Mobile Sensor Deployment and Coverage Using Multi-Agent-based Collective Formation Schemes

Transcription

1 Internatonal Journal of Performablty Engneerng, Vol. 8, No. 2, March, 2012, pp RAMS Consultants Prnted n Inda Moble Sensor Deployment and Coverage Usng Mult-Agent-based Collectve Formaton Schemes 1. Introducton ZHAO CHENG 1, LI BAI 1 *, DHRITIMAN CHOUDHURY 1, SAROJ BISWAS 1 and JIE WU 2 1 Department of Electrcal and Computer Engneerng, Temple Unversty, PA Department of Computer Scence, Temple Unversty, PA 19027, USA (Receved on October 30, 2010; Revsed on Aprl 18, 2011) Abstract: In ths paper, we present a novel moble wreless sensor deployment and network coverage technque usng mult-agent-based collectve formaton control. Most of the exstng approaches of sensor deployment are focused on centralzed methods, whch restran the sensor nodes to mantan fxed dstance among all neghborng nodes. Therefore, these approaches have some drawbacks, vulnerabltes and nflexblty, especally when some of the sensor nodes are not functonng due to unexpected node falure, e.g., power loss. As a result, sensor coverage could be compromsed or dmnshed. To address these problems, we propose to ncorporate an attractve/repulsve (AR) collectve formaton model to control the dynamcs of wreless sensor nodes whch are consdered as autonomous agents. We show that sensor deployment wth AR model provdes robustness and flexblty. When some sensor nodes are lost unexpectedly, movement of neghborng nodes are relatvely localzed so that the least amount of energy wll be used to regan control of the network coverage. Consequently, the proposed method can sgnfcantly mprove the tme-effcency, network stablty and sensng coverage when sensor nodes are deployed to explore harsh terrans and unpredctable envronments. Keywords: Moble sensor network, collectve formaton, sensng coverage, mult-agent system Sensor management n moble sensor networks has attracted a lot of attenton and research nterest n recent years. The self organzng characterstcs of collectve moton n nature, such as a flock of brds, a school of fsh, or a swarm of locusts, are resourceful stmul to the study of sensor management. Collectve moton has been studed n [6], [12], [16], [18], [19] n whch agents (or self-propelled partcles) move agreeng upon certan quanttes of nterest, such as poston, temperature, and voltage etc. These quanttes of nterest have mplcatons on general desgn of moble sensor networks, sensor network data fuson, atttude algnment of satellte clusters, congeston control of communcaton networks and mult-agent formaton control [1], [2], [9], [11]. A flockng model of multagent formaton was proposed by Reynolds [15] where three heurstc rules were assumed: () separaton: steer to avod crowdng and collson, () algnment: steer towards an average headng, and () coheson: steer to move towards the average poston. These rules have been proven effectve and are often used n the desgn of bogroup dynamc models (see [5], [13], [14]). In 2004, Olfat-Saber [13] developed a theoretcal framework for flockng dynamcs. In ths work, a mass model s proposed based on the Newton s law of moton whch * Correspondng author s emal: lba@temple.edu 141

2 142 Z. Cheng, L. Ba, D. Choudhury, S. Bswas and J. Wu shows that the movng agents eventually gather and form a lattce. The case of flockng n free-space wth multple obstacles s also consdered. Smulaton results showed that the formed lattce had good performance n avodng complex obstacles. Stablty of formaton was also proved usng Lyapunov s approach based on collectve potental of the agents. Vcsek [17] dscussed the unversal patterns of formaton of lvng organsms as well as non-lvng objects, such as nteractng robots. He focused on one of the most common and spectacular manfestaton of coordnated actons, whch descrbe the essental aspects of collectve moton of a wde selecton of systems rangng from colones of tssue cells to flocks of brds to collectvely movng robots. It has the potental of mprovng the nterpretaton of collectve behavor n both lvng and nert systems to understandng the nterrelaton of the systems by learnng smlar phenomena n the two domans of nature. Recently, Juanco [8] proposed a modfed knematc model wth AR par-wse nteracton, and showed an nterestng smulaton that shows that agents form several stellate patterns due to changes of dstrbuton of preferred par-wse length. He also gave the defnton and analyss of order parameter as a measure of pattern meta-stablty. Although the varaton among movng agents s a fact rooted n natural and artfcal swarm systems, the role of dversty n the self-organzed pattern formaton has not been prevously explored. In 2009, Chen and Cheng [3] ntroduced the AR functonal lnk between each par of agents. By changng the slope of an AR functon, they found a dramatc transton between two dfferent formaton patterns. In the lqud-lke pattern, the outer agents are sparsely dstrbuted whle the nner ones are packed densely. In contrast, agents mantan a constant dstance from each other n a crystal-lke pattern. A sensor network s a networked system that s composed of a large number of sensor nodes [1]. Moble sensor networks are sensor networks n whch sensor nodes have the capablty of moton under ther own control. In 2004, Ogren [11] presented a stable control strategy for sensor nodes to move and reconfgure cooperatvely n response to a sensed and dstrbuted envronment. Gradent clmbng strateges are appled to artfcal potentals to drve the sensor nodes. Nguyen [10] ntroduced the concept of an artfcal potental feld to gude the movement of a robot. In hs research, the robot was bult wth artfcal potental felds whch was able to navgate tself to a partcular locaton through a path wth obstacles. Later, the artfcal-potental fled approach was appled n the management of a moble sensor network to mprove sensor performance. Heo [7] proposed and analyzed dstrbuted energy-effcent deployment algorthms for moble sensor networks. In 2008, Ma [9] ntroduced non-newton s model and dscussed sensor coverage problems n sensor management of moble sensor networks. Ths research s based on smulatons and analyss n three cases: spatal coverage, spatal mgraton and spatal retreats. In ths paper, we propose the use of AR dynamcs for moble wreless sensor deployment to maxmze sensor coverage. We manly used smulaton to show performance and robustness of dstrbuted sensor deployment schemes. We demonstrated that the crystal-lke deployment scheme s sutable for explorng open areas wthout many obstacles, however lqud-lke deployment scheme performs better n a narrow and short pathway or envronment wth many obstacles. We show that even f some sensor nodes are lost unexpectedly, movements of neghborng sensor nodes are relatvely localzed so that the least amount of energy wll be used to regan control of the network coverage. Our proposed method can sgnfcantly mprove the tme-effcency, network stablty and sensng coverage when deployng sensor nodes to explore harsh terrans and unpredctable envronments.

3 Moble Sensor Deployment and Coverage Usng Mult-Agent-based Collectve Formaton Schemes 143 The rest of the paper s organzed as follows: Secton 2 detals some background research, Secton 3 provdes some smulaton results and dscussons, and Secton 4 presents the concluson and some deas for future work. 2. Models of Collectve Moton There are three popular types of models descrbng relatonshp between sensors nodes n collectve groups: ) the Unt-vector Model, ) the Knematc Model and ) the Mass Model. The Unt-vector Model s a steerng control model, n whch the drecton of each agent s determned by a control algorthm. However, n the thrd model, the Mass Model, agents obey the Newton s law of moton: both drecton and speed of each agent s updated at each step. Smlarly the second model, named as Knematc Model, s also a system whose agents have both ther drecton and scalar veloctes updated accordng to the control algorthm. But, they are stll dfferent, snce the control nput s appled drectly on the velocty of each agent n Knematc Model nstead of acceleraton (dervatve of velocty) n the Mass Model. Thus, the Knematc Model can also be vewed as an approxmaton of Mass Model. Note that the Knematc Model s a frst-order system whle the Unt-vector Model and the Mass Model are second-order systems. Generally, the second-order systems have better stablty performance than frst-order systems. The Mass Model s also the closest one to movement mechansm of a real object over all three models. Because of these reasons, the Mass Model s adopted and dscussed n the followng sectons. Consder a group of n movng agents (or partcles) n a square cell wth perodc boundary condton wth equatons of moton. It s common to study the communcaton problems among agents by means of a graph. A graph G s defned as a par (V,E) consstng of a set of vertces V = {1,2N} and a set of edges E {(, j) :, j V, j}. The movng agents are the vertces of the graph G and the connecton between a par of m agents s an edge defned n G. Let x, v, u R denote the poston, velocty and control V m nput of node for all, respectvely n the Eucldean space, R. Let r denote the nteracton range between two agents, then the set of spatal neghbors of agent s defned by N j V : x x r (1) where s the Eucldean norm n m R. Defne j N as the total number of elements n the set N. Fg. 1 shows the parwse nteracton of the agent 1. The agents 2 and 3 are neghborng agents of agent 1 snce ther dstance to agent 1 s less than r. The parwse force functon between two sensor nodes s represented as whch x j between agents and j. It reaches the global x = d (see Fg. 2) whch s the equlbrum pont, and r s the scope of an vares accordng to the nterval mnmum 0 at agent. The rato of the scope and the equlbrum pont of an agent s denoted by k r / d.

4 144 Z. Cheng, L. Ba, D. Choudhury, S. Bswas and J. Wu Fgure 1: Parwse Interacton of an Agent Fgure 2:Parwse Force (x) The equaton of moton s defned as, where x = v v = u V and the control nput of agent consstng of three terms (2) where u = w 1 j j jv gradentterm w ( x x n j s a unt vector pontng from ) n w ( v 2 j jn consensusterm k ( x x ) k ( v v )) 3 x r v r navgatonterm x to x j, n x v ) j j =, V x j x x (3), w 1, w 2 and w 3 are weghts for gradent term, consensus term and navgaton term, respectvely. The gradent term s the Attractve/Repulsve force determned by the dstance between two partcles, whch gudes two partcles to move to a desred dstance at x=d and avod collson (due to repulsve force). The consensus term ensures the velocty consensus wth other partcles, whle the navgaton term s actually proportonal feedback control based on desred dstance and/or velocty. Note that the navgaton term s not necessary f there s no poston or velocty goal assgned to the sensors/partcles. In ths paper, we consder the control nput dependng on the gradent and the consensus terms only,.e., w 1, w 2 1 and w Note that the velocty consensus term s only requred for smulatons presented n Fg For other smulatons dscussed n ths paper, ths term s not requred. Snce the consensus term takes effect only when partcles are movng wth unsynchronzed speed, t wll be zero when partcles are stable. To defne the functon, we consder two pecewse functons ) an exponental functon (attractve force) and ) a second-order polynomal functon (repulsve force) as ( x) = b c 2 Ax Bx a, ( x d) x d exp c 2, x [0, d), x [ d, ). (4) 2 The second-order polynomal Ax Bx a must satsfy the followng condtons to ensure contnuty and dfferentablty of (x) at x = d :

5 Moble Sensor Deployment and Coverage Usng Mult-Agent-based Collectve Formaton Schemes 145 ( x ) = ( x) xd x = d ( x ) = ( x) xd x = d The parameters A and B can be easly determned by three other parameters a, b, and c, through ensurng contnuty and dfferentablty at x = d,.e., a b a b A d 2 B 2 cd d c Parameters a and b are related to the sze of smulaton and determne the maxmum value of repulson and attracton force, respectvely. To make a square area wth edge value of 200, we let a=5 and b=0.2. We have two free parameters c and d. The parameter d s an equlbrum pont of two agents and determnes whether two agents attract or repel. The parameter c determnes r (cutoff dstance),.e., the dstance wthn whch neghborng agents nteract. To ensure stablty, we need to fnd the approprate range for parameter c. A good approxmaton (see [4]) of collectve behavor to ensure stablty requrements s obtaned when we have a = 5, b = 0.2 and 5 c 17. Also, Cheng [4] found that two dstnct patterns are determned by the value of k 1) 1 < k < 2: agents form collectve behavor as a crystal-lke pattern, and 2) k 2: agents form collectve behavor as a lqud-lke pattern. 3. Smulatons of Sensor Management Collectve behavor has many applcatons. Here, we wll demonstrate ts use n moble wreless sensor network deployment by tunng the parameter k for ) lqud-lke deployment scheme (k > 2) and ) crystal-lke deployment scheme (1 < k < 2). Through smulaton, we wll make a detaled analyss for spatal coverage, obstacle avodance and fault-tolerant of two deployment schemes. We manly focus on the comparson between a lqud-lke deployment scheme and a crystal-lke deployment scheme. Three case studes are presented based on smulatons n a 2D square cell wth elastc boundares, 1) area covered analyss, 2) fault tolerance analyss, and 3) obstacle avodance analyss. In the followng smulatons, N partcles are generated wth random velocty and poston. The agents are synchronzed and move autonomously accordng to equaton (3) at each smulaton step t. 3.1 Area Covered Analyss Frst, we look at the sensor coverage of our proposed method. Let N agents deploy n a small boundary area. Intally, we expect that agents wll expand throughout the whole area (due to repulsve force) and fnally fll the entre square area (see Fg. 3). In ths fgure, each dot represents a partcle and the edge connectng two partcles s the connecton pared between two neghborng partcles. Sub-fgures are presented to show the progresson of sensor node expanson at a) t = 0, b) t = 50, c) t = 250 and d) t = 1,250. In ths smulaton, we can see that agents are ntalzed n a small square area but expand and cover the whole area quckly (roughly at t = 250). Ths expanson s manly domnated by the repulsve force defned n equaton (3) because agents are ntalzed n a small contaned area. For a large sensor coverage area, we can see how agents can quckly reorganze and change sensor coverage destnatons as shown n Fg. 4. They represent a crystal-lke deployment scheme and a lqud-lke deployment scheme, respectvely. The objectve of ths smulaton s to show robustness of the proposed sensor deployment scheme. The smulaton s set up n such a way that a group of nodes are drected to a specfc drecton

6 146 Z. Cheng, L. Ba, D. Choudhury, S. Bswas and J. Wu and s requested to change the sensor coverage area. The trajectory of the group shows a stable and rgd sensor coverage area. However, we can see a sgnfcant dfference at the vertex of drecton changes for both deployment schemes. The crystal-lke scheme requres a larger area to redrect the deployment drecton. In contrast, the lqud lke scheme s not affected by the change of drecton. It suggests that the lqud lke deployment scheme mght be more robust for sensor coverage problems that requre many drecton changes. (a) t = 0 (b) t = 200 (c) t = 500 (d) t = 4000 Fgure 3: Moble Sensors Span and fll a Square Area. (a) A crystal-lke deployment scheme (b) A lqud-lke deployment scheme Fgure 4: Moble Sensors Movement n Dfferent Deployment Schemes Also, agents are compact n a lqud-lke deployment scheme whereas agents have flexblty n crystal-lke deployment scheme. For ths reason, agent paths n the crystallke deployment scheme are wder due to repulsve behavor among agents. Velocty s another factor that plays a vtal role n drecton changes. The combnaton of velocty and repulsve force affects the agents movements n crystal-lke deployment. Consequently, we see that agents can move more robustly n the lqud lke deployment scheme.

7 Moble Sensor Deployment and Coverage Usng Mult-Agent-based Collectve Formaton Schemes Fault Tolerance Analyss For ths study, we evaluate how network topology recovers from random sensor node falures. It gves us the ablty to understand how fault tolerant two deployment schemes are. In ths smulaton, some sensor nodes were removed randomly at t = 2,000, and we observe that the remanng sensors regroup quckly and regan control of the whole sensor coverage area. Total smulaton tme s 5,000 steps. Fgure 5 llustrates how the network topology recovers n the crystal-lke deployment scheme. At t = 2,000, some sensor nodes are randomly removed. Sensor nodes regrouped and reconstructed the formaton, however some nodes are stll not connected at t = 5,000. (a) t = 1980 (b) t = 2000 (c) t = 5000 Fgure 5: Fault Tolerance of Moble Sensor Network n Crystal-lke Deployment Scheme Fgure 6 shows that the network topology recovers n the lqud-lke deployment scheme. At t = 2,000, some sensor nodes are randomly removed. We observe that the remanng sensor nodes regrouped and reconstructed the formaton. At t = 5,000, the remanng sensors are fully connected wth neghborng nodes. (a) t = 1,980 (b) t = 2,000 (c) t = 5,000 Fgure 6: Fault Tolerance of Moble Sensor Network n Lqud-lke Deployment Scheme From these two smulatons, we see the lqud-lke deployment scheme has better recoverablty because sensor nodes are reconnected quckly. In comparson, the crystallke deployment scheme takes much longer tme to reconstruct the formaton and to regan network coverage. The man reason s that the lqud lke deployment scheme has the longer dstance r between two adjacent nodes n the outskrts of the formaton where they nteract. Also, the lqud-lke schemes allow any formaton shape. Even when some nodes are removed abruptly, the blank spaces are quckly flled up by the rest of the nodes. Repulsve force plays another vtal role for stablzaton of the formaton. Nodes n the crystal-lke deployment scheme are too close to each other, and are not aware of nodes beng mssng. In order to fll up the blank space, partal group nodes cannot complete the recovery process unless all nodes know that they need to fll the empty coverage gap. Hence, t wll take longer tme to complete the whole system to regan control of network coverage.

8 148 Z. Cheng, L. Ba, D. Choudhury, S. Bswas and J. Wu 3.3 Obstacle Avodance Analyss Ths subsecton presents the collectve behavor of sensor nodes when an area of coverage has obstacles. We can see that nodes behave dfferently among two sensor deployment schemes. Fgure 7 represents how effcent and effectve sensor nodes deploy n the whole coverage area wth an obstacle. For our smulaton, a 2D elastc plane of 8x8 unts s chosen. The wall of the square s consdered to be an elastc boundary. In other words, sensor nodes wll experence a backward force when nodes encounter the wall. The square plane s partally splt by a wall (or obstacle) of length 5x1 unts whch provdes a narrow pathway. The objectve of the sensor group s to cover up the whole space by avodng the obstacle. Each dot n the smulaton represents a sensor node, and the crcle surroundng each dot represents the nteracton range of the node agent. Intally, agents are concentrated at the top rght corner of the plane. Due to the velocty of agents and repulsve behavor among the agents, they wll start coverng the square plane avodng the obstacle. The results are depcted n Fg. 7, where a) t = 0, agents are ntalzed and randomly dstrbuted n the top rght corner; b) t=200, agents start to expand due to the velocty and repulsve forces among neghborng node agents; c) t = 500, agents start to cover up more than 50% of the area; d) t = 2,000, agents cover more than 75% of the area; e) t = 5,000, agents cover most of the area; and f) t = 8,000, agents cover almost the whole area and nodes stop movng. Every node connects wth at least another agent. We have also tested other scenaros of obstacles and coverage area by changng the x- axs length from -2 to 2 unts y-axs blockng length from 0-7 unts wth ncrements of 0.5 unt. The complete coverage tme vares slghtly due to dfferent settng of the obstacle; however the smulatons runs demonstrate that moble sensors wll eventually cover the whole area. In ths paper, we manly focused on two deployment schemes. Formaton patterns of nodes depend on the parameter c. For ths example, we can tune the agent group as the crystal-lke deployment scheme by settng c = 0.06, and as lqud-lke deployment scheme by settng c = Parameter d also plays the key roles n deployment scheme. We can use parameter d to determne the nterval among the node agents. Parameter d s set as 1 and 2 for crystal-lke and lqud-lke deployment schemes, respectvely. Fgure 8 shows how a sensor group can cover the area wth an obstacle quckly by turnng parameter c from crystal-lke deployment (c = 0.06) schemes to lqud-lke deployment schemes (c = 0.2). Parameter d can also vary wth respect to parameter c. When parameter c = 0.06, the smulaton tme to cover up the area wth an obstacle s long. From the smulaton, we can conclude that we need a longer tme to cover an area wth an obstacle f we use a crystal lke deployment scheme. As we change the parameter c close to 0.20, the tme to cover the area becomes faster and wth smaller fluctuatons. Ths sgnfes that, we can deploy moble sensor nodes n lqud-lke deployment schemes to gan quck coverage, even wth an obstacle. 4. Concluson and Future Work In ths paper, we developed a moble wreless sensor deployment technque usng a multagent-based AR collectve formaton control approach. The proposed approach and analyss embodes several tunable parameters for dfferent deployment schemes n order to ncorporate the formaton of the model to control the dynamcs of the sensor nodes, whch consdered as multple autonomous agents. In our future work, we can consder problems such as coverage areas wth more obstacles. Also, we can develop more performance-based metrcs to analyze two dfferent deployment schemes.

9 Moble Sensor Deployment and Coverage Usng Mult-Agent-based Collectve Formaton Schemes 149 (a) t = 0 (b) t = 200 (c) t = 500 (d) t = 2000 (e) t = 5000 (f) t = 8000 Fgure 7: Moble Sensor Nodes Fll n Partally Blocked Area. Fgure 8: Complete Overage Tme for Moble Wreless Sensor Nodes References [1] Akyldz. I. F. and H. Rudn. Computer networks, the journal, and computer networks, the technology. Computer Networks-the Internatonal Journal of Computer and Telecommuncatons Networkng 2002; 40: [2] Bertsekas. D. P. and J. N. Tstskls. Comments on coordnaton of groups of moble autonomous agents usng nearest neghbor rules. IEEE Transactons on Automatc Control 2007; 52: [3] Chen. M. Z. Q., Z. Cheng, H.-T. Zhang, T. Zhou, and I. Postlethwate. Collectve aggregaton pattern dynamcs control va attractve/repulsve functon. n Complex Scences, ser. Lecture Notes of the Insttute for Computer Scences, Socal Informatcs and Telecommuncatons Engneerng, J. Zhou, Ed. Sprnger-Verlag, 2009; 5: [4] Cheng. Z., Collectve behavor and sensor network a mult-agent dynamc system approach. Master s thess, Temple Unversty, August [5] Gaz. V. and K. M. Passno. Stablty analyss of swarms. IEEE Transactons on Automatc Control 2003; 48:

10 150 Z. Cheng, L. Ba, D. Choudhury, S. Bswas and J. Wu [6] Gregore. G. and H. Chate, Onset of collectve and cohesve moton. Phys Rev Letter, 2004; 92: [7] Heo N. and P. K. Varshney. Energy-effcent deployment of ntellgent moble sensor networks. IEEE Transactons on Systems Man and Cybernetcs Part A-Systems and Humans 2005; 35: [8] Juanco. D. E. O. Self-organzed pattern formaton n a dverse attractve-repulsve swarm. EPL 2009; 86 (4). [9] Ma. K., Y. Y. Zhang, and W. Trappe. Managng the moblty of a moble sensor network usng network dynamcs. IEEE Trans. on Parallel & Dstrbuted Systems 2008; 19: [10] Nguyen. B., Y.-L. Chuang, D. Tung, C. Hseh, Z. Jn, L. Sh, D. Marthaler, A. Bertozz, and R. Murray. Vrtual attractve-repulsve potentals for cooperatve control of second order dynamc vehcles on the caltech mvwt. Proc Am. Control Conf 2005; pp [11] Ogren. P., E. Forell, and N. E. Leonard. Cooperatve control of moble sensor networks: Adaptve gradent clmbng n a dstrbuted envronment. IEEE Transactons on Automatc Control 2004;. 49: [12] Olfat-Saber. R. Flockng for mult-agent dynamc systems: Algorthms and theory. IEEE Transactons on Automatc Control 2006; 51: [13] Olfat-Saber. R. and R. M. Murray. Consensus problems n networks of agents wth swtchng topology and tme-delays. IEEE Trans. on Automatc Control 2004; 49: [14] Ren. W. and R. W. Beard. Consensus seekng n multagent systems under dynamcally changng nteracton topologes. IEEE Trans. on Automatc Control 2005; 50: [15] Reynolds. C. W. Flocks, herds, and schools: A dstrbuted behavoral model. Computer Graphcs 1987; 21: [16] Vcsek. T., A. Czrok, E. Ben-Jacob, I. I. Cohen, and O. Shochet. Novel type of phase transton n a system of self-drven partcles. Phys Rev Letter 1995; 75: [17] Vcsek. T. Unversal patterns of collectve moton from mnmal models of flockng. Second IEEE Internatonal Conference on Self-Adaptve and Self-Organzng Systems, SASO 08, 2008; pp [18] H.-T. Zhang, M. Z. Q. Chen, and T. Zhou. Improve consensus va decentralzed predctve mechansms. EPL 2009; 86 (4). [19] Zhang. H.-T., M. Z. Chen, G.-B. Stan, T. Zhou, and J. M. Macejowsk. Collectve behavor coordnaton wth predctve mechansms. IEEE Crcut & Systems Magazne 2008; 3: Zhao Cheng receved hs Master's degree from Temple Unversty. He s currently a PhD canddate n Dartmouth College. Hs research nterest ncludes, cooperatve control for mult-agent systems, moble sensor networks, robotcs, medcal/acoustc magng. L Ba s an Assocate Professor at Temple Unversty. Hs research nterests are performablty n dstrbuted systems, such as relablty, securty and montorng, and mult-agent systems. Dhrtman Choudhury receved hs Master s degree from Temple Unversty n Electrcal and Computer Engneerng. Currently he s workng n Semens moblty under a project wth MTA. Hs area of nterest ncludes dstrbuted sensor networks, robotcs, telemedcne systems, mult-agent systems. Saroj Bswas s a Professor of Electrcal and Computer Engneerng at Temple Unversty, Phladelpha. He receved Ph.D. n Electrcal Engneerng from the Unversty of Ottawa, Canada, n Hs current research nterests focus on multagent systems, ntellgent control, optmal and nonlnear control, and power systems. Dr. Bswas s a member of IEEE and Sgma X. Je Wu s the char and a professor n the Department of Computer and Informaton Scences, Temple Unversty. Pror to jonng Temple Unversty, he was a program drector at Natonal Scence Foundaton. Hs research nterests nclude wreless networks and moble computng, routng protocols, fault-tolerant computng, and nterconnecton networks. Dr. Wu s a Fellow of the IEEE.