Title: MEETING THE NEED FOR ROBUSTIFIED NONLINEAR SYSTEM THEORY CONCEPTS
1MEETING THE NEED FOR ROBUSTIFIED NONLINEAR
SYSTEM THEORY CONCEPTS
Daniel Liberzon
Coordinated Science Laboratory and Dept. of
Electrical Computer Eng., Univ. of Illinois at
Urbana-Champaign
Plenary lecture, ACC, Seattle, 6/13/08
1 of 36
2TALK OVERVIEW
Context
Concept
2 of 36
3TALK OVERVIEW
Context
Concept
Stability of switched systems
Observability
Minimum-phase property
Adaptive control
2 of 36
4INFORMATION FLOW in CONTROL SYSTEMS
Plant
Controller
3 of 36
5INFORMATION FLOW in CONTROL SYSTEMS
- Limited communication capacity
- Need to minimize information transmission
3 of 36
6BACKGROUND
Previous work
Brockett, Delchamps, Elia, Mitter, Nair, Savkin,
Tatikonda, Wong,
- Deterministic stochastic models
- Tools from information theory
- Mostly for linear plant dynamics
- Unified framework for
- quantization
- time delays
- disturbances
4 of 36
7OUR APPROACH
(Goal treat nonlinear systems handle
quantization, delays, etc.)
Caveat This doesnt work in general, need
robustness from controller
5 of 36
8QUANTIZATION
6 of 36
9QUANTIZATION and INPUT-to-STATE STABILITY
7 of 36
10QUANTIZATION and INPUT-to-STATE STABILITY
assume glob. asymp. stable (GAS)
7 of 36
11QUANTIZATION and INPUT-to-STATE STABILITY
no longer GAS
7 of 36
12QUANTIZATION and INPUT-to-STATE STABILITY
7 of 36
13QUANTIZATION and INPUT-to-STATE STABILITY
7 of 36
14QUANTIZATION and INPUT-to-STATE STABILITY
quantization error
Assume
Solutions that start in enter and remain
there
This is input-to-state stability (ISS) w.r.t.
measurement errors
7 of 36
15LINEAR SYSTEMS
8 of 36
16DYNAMIC QUANTIZATION
9 of 36
17DYNAMIC QUANTIZATION
Hybrid quantized control is discrete state
9 of 36
18DYNAMIC QUANTIZATION
Hybrid quantized control is discrete state
9 of 36
19DYNAMIC QUANTIZATION
zooming variable
Zoom out to overcome saturation
9 of 36
20DYNAMIC QUANTIZATION
After the ultimate bound is achieved, recompute
partition for smaller region
Can recover global asymptotic stability
9 of 36
21QUANTIZATION and DELAY
Architecture-independent approach
Delays possibly large
Based on the work of Teel
10 of 36
22QUANTIZATION and DELAY
11 of 36
23SMALL GAIN ARGUMENT
12 of 36
24FINAL RESULT
Need
13 of 36
25FINAL RESULT
Need
small gain true
13 of 36
26FINAL RESULT
Need
small gain true
Can use zooming to improve convergence
13 of 36
27State quantization and completely unknown
disturbance
14 of 36
28State quantization and completely unknown
disturbance
14 of 36
29State quantization and completely unknown
disturbance
After zoom-in
Issue disturbance forces the state outside
quantizer range
Must switch repeatedly between zooming-in and
zooming-out
Result for linear plant, can achieve ISS w.r.t.
disturbance
(ISS gains are nonlinear although plant is
linear cf. Martins)
14 of 36
30TALK OVERVIEW
Context
Concept
Input-to-state stability
Stability of switched systems
Observability
Minimum-phase property
Adaptive control
15 of 36
31OBSERVABILITY and ASYMPTOTIC STABILITY
Barbashin-Krasovskii-LaSalle theorem
is glob. asymp. stable (GAS) if s.t.
16 of 36
32SWITCHED SYSTEMS
17 of 36
33Theorem (common weak Lyapunov function)
Switched linear system is GAS if
Want to handle nonlinear switched systems and
nonquadratic weak Lyapunov functions
Need a suitable nonlinear observability notion
18 of 36
34OBSERVABILITY MOTIVATING REMARKS
Several ways to define observability (equivalent
for linear systems)
Benchmarks
- observer design or state norm estimation
- detectability vs. observability
- LaSalles stability theorem for switched systems
Joint work with Hespanha, Sontag, and Angeli
No inputs here, but can extend to systems with
inputs
19 of 36
35STATE NORM ESTIMATION
This is a robustified version of
0-distinguishability
20 of 36
36OBSERVABILITY DEFINITION 1 A CLOSER LOOK
21 of 36
37DETECTABILITY vs. OBSERVABILITY
22 of 36
38OBSERVABILITY DEFINITION 2 A CLOSER LOOK
Definition
and
Theorem This is equivalent to definition 1
(small-time obs.)
23 of 36
39STABILITY of SWITCHED SYSTEMS
Theorem (common weak Lyapunov function)
Switched system is GAS if
24 of 36
40MULTIPLE WEAK LYAPUNOV FUNCTIONS
Example Popovs criterion for feedback systems
25 of 36
41TALK OVERVIEW
Context
Concept
Control with limited information
Input-to-state stability
Stability of switched systems
Observability
26 of 36
42MINIMUM-PHASE SYSTEMS
Robustified version LSontagMorse
output-input stability
- implies minimum-phase for nonlinear systems
(when applicable)
- reduces to minimum-phase for linear systems
(SISO and MIMO)
27 of 36
43UNDERSTANDING OUTPUT-INPUT STABILITY
OSS (detectability)
w.r.t. extended output,
28 of 36
44CHECKING OUTPUT-INPUT STABILITY
For systems affine in controls, can use structure
algorithm for left-inversion to check the
input-bounding property
Input bounding
OUTPUT-INPUT STABLE
Detectability
29 of 36
45CHECKING OUTPUT-INPUT STABILITY
For systems affine in controls, can use structure
algorithm for left-inversion to check the
input-bounding property
Example 2
Output-input stability allows to distinguish
between the two examples
30 of 36
46 FEEDBACK DESIGN
Example global normal form
Output-input stability guarantees closed-loop GAS
It is stronger than minimum-phase ISS internal
dynamics
In general, global normal form is not needed
31 of 36
47 CASCADE SYSTEMS
For linear systems, recover usual detectability
32 of 36
48ADAPTIVE CONTROL
Linear systems
33 of 36
49ADAPTIVE CONTROL
Plant
Design
Controller
model
Nonlinear systems
- plant is minimum-phase
- system inside the box is output-stabilized
- controller and design model are detectable
If
then the closed-loop system is detectable through
e (tunable Morse 92)
34 of 36
50ADAPTIVE CONTROL
Plant
Design
Controller
model
Nonlinear systems
- plant is output-input stable
- system inside the box is output-stabilized
- controller and design model are detectable
If
then the closed-loop system is detectable through
e (tunable Morse 92)
34 of 36
51ADAPTIVE CONTROL
Plant
Design
Controller
model
Nonlinear systems
- plant is output-input stable
- system in the box is output-stabilized
- controller and design model are detectable
If
then the closed-loop system is detectable through
e (tunable Morse 92)
34 of 36
52ADAPTIVE CONTROL
Plant
Design
Controller
model
Nonlinear systems
34 of 36
53TALK SUMMARY
Context
Concept
35 of 36
54ACKNOWLEDGMENTS
Special thanks go to
- Colleagues, students and staff at UIUC
- Financial support from NSF and AFOSR
36 of 36