behavior of distributed algorithms withcommunication constraints.

Figure 3: Relative positioning of the robots in formation.

For the presented example application, the mobilerobots are modeled with the kinematics of a differen-tial drive robot as first order system (cf. equation 1);

xi = vi ·cosΘi

yi = vi ·sinΘi

Θi = ωi (1)

xi , yi , andΘi denote the pose of the roboti. vi isthe translational velocity andωi the turn rate. On eachof the robots a combination of the following positioncontroller and a controller for obstacle avoidance isimplemented. The controller switches between dif-ferent behaviors depending on the current conditions.Without obstacles in the defined obstacle avoidancezone and the robot’s orientation is not towards thegoal (Θi 6= Θgi) the following controller applies:

Θi = −(Θi −Θgi) =r i

Li(uri −uli )

⇒(uri −uli ) = −Li

r i(Θi −Θgi) (2)

r i denotes the radius of thei-th robots’ wheels,L is thelength between the wheels,uri is the left wheel speed,uri is the right wheel speed respectively andΘgi is thedesired orientation towards the currently defined goal.

If the robot is oriented towards the goal(Θi −Θgi) < to (to - threshold for accuracy of orientationof robot towards goal),vi is aligned with the straightline between the robot’s position and the goal posi-tion. Therefore,vi only applies to ˙xi in the robot co-ordinate frame ˙xi = v andyi = 0. The following con-troller can be applied:

vi = −xi =r i

2(uli +uri )

⇒uri = uli = −xi

r iΘ (3)

In the robot coordinate framexi is under the abovegiven conditions equal to the distance between robotand current goal and it becomes zero if the desired

A SIMULATION SETUP FOR COMMUNICATION HARDWARE IN THE LOOP EXPERIMENTS

315

goal is reached. If an obstacle is in the defined sens-ing range, the controller is adapted according to thefollowing rules: First a vectorFoi is calculated.Foipoints in opposite direction of the nearest obstacle tothe robot and its length increases indirect proportionalwith the distance to the next obstacle. Then this vectoris combined with the normalized vector in goal direc-tion Fgi to a new goal direction vector incorporatingan obstacle avoidance component and a new desiredheadingΘoai (cf. Figure 4):

Foai = (Foi +Fgi) (4)

Θoai = arctan2(yoai,xoai) (5)

Finally, this value is inserted in the controller definedin Equation 2:

Θi = −(Θi −Θoai) =r i

Li(uri −uli )

⇒(uri −uli ) = −Li

r i(Θi −Θoai) (6)

Figure 4: Overview of the different values for the obstacleavoidance controller.

After orientation towardsΘoai the robot alwaysmoves for a small time period in this direction to avoidoscillations in reorienting due to the limitation of theobstacle sensing to 180 degrees. This translationalmovement is only done in cases where definitely nocollision can occur.

The experiments with this cooperative collisionavoidance algorithm were done withn = 4 robots.The results can be seen in Fig. 5 and Fig. 6. Fig.5 shows how the four robots move with respect toeach other over time while the three robots follow theleader robot driving a predefined rhombus trajectoryfor a certain experiment time. In each plane at a cer-tain time the position of the robots at this times can beseen. The 3D plot of the trajectories shows the reori-entation of the formation at the edges of the rhombus

over time and it can be seen that there was now col-lision because none of the trajectories is touching orcrossing each other. Fig. 6 shows the minimum dis-tance inside the group of mobile robots. The relativeposition of the three following robots was designed tohave a distance minimum of 1.7m between all robotswhen they are moving in perfect formation. In ad-dition the robots should never get closer then 0.4m.Fig. 6 shows that the algorithm satisfies these require-ments. The peaks in the graph occur always when theformation is reorienting at the edges of the rhombusdriven by the leader robot.

−10−5

05

10

−20−10

010

20

0

0.5

1

1.5

2

2.5

3

3.5

4

x 105

x−position [m]y−position [m]

time

[ms]

Figure 5: Position of each robot while driving in formation.

0 0.5 1 1.5 2 2.5 3 3.5 4

x 105

0.5

1

1.5

2

2.5

3

3.5

time [ms]

min

imum

dis

tanc

e [m

]

Figure 6: Minimum distance occurring between all robotsinside the formation.

4 APPLICATION AREAS ANDCONSTRAINTS

This simulation setup is designed for real communi-cation hardware in-the-loop simulations of networkedrobotics scenarios. The advantage of this specialsetup is, that the real communication protocol stackis used which saves the very complex simulation ofthe protocol stack and the physical behavior of thelink. A well suited application area of this system isin the simulation of swarms, multi-robot teams, and

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

316

formation driving. In these scenarios, control algo-rithms can be tested and evaluated under the influenceof real communication link behavior (limitations inmedium access etc.) and also different communica-tion protocols can be tested easily. Of course, alsosome limitations exist. As the hardware nodes are lo-cated quite close to each other, long distance com-munications and the consequential channel behaviorcannot be simulated. Nevertheless, for detailed sim-ulations of the interaction of communication proto-cols, control engineering and the underlying systemin multi-robot teams and formation scenarios, the pre-sented architecture is suites very well and allows aneasy and fast setup of significant simulations.

5 CONCLUSIONS

In order to simulate the behavior of networked multi-robot systems in general a model for the communica-tion channel has to be to implemented and verified.In most cases this is only possible with simplifica-tions and limitations and the simulated channel is notrepresenting a real communication channel anymore.Therefore, the conclusions drawn from the simulationof the tested algorithm might not be as meaningful asdesired.

The presented approach offers a possibility to re-alize easily a meaningful simulation with real com-munication hardware for network robotic scenarios.It provides the exact behavior of the complete, com-plex communication stack without any approximationor simplifications. Thus, the behavior of multi-robotalgorithms can be directly investigated with all thechanges in the communication and data flow betweenthe robots. In the combination with a simulator likeUSARSim is is possible to simulate network roboticsystems with basic physics and real communication.

The setup of a hardware communication in theloop simulation is much easier than the setup of ameaningful communication channel simulation com-bined with a multi-robot simulation. There are evenless uncertainties in the behavior of the system whenyou later go to real hardware.

Therefore, it is very easy to test the behavior ofalgorithms for typical applications of network controlsystems like teleoperation of robots or robot forma-tion driving with a real communication channel be-fore going to the real hardware. Due to the standard-ized interfaces which are used, such kind of setupsalso allow for an easy evaluation of different type ofwireless communication systems like e.g. WLAN,UMTS, HSDPA/HSUPA, Bluetooth, WiMax. Espe-cially testing of swarm behavior is very meaningful,

because like in the real system naturally the nodes inthe communication hardware in the loop simulationare very close to each other.

REFERENCES

Bagrodia, R., Tang, K., Goldman, S., and Kumar, D. (2006).An accurate, scalable communication effects serverfor the fcs system of systems simulation environment.In Proceedings of the Winter Simulation Conference.

Bullo, F., Cortes, J., and Martınez, S. (2008).Dis-tributed Control of Robotic Networks. Princeton Se-ries in Applied Mathematics. Princeton UniversityPress, Princeton, NJ.

Carpin, S., Lewis, M., Wang, J., Balakirsky, S., and Scrap-per, C. (2007). Usarsim: a robot simulator for researchand education. In2007 IEEE International Confer-ence on Robotics and Automation (ICRA’07).

Carson, M. and Santay, D. (2003). Nist net: a linux-basednetwork emulation tool.SIGCOMM Computer Com-munication Review archive, 33(3):111 – 126.

Doshi, S. R., Lee, U., and Bagrodia, R. L. (2007). Wirelessnetwork testing and evaluation using real-time emula-tion. ITEA Journal of Test and Evaluation, 28(2).

Gerkey, B., Vaughan, R. T., and Howard, A. (2003). Theplayer/stage project: Tools for multi-robot and dis-tributed sensor systems. In11th International Confer-ence on Advanced Robotics (ICAR), pages 317–323,Coimbra, Portugal.

Liu, C. and Kaiser, J. (2005). A survey of mobile ad hocnetwork routing protocols. Technical report, Univer-sity of Magdeburg, TR-4: Middleware for NetworkEccentric and Mobile Applications (MINEMA).

Lopez, I., Piovesan, J., Lee, C. A. D., Martinez, O., Spong,M., and Sandoval, R. (2006). Practical issues in net-worked control systems. InProceedings of the Amer-ican Control Conference.

Sankaranarayanan, G., Potter, L., and Hannaford, B. (2007).Measurement and emulation of time varying packetdelay with applications to networked haptic virtual en-vironments. InRoboComm - First International Con-ference on Robot Communication and Coordination.

Stipanovic, D. M., Hokayem, P. F., Spong, M. W., andSiljak, D. D. (2007). Cooperative avoidance controlfor multiagent systems.Journal of Dynamic Systems,Measurement, and Control, 129(5):699–707.

Walsh, G., Hong, Y., and Bushnell, L. (2002). Stabilityanalysis of networked control systems.IEEE Transac-tions on Control Systems Technology, 10(3):438–446.

Wei, Z., Branicky, M., and Phillips, S. (2001). Stability ofnetworked control systems.Control Systems Maga-zine, IEEE, 21(1):84 99.

Xu, K., Tang, K., Bagrodia, R., Gerla, M., and Bereschin-sky, M. (2003). Adaptive bandwidth management andqos provisioning in large scale ad hoc networks. InProceedings of MILCOM’03.

A SIMULATION SETUP FOR COMMUNICATION HARDWARE IN THE LOOP EXPERIMENTS

317