Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
IEEE Control Systems Society
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
IEEE – Institute of Electrical and Electronics Engineers
Formed in 1963 by merger of the American Institute ofElectrical Engineers (founded 1884) and the Institute ofRadio Engineers (founded 1912)
IEEE is the world’s leading association for theadvancement of technology, has more than 375,000members located in 160 countries; 45% of its membersare from outside the US
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
CSS – Control Systems Society
CSS is one of 38 Societies of IEEE, with a total numberof 8,338 members, founded in 1956
Number of represented countries: 116 USA members: 3,553 members
R. Tempo, “Internationalizationand Globalization” IEEEControl Systems Magazine,February 2010
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
IEEE Control Systems Society
President: Roberto Tempo
VPPA: Christos G. Cassandras VPTA: Shuzhi Sam Ge VPMA: Shinji Hara VPFA: Pradeep Misra VPCA: Maria Elena Valcher
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
CSS Transactions and Magazine
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
2010 CSS Sponsored Conferences
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
CSS Technical Committees Aerospace Controls Automotive Controls Behavioral Systems and Contr. Th. Computational Aspects of Contr. Sys. Control Education Discrete Event Systems Distributed Parameter Systems Hybrid Systems Industrial Process Contr Intelligent Control Manufacturing Autom. Robotic Contr. Networks and Communications Nonlinear Systems and Controls
Systems Biology System Id & Adaptive Control Systems with Uncertainty Variable Structure and SMC
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
President’s Messages IEEE CSM 2010
1. “Internationalization and Globalization” February
2. “Research, a Never-Ending Story” April
3. “From Hard Copy to Electronic” June
4. “Handwriting, Typing and LaTeX” August
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithms for Systems and Control: Theory and Applications
Roberto TempoIEIIT-CNR
Politecnico di Torino, [email protected]
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithms: A Success Story
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
A Success Story
Randomized Algorithms (RAs) are successfully usedin various areas outside control
1. CS: Sorting problems (e.g., QuickSort algorithm)
2. CS: Data structuring, search trees, graph algorithms
3. Mathematics of finance: Computation of integrals
4. Genomics: String matching and classification
5. Robotics: Motion and path planning problems
6. Control of Unmanned Aerial Vehicles (UAV)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Monte Carlo and Las Vegas Algorithms
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Monte Carlo and Las Vegas
Monte Carlo was invented by Metropolis, Ulam, vonNeumann, Fermi, … (Manhattan project)
Las Vegas first appeared in computer science in the lateseventies
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithm: Definition
Randomized Algorithm (RA): An algorithm that makesrandom choices during its execution to produce a result
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithm: Definition
Randomized Algorithm (RA): An algorithm that makesrandom choices during its execution to produce a result
Example of a “random choice” is a coin tossheads or tails
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithm: Definition
Randomized Algorithm (RA): An algorithm that makesrandom choices during its execution to produce a result
For hybrid systems, “random choices” could beswitching between different states or logical operations
For uncertain systems, “random choices” require (vectoror matrix) random sample generation
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Monte Carlo Randomized Algorithm
Monte Carlo Randomized Algorithm (MCRA): Arandomized algorithm that may produce incorrect results,but with bounded error probability
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Monte Carlo Randomized Algorithm
Monte Carlo Randomized Algorithm (MCRA): Arandomized algorithm that may produce incorrect results,but with bounded error probability
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Example of Monte Carlo: Area/Volume Estimation
Estimate the volume of the red area: Generate N sampleuniformly in the rectangle; count how many fall withinthe red area (M), estimated area = M/N
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Las Vegas Randomized Algorithm
Las Vegas Randomized Algorithm (LVRA): Arandomized algorithm that always produces correctresults, the only variation from one run to another is therunning time
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Las Vegas Randomized Algorithm
Las Vegas Randomized Algorithm (LVRA): Arandomized algorithm that always produces correctresults, the only variation from one run to another is therunning time
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Example of Las Vegas
Example of Las Vegas Randomized Algorithm: RandomSurfer Model
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Surfer Model
Web representation with incoming and outgoing links
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Surfer Model
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Surfer Model
Pick an outgoing link at random
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Surfer Model
Arriving at a new web page
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Surfer Model
Pick another outgoing link at random
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Surfer Model
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Surfer Model
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Surfer Model
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Example of Las Vegas: RQS
Problem: Sorting N real numbers
Algorithm: RandQuickSort (RQS)
RQS is implemented in a C library of Linux for sortingnumbers[1-2]
[1] C.A.R. Hoare (1962) – [2] D.E. Knuth (1998)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
A Success Story in CS
Problem: Sorting N real numbers
Algorithm: RandQuickSort (RQS)
RQS is implemented in a C library of Linux for sortingnumbers
Sorting Problem
given N real x1 x2 x3 sort them in
numbers x4 x5 x6 increasing order
S1
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RandQuickSort (RQS)
The idea is to divide the original set S1 into two setshaving (approximately) the same cardinality
This requires finding the median of S1 (which may bedifficult)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RandQuickSort (RQS)
RQS is a recursive algorithm consisting of two phases
1. randomly select a number xi (e.g. x4)2. deterministic comparisons between xi and other (N-1) numbers
x2 x3 < x4 < x1 x5 x6
numbers smaller than x4 numbers larger than x4
S3S2
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RQS: Binary Tree Structure
We use randomization at each step of the (binary) tree
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Running Time of RQS
Because of randomization, running time may bedifferent from one run of the algorithm to the next one
RQS is very fast: Average running time is O(N log (N)) This is a major improvement compared to brute force
approach Average running time is highly probable…
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RQS: The Las Vegas Viewpoint
Average running time is O(N log(N)) This running time holds for every input The average running time holds with probability at least
1-1/N Hint: Use the so-called Chernoff bound to prove this Improvements for RQS to avoid achieving the worst
case running time O(N2)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Find Algorithm
Find Algorithm: Find the k-th smallest number in a set Basically it is a RQS but it terminates when the number
is found Average running time of Find is O(N)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Sorting of Switched Systems
Motivations: Deciding which systems are more stablethan others is useful information for the controller
This requires finding a LVRA which provides amatrix sorting for the N Lyapunov equations
Matrix version of RandQuickSort is developed[1]
Technical difficulty: The equations may be notcompletely sortable because of sign indefiniteness
[1] H. Ishii, R. Tempo (2007)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RQS for Matrices: Trinary Tree
We use randomization at each step of the (trinary) tree
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Unmanned Aerial Vehicles (UAVs)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Italian National Project for Fire Prevention
This activity is supported by the Italian Ministry forResearch within the National Project
Study and development of a real-time land control and monitoring system for fire prevention
Five research groups are involved together with agovernment agency for fire surveillance and patrollocated in Sicily
The aerial platform is based on the MicroHawkconfiguration, developed at the Aerospace EngineeringDepartment, Politecnico di Torino, Italy
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
MH1000 Platform - 1
Platform features- wingspan 3.28 ft (1 m)- total weight 3.3 lb (1.5 kg)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
MH1000 Platform - 2
Main on-board equipment- various sensors and two cameras (color and infrared)
DC motor Remote piloting and autonomous flight Flight endurance of about 40 min Flight envelope
- min/max velocity: 33 ft/s (10 m/s) – 66 ft/s (17 m/s)- average velocity: 43 ft/s (14 m/s)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Flight Envelope (Limits)
Wing loading effect total weight
Propeller sizing effectPropulsive constraint (blu) maximum flight speed
Aerodynamic constraint (red) minimum flight speed (stall effect)
velocity: 33 ft/s (10 m/s) – 66 ft/s (17 m/s)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
DC motor: Hacker B20-15L (4:1)
controller: Hacker Master Series 18-B-Flight
battery: Kokam 2000HD (3x)
receiver: Schulze Alpha840W
servo: Graupner C1081 (2x)
weight: 58 g
dimensions: Ø 20 x 40 mm
Kv: 3700 rpm/volt
weight: 21 g
dimensions: 33 X 23 X 7 mm
current drain: 18 A
weight: 160 g
dimensions: 79 X 42 X 25 mm
capacity: 2000 mAh
weight: 13.5 g
dimensions: 52 X 21 X 13 mm
8 channels
weight: 13 g
dimensions: 23 X 9 X 21 mm
torque: 12 Ncm
Basic on-board Systems
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Prototype Manufacturing - 1
raw materialpolistyrene
kevlar
fiberglass
carbon fiber
epoxy resinplywood
balsa wood
glue
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
working instrumentshot wire foam cutting machine
lifting surfaces outline
slide outlinefuselage reference
Prototype Manufacturing - 2
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
prototype
easy constructionrapid manufacturing
bad model reproducibilityinaccurate geometry
Prototype Manufacturing - 3
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
State Space Model
State space formulation obtained by linearization of thefull (12 coupled nonlinear ODE) model
x(t) = A(∆) x(t) + B(∆) u(t)
u(t) = - K x(t)where x = [V, α, q, θ]T (V flight speed, α angle ofattack, q and θ pitch rate and angle), ∆ uncertainty
Consider longitudinal plane dynamics stabilization Control u is the symmetrical elevon deflection
.
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Uncertainty Description - 1
We consider structured parameter uncertainties affectingplant and flight conditions, and aerodynamic database
Uncertainty vector ∆ = [δ1,..., δ16] where δi∈ [δi-, δi
+] Key point: There is no explicit relation between state
space matrices A and B and uncertainty ∆ This is due to the fact that state space system is obtained
through linearization and off-line flight simulator The only techniques which could be used in this case are
simulation-based which lead to randomized algorithms
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Uncertainty Description - 2
We consider random uncertainty ∆ = [δ1,..., δ16]T
The pdf is either uniform (for plant and flightconditions) or truncated Gaussian (for aerodynamicdatabase uncertainties)
Flight conditions uncertainties need to take into accountlarge variations on physical parameters
Uncertainties for aerodynamic data are related toexperimental measurement or round-off errors
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Plant and Flight Condition Uncertainties
parameter pdf δi % δi- δi
+ #
flight speed [ft/s] U 42.65 ± 15 36.25 49.05 1
altitude [ft] U 164.04 ± 100 0 328.08 2
mass [lb] U 3.31 ± 10 2.98 3.64 3
wingspan [ft] U 3.28 ± 5 3.12 3.44 4
mean aero chord [ft] U 1.75 ± 5 1.67 1.85 5
wing surface [ft2] U 5.61 ± 10 5.06 6.18 6
moment of inertia [lb ft2] U 1.34 ± 10 1.21 1.48 7
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Aerodynamic Database Uncertainties
parameter pdf δi σi #CX [-] G -0.01215 0.00040 8CZ [-] G -0.30651 0.00500 9Cm [-] G -0.02401 0.00040 10CXq [rad-1] G -0.20435 0.00650 11CZq [rad-1] G -1.49462 0.05000 12Cmq [rad-1] G -0.76882 0.01000 13CX [rad-1] G -0.17072 0.00540 14CZ [rad-1] G -1.41136 0.02200 15Cm [rad-1] G -0.94853 0.01500 16
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Standard Deviation and Velocity
Standard deviation is experimentally computed from the velocity
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Gain Synthesis (RGS)
Specs in the complex plane
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithm 1 (RGS) Uniform pdf for controller
gains K in given intervals Accuracy and confidence
ε =4 ·10-5 and η = 3 · 10-4
Number of random samples is computed with “Log-over-Log” Bound obtaining N = 200,000
We obtained s = 5 gains Ki
satisfying specification property S1
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Random Gain Set
gain set KV Kα Kq Kθ
K1 0.00044023 0.09465000 0.01577400 -0.00473510
K2 0.00021450 0.09581200 0.01555500 -0.00323510
K3 0.00054999 0.09430800 0.01548200 -0.00486340
K4 0.00010855 0.09183200 0.01530000 -0.00404380
K5 0.00039238 0.09482700 0.01609300 -0.00417340
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithm 2 (RSRA) Take Krand from Phase 1 Accuracy and confidence
ε = η = 0.0145 Number of random
samples is computed with Chernoff Bound obtaining N =5,000
Empirical probability is defined using an indicator function
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Empirical Probability of Stability for Phase 2
gain set empirical probability
K1 88.56%
K2 90.60%
K3 89.31%
K4 93.86%
K5 85.14%
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Probability Degradation Function
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Root Locus Plot
Root locus for K2 (left) and K4 (right)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Bandwidth Criterion
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithm 3 (RPRA)
Take Krand from Phase 1 Numer of random samples
is computed with the Chernoff Bound obtaining N =5,000
Empirical probability is defined using an indicator function
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Empirical Probability of Performance for Phase 3
gain set empirical probability
K1 93.58%
K2 95.16%
K3 90.80%
K4 84.78%
K5 96.06%
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Probability Degradation Function
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Bandwidth Criterion
Bandwidth criterion for K1 (left) and K3 (right)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Gain Selection
Multi-objective criterion as a compromise between different specifications
Finally we selected gain K1 as the best compromise between all the specs and criteria
Detailed results available in[1]
[1] L. Lorefice, B. Pralio and R. Tempo (2009)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Conclusions: Flight Tests in Sicily - 1
Evaluation of the payload carrying capabilities andautonomous flight performance
Mission test involving altitude, velocity and headingchanging was performed in Sicily
Checking effectiveness of the control laws forlongitudinal and lateral-directional dynamics
Flight control design based on RAs for stabilization andguidance
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Conclusions: Flight Tests in Sicily - 2
Satisfactory response of MH1000
Possible improvements by iterative design procedure
Stability of the platform is crucial for the video qualityand in the effectiveness of the surveillance andmonitoring tasks
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Color Camera: Right Turn
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Color Camera: Landing Phase
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RACTRandomized Algorithms Control Toolbox
http://ract.sourceforge.net
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RACT
RACT: Randomized Algorithms Control Toolbox forMatlab
RACT has been developed at IEIIT-CNR and at theInstitute for Control Sciences-RAS, based on a bilateralinternational project
Members of the projectAndrey Tremba (Main Developer and Maintainer) Giuseppe Calafiore Fabrizio Dabbene Elena Gryazina Boris Polyak (Co-Principal Investigator) Pavel Shcherbakov Roberto Tempo (Co-Principal Investigator)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RACT
Main features Define a variety of uncertain objects: scalar, vector and
matrix uncertainties, with different pdfs Easy and fast sampling of uncertain objects of almost
any type Sequential randomized algorithms for feasibility of
uncertain LMIs using stochastic gradient and localizationmethods (ellipsoid or cutting plane)
Non-sequential randomized algorithms for optimizationof convex problems
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RACT
Under construction: Non-sequential randomizedalgorithms for feasibility and optimization of non-convex problems
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
RACT
RACT: Randomized Algorithms Control Toolbox forMatlab
http://ract.sourceforge.net
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithms for Systems and Control Applications
IEIIT-CNR
CCC Beijing ©RT 2010
Randomized Algorithms for Systems and Control Applications - 1
Aerospace control andunmanned aerial vehicles(UAVs)[1,2,3]
Multi-agent systemsand consensus[4,5]
[1] C.I. Marrison and R.F. Stengel (1998)[2] B. Lu and F. Wu (2006)[3] L. Lorefice, B. Pralio and R. Tempo (2009)[4] H. Ishii and R. Tempo (2010)[5] L. Pallottino, V.G. Scordio, E. Frazzoli and A. Bicchi (2007)
IEIIT-CNR
CCC Beijing ©RT 2010
Randomized Algorithms for Systems and Control Applications - 2
Network congestion control[1]
Quantized and switchedsystems[2-3]
Fault detection, isolation,vision-based control[4-5]
[1] T. Alpcan, T. Basar and R. Tempo (2005)[2] H. Ishii, T. Basar and R. Tempo (2004)[3] H. Ishii, T. Basar and R. Tempo (2005)[4] S. Kanev and M. Vehaegen (2006)[5] W. Ma, M. Sznaier and C.M. Lagoa (2007)
IEIIT-CNR
CCC Beijing ©RT 2010
Randomized Algorithms for Systems and Control Applications - 3
Embedded and electriccircuits[1,2]
Advanced driver assistancesystems[3]
[1] C. Alippi (2002)[2] C.M. Lagoa, F. Dabbene and R. Tempo (2008) [3] O.J. Gietelink, B. De Schutter, and M. Verhaegen (2005)
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Main References
R. Tempo, G. Calafiore and F. Dabbene, “Randomized Algorithms for Analysis and Control of Uncertain Systems,” Springer-Verlag, London, 2005
F. Dabbene and R. Tempo, “Probabilistic and Randomized Tools for Control Design,” The Control Handbook, Taylor & Francis, 2010 (to appear)
Additional documents, papers, etc, please consulthttp://staff.polito.it/roberto.tempo/
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Conclusions
IEIIT-CNR
Northeastern University, Shenyang © RT 2010
Randomized Algorithms: A Success Story
Randomized algorithms are a success story for systems and control