fundamental issues in networked control systems –...

© June 26, 2009 , P. R. Kumar

Fundamental issues in networked control systems – 3:Architecture and Theory

P. R. Kumar

Dept. of Electrical and Computer Engineering, and Coordinated Science Lab University of Illinois, Urbana-Champaign

Email: [email protected]: http://decision.csl.illinois.edu/~prkumar

3rd WIDE PhD School on Networked Control Systems, Siena, Italy July 7-9, 2009

See last page and http://creativecommons.org/licenses/by-nc-nd/3.0/

1/47

© June 26, 2009 , P. R. Kumar

2/47

Networked Cyber-Physical Systems

Wireless networks

Networked Embedded

Control

Sensor Networks

© June 26, 2009 , P. R. Kumar

3/47

The Third Generation of Control Systems

  First generation: Analog Control Systems –  Technology: Electronic Feedback Amplifiers –  Theory: Frequency domain analysis: Bode, Evans

  Second Generation: Digital Control –  Technology: Digital computers –  Theory: State-space design, Kalman filter, H∞, Real-Time

Scheduling

  Third generation: Networked Embedded Control Systems –  Embedded computers –  Wireless and wireline networking –  Software

© June 26, 2009 , P. R. Kumar

Intertwined histories of communication,computation and control

  “…the era of cyberspace and the Internet, with its emphasis on the computer as a communications device and as a vehicle for human interaction connects to a longer history of control systems that generated computers as networked communications devices.”− David Mindell in “Feedback, Control and Computing before Cybernetics,” 2002

4/47

  Perhaps the most exciting developments in the information area relate to the large-scale digital computing machines.”− Claude Shannon, 1947

  “I think I can claim credit for transferring the whole theory of the servomechanism bodily to communication engineering.”− Norbert Wiener, 1956

© June 26, 2009 , P. R. Kumar

5/47

The themes

  Time

  Information

  Abstractions

  Analysis

  Applications

© June 26, 2009 , P. R. Kumar

6/47

Challenge of Abstractions and Architecture

Session Layer Session Layer

Presentation Layer Presentation Layer

Application Layer Application Layer

Transport Layer Transport Layer

Network Layer Network Layer

Data Link Layer Data Link Layer

Physical Layer Physical Layer

Internet Digital Communication

Source Coding

Channel Coding

  What are the abstractions and architecture for convergence with communication and computing?

Hardware Software

Serial computation

von NeumannBridge

  Goal is to enable rapid design and deployment –  Critical Resource: Control Designerʼs Time

  Standardized abstractions and architecture –  Minimal reconfiguration and reprogramming

  Hopefully leading to massive proliferation

© June 26, 2009 , P. R. Kumar

7/57

Architecture and computing, communication and control

  Building software systems on a standard platform   Standard abstractions and

architecture   Program as data

  Interfaces   Ease of customization   Modular design of software   Portability – High level languages   Reusability – Component libraries

  Building computer networks around a standard architecture

–  Standard software architecture   TCP/IP is de facto standard

–  Layers of protocols   Breakdown networking into sub-problems   Solve sub-problems in different layers   Compose solutions into a working stack

–  Mechanism: Encapsulation of packets

  From general purpose Computing

IBM 360: General purpose computer   First commercially successful model

  360 – All round capability

What should be the architecture of third generation systems?

  To general purpose Networking

–  Example: PID Controller »  Generic tunable component   Easy to integrate

–  Key concept: Standard interfaces

  To general purpose systems –  Building operational control applications

with generic components in a standard framework

© June 26, 2009 , P. R. Kumar

8/47

Information TechnologyConvergence Lab: The Systems Vision Sensors

Automatic Control

Wireless Ad Hoc Network

Planning and Scheduling

(Baliga,Graham,Huang& K ʻ02)

© June 26, 2009 , P. R. Kumar

Convergence Lab

9/47

© June 26, 2009 , P. R. Kumar

10/44

Abstraction of Virtual Collocation (Graham, Baliga & K ʻ05)

Data Fusion

Actuator1

Vision Server1

Vision Server2

Controller2

Actuator2 ActuatorN

ControllerN

Planner1

Tracker1 SET1

SEP1

Supervisor SESu

Multi-input Multi Output(MIMO) System

© June 26, 2009 , P. R. Kumar

11/47

Application Layer

The Abstraction Layers TrajectoryPlanner

KalmanFilter

DeadlockAvoidance

Set PointGenerator

DiscreteEvent

Scheduler

ImageProcessing

Network Layer

Transport Layer System Layer

Link Layer

(Graham, Baliga & K ʻ05)

© June 26, 2009 , P. R. Kumar

12/47

Application Layer

The Abstraction Layers TrajectoryPlanner

KalmanFilter

DeadlockAvoidance

Set PointGenerator

DiscreteEvent

Scheduler

ImageProcessing

Network Layer


Link Layer

(Graham, Baliga & K ʻ05)

© June 26, 2009 , P. R. Kumar

13/47

The Abstraction Layers

Application Layer

Network Layer


Link Layer Discrete Event

Scheduler

Kalman filter

TrajectoryPlanner

Car

controller

Model PredictiveController

Set PointGeneration

ImageProcessing

Control LawOptimization

  Middleware manages the Components

(Baliga, Graham, & K ʻ05)

© June 26, 2009 , P. R. Kumar

14/47

The Abstraction Layers

Discrete Event

Scheduler

Kalman filter

TrajectoryPlanner

Car

controller

Model PredictiveController

Set PointGeneration

ImageProcessing

Control LawOptimization

Etherware –  Location

independence –  Semantic

addressing of components

–  System startup and upgrade during execution

–  Automatic migration of components for performance

System Layer

Transport Layer

Network Layer

MAC

Physical Layer

Application Layer

Serv

ice

2

Serv

ice

3

Cloc

k

  Middleware manages the Components

  Minimal reconfiguration and reprogramming (Baliga, Graham, & K ʻ05)

© June 26, 2009 , P. R. Kumar

15/47

Collision avoidance(Schuetz, Robinson & K ʻ05)

http://decision.csl.uiuc.edu/~testbed/videos/CollisionAvoidance.mpg

© June 26, 2009 , P. R. Kumar Sensor

Actuator

Control Law Design

X X

State

X X

Wirelessnetwork

Wirelessnetwork

State Estimator

Actuator Buffer

Time Translation Service

Receding Horizon Control

State Estimator

16/47

© June 26, 2009 , P. R. Kumar

X X

State Estimator

Actuator Buffer



State Estimator

Sensor

Actuator

Control Law Design

Wirelessnetwork

Wirelessnetwork

State

17/47

© June 26, 2009 , P. R. Kumar

X X

X

State Estimator

Actuator Buffer



State Estimator

Sensor

Actuator

Control Law Design

Wirelessnetwork

Wirelessnetwork

State

18/47

© June 26, 2009 , P. R. Kumar

X X

X

State Estimator

Actuator Buffer



State Estimator

Sensor

Actuator

Control Law Design

Wirelessnetwork

Wirelessnetwork

State

19/47

© June 26, 2009 , P. R. Kumar

Local Temporal Autonomy   Components able to tolerate failures of other components for some time

  Example: Insulating Controller from Sensor and Communication Network

Sensor Controller If sensor or network fails, controller fails too

Now controller has Local Temporal Autonomy Sensor Controller State

Estimator

Controller Actuator If Controller or Networkfails, Actuator fails too Now Actuator has LocalTemporal Autonomy Controller Actuator Block

Computation ActuatorBuffer

  Example: Insulating Actuator from Controller and Communication Network

  Converts Dependency relationships to Use If Available relationships   Makes possible other facilities such as

–  Automatic Restart of Failed Components –  Migration of Components –  Component Upgrade at Runtime

Reliability, robustness System integration, InitializationEvolution and Scalability 20/47

© June 26, 2009 , P. R. Kumar

21/47

Example of capabilities:Component Migration

  Designer should not have todeal with such low level issues

–  Designerʼs time is thecritical resource

Communicate pixels?

Excessive communication

overhead

Or compute position?

Migrate KalmanFilter to

Computer 2: Done through

Memento Design pattern

Kalman filter

Computer 2

Car controller

Computer 1

© June 26, 2009 , P. R. Kumar

22/47

Component Migration at Run-Time

  Migrate a controller to another location while system is running

http://decision.csl.uiuc.edu/~testbed/videos/migration.mpg

© June 26, 2009 , P. R. Kumar

Forcing functions and application design: Delays and Losses

Controller Sensor Actuator

Forcing function   Wireless channels have unpredictable delays and high losses

State estimator

Control buffer

Application design constraint   Use state estimators and control buffers

Etherware design   Assume components can tolerate some delays (no hard deadlines)

Benefits   Application can tolerate some delays – soft real time   Can use general purpose platforms (OS, languages, etc)

Wireless links

23/47

© June 26, 2009 , P. R. Kumar

Robustness

Controller Sensor Actuator

Forcing function   Components must have high availability

Component failure

Application design constraint   Check-point component state for quick restarts

Etherware design   Components implement an explicit state check-pointing interface

Benefits   Initialization can be performed using default initial state   Checkpoints can be used for component upgrades and migration

Checkpoint Re-initialize

24/47

© June 26, 2009 , P. R. Kumar

25/47

Five themes

  How much traffic can wireless networks carry?

  How should networks compute functions of distributed data?

  How can clocks be synchronized over networks?

  What are the appropriate abstractions and architecture?

  Where to locate control laws?

  How to establish properties of overall hybrid systems?

© June 26, 2009 , P. R. Kumar

26/47

Controller

Challenge of locating and optimizing control laws

Plant

Δ

Where to locate or migrate Controller? What Control Law to use?

  Much success over last 50 years in parametric and nonparametric control law design –  Optimal gains (LQG, H-inf) –  Adaptive control –  Backstepping

  Many open problems

  Optimizing with respect to delays, failures, structures, etc –  Migration, Reconfiguration, etc

  What are the appropriate theories to take advantage of the next generation capabilities

  Currently bottleneck is Theory, not Technology

© June 26, 2009 , P. R. Kumar

27/47

Location and optimization of control over a lossy network

C

  What is the best location for controller that calculates control law?

  And what should that control law be?

S

A

  LQG Problem: Sensor

measurements Actuation

commands

Min

Non-degeneracy assumptions

  Lossy network –  IID Packet drops

y(k) y(k) y(k) y(k)

u(k) X

  Non-classical information pattern –  Memory of past actions is lost

  Witsenhausen (1968) –  Computation of optimal controller is

intractable

Plant

y(k)

u(k)

© June 26, 2009 , P. R. Kumar

28/47

The Long Packet Assumption (LPA)(Robinson & K ʻ06)

  Theorem –  Suppose that “long packets” containing all history are transmitted

–  Then optimal location for Controller is at Actuator

–  The optimal control law is:

Provides a lowerbound on cost without long packets

  Even without LPA, optimal controller can be calculated when control is at actuator (Sinopoli, Schenato, Franceschetti et al ʻ04, Imer et al ʻ06) –  No Witsenhausen problem when controller is placed at actuator –  Can compare this suboptimal controller with lower bound

Slight strengthening ofseparation theorem

© June 26, 2009 , P. R. Kumar

29/47

The Long Packet Assumption (LPA)

  Theorem (Robinson & K ʻ06) –  Suppose that “long packets” containing all history are transmitted –  Then optimal location for Controller is at Actuator

–  ∃ policy with finite cost and iff

–  The optimal long-term average quadratic cost per is:

It is a lower bound on cost without long packets. –  Optimal controller is

  Even without LPA, optimal controller can be calculated when control is at actuator (Sinopoli, Schenato, Franceschetti et al ʻ04, Imer et al ʻ06) –  No Witsenhausen problem when controller is placed at actuator –  Can compare this suboptimal controller with lower bound

Slight strengthening ofseparation theorem

uk* = −[R + ′B WB]−1 ′B WAx̂k|k−Dk−1

© June 26, 2009 , P. R. Kumar

30/47

Near optimality of placing controller at actuator

  Problem is effectively solved

© June 26, 2009 , P. R. Kumar

31/47

Five themes

  How much traffic can wireless networks carry?

  How should networks compute functions of distributed data?

  How can clocks be synchronized over networks?

  What are the appropriate abstractions and architecture?

  Where to locate control laws?

  How to establish properties of overall hybrid systems?

© June 26, 2009 , P. R. Kumar

32/47

The themes

  Time

  Information

  Abstractions

  Analysis

  Applications

© June 26, 2009 , P. R. Kumar

33/47

Automatic traffic control: Safety and Liveness guarantees

http://decision.csl.uiuc.edu/~testbed/videos/city_7cars.mpg (Giridhar, Graham, Baliga & K ʻ03)

  Hybrid System: –  Bridging between discrete

and continuous models

1 2 3 4 5 6 7

9 10 11 12

82 81

83

80

86 85

22 21

66 92

91

71 84 72 73

74 75

79 78

76 77

30 29

23 24

28 27

25 26

70 69

87 88

68 67

89 90

65 93

64 94

63 95

62 96

61 97

60 98

47 50

40 51

43 48

42 49

45 46

44 47

31 20 32

35

54

19

34 53

18 8

17

52

33

© June 26, 2009 , P. R. Kumar

34/47

Analyzing the whole CPS system: System Safety and Liveness

  Theorem (Baliga & K ʼ05) –  Directed graph model of road network

»  Each bin has in-degree 1 or out-degree 1 »  System has no occupied cycles initially

–  Road width: »  Initial condition: »  Intersection angles , and road lengths: » Multiple cars with appropriate spacing

–  Car control model: Kinematic model with turn radii R and R

–  Real time renewal tasks: HST scheduling with

–  Then cars can be operated » Without collisions (Safety) or » Gridlocks (Liveness)

http://decision.csl.uiuc.edu/~testbed/videos/city_7cars.mpg

1 2 3 4 5 6 7

9 10 11 12

82 81

83

80

86 85

22 21

66 92

91

71 84 72 73

74 75

79 78

76 77

53 29

53 24

28 53

53 53

70 69

87 88

68 67

89 90

65 93

64 53

63 95

53 96

61 97

60 98

41 53

53 36

53 53

42 53

45 46

53 53

31 20 53

35

54

19

53 53

18 8

17

52

53

W = R(1− cosβ(2 cosα − 1)(d,θ ) : d + R(1 − cosθ ) < W

≤ γ L = (2γ RR ) (R − R )

CiD

i≤ 1∑

© June 26, 2009 , P. R. Kumar

36/47

An application: Intelligent Intersections   Stop signs and Traffic lights are wasteful

–  At suburban or rural intersections –  At night, when traffic load is very low

  Goal: Intelligent intersections where vehicles coordinate their movement through a combination of centralized and distributed real-time decision-making

–  Lower fuel consumption –  Lower traffic delays –  Greater throughput

  Assign time-slots to each car and design distributed algorithm which ensures safety

  “Electronic equivalent of a Traffic light”

Kowshik, Caveney & K ʻ2007

© June 26, 2009 , P. R. Kumar

37/47

Intelligent Intersections   Stop signs &Traffic lights wasteful

–  Suburban or rural intersections –  At night, when traffic light

  Goal: Intelligent intersections   Cars negotiate with intersection

–  Lower fuel consumption –  Lower traffic delays –  Greater safety

(Kowshik, Caveney & K ʼ05)

© June 26, 2009 , P. R. Kumar

38/33

But are we solving the right problem?

  Human beings seem to have decided to dedicate one hour per day for travel

  So should we we actually make it easier for people to travel greater distances in the same time?

From Frank Kellyʼs talk at StochNet 2006

© June 26, 2009 , P. R. Kumar

39/47

The oncoming theoretical convergence

  1950 — 2000 and continuing: Substantial progress in several individual disciplines

–  Computation: ENIAC (1946), von Neumann (1944), Turing,.. –  Sensing and inference: Fisher, Wiener (1949),… –  Actuation/Control: Bode, Kalman (1960),… –  Communication: Shannon (1948), Nyquist,… –  Signal Processing: FFT, Cooley-Tukey (1965),…

  2000 — onwards –  A gradual fusion of all these fields –  But still knowledge of all these fields may be important –  Pedagogical as well as research challenges

  Larger grand unification of sensing, actuation, communication and computation

Post Maxwell,von Neumann,

Shannon, Bardeen-Brattain world

Age of system building Nodes can Compute Communicate Sense and Actuate

© June 26, 2009 , P. R. Kumar

References-1   Scott Graham, Girish Baliga and P. R. Kumar, “Abstractions,

Architecture, Mechanisms, and a Mid- dleware for Networked Control.” To appear in IEEE Transactions on Automatic Control, August 2009.

  G. Baliga, S. Graham, L. Sha, and P. R. Kumar, “Service continuity in networked control using etherware.” IEEE Distributed Systems Online, 15 pages, vol. 5, No. 8, September 2004. ISSN: 1541- 4922. http://csdl2.computer.org/comp/mags/ds/2004/09/o9002.pdf

  Girish Baliga and P. R. Kumar, Middleware Architecture for Federated Control Systems. Presented at Midd leware 2003, Rio de Janeiro, Brazil, June 16–June 20, 2003. Published in IEEE Distributed Systems online, June 2003. Link: http : dsonline.computer.org0306f balrint.htm

  Scott Graham and P. R. Kumar, “The Convergence of Control, Communication, and Computation,” Proceedings of PWC 2003: Personal Wireless Communication, Lecture Notes in Computer Science, vol. 2775/2003, Springer-Verlag, Heidelberg, Germany, pp. 458–475, 2003.

40/47

© June 26, 2009 , P. R. Kumar

References-3   Girish Baliga, Scott Graham, Lui Sha, and P. R. Kumar,“EtherWare:

Domainware for Wireless Control Networks,” Proceedings of The 7th IEEE International Symposium on Object-oriented Real-time Distributed Computing, Vienna, Austria, pp. 155-162, May 12-14, 2004.

  Scott Graham, Girish Baliga, and P. R. Kumar, “Issues in the Convergence of Control with Communication and Computing: Proliferation, Architecture, Design, Services, and Middleware,” Proceedings of the 43rd IEEE Conference on Decision and Control, Paradise Island, Bahamas, pp. 1466–1471, December 14-17, 2004.

  C. L. Robinson, G. Baliga, S. Graham, and P. R. Kumar, “Design Patterns for Robust and Evolv- able Networked Control,” Proceedings of Third Annual Conference on Systems Engineering Research (CSER) 2005, Hoboken, NJ, March 23-25, 2005.

  Girish Baliga and P. R. Kumar, “A Middleware for Control over Networks,” Proceedings of the 44th IEEE Conference on Decision and Control, pp. 482–487, Seville, Spain, Dec. 12–15, 2005.

41/47

© June 26, 2009 , P. R. Kumar

References-4   Girish Baliga, Scott Graham and P. R. Kumar, “Middleware and

Abstractions in the Convergence of Control with Communication and Computation,” Proceedings of the 44th IEEE Conference on Decision and Control, pp. 4245–4250, Seville, Spain, Dec. 12–15, 2005.

  Girish Baliga , Scott Graham , Carl A. Gunter and P. R. Kumar, “Reducing Risk by Managing Software Related Failures in Networked Control Systems” Proceedings of the 45th IEEE Conference on Decision and Control, , pp. 2866–2871, San Diego, Dec. 13–15, 2006.

  C. L. Robinson, H.-J. Schuetz, G. Baliga and P. R. Kumar, “Architecture and Algorithm for a Labora- tory Vehicle Collision Avoidance System.” 2007 IEEE International Symposium on Intel ligent Control (ISIC2007), Singapore, October 1-3, 2007.

  Craig Robinson and P. R. Kumar, “Optimizing Controller Location in Networked Control Systems with Packet Drops.” IEEE Journal on Selected Areas in Communications, Special Issue on Control and Communications, pp. 661–671, vol. 26, no. 4, May 2008.

42/47

© June 26, 2009 , P. R. Kumar

References-5   Tanya Crenshaw, C. L. Robinson, Hui Ding, P. R. Kumar, Lui Sha, “A

Pattern for Adaptive Behavior in Safety-Critical, Real-Time Middleware.” Proceedings of the 27th IEEE Real-Time Systems Symposium, pp. 127–13, Rio de Janeiro, Brazil, December 5–8, 2006.

  Craig L. Robinson and P. R. Kumar, “Control over Networks of Unreliable Links: Controller Location and Performance Bounds.” Proceedings of ConCom: Control over Communication Channels, Limassol, Cyprus, April 20, 2007.

  C. L. Robinson and P. R. Kumar, “Sending the Most Recent Observation is not Optimal in Networked Control: Linear Temporal Coding and Towards the Design of a Control Specific Transport Protocol.” Proceedings of the 46th IEEE Conference on Decision and Control, pp. 334–339, New Orleans, Dec. 12– 14, 2007.

  Tanya Crenshaw, Elsa Gunter, C. L. Robinson, Lui Sha and P. R. Kumar, “The Simplex Reference Model: Limiting Fault-Propagation due to Unreliable Components.” Proceedings of the Real-Time Systems Symposium 2007 (RTSS 2007), pp. 400–412, Tucson, Dec. 3–6, 2007.

43/47

© June 26, 2009 , P. R. Kumar

References-6   C. L. Robinson and P. R. Kumar, “Networked Control Systems with

Packet Delays and Losses.” Proceedings of the 47th IEEE Conference on Decision and Control, pp. 4602–4607, Cancun, Dec. 9–11, 2008.

  Kyoung-Dae Kim and P. R. Kumar, “Architecture and Mechanism Design for Real-Time and Fault- Tolerant Etherware for Networked Control.” Proceedings of the 17th World Congress, The International Federation of Automatic Control, pp. 9421–9426, Seoul, Korea, July 6-11, 2008.

  Scott R. Graham, “Issues in the Convergence of Control with Communication and Computation,” Ph. D. Thesis, July 8, 2004, University of Illinois, Urbana-Champaign.

  Girish Bantwal Baliga, “A Middleware Framework for Networked Control Systems,” Ph. D. Thesis, August 12, 2005, Department of Computer Science, University of Illinois, Urbana-Champaign.

  Girish Bantwal Baliga, “A Software Architecture for Federated Control Systems,” December 2002. MS Thesis. Department of Computer Science, University of Illinois, Urbana-Champaign.

44/47

© June 26, 2009 , P. R. Kumar

References-7   Craig Robinson, “Policies and Mechanisms for Networked Control

Systems,” Ph. D. Thesis, February 19, 2008. Department of Industrial and Enterprise Systems Engineering, University of Illinois, Urbana-Champaign.

45/47

fundamental issues in networked control systems –...

Documents