SENSOR NETWORKS ECE 654 Irene Ioannou. Sensor networks communication architecture.
N.E. Leonard – U. Pisa – 18-20 April 2007 Slide 1 Cooperative Control and Mobile Sensor Networks...
-
Upload
akira-duston -
Category
Documents
-
view
212 -
download
0
Transcript of N.E. Leonard – U. Pisa – 18-20 April 2007 Slide 1 Cooperative Control and Mobile Sensor Networks...
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 1
Cooperative Control and Mobile Sensor Networks
Application to Mobile Sensor Networks, Part II
Naomi Ehrich Leonard
Mechanical and Aerospace EngineeringPrinceton University
and Electrical Systems and Automation University of Pisa
[email protected], www.princeton.edu/~naomi
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 2
Key References
[1] Leonard, Paley, Lekien, Sepulchre, Fratantoni, Davis, “Collective motion, sensor networks and ocean sampling,” Proc. IEEE, 95(1), 2007.
[2] F. Lekien and N.E. Leonard, “Non-Uniform Coverage and Cartograms,” preprint, [Online] http://www.princeton.edu/~naomi
[3] D. Paley, F. Zhang and N.E. Leonard. Cooperative control for ocean sampling: The Glider Coordinated Control System. IEEE Transactions on Control System Technology, to appear.
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 3
5 Scripps Spray Gliders 10 WHOI Slocum Gliders
AOSN-II Glider Plan
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 4
AOSN-II Glider Measurements
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 5
Sampling Metric: Objective Analysis Error
Scalar field viewed as a random variable:
Data collected consists of
is OA estimate that minimizes
is a priori mean. Covariance of fluctuations around mean is
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 6
QuickTime™ and aCinepak decompressor
are needed to see this picture.
(with F. Lekien, D. Paley)
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 7
Coverage Metric: Objective Analysis ErrorRudnick et al, 2004
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 8
Coverage Metric: Objective Analysis
Error
QuickTime™ and aCinepak decompressor
are needed to see this picture.
Francois Lekien
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 9
Computation of Optimal Trajectories
Box:
Trajectories:
Constraint:
Optimality Trajectories:
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 10
Maximize Information in Measurements
Direct optimization of sampling metric leads to overly complex patterns of data distribution.
Direct optimization of sampling metric along “ideal tracks” provides practical basis for automatic steering and inter-vehicle coordination.
Family of closed loops as candidate ideal tracks is sufficiently large for multi-scale patterns and information rich sampling plans.
Adaptations can be made as changes in ideal tracks.
A set of ideal tracks with prescribed relative glider spacing is called a set of Glider Coordinated Trajectories (GCT).
adapt
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 11
Optimal Solution for Ellipses
For ellipses, the optimum is at
Corresponds to one glider per regionof area
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 12
Optimization
Optimal elliptical trajectories for two vehicles on square spatial domain. Feedback control used to stabilize vehicles to optimal trajectories.
Optimal solution corresponds to synchronized vehicles.
Flow shown is 2% of vehicle speed.
No flow. Metric = 0.018
Horizontal flow. Metric = 0.020
Vertical flow. Metric = 0.054
No heading coupling. Metric = 0.236Leonard, Paley, Lekien, Sepulchre,
Fratantoni, Davis, Proc. IEEE, Jan. 2007.
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 13
Glider Plan
GCT for increased sampling in southwest corner of ASAP box.
Candidate default GCT with grid for glider tracks. Adaptation
SIO gliderWHOI glider
km
km
km
km
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 14
Virtual Pilot Experiment: March ‘06; Ocean: August ‘03
QuickTime™ and a decompressor
are needed to see this picture.
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 15
Maximize Information in Measurements
HOPS
ROMS
NCOM
QuickTime™ and aCinepak decompressor
are needed to see this picture.
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 16
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 17
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 18
Glider Coordinated Trajectories (GCT)
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 19
Glider Planner Status
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 20
August 2006 Adaptations
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 21
Monterey Bay, CA, August 2006
QuickTime™ and a decompressor
are needed to see this picture.
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 22
OA error map, 2006
QuickTime™ and aCinepak decompressor
are needed to see this picture.
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 23
ASAP Sampling Performance, August 2006
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 24
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 25
Comparing Actual Glider Motion and NCOM Prediction
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 26
Comparing Actual Glider Motion and HOPS Prediction
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 27
Evaluating Glider Motion Predictions
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 28
Non-uniform metric optimization
There exist many methods to cover a regionuniformly. Example: Cortes and Bullo, 2005.
These methods do not extend easily tonon-uniform and dynamic metrics.
Find a transformation from the physical space where volume elements are given by the non-uniform OA metric (i.e., dA=B(x,y) dx dy) to a virtual space with the Euclidian metric (i.e., dA=dx dy). Examples: the cartograms of Gastner and Newman, 2004.
Reproduced from Cortes & Bullo, SIAMJ. Control Optim, 44(5), 1543—1574, 2005.
Reproduced from Gastner & Newman, Proc. National Academy US, 101(20), 7499—7504, 2004.
Francois Lekien
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 29
Non-uniform metric optimization
Step #1: Extend the function to a larger domain with homogeneous Neumann boundary conditions:
Step #2: Solve the diffusion equation:
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 30
Non-uniform metric optimization
Physical space: non-uniform metric Transformed space: Euclidian metric
QuickTime™ and aBMP decompressor
are needed to see this picture.
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 31
OA error map(Cartesian space with nonlinear metric)
After transformation(Curved space with Euclidean metric)
QuickTime™ and aCinepak decompressor
are needed to see this picture.
QuickTime™ and aCinepak decompressor
are needed to see this picture.
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 32
Data Flow: Actual and Virtual ExperimentsPrinceton Glider Coordinated Control System (GCCS)
N.E. Leonard – U. Pisa – 18-20 April 2007Slide 33
Princeton Glider Coordinated Control System