Autonomous Adaptive Resource Management in Sensor ... - CiteSeerX

Autonomous Adaptive Resource Management in Sensor Network Systems for Environmental Monitoring Ashit Talukder Jet Propulsion Laboratory, California Institute Of Technology 4800 Oak Grove Drive Pasadena, CA 91109 818-354-1000 [email protected]

Anand Panangadan CHLA/USC 4650 Sunset Blvd, MS 139 Los Angeles, CA 90027 323-361-2413 [email protected]

Abstract—The paper describes the use of model predictive control (MPC) as a framework for optimal resource management in environmental monitoring sensor networks. The MPC formulation adapts sensor and network parameters (such as sensor sampling rates, and routing of data) that impact the utilization of the system resources (such as energy reserves at off-shore in-situ sensors, and wireless bandwidth). The control parameters are optimized so as to maximize a measure of the total information extracted from the system. This information measure takes into account the spatio-temporal events of interests that are detected in the environment. The approach is illustrated on a coastal monitoring and forecast system that is in operation in the New York harbor and surrounding area. Offline results using actual modeled data from in-situ sensory measurements demonstrate how the sensor parameters can be adapted to maximize observability of a freshwater plume while ensuring that individual system components operate within their physical limitations. 1 2

processes take place at a uniform rate, then either the system will have to restrict the information gathering rate to a low level to guarantee an extended lifetime, or sacrifice lifetime to ensure high resolution sensing. Ideally, we would like the

t t+1 Extent of Spatial Event Figure 1 - Sensors embedded in the environment. The spatial extent of the event at two control steps is shown. The vertical bars represent the sampling rates at these control steps.

TABLE OF CONTENTS 1. INTRODUCTION ......................................................1 2. RELATED WORK ...................................................2 3. MODEL PREDICTIVE CONTROL ............................3 4. MPC FOR A COASTAL MONITORING SENSOR NETWORK ...................................................................................3 5. RESULTS ................................................................6 6. EXTENSIONS OF THE MPC MODEL ......................6 7. CONCLUSIONS .......................................................6 REFERENCES .............................................................7 BIOGRAPHIES ............................................................8

system to be designed such that there is an intelligent tradeoff between the rate of energy utilization and the information output of the sensor network. The rate of resource utilization at a sensor node is dependent on certain controllable parameters. For instance, the sampling rate affects energy consumption due to the demands of operating the sensor (especially those sensors that have mechanical components) and the need to transmit the resulting data. If a critical event is detected, the sensors are expected to provide information at a high resolution (high sampling rate) while at other times the sensors can operate at a much lower rate, thus conserving energy for critical periods. Thus, if we explicitly control these parameters in response to the changing event criticality we can significantly extend the lifetime of the sensor network. In this work, we adapt these controllable parameters to the internal resources of the sensor system and the application requirements in a mathematical rigorous way.

1. INTRODUCTION Wireless sensor networks have been deployed for real-time monitoring in a variety of physical environments [1-3]. However, the limited resources available at a sensor node severely limit the lifetime and utility of such a system. The small size of a typical node restricts the battery size. On the other hand, wireless communication is an energy intensive process. This implies that if the communication and sensing 1 1 2

Thomas Herrington, Alan Blumberg, Nickitas Georgias Stevens Institute of Technology, Hoboken, NJ

1-4244-1488-1/08/$25.00 ©2008 IEEE. IEEEAC paper #1348, Final, Version 1, Updated Jan 14, 2008

1 Authorized licensed use limited to: University of Southern California. Downloaded on December 4, 2009 at 22:17 from IEEE Xplore. Restrictions apply.

We introduce a general mathematical control framework which will enable the resources of a sensor network to be regulated to guarantee a certain system lifetime while continuing to capture and measure information according to the needs of the application. As on-board computation in general uses significantly lower energy resources compared to wireless communication, it is beneficial to implement a sophisticated controller if it brings about significant savings in the amount of energy expended. A model-based controller is the most appropriate mechanism for this situation since the internal resources of the sensor system can be parameterized in advance and the expected changes in the environment are usually known for a given application. Model Predictive Control (MPC) is an established technique for controlling complex continuous systems. MPC assumes that a model that describes how the system state responds to control inputs is available. At each control iteration, the values of the controlled inputs are obtained by solving an optimization problem that utilizes this state model. Limits on the range of the control, and other domain-specific requirements can be specified naturally as equality and inequality constraints in the optimization step. This flexibility in problem specification and the ability to derive optimal control are some of the chief advantages of this control technique. Successful applications of MPC include chemical process control and resource management in the battlefield [4-7].

scheduling sleep/wake cycles for the sensors [11, 12]. Another method is to enable sensors to adapt their radio range to maintain coverage while reducing the energy spent in transmissions [13, 14]. In these approaches, it is assumed that multiple sensors can cover the same area. In our target applications, all the sensors are expected to be operational and it is their relative sampling rates that are adapted. Rakhmatov and Vrudhula present a mathematical model that specifically describes battery lifetime and this model is used in designing energy efficient protocols [15]. Bandyopadhyay, Tian et al. study the effect of implementing energy conserving transmission schemes on the maximum spatial and temporal sampling rates possible from a sensor network [16]. A resource allocation scheme that is based on combinatorial auctions is described in [17]. This approach is similar to ours in that a centralized scheme is used, but the focus of that work is not on energy conservation. The energy expended in a multi-hop network can be reduced by implementing energy efficient routing schemes [18-21]. In our paper, we use a simple model for the energy required to transfer data over a wireless link. Adaptive sampling refers to the explicit control of sensor sampling rates and locations in order to conserve system resources or to respond to changing environmental conditions. This approach has been applied to minimize energy usage in a sensor network [22-27]. These approaches define a stochastic model of the environment and minimize the resources used to sample from this event model. These works model the environment but not the dynamics of the sensing system itself. Thus, though methods exist to explicitly control resource utilization in specific sensor networks, there is no general framework for optimal resource management that integrates both the applicationspecific information demands from the environment and the physical resource constraints imposed by the network.

We illustrate the application of MPC to sensor networks using an existing coastal monitoring network (the New York Harbor Observing and Predicting System, NYHOPS). Here, a “critical event” corresponds to a time varying spatial phenomena such as an expanding freshwater plume after heavy rainfall. The control parameters that will be adapted are the sampling rates of each sensor and the routes taken by data during wireless communication.

Optimization techniques developed specifically for use in sensor network applications have largely been restricted to discrete constraint satisfaction and optimization problems [28-30]. (In a constraint satisfaction problem, a set of variables is to be assigned values from a finite space such that the assignment does not violate any of a set of constraints specified as the allowable values of subsets of variables). However, formulations based on optimization of continuous functions (such as MPC presented in this paper) are more suited to the control of continuous parameters such as sampling rate at a sensor node. In our approach, we assume that sensors can be directly controlled by a central controller. However, distributed versions of MPC [31-33] have been studied if our approach has to be extended to a completely distributed scenario.

We first describe related work on sensor network control mechanism. We then introduce the standard model predictive controller framework. We next describe how this framework can be applied to the NYHOPS coastal monitoring network. Offline resource management results of the MPC on sensor web parameters in response to actual coastal plume events captured by the NYHOPS sensor web system are then presented.

2. RELATED WORK Resource conservation to prolong system lifetime is studied by the wireless sensor network community as this is critical to ensure that wireless embedded systems remain in operation long enough to be useful in real-world applications. Energy minimization has been incorporated into the deployment and coverage of sensor networks [810]. One way to conserve energy while still ensuring that the area to be monitored is sufficiently covered is by 2

Authorized licensed use limited to: University of Southern California. Downloaded on December 4, 2009 at 22:17 from IEEE Xplore. Restrictions apply.

Subject to:

3. MODEL PREDICTIVE CONTROL

Resource constraints: None of the system’s resources run out before a pre-defined minimum lifetime

Model Predictive Control (MPC) uses a process model that describes the dynamics of the open loop system. The goal of the controller is to determine the set of control inputs that when applied to the process model will cause the controlled outputs to most closely match a given set of desired outputs in the future. Let X (t ) denote the state of the system model

Physical constraints:

the sensors stay within their

at time t , and U (t ) and Y (t ) denote the corresponding inputs and outputs. Traditionally, the state model is assumed to be linear because of the computational efficiency associated with the solution of linear systems:

X (t + 1) = AX (t ) + BU (t ) Y (t ) = CX (t ) + DU (t ) where A, B, C , D are constant matrices representing the coefficients of the linear model. The sequence of future control signals is computed to minimize a cost function F

min F ( X (t ), Y (t ) ) U (t )

where Y (t ) = {Y (t + 1| t ),

, Y (t + T | t )}

such that

G ( X (t + j | t ), U (t + j | t ), Z (t + j | t ) ) ≤ 0

H ( X (t + j | t ), U (t + j | t ), Z (t + j | t ) ) = 0 ∀j : j = 0,1,… , T − 1 where G and H are linear functions denoting the inequality and equality constraints respectively. The constraints of many controllable systems are also affected by external factors which we represent by Z (t ) . The predicted future outputs

Y (t ) = {Y (t + 1| t ),

, Y (t + T | t )}

for

Figure 2 - Location of sensors (red dots) and dataloggers (green dots) in NYHOPS simulation.

the

operational bounds

prediction horizon T are calculated at each instant t using the process model. This optimization problem is solved using a (computationally expensive) quadratic programming solver in standard MPC formulations. Only the current control signal U (t | t ) is applied to the process (receding

The details of the MPC for spatiotemporal sensor webs is provided next, where observed events evolve over space and time.

horizon). At the next sampling instant, Y (t + 1) is measured and the procedure is repeated to calculate U (t + 1| t + 1) .

4. MPC FOR A COASTAL MONITORING SENSOR NETWORK

The problem of managing system resources in a sensor network can be framed as a MPC problem. The general formulation takes the form of the following optimization problem:

The New York Harbor Prediction and Observation System (NYHOPS) comprises of a network of sensors to monitor coastal and ocean parameters in the densely populated regions of the Hudson-Raritan Estuary and the New Jersey Atlantic Ocean shoreline [34]. The modeled area is shown in Figure 2. The readings from the sensors are provided to a predictive model of the environment, the ECOMSED/POM

Determine the control inputs such that information extracted from the environment is proportional to event criticality 3


POM parameters, boundary conditions

Sensor data

ECOMSED Model

Sensor Network Resource Levels

Event Detection

Forecast

MPC Controller

Region of interest

Optimal control

Figure 3 – Integrating sensor network control into a coastal monitoring application. model [35]. The ECOMSED model describes the physical properties of the entire water mass in the NY/NJ harbor area using a set of differential equations. This model uses the sensor readings as input and outputs predictions of environmental parameters for the entire area. The model outputs can be used to detect regions of interest such as freshwater plumes. These regions can be studied in greater detail by increasing the resolution of sensors that are close to that area. Thus, the goal of a sensor network optimization protocol is to regulate the operation of the sensors so as to maximize the measurement accuracy at all points in areas of interest while ensuring that the physical limitations of the sensor network are not exceeded. The relationship between the optimal MPC-based control framework and the modules in the NYHOPS system is shown in Figure 3.

be optimized represents the sensing error and the energy required for data transfer. The constraints represent the physical limitations of the system such as bandwidth of the wireless networks. Let S represent the set of sensors and let ps , s ∈ S , represent the position of a sensor s . We assume that the errors in a sensor s ’s measurements, are random, zero-mean Gaussian with variance

σ s2 .

Let the

number of independent measurements that are averaged to generate one sample (the sensor sampling rate) be us . Then, the variance of error of the averaged samples is

σ s2

.

us We assume that the sensor’s measurements at location ps

are correlated with values that would be obtained at location p . This correlation decreases linearly with the distance

In addition, the sensors that are deployed along the coast or in the NY/NJ harbor have to transmit their measurements to a central data acquisition/control computer. We simulate the use of dataloggers as intermediate relay stations situated between the sensors and the central computer requires. A datalogger compresses data files, establishes a connection to the Internet via a local ISP, and pushes the data to the data acquisition server. The data transmission to the remote datalogger is through a line-of-sight serial radio modem system. A sensor can establish a 1200 baud, two-way simplex communication link with any of the remote dataloggers. The bandwidth across the network can be optimized by intelligently assigning sensors to dataloggers. If two high throughput sensors need to transmit simultaneously, they should be assigned to different dataloggers. In other instances, the assignment should be such that total energy cost of transmission is reduced. The actual transmission schedule (such as the transmission slots in a Time Division Multiple Access (TDMA) scheme) can be calculated analytically after the datalogger assignment is completed.

d ( p, ps ) between p and ps . The variance of error at location p when sensor s is used for measurements is

σ s2 us

+ d ( p, ps ) σ D2

proportionality that correlation decreases multiple sensors are model) to determine

σ D is

where

a

constant

of

quantifies the rate at which the with distance. If measurements from optimally fused (as in the temporal the value at a location p , then the

variance of error, σ p , is given by 2

1

σ

2 p

(u )

=∑ s∈S

1

σ

2 s

us

+ d ( ps , p )σ D2

where u = ( us ) , s ∈ S is the vector of sensor sampling rates. Let R denote the entire expanse of the region being modeled. As a measure of the overall accuracy of the sensor network for a particular set of sensing rates, we will use the mean variance of error over all points in R :

Mathematical Model We now describe how a MPC formulation can be used to regulate the operation of the sensors and their communications in the NYHOPS network. The function to 4


transmitting data over wireless links from the sensors to the dataloggers. We assume that the energy expenditure is proportional to the volume of data (the sampling rate) and the square of the distance between the transmitter and receiver (an assumption that is valid for radio transmission). Then the total energy expenditure is proportional to the function fW :

1 1 ∑ 2 | R | p∈R σ p ( u )

f ( u, R ) =

The extent of the region can change over time. For instance, the region may represent a freshwater plume that changes shape and location which this can be predicted by the ECOMSED/POM model. The MPC framework can use a predictive model of the system to optimize control not just for a single control step, but also for a finite number of control steps in the future. Denote by RM (t ) the extent of

M

fW ( u, a ) = ∑∑ us as ,m d 2 ( s, m ) m =1 s∈S

the region at time t as predicted by the spatial model, M . A new objective function can then be defined that sums the objective functions for every control step that has to be modeled.

where is the distance between sensor s and datalogger m . We integrate the components related to dynamic regions and constraints imposed by remote dataloggers. This leads to a multiobjective optimization problem where the two objectives are minimizing the sampling variance ( f M ) and

f T ( u ( t + 1) , u ( t + 2 ) ,… , u ( t + T ) ) = t +T

∑ f ( u ( t ') , R ( t ') )

the energy expended during wireless data transfer ( fW ).

M

We use the lexicographic approach to multiobjective optimization. The lexicographic approach optimizes the functions in decreasing order of importance. We assume for the NYHOPS system, it is more important to minimize the sensor sampling error than to minimize the energy expenditure. In this case, we can solve the optimization problem in the following two steps.

t '=t +1

Here, t denotes the current time and T the number of future control steps to be optimized. We now present the formulation that models the limitations of the data links between dataloggers and sensors. When this is incorporated into an optimization framework, then the data transfer schedule that maximizes throughout can be determined. Let bs (proportional to its sampling rate us )

( u1 , a1 ) = arg min ⎡⎣ f M ( u, RM )⎤⎦ u∈U |S|

denote the total data generated by sensor s . Let there be M dataloggers. Each sensor can establish a connection with exactly one datalogger. Denote by binary variable as ,m if sensor s is connected to the m-th datalogger, i.e.,

M

us > 1, ∀s ∈ S ; ∑ as ,m = 1,∀s ∈ S m =1

∑b a

s s ,m

as ,m = 1 if they are connected and as ,m = 0 if they are not. Let Dm denote the bandwidth of the data link to the m-th

RM denotes the extent of the region of interest as predicted by the spatial model, M . Note that this is a mixed integer

datalogger (for example, 1200 baud). Then, the network limitations can be expressed by the following constraints.

∑b a

s s ,m

programming problem. The multiobjective optimization is

≤ Dm , ∀m = 1, 2,… , M

step

in

the

u∈U |S |

M

s ,m

= 1,∀s ∈ S

M

∑a

m =1

m =1

as ,m ∈ {0,1}, ∀m = 1, 2,… , M , ∀s ∈ S

that it is physically impossible to establish a communication link between sensor s and datalogger m . The above constraints lead to an integer programming formulation as the as , m are integers. A given assignment of sensors to determines

the

energy

expended

s ,m

= 1,∀s ∈ S ; ∑ bs as ,m ≤ Dm , ∀m = 1, 2,… , M s∈S

The optimal control parameters are then obtained as u1 (sampling rates) and a 2 (assignment of sensors to

In practice, some of the as ,m will be preset to 0 indicating

dataloggers

second

a 2 = arg min ⎡⎣ fW ( u1 , a ) ⎤⎦

s∈S

∑a

≤ Dm , ∀m = 1, 2,… , M

s∈S

dataloggers).

while 5


5. RESULTS We now present results from executing the MPC on a simulation of the NYHOPS system. We restrict the area to the NY/NJ harbor area. There are 606 locations within this area which are included in the grid model of the ECOMSED model. We identified the location of 21 sensors from the NYHOPS website for use in our simulation (Figure 2). We model these sensors as having the same variance with respect to the monitored phenomena. We simulate a freshwater plume that moves out of the mouth of the Raritan River into the open ocean. This plume constitutes the region of interest that has to be monitored with high fidelity using the NYHOPS sensor network. In addition, we chose 5 points on the coastline to represent the locations of the dataloggers. Figure 4 shows the sampling rates and assignments of sensors to dataloggers at one specific time point as the plume expands. The points used to approximate the plume are shown by small red dots. As expected, the sampling rates of sensors close to the plume are high. As the dataloggers close to the plume are also able to associate with only a few of these sensors because of their high sampling rates, the dataloggers that are farther away from the plume accommodate the remaining sensors.

6. EXTENSIONS OF THE MPC MODEL Figure 4 - Sampling rates of sensors as determined by the MPC controller. Size of dots representing sensors is proportional to sensor sampling rate. Small red dots indicate the extent of the plume. Datalogger locations are indicated by unfilled squares. Colors are used to indicate assignment of a sensor to a datalogger.

The MPC controller that was described for the NYHOPS system considered critical regions that were determined by an event detection module (Figure 3). The event detection used the forecast provided by the predictive ECOMSED model. However, the event detection can incorporate other data sources such as satellite imagery to provide a more accurate prediction of regions of interest. In addition, if these data sources are controllable, then these control parameters can be included in the optimization framework. For instance, the paths of relevant remote sensing satellites can be modeled as part of the optimization step. The optimal control can then include desired remote sensing tasks for the satellite.

a geographically defined area of interest that varied over time. The simulation results indicate that our proposed control algorithm can be effective in a variety of sensor network applications. The technique can also be extended to take into account data sources such as satellite images and generate scheduling controls for remote sensing satellites.

7. CONCLUSIONS

ACKNOWLEDGEMENT

The paper demonstrated the use of model predictive control as a general framework for resource management in sensor networks. The technique adapts the system parameters that affect the resolution of the sensed data (sampling rates of the sensors) and the rate of system resource utilization (data routes in wireless transmission) such that these two competing factors are optimized taking into account the detection of interesting events. This optimization uses the established mathematical framework of model predictive control. We illustrated the approach on a coastal environment monitoring system. The sensor sampling rates and effective network throughput were varied in response to

This work has been sponsored by the NASA Applied Information Systems Technology Program under AISTQRS Award # AIST-QRS-06-0017.


[11] C.-f. Hsin and M. Liu, "Network coverage using low duty-cycled sensors: random & coordinated sleep algorithms," presented at Proceedings of the third international symposium on Information processing in sensor networks (IPSN), Berkeley, California, USA, 2004.

REFERENCES [1] D. Estrin, L. Girod, G. J. Pottie, and M. Srivastava, "Instrumenting the world with wireless sensor networks," presented at International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Salt Lake City, 2001.

[12] C. Gui and P. Mohapatra, "Virtual patrol: a new power conservation design for surveillance using sensor networks," presented at Proceedings of the Fourth International Symposium on Information Processing in Sensor Networks (IPSN), Los Angeles, California, USA, 2005.

[2] E. Szewczyk, A. Mainwaring, J. Polastre, J. Anderson, and D. Culler, "An Analysis of a Large Scale Habitat Monitoring Application," presented at 2nd International Conference On Embedded Networked Sensor Systems, Baltimore, MD, 2004.

[13] Z. Zhou, S. Das, and H. Gupta, "Variable radii connected sensor cover in sensor networks," presented at First Annual IEEE Communications Society Conference on Sensor and Ad Hoc Communications and Networks (SECON), Santa Clara, California, USA, 2004.

[3] J. D. Lundquist , D. R. Cayan, and M. D. Dettinger, "Meteorology and Hydrology in Yosemite National Park: A Sensor Network Application," presented at Information Processing in Sensor Networks, Palo Alto, CA, 2003. [4] D. J. Sandoz, M. J. Desforges, B. Lennox, and P. R. Goulding, "Algorithms for industrial model predictive control," Computing & Control Engineering Journal 11, pp. 125-34, 2000.

[14] X. Li, T. Nguyen, and R. Martin, "Using Adaptive Range Control to Maximize 1-Hop Broadcast Coverage in Dense Wireless Networks," presented at First Annual IEEE Communications Society Conference on Sensor and Ad Hoc Communications and Networks (SECON), Santa Clara, California, USA, 2004.

[5] V. Gopal and L. T. Biegler, "Large scale inequality constrained optimization and control," Control Systems Magazine, IEEE, vol. 18, pp. 59-68, 1998.

[15] D. Rakhmatov and S. Vrudhula, "Energy Management for Battery-Powered Embedded Systems," ACM Transactions on Embedded Computing Systems, vol. 2, pp. 277-324, 2003.

[6] J. Jelinek and D. Godbole, "Model predictive control of military operations," in Proceedings of the 39th IEEE Conference on Decision and Control (Cat. No.00CH37187); Piscataway, NJ, USA : IEEE, 2000, 5 vol. (lxiii+li+5229) p. (2562-7 vol.3) Part No: vol.3.

[16] S. Bandyopadhyay, Q. Tian, and E. J. Coyle, "Spatiotemporal sampling rates and energy efficiency in wireless sensor networks," IEEE/ACM Transactions on Networking, vol. 13, pp. 1339-1352, 2005.

[7] J. E. Tierno, "Distributed autonomous control of concurrent combat tasks," presented at American Control Conference, 2001. Proceedings of the 2001, 2001.

[17] B. N. Chun, P. Buonadonna, A. AuYoung, C. Ng, D. C. Parkes, J. Shneidman, A. C. Snoeren, and A. Vahdat., "Mirage: A Microeconomic Resource Allocation System for SensorNet Testbeds," presented at Proceedings of the 2nd IEEE Workshop on Embedded Networked Sensors (EmNetS), Sydney, Australia, 2005.

[8] F. Y.-S. Lin and P.-L. Chiu, "Energy-efficient sensor network design subject to complete coverage and discrimination constraints," presented at Second Annual IEEE Communications Society Conference on Sensor and Ad Hoc Communications and Networks, Santa Clara, California, USA, 2005.

[18] M. Kalantari and M. A. Shayman, "Energy Efficient Routing in Wireless Sensor Networks," presented at Conference on Information Sciences and Systems, Princeton University, 2004.

[9] M. Maleki and M. Pedram, "QoM and lifetimeconstrained random deployment of sensor networks for minimum energy consumption," presented at Fourth International Symposium on Information Processing in Sensor Networks (IPSN), Los Angeles, 2005.

[19] Y. Xu, J. Heidemann, and D. Estrin, "Adaptive EnergyConserving Routing for Multihop Ad Hoc Networks," USC/Information Sciences Institute 527, 2000.

[10] X. Zhang and S. B. Wicker, "How to distribute sensors in a random field?," presented at Third International Symposium on Information Processing in Sensor Networks (IPSN), Berkeley, California, USA, 2004.


[20] T. Melodia, D. Pompili, and I. F. Akyildiz, "Optimal local topology knowledge for energy efficient geographical routing in sensor networks," presented at Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies (INFOCOM), Hong Kong, 2004.

[30] W. Zhang, G. Wang, and Z. Xing, "Distributed stochastic search and distributed breakout: Properties, comparison and applications to constraint optimization problems in sensor networks," Artificial Intelligence, vol. 161, pp. 55-87, 2005. [31] E. Camponogara, D. Jia, B. H. Krogh, and S. Talukdar, "Distributed Model Predictive Control," IEEE Control Systems Magazine, pp. 44-52, 2002.

[21] B. Krishnamachari and F. Ordonez, "Analysis of energy-efficient, fair routing in wireless sensor networks through non-linear optimization," presented at IEEE 58th Vehicular Technology Conference, 2003.

[32] W. B. Dunbar and R. M. Murray, "Model Predictive Control of Coordinated Multi-Vehicle Formations," presented at IEEE Conference on Decision and Control, 2002.

[22] Y. T. Hou, Y. Shi, and H. Sherali, "Rate allocation in wireless sensor networks with network lifetime requirement," presented at Proceedings of the 5th ACM International symposium on Mobile ad hoc networking and computing, Tokyo, 2004.

[33] A. N. Venkat, J. B. Rawlings, and S. J. Wright, "Stability and optimality of distributed model predictive control," presented at CDC-ECC Joint Conference, Seville, Spain, 2005.

[23] W.-P. Chen and L. Sha, "An energy-aware data-centric generic utility based approach in wireless sensor networks," presented at Third International Symposium on Information Processing in Sensor Networks (IPSN), 2004.

[34] T. O. Herrington, M. S. Bruno, and J. W. Rankin, "The New Jersey Coastal Monitoring Network: A Real-Time Coastal Observation System," Journal of Marine Environmental Engineering, vol. 6, pp. 69-82, 2000.

[24] N. Sadagopan and B. Krishnamachari, "Maximizing Data Extraction in Energy-Limited Sensor Networks," presented at Proceedings of IEEE INFOCOM, 2004.

[35] A. F. Blumberg, L. A. Khan, and J. P. St John, "Three dimensional hydrodynamic model of New York Harbor region," Journal of Hydraulic Engineering-Asce, vol. 125, pp. 799-816, 1999.

[25] R. Willett, A. Martin, and R. Nowak, "Backcasting: adaptive sampling for sensor networks," presented at Proceedings of the 3rd ACM/IEEE Symposium on Information Processing in Sensor Networks (IPSN), 2004. [26] B. Zhang, G. Sukhatme, and A. Requicha, "Adaptive Sampling for Marine Microorganism Monitoring," presented at.IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2004.

BIOGRAPHIES Ashit Talukder graduated with a Ph.D. in Electrical and Computer Engineering from Carnegie Mellon University in 1999. His research interests and expertise include pattern recognition, image and signal processing, machine learning, AI, controls, computer vision, optimization, and mathematics, applied to data mining, robotics, autonomous systems and sensors, sensor networks, biometrics, ATR, and human-machine interaction.

[27] A. Jain and E. Chang, "Adaptive sampling for sensor networks," presented at Proceeedings of the First international workshop on Data management for sensor networks, Toronto, 2004. [28] R. Béjar, C. Domshlak, C. Fernàndez, C. P. Gomes, B. Krishnamachari, B. Selman, and M. Valls, "Sensor networks and Distributed CSP: Communication, Computation and Complexity," Artificial Intelligence, vol. 161, pp. 117-147, 2005.

He is currently a Senior Researcher (Level A) at Jet Propulsion Laboratory / NASA, California Institute of Technology, in Pasadena, California, and also holds a joint appointment as a research adjunct faculty member at University of Southern California and senior researcher at CHLA. He is a principal investigator and technical lead on several NASA, NIH, NSF, DoD and JPL-funded projects. He has over 50 refereed journal and conference publications and authored a book chapter.

[29] A. Petcu and B. Faltings, "A scalable method for multiagent constraint optimization," presented at Proceedings of the 19th International Joint Conference on Artificial Intelligence (IJCAI), 2005.


Dr. Talukder is on the organizing committee of the Annual SPIE Defense and Security Symposium. He is a Technical Program Committee member of the International Conference on Digital Telecommunications ICDT 2006 and The First International Conference on Advances in Computer-Human Interaction ACHI 2008. He has a patent for a biomedical data analysis and visualization system which was nominated for the NASA software of the year, and holds a provisional patent for a remote biometric identification system. He is a recipient of the NASA Space Act Award and a PACE award at Iowa State University. He has chaired several SPIE conference sessions, and given three keynote addresses and several invited talks at international conferences. He is a reviewer on several technical journals.

Anand Panangadan received the Ph.D. degree in computer science from the University of California, Los Angeles in 2002. He received the B.Tech. degree in Computer science and Engineering from the Indian Institute of Technology, Bombay in 1996. Dr. Panangadan is currently a Research Associate at the Saban Research Institute of the Childrens Hospital Los Angeles. He is developing algorithms for energy conservation in wireless sensor networks. These algorithms are to be deployed in remote human health monitoring systems and environmental sensor networks. He has also designed data compression algorithms that take into account the lossy nature of wireless transmission. Prior to joining the Childrens Hospital Los Angeles, Dr. Panangadan was a post-doctoral researcher at the Robotics Research Lab of the University of Southern California. While there, he developed probabilistic models for describing human movement and region detection algorithms for sensor networks. His Ph.D. dissertation describes a framework for building autonomous agents that can collectively construct arbitrary structures in a simulated environment
