Proceedings of the 2006 IEEEInternational Conference on Robotics and Biomimetics

December 17 - 20, 2006, Kunming, China

Communication Mechanism Study of a Multi-RobotPlanetary Exploration System *

Zheng Zhang"2, Shugen Ma 1,2,3, Zhenli Lu"4, Binggang Cao1 Robotics Laboratory, Shenyang Institute of Automation, 114 Nanta Street, Shenyang 110016 P.R.China

2 School of Mechanical Engineering, Xi'an Jiaotong University, 28 Xianning West Street, Xi'an 710049 P.R.China3 Organization for Promotion of the COE Program, Ritsumeikan Univeristy, Kusatsu, 525-8577 Japan

4 Shenyang Ligong University, Shenyang, 110168 P. R. Chinazhangzh@mail.xjtu.edu.cn Shugen.Magieee.org luzhl@sia.ac.cn inte-caogmail.xjtu.edu.cn

Abstract - Communication mechanism is a key problem formulti-robot research. The communication characteristics of areconfigurable planetary multi-robot system are analyzed.According to the function requirement of the multi-robot system,enlightened by the principle of TDMA (Time Division MultipleAddress) and rotational mechanism, a novel multi-robot wirelesscommunication mechanism is proposed. The mechanism and thecommunication protocol are also described in detail. Simulationand experimentation both prove the feasibility and practicabilityof the proposed mechanism.

Index Terms -Multi-robot coordination; Communicationmechanism; Wireless communication; Planetary Exploration

I. INTRODUCTION

The research field of multi-robot system is that how themultiple robots to accomplish a certain mission by theirinteractions. Such missions are often unable or difficult to befulfilled by a single robot. Therefore, communicationbecomes one of the key problems of multi-robot systems. Thecommunication mechanism of a multi-robot system is tightlyrelated to the coordination of the system. The communicationsbetween multiple robots ensure their interactions and theircoordination or cooperation.

Balch and Arkin[1] considered that communicationbetween the robots can multiply their capabilities andeffectiveness. Speech act theory regards the communicationsbetween robots as actions. This theory supposes that theexecutive languages of agents are just like their actions topromote their intentions. It provided and affected methods andthoughts for many agent communication languages developedrecent years. The best known ACL is KQML, developed bythe ARPA knowledge sharing effort (KSE)[2].The KSE iscomprised of two parts: the knowledge query andmanipulation language(KQML); and the knowledgeinterchange format (KIF). Klavins[3] studied the complexityof multi-robot system communication and proposed aCommon Control Language (CCL).

However, most of the theories and results aboutmultiagent systems aim at the coordination and thecommunications between two software multiagents. Thesetheories and results are difficult to suit the multi-robot

systems. Therefore, the general purpose communicationlanguages for multi-robot systems are unavailable. Fordifferent kinds of multi-robot systems, we have to design theircommunication mechanisms and communication protocolsaccording to the characteristics of systems and some up todate theories and results.

Fig.1 Reconfigurable planetary robot system(RPRS)

This paper investigates a reconfigurable planetary robotsystem (RPRS) [4][5][6]. HIROSE proposed RPRS anddeveloped a TIT model. We developed another new model inShenyang Institute of Automation(SIA). RPRS consists of sixwheels to support the main body, on which the main controlsystem of RPRS, solar battery cell, communication devicesand other equipments needed for the planetary exploringcarried. Each of the wheels is a single autonomous child robot,and is detachable from the main body, as shown in Fig. 1. Themain body can be regarded as a father robot. When a childrobot is in connected mode, it acts as one of the wheels of thefather robot. When it is in autonomous mode, it candisconnect automatically from the main body and then rove onthe planetary ground to sample or manipulate. When a childrobot blocked by an obstacle, it will send some messages toget help from other child robot or father robot. Two or morechild robots can connect automatically to form a new robot.The new one has more ability to get across the obstacle. InRPRS, the reconfigurations are required not only between

This work is partially supported by the National High Technology Research and Development Program of China (2002AA422130).

1-4244-0571-8/06/$20.00 C)2006 IEEE 49

child robots but also between child robots with father robot.The physical model of the child robot is shown in fig.2.

In this paper, we analyze the communication requirementsand characteristics of RPRS, then, we build a message basedcommunication model and propose a new mechanism ofmulti-robot communication. The advantage of the proposedmechanism is that some facility and complexity coordinationstrategies can be realized under the condition of simplecommunication hardware. Finally, the practicability of theproposed communication mechanism is proved both bysimulation and experiment.

Fig.2 Photo of child robot

II. COMMUNICATION ANALYSIS OF RPRS

Some occasions communication required in RPRS are asfollowing.

Firstly, when all the child robots connected to father robotand become one of the supporting wheels, the realization ofthe motion of father robot depends on the coordination of allthe connected child robots.

Secondly, when child robots leave from father robot toperform some kind of exploration missions, they need tocommunicate and exchange messages continuously each otheror with father robot, so that the mission can be fulfilledsuccessfully.

Thirdly, when two or more child robots reconfigured as anew robot, communication within these child robots isindispensable.

On the first instance, child robots support father robot andprovide its mobility, at the same time, the energy of childrobots come from the devices on father robot. Therefore, thecables connecting father robot with child robots areindispensable. So that, the communication can be realized bythis cables or by another added cables that between fatherrobot and child robots. On the second instance, because of thevariability of the relative position between all robots, wirelesscommunication is the only choice. On the third instance, linecommunication and wireless communication are both suitable.If we select wire communication, we need to design thecommunication interface between child robots with fatherrobot, so that the cable on the first instance can be reused.

Consequently, there are not only wire communications butalso wireless communication in RPRS. For the reconfiguredchild robots or the reconfigured child robot and father robot,cable can assure that the communication is real time. For therobot moves respectively, a kind of wireless communicationmechanism should be designed to ensure the effectiveworking of RPRS.

The following will focus on the study of the wirelesscommunication mechanism between the mobile robots ofRPRS. In theory, multiple mobile robot communication can beregarded as multiple communications or a wireless local areanetwork. One of the characteristics of multiple mobile robotcommunication is wireless communication determined by themobility of robots. Another characteristic is that thecommunication is multipoint to multipoint communication.Furthermore, on different occasions in RPRS, thecommunication may be one point to multi-point or multipointto multipoint. This just likes a team, and there is a leader inthe team. Each member of the team can contact with eachother, or they can only contact with the leader.

III. THE CONTROL SYSTEM STRUCTURE AND THECOMMUNICATION HARDWARE

A. Control system structure ofRPRS

PCI PC2

IFather ro-botI.. ..

.......HE.._..L- -

'-V

/ IChild robot2

/ I II I____.

"I.,.

IChildr-.~Chl obotiI

I~~~II

Fig.3 Overall control structure ofRPRSThe overall control structure of RPRS is shown in Fig.3.

Father robot equipped with two high performance computersPCI and PC2. PCI and PC2 can be a back-up system for eachother. While these two computers are normal, they take chargeof the tasks of themselves respectively. While one of them fail

50

to function, the other one will immediately take over thefailure computer's tasks, in case of the breakdown of RPRS.Therefore, the reliability of RPRS is guaranteed.

The main functions of father robot include the monitoringof the whole system, task decomposition, generating of thecoordination strategy, and the communication between fatherrobot with child robots and space station. Each child robotequipped powerful Microprogrammed Control Unit (MCU)which takes charge of the controlling of each joint of the armand adjusting the joint angle to perform a certain task. TheMCU goes with some sensors and wireless communicationfacility to provide communication and coordination betweenchild robots.

B. Control structure ofchild robot

The control structure of child robot is mostly composed ofcentral planning module and executive module, as shown infig.4. Central planning module includes control submodule,calculate submodule and wireless communication submodule.Executive module consists of three submodules that are jointexecutive submodule, paw executive submodule and wheelexecutive submodule[7][8].

Child robotcontrol system

Wireless Father robot or othercommunication child robots

submodule submodule Communication planningCalculateCon ommCentralosubmodule module

CAN bus

Joint executive Joint executive Wheel Pawsubmodule 1 submodule 4 executive executive

submodule submodule

Fig.4 Control structure of the child robot in RPRS

C. Communication hardware

Fujitsu MCU, MB9OF549, is selected as the controller ofchild robot. It includes two CAN bus interfaces. The serialport is provided both synchronization communication andasynchronism communication. It is also integrated with a 8/10bits A/D convertor and a 16 bits counter. The interfacebetween CAN controller with the physical bus is 82C250which accomplish the communication with bus.

Serial port of MAX232 PTR2000father robot

Wirelesscommunication:

Serial port of PTR2000child robot

Fig.5 Hardware structure of robot wireless communication

A wireless module, PTR2000, is adopted to realizecommunication between the robots, as shown in Fig.5. Theserial port in PC of father robot communicates with PTR2000by a MAX232 convertor. The MCU communicate withPTR2000 by the serial port of child robot. Therefore, twoPTR2000 modules connect father robot and child robot bywireless communication.

Furthermore, the condition we have to face is that all thechild robots and father robot must communicate each other byan only wireless communication channel.

IV. COMMUNICATION MECHANISM

Planetary exploration mission is different from others.RPRS is a special multi-robot system. Therefore, designing aflexible and reliable communication mechanism andcommunication protocol is inevitable.

A. Communication model

In multiagent systems domain, there are threecommunication models, such as blackboard model, point topoint model and message based model.

The advantages of blackboard model are centralizedcontrol, shared data structure and high effective when it isused to solve simple tasks. Since the central control performedby a scheduler, the complexity and performance of schedulerusually become a bottleneck of the communication system.Therefore, low dependability of this model makes it unsuitablefor some high security systems.

TCP/IP protocol is commonly adopted in point to pointcommunication model. Communication linkage is builtbetween every two agents in this model. The networktechnique is used to replace the central control model, so thatthe expandability is enhanced and a flexible communicationmechanism is feasible in this model.

In message based multiagent communication systems, amessage sent directly from one agent to another agent. Socache memory is no longer needed. Broadcast is a special casethat the message is sent to all the other agents. Usually, thebroadcast messages include an agent address by which theinformation appointed agent will receive this message and allother agents will discard it.

Message based dialogue communication is the basis ofbuilding a flexible and complicated coordination strategy inmulti-robot systems. Each agent uses a specified protocol tointerchange information, so that the communicationmechanism and coordination mechanism can be built.

Because of the limitation of the MCU of child robot,TCP/IP based multi-robot communication mechanism isunfeasible. Therefore, the message based model is used in ourcommunication mechanism[9].

B. Multiaddress communication method

A shared communication medium (SCM) is characterizedby multiple entities that use this medium by reading andwriting from and to it. Data writers perform write operationson the SCM. Data readers perform read operations and

51

retrieve values written by data writers. An example of such aSCM is the network in a distributed system such as Ethernet ina distributed application or a CAN bus in an automotiveapplication[10]. One well understood collision-avoidancescheme is Time Division Multiple Access (TDMA). TDMAsystems divide the transmission time into time slots, and ineach slot only one user is allowed to either transmit or receive.

TDMA shares a single carrier frequency with severalusers, where each user makes use of nonoverlapping timeslots.

Although frequency Division Multiple Access (FDMA)and Code Division Multiple Access (TDMA) are the othertwo techniques in multiple access communication system, theyare not suitable in RPRS.

The thought of time division multiple access can be usedin RPRS. Some modification of TDMA should be made toadapt the requirements of RPRS.

In the following section, a new distributed multi-robotcommunication mechanism is introduced, which enlightenedby the principle of TDMA and rotational mechanism. Withsuch a mechanism, multiple robots share one single frequencychannel to realize a kind of message based communication.C. Wireless communication mechanism ofRPRS

In this paper, we designed an appropriate proposal andmechanism of RPRS communication by which the wirelesscommunication problems of high error rate and time delay aresolved. In following, we will introduce the original wirelesscommunication mechanism by four child robots of RPRS.

TiRobotl

T4Robot4

T2Robot2

T3Robot3

Fig.6 Rotational communication mechanism

Fig.7 Rotational transfer seriesAt first, we designate the four robots as RI, R2, R3 and

R4 respectively. The four robots are assigned the right tomonopolize the communication channel in turn. Therefore, aseries of time slots are divided for each of the four robots. In atime slot, there is only one robot can send out its ownmessages. In others, it has to wait for and receive messagesfrom other robot to itself. Simultaneously, the length of the

time slots is protean as a result of the different messages fromdifferent robots. It just like every person gives a speech byturns in a meeting. However, each robot must send itsmessage when its turn coming, otherwise, it will be regardedas disabled.

Once the communication cycle beginning, RI will be thefirst robot to send message, the three others will be acceptingstate. The message sent from RI can be received by R2, R3and R4 at the same. After completing its sending, RI willswitch to accepting state. R2 will occupy the wireless channelto switch to sending state. After accomplishing its sending, R2will switch to accepting and the next time slot will be the turnfor R3 to send. After accomplishing its sending, R3 will giveup the channel and transform into accepting state. Then, R4will occupy the channel. After R4 finishing sending, RI willoccupy the channel send time that stand for starting of thenext cycle. The rotational communication mechanism formulti-robot systems is shown in Fig.6.

The time sequence of the four robots occupy the channelis shown in Fig.7. In this figure, we can find that Tmax is theupper limit of the time each robot using the channel for a time,and that Ttrans(5ms) is the transform time for a robot switchfrom accepting state to sending state or reverse. After R4sending, if RI doesn't send any message until Tmax haspassed, R2 will automatically start its sending and the right ofRI this time will be bereaved simultaneously. If the samesituation appears three times continuously on RI, RI will beregarded as a disabled robot by other robots. And then, thecommunication sequence will adjust to suit for the other threerobots which no longer includes RI again.D. Communication protocol

The message format a robot sending is as Fig.8. Thedetailed explaining for each part of the message is asfollowing.

Header Destination Source Number Type Length Content End

Fig.8 Message format of the wireless communication

Header. Three bytes, OxfOxaOx5, stand for the beginningof a message. This is because of that OxaOx5 following Oxfhardly appear[1 1].

Destination. One byte stands for which robot thismessage aim at. 'O' stands father robot. '1' to '6' stands forevery child robot respectively. '7' indicate a broadcastmessage that all the robots need to solve. '8' and '9' leave touse later.

Source. One byte denotes which robot this message comefrom.

Number. Two bytes explain the serial number ofmessages sending from robot that the source number indicates.For example, the first message sending from RI will bemarked as '00', the second message from RI be marked as'01'. The number will be increased by 1 each time until '99'reach. Then the number will begins from '00' again and thenext cycle starts.

52

Type. One byte shows the type of the message, '0' isseeking help, '1' is invitation of coordination, '2' iscoordination instruction, '3' is broadcast information, '4' isinformation of failure.

Length. Four bytes indicate the length of the informationby which the content of the information can be verified.

Content. The material communication content of themessage can be different type here. This part is the kernel ofthe message, while all other parts are designed to transfercontent to its destination successfully. The format of this partcan be described as following,<command><parameters><command><parameters><command><parameters>. Here, 'command' and 'parameters' arelanguages that all the communication robots can identify, suchas aht movex(dx),movey(dy),movez(dz), ormoveto(xposition,yposition,zposition),moveto(xposition,yposition,zposition).

End. Two bytes, 0x170xl9, denote the end of thismessage.

V. SIMULATION AND EXPERIMENTAL RESULTS

A. Simulation ofthe mechanism

In order to validate the proposed wireless communicationmechanism, we developed a multithreading based multi-robotcommunication simulation program, as shown in Fig.9. In thisfigure, there are four child robots participating in therotational communication mechanism. For each robot, there isa list box corresponding to it to display the messages thisrobot received, and there is an edit box to display themessages waiting to be sent. At the same time, in the leftbottom of the interface, the total number of the wirelesschannel be occupied can be seen.

Fig.9 Simulation of the communication mechanismA trouble management method is imported in the

simulation program. When RI is failure, the program canautomatically overleap it and a new communication cycle willbe built within the other three robots. So that, thecommunicating robot number decreases to three.

The simulation results shows that the proposed wirelesscommunication mechanism not only can be running withinfour child robots but also within three robots. Whenever somecommunication failures occurrence on one of the robots, thecommunication mechanism automatically can skip it over andcan built a new communication cycle to avoid the situation ofwaiting around or collapse of the mechanism.

B. Multi-robot communication experiment

We rebuilt the mechanical structure of RPRS child robotand made use of the reconstructed robot to do a multi-robotcommunication experiment.

Fig. 10 Initial state of the communication experiment

Fig. 11 Multi-robot communication experimentThree child robots are used to perform an action sequence

to verify the feasibility of the mechanism. The robots are asshown in Fig. 10 in the initial states. They are placed as a rowin equal space between each other. They are numbered as RI,R2 and R3 from left to right respectively.

At the beginning of the experiment, RI firstly acquire theaccess of the communication channel, then RI send its firstmessage to tell the other two robots that it will perform anaction to adjust its own joint angles so as to make its end link

53

point to left, as shown in Fig.11(a). When RI finish thisaction, it will inform the other robots. Then R2 starts toperform the same action, RI and R3 are waiting. After R2finish its action as shown in Fig. 11(b), R3 will start the sameaction and RI and R2 are waiting.

When R3 finish its action, first half of the sequence actioncycle is fulfilled, as shown in Fig. 11(c). And then, RI start toperform a reversed action to resume its initial state, as shownin Fig. 11(d). Afterwards, R2 and R3 resume their statesaccording to the same order as the first half cycle, as shown inFig.1 1(e) and (f). Till this time, a whole cycle is finished.Fig. 11(g) to (1) show the next cycle of such an actionsequence.

In the communication experiment as shown in Fig. 11, themulti-robot communication mechanism described in Fig.6 andFig.7 is adopted, the baud rate is 9600bps, andTtrans=5ms,Tmax=600ms. However, a child robot cannotcomplete such an adjustment of joint angle within 600milliseconds. Therefore, a robot cannot occupy the channel atall times when it is performing the actions. So that, there aretwo cycles, one is the action sequence cycle, and the other isthe communication mechanism cycle. The two cycles areasynchronous. In the experiment, the communicationmechanism was running in the three robots successively whenthey perform the action sequence.

The experiment shows that the communication mechanismis feasible and practicable.

[3] E. Klavins, "Communication Complexity of Multi-Robot Systems",Proceedings of the 5th International Workshop on AlgorithmicFoundations of Robotics. Berlin, Germany: Springer-Verlag, 2002. 275-291.

[4] A. Kawakami, A. Torii, K. Motomura and S. Hirose, "SMC Rover:Planetary Rover with Transformable Wheels". SICE 2002 proceedings ofthe 41st SICE Annual Conference, Osaka, Japan, 5-7 August, 2002. vol. 1,pp.157-162.

[5] R. Damoto, A. Kawakami and S. Hirose, "Study of super-mechanocolony: concept and basic experimental set-up", Advanced Robotics, vol.15, no. 4, pp. 391-408, April 2001.

[6] Z. Zhang, S. G. Ma, B. Li, L.P. Zhang and B.G. Gang, "OpenGL BasedExperimental Platform for Simulation of Reconfigurable Planetary RobotSystem", Journal of System Simulation, vol.17, no.4, pp. 885-888, April2001.

[7] M. H. Wang, S. G. Ma and B. Li, "Design and Implementation of aReconfigurable Planetary Robot Control System", Robot, vol.27, no.3, pp.273-277, March 2001.

[8] L.P. Zhang, S. G. Ma, B. Li, Z. Zhang and B.G. Gang, "KinematicModeling and Trajectory Planning for Reconfigurable Planetary RobotSystem", Journal of Xi'an Jiaotong University, vol.39, nol, pp. 87-91,January 2005.

[9] H. Wang, G. Z. Yan and Z. J. Gao, "Communication of Muti-robotSystem Based on TCP/IP", Computer Automated Measurement &Control, vol.11, no3, pp. 205-206, March 2003.

[10]F. Sebastian. "Describing Multidimensional Schedules for Media-AccessControl in Time-Triggered Communication". In Proc. of the 10th IEEESymposium on Computers and Communications (ISCC'05), 2005.

[1 I]G.Y. Nie, "Point-to-Multipoint Multitasking Wireless Communication",Microcontroller & Embedded System, no3, pp. 24-26, March 2003.

VI.CONCLUSION AND FUTURE WORK

The communication mechanism of RPRS multi-robotsystem is an important factor for the system performance.Because of the complexity of RPRS control system and thehardware limitations of child robots, the multiple robots haveto communicate each other by a single wireless channel. Inthis paper, a novel wireless communication mechanism isproposed to solve the problem. The communicationmechanism and protocol are introduced in detailed. Both acomputer based digital simulation and a physicalexperimentation validates the feasibility and practicability ofthe mechanism. The mechanism can also be used on the TITmodel.

In the future, the communication mechanism proposed inthis paper should be combined into the coordination strategyto build a robust control architecture of RPRS.

ACKNOWLEDGMENT

The authors would like to thank Prof. Shigeo Hirose, DrAtsushi Kawakami, and Mr. Kazuhiro Motomura in TokyoInstitute of Technology for cooperating in this research.

REFERENCES[1] T. Balch and R. C Arkin, "Communication in reactive multiagent robotic

systems", Autonomous Robots, vol. 1, no. 1, pp. 27-52, January 2001.[2] M. Wooldridge, "An Introduction to MultiAgent Systems", Publishing

House ofElectronics Industry, China. 2003.118-132.

54