Controllability Reachability Invariance - Handsout

Model Predictive Control
Reachability and Invariance
F. Borrelli∗ , M. Morari, C. Jones†
∗
UC Berkeley
Institut für Automatik

ETH Zürich
†
EPFL
Fall Semester 2014 (revised September 2014)
Table of Contents
1. Polyhedra and Polytopes

1.1 General Set Definitions and Operations
1.2 Basic Operations on Polytopes
2. Reachable Sets
2.1 Pre and Reach Sets Definition
2.2 Pre and Reach Sets Computation
2.3 Controllable Sets
2.4 N -Step Reachable Sets
3. Invariant Sets
3.1 Invariant Sets
3.2 Control Invariant Sets
Reachability and Invariance F. Borrelli, M. Morari, C. Jones - Fall Semester 2014 (revised September 2014) i
1 1. Polyhedra and Polytopes
Outline

2. Reachable Sets
3. Invariant Sets
Reachability and Invariance F. Borrelli, M. Morari, C. Jones - Fall Semester 2014 (revised September 2014) ii
1 1. Polyhedra and Polytopes 1.1 General Set Definitions and Operations
Table of Contents

Reachability and Invariance F. Borrelli, M. Morari, C. Jones - Fall Semester 2014 (revised September 2014)
Definitions (Polyhedra and polytopes)

A polyhedron is the intersection of a finite number of closed halfspaces:
Z = {z | a1> z ≤ b1 , a2> z ≤ b2 , . . . , am
>
z ≤ bm }
= {z | Az ≤ b}
where A := [a1 , a2 , . . . , am ]> and b := [b1 , b2 , . . . , bm ]> .
A polytope is a bounded polyhedron.

Polyhedra and polytopes are always convex.
z2 z2
z1 z1
An (unbounded) polyhedron A polytope

Reachability and Invariance F. Borrelli, M. Morari, C. Jones - Fall Semester 2014 (revised September 2014) 1-3
General Set Definitions and Operations
An n-dimensional ball pB(x0 , ρ) is the set

n
B(x0 , ρ) = {x ∈ R | kx − x0 k2 ≤ ρ}. x0 and ρ are the center and the
radius of the ball, respectively.
The convex combination of Pkx1 , . . . , xk is defined as the point
λ1 x1 + . . . + λk xk where i=1 λi = 1 and λi ≥ 0, i = 1, . . . , k.
The convex hull of a set K ⊆ Rn is the set of all convex combinations of
points in K and it is denoted as conv(K ):
conv(K ) ,{λ1 x1 + . . . + λk xk | xi ∈ K , λi ≥ 0, i = 1, . . . , k,
k
X
λi = 1}.
i=1
Polyhedra Representations
An H-polyhedron P in Rn denotes an intersection of a finite set of closed
halfspaces in Rn :
P = {x ∈ Rn : Ax ≤ b}
In Matlab: P = Polytope(A,b)
A two-dimensional H-polyhedron
a1x ·b1
a2x ·b2
a4x ·b4
a3x ·b3
a5x ·b5
Inequalities which can be removed without changing the polyhedron are called
redundant. The representation of an H-polyhedron is minimal if it does not
contain redundant inequalities.
A V-polytope P in Rn is defined as
P = conv(V )
for some V = [V1 , . . . , Vk ] ∈ Rn×k .

Any H-polytope is a V-polytope and viceversa.
A polytope P ⊂ Rn , is full-dimensional if it is possible to fit a non-empty
n-dimensional ball in P
If kAi k2 = 1, where Ai denotes the i-th row of a matrix A, we say that the
polytope P is normalized.
The faces of dimension 0 and 1 are called vertices and edges, respectively.
10 10
8 8
6 6
4 4
2 2
x2
x2
0 0
−2 −2
−4 −4
−6 −6
−8 −8
−10 −10
−10 −8 −6 −4 −2 0 2 4 6 8 10 −10 −8 −6 −4 −2 0 2 4 6 8 10
x1 x1
(a) V-representation. (b) H-representation.
Polytopal Complexes
A set C ⊆ Rn is called a P-collection (in Rn ) if it is a collection of a finite number

of n-dimensional polytopes, i.e.
C = {Ci }N
i=1 ,
C
where Ci := {x ∈ Rn : Cix x ≤ Cic }, dim(Ci ) = n, i = 1, . . . , NC , with NC < ∞.
In Matlab: Q = [P1, P2, P3], R = [P4, Q, [P5, P6], P7]
Functions on Polytopal Complexes
A function h(θ) : Θ → Rk , where Θ ⊆ Rs , is piecewise affine (PWA) if there

exists a strict partition R1 ,. . . ,RN of Θ and h(θ) = H i θ + k i , ∀θ ∈ Ri ,
i = 1, . . . , N .
A function h(θ) : Θ → Rk , where Θ ⊆ Rs , is piecewise affine on polyhedra
(PPWA) if there exists a strict polyhedral partition R1 ,. . . ,RN of Θ and
h(θ) = H i θ + k i , ∀θ ∈ Ri , i = 1, . . . , N .
A function h(θ) : Θ → R, where Θ ⊆ Rs , is piecewise quadratic (PWQ) if
there exists a strict partition R1 ,. . . ,RN of Θ and h(θ) = θ0 H i θ + k i θ + l i ,
∀θ ∈ Ri , i = 1, . . . , N .
A function h(θ) : Θ → R, where Θ ⊆ Rs , is piecewise quadratic on polyhedra
(PPWQ) if there exists a strict polyhedral partition R1 ,. . . ,RN of Θ and
h(θ) = θ0 H i θ + k i θ + l i , ∀θ ∈ Ri , i = 1, . . . , N .
1 1. Polyhedra and Polytopes 1.2 Basic Operations on Polytopes
Table of Contents

Basic Operations on Polytopes
Convex Hull of a set of points V = {Vi }N n

i=1 , with Vi ∈ R ,
V
NV
X NV
X
n
conv(V ) = {x ∈ R :x= αi Vi , 0 ≤ αi ≤ 1, αi = 1}. (1)
i=1 i=1
In Matlab: P=hull(V ), V matrix containing vertices of the polytope P

Vertex Enumeration of a polytope P given in H-representation. (dual of the
convex hull operation)
In Matlab: V=extreme(P)
Used to switch from a V-representation of a polytope to an H-representation.

Polytope reduction is the computation of the minimal representation of a
polytope. A polytope P ⊂ Rn , P = {x ∈ Rn : Ax ≤ b} is in a minimal
representation if the removal of any row in Ax ≤ b would change it (i.e., if
there are no redundant constraints).
In Matlab: P = Polytope(A,b,normal,minrep), minrep=1

The Chebychev Ball of a polytope P corresponds to the largest radius ball
B(xc , R) with center xc , such that B(xc , R) ⊂ P.
In Matlab: P.xCheb, P.rCheb

10
2
x2
−2
−4
−6
−8
−10
−10 −8 −6 −4 −2 0 2 4 6 8 10
x1

Projection Given a polytope
P = {[x 0 y 0 ]0 ∈ Rn+m : Ax x + Ay y ≤ b} ⊂ Rn+m the projection onto the
x-space Rn is defined as
projx (P) := {x ∈ Rn | ∃y ∈ Rm : Ax x + Ay y ≤ b}.
In Matlab: Q = projection(P,dim)
Affine Mappings and Polyhedra
Consider a polyhedron P = {x ∈ Rn | Hx ≤ k}, with H ∈ RnP ×n and an

affine mapping f (z)
f : z ∈ Rn 7→ Az + b, A ∈ Rn×n , b ∈ Rn
Define the composition of P and f as the following polyhedron
P ◦ f , {z ∈ Rn | Hf (z) ≤ k} = {z ∈ Rm | HAz ≤ k − Hb}
Useful for backward-reachability
Affine Mappings and Polyhedra

Consider a polyhedron P = {x ∈ Rn | Hx ≤ k}, with H ∈ RnP ×n and an
affine mapping f (z)
f : z ∈ Rn 7→ Az + b, A ∈ Rn×n , b ∈ Rn
Define the composition of f and P as the following polyhedron
f ◦ P , {y ∈ Rn | y = Ax + b ∀x ∈ Rn , Hx ≤ k}
The polyhedron f ◦ P in can be computed as follows. Write P in

V-representation P = conv(V ) and map the vertices V = {V1 , . . . , Vk }
through the transformation f . Because the transformation is affine, the set
f ◦ P is the convex hull of the transformed vertices
f ◦ P = conv(F ), F = {AV1 + b, . . . , AVk + b}.
Useful for forward-reachability
2 2. Reachable Sets
Outline
2. Reachable Sets
3. Invariant Sets
2 2. Reachable Sets 2.1 Pre and Reach Sets Definition
Table of Contents
2. Reachable Sets
Set Definition
We consider the following two types of systems autonomous systems:
x(t + 1) = fa (x(t)), (2)
and systems subject to external inputs:
x(t + 1) = f (x(t), u(t)). (3)
Both systems are subject to state and input constraints
x(t) ∈ X , u(t) ∈ U, ∀ t ≥ 0.
The sets X and U are polyhedra and contain the origin in their interior.
Reach Set Definition
For the autonomous system (2) we denote the one-step reachable set as
Reach(S) , {x ∈ Rn : ∃ x(0) ∈ S s.t. x = fa (x(0))}
For the system (3) with inputs we denote the one-step reachable set as
Reach(S) , {x ∈ Rn : ∃ x(0) ∈ S, ∃ u(0) ∈ U s.t. x = f (x(0), u(0))}
Pre Set Definition
“Pre” sets are the dual of one-step reachable sets. The set
Pre(S) , {x ∈ Rn : fa (x) ∈ S}
defines the set of states which evolve into the target set S in one time step for the
system (2).
Similarly, for the system (3) the set of states which can be driven into the target
set S in one time step is defined as
Pre(S) , {x ∈ Rn : ∃u ∈ U s.t. f (x, u) ∈ S}
2 2. Reachable Sets 2.2 Pre and Reach Sets Computation
Table of Contents
2. Reachable Sets
Pre Set Computation - Autonomous Systems
Assume the system is linear and autonomous
x(t + 1) = Ax(t)
Let
S = {x : Hx ≤ h}, (4)
Then the set Pre(S) is
Pre(S) = {x : HAx ≤ h}
Note that by using polyhedral notation, the set Pre(S) is simply S ◦ A.
Reach Set Computation - Autonomous Systems
The set Reach(S) is obtained by applying the map A to the set S.

Write S in V-representation
S = conv(V ) (5)
and map the set of vertices V through the transformation A.
Because the transformation is linear, the reach set is simply the convex hull of the
transformed vertices
Reach(S) = A ◦ S = conv(AV ) (6)
Pre Set Computation - System with Inputs
Consider the system

x(t + 1) = Ax(t) + Bu(t)
Let
S = {x | Hx ≤ h}, U = {u | Hu u ≤ hu }, (7)
The Pre set is

n HA HB x h
Pre(S) = x ∈ R | ∃u ∈ R s.t. ≤
0 Hu u hu
which is the projection onto the x-space (with dimension Rn ) of the polyhedron

HA HB x h
T := { ≤ }.
0 Hu u hu
In Matlab: Q = projection(T ,n)
2 2. Reachable Sets 2.3 Controllable Sets
Table of Contents
2. Reachable Sets
2 2. Reachable Sets 2.3 Controllable Sets
Contorllable Sets
Definition (N -Step Controllable Set KN (O))
For a given target set O ⊆ X , the N -step controllable set KN (O) is defined as:
KN (O) , Pre(KN −1 (O)) ∩ X , K0 (O) = O, N ∈ N+ .
All states x0 ∈ KN (O) can be driven,through a time-varying control law, to the

target set O in N steps, while satisfying input and state constraints.
Definition (Maximal Controllable Set K∞ (O))

For a given target set O ⊆ X , the maximal controllable set K∞ (O) for the
system x(t + 1) = f (x(t), u(t)) subject to the constraints x(t) ∈ X , u(t) ∈ U is
the union of all N -step controllable sets contained in X (N ∈ N).
2 2. Reachable Sets 2.4 N -Step Reachable Sets
Table of Contents
2. Reachable Sets
2 2. Reachable Sets 2.4 N -Step Reachable Sets
N -Step Reachable Sets
Definition (N -Step Reachable Set RN (X0 ))

For a given initial set X0 ⊆ X , the N -step reachable set RN (X0 ) is
Ri+1 (X0 ) , Reach(Ri (X0 )), R0 (X0 ) = X0 , i = 0, . . . , N − 1
All states x0 ∈ X0 can will evolve to the N -step reachable set RN (X0 ) in N steps
Same definition of Maximal Reachable Set R∞ (X0 ) can be introduced.
3 3. Invariant Sets
Outline
2. Reachable Sets
3. Invariant Sets
3.1 Invariant Sets
3 3. Invariant Sets 3.1 Invariant Sets
Table of Contents
3. Invariant Sets
3.1 Invariant Sets
Invariant Sets
Invariant sets
are computed for autonomous systems
for a given feedback controller u = g(x), provide the set of initial states
whose trajectory will never violate the system constraints.
Definition (Positive Invariant Set)

A set O ⊆ X is said to be a positive invariant set for the autonomous
system x(t + 1) = fa (x(t)) subject to the constraints x(t) ∈ X , if
x(0) ∈ O ⇒ x(t) ∈ O, ∀t ∈ N+
Definition (Maximal Positive Invariant Set O∞ )

The set O∞ is the maximal invariant set if O∞ is invariant and O∞ contains all
the invariant sets contained in X .
Invariant Sets
Algorithm
Input: fa , X
Output: O∞
1 let Ω0 = X ,
2 let Ωk+1 = Pre(Ωk ) ∩ Ωk
3 if Ωk+1 = Ωk then O∞ ← Ωk+1
4 else go to 2
The algorithm generates the set sequence {Ωk } satisfying Ωk+1 ⊆ Ωk , ∀k ∈ N and
it terminates when Ωk+1 = Ωk so that Ωk is the maximal positive invariant set
O∞ for x(t + 1) = fa (x(t)).
3 3. Invariant Sets 3.2 Control Invariant Sets
Table of Contents
3. Invariant Sets
3.1 Invariant Sets
Control Invariant Sets
Control invariant sets

are computed for systems subject to external inputs
provide the set of initial states for which there exists a controller such that
the system constraints are never violated.
Definition (Control Invariant Set)

A set C ⊆ X is said to be a control invariant set if
x(t) ∈ C ⇒ ∃u(t) ∈ U such that f (x(t), u(t)) ∈ C, ∀t ∈ N+
Definition (Maximal Control Invariant Set C∞ )

The set C∞ is said to be the maximal control invariant set for the system
x(t + 1) = f (x(t), u(t)) subject to the constraints in x(t) ∈ X , u(t) ∈ U, if it is
control invariant and contains all control invariant sets contained in X .
Control Invariant Sets

Same geometric condition for control invariants holds: C is a control invariant set
if and only if
C ⊆ Pre(C) (8)
Algorithm
Input: f , X and U
Output: C∞
1 let Ω0 = X ,
2 let Ωk+1 = Pre(Ωk ) ∩ Ωk
3 if Ωk+1 = Ωk then C∞ ← Ωk+1
4 else go to 2
The algorithm generates the set sequence {Ωk } satisfying Ωk+1 ⊆ Ωk , ∀k ∈ N and
it terminates if Ωk+1 = Ωk so that Ωk is the maximal control invariant set C∞ for
the constrained system.
Invariant Sets and Control Invariant Sets
The set O∞ (C∞ ) is finitely determined if and only if ∃ i ∈ N such that

Ωi+1 = Ωi .
The smallest element i ∈ N such that Ωi+1 = Ωi is called the
determinedness index.
For all states contained in the maximal control invariant set C∞ there exists a
control law, such that the system constraints are never violated.

Controllability Reachability Invariance - Handsout

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Controllability Reachability Invariance - Handsout

Caricato da

Copyright:

Formati disponibili

Model Predictive Control

Reachability and Invariance

F. Borrelli∗ , M. Morari, C. Jones†

Institut für Automatik

Fall Semester 2014 (revised September 2014)

1. Polyhedra and Polytopes

1. Polyhedra and Polytopes

1 1. Polyhedra and Polytopes 1.1 General Set Definitions and Operations

1. Polyhedra and Polytopes

Definitions (Polyhedra and polytopes)

where A := [a1 , a2 , . . . , am ]> and b := [b1 , b2 , . . . , bm ]> .

A polytope is a bounded polyhedron.

An (unbounded) polyhedron A polytope

1 1. Polyhedra and Polytopes 1.1 General Set Definitions and Operations

General Set Definitions and Operations

An n-dimensional ball pB(x0 , ρ) is the set

1 1. Polyhedra and Polytopes 1.1 General Set Definitions and Operations

for some V = [V1 , . . . , Vk ] ∈ Rn×k .

(a) V-representation. (b) H-representation.

1 1. Polyhedra and Polytopes 1.1 General Set Definitions and Operations

A set C ⊆ Rn is called a P-collection (in Rn ) if it is a collection of a finite number

where Ci := {x ∈ Rn : Cix x ≤ Cic }, dim(Ci ) = n, i = 1, . . . , NC , with NC < ∞.

In Matlab: Q = [P1, P2, P3], R = [P4, Q, [P5, P6], P7]

Functions on Polytopal Complexes

A function h(θ) : Θ → Rk , where Θ ⊆ Rs , is piecewise affine (PWA) if there

1 1. Polyhedra and Polytopes 1.2 Basic Operations on Polytopes

1. Polyhedra and Polytopes

Basic Operations on Polytopes

Convex Hull of a set of points V = {Vi }N n

In Matlab: P=hull(V ), V matrix containing vertices of the polytope P

1 1. Polyhedra and Polytopes 1.2 Basic Operations on Polytopes

Basic Operations on Polytopes

In Matlab: P = Polytope(A,b,normal,minrep), minrep=1

In Matlab: P.xCheb, P.rCheb

Basic Operations on Polytopes

projx (P) := {x ∈ Rn | ∃y ∈ Rm : Ax x + Ay y ≤ b}.

1 1. Polyhedra and Polytopes 1.2 Basic Operations on Polytopes

Affine Mappings and Polyhedra

Consider a polyhedron P = {x ∈ Rn | Hx ≤ k}, with H ∈ RnP ×n and an

Define the composition of P and f as the following polyhedron

P ◦ f , {z ∈ Rn | Hf (z) ≤ k} = {z ∈ Rm | HAz ≤ k − Hb}

Useful for backward-reachability

Affine Mappings and Polyhedra

Define the composition of f and P as the following polyhedron

The polyhedron f ◦ P in can be computed as follows. Write P in

f ◦ P = conv(F ), F = {AV1 + b, . . . , AVk + b}.

Useful for forward-reachability

1. Polyhedra and Polytopes

2 2. Reachable Sets 2.1 Pre and Reach Sets Definition

We consider the following two types of systems autonomous systems:

x(t + 1) = fa (x(t)), (2)

and systems subject to external inputs:

x(t + 1) = f (x(t), u(t)). (3)

Both systems are subject to state and input constraints

Reach Set Definition

Reach(S) , {x ∈ Rn : ∃ x(0) ∈ S s.t. x = fa (x(0))}

Reach(S) , {x ∈ Rn : ∃ x(0) ∈ S, ∃ u(0) ∈ U s.t. x = f (x(0), u(0))}

2 2. Reachable Sets 2.1 Pre and Reach Sets Definition

Pre Set Definition

Pre(S) , {x ∈ Rn : ∃u ∈ U s.t. f (x, u) ∈ S}

2 2. Reachable Sets 2.2 Pre and Reach Sets Computation

Pre Set Computation - Autonomous Systems

Assume the system is linear and autonomous