InvariantSets

InvariantSets.jl brings some of the set computational functionalities provided by MATLAB plugin Multi-Parametric Toolbox 3 (MPT) to the Julia Programming Language and extends it with lazy set computation features. This package enables the user to compute, approximate and display invariant sets using a similar notation as in MPT but without needing a costly MATLAB license.

InvariantSets.jl builds upon LazySets.jl which provides "lazy" and concrete set computation (with Polyhedra.jl and CDDLib.jl as polyhedral computation backend). As a result, if needed, the full computational power of LazySets.jl can be leveraged if needed.

Installation

Content

Set Types

Set Type	Constructor	Description
HPolygon/HPolytope/HPolyhedron	HPolytope(A, b)	P = {x∈ℝⁿ: Ax≤b}
VPolygon/VPolytope/VPolyhedron	VPolytope(V)	V∈ℝⁿˣᵐ
Ballp/ Ball1 / Ball2/ BallInf	Ball1(center, radius)
Ellipsoid	Ellipsoid(center, shape_matrix)
Zonotope	Zonotope(center, generator)
Hyperrectangle	Hyperrectangle(center, radius)
	Hyperrectangle(low=min, high=max)
Hyperplane / HalfSpace	HalfSpace(a,b)
Singleton	Singleton(element)
SingleEntryVector	SingleEntryVector(dim, idx, value)
Interval	Interval(start, end)

Concrete Set Operations

In LazySets.jl the common operator +, * and ∩ correspond to lazy set operations. This is the same in InvariantSets.jl, in addition, the concrete set operators +ᶜ, -ᶜ, *ᶜ and ∩ᶜ. In most IDEs, the superscript ᶜ can be written with \^c → tab.

using Polyhedra
using CDDLib

A = [1.0  0.0;
     0.0  1.0;
    -1.0  0.0;
     0.0 -1.0;
     1.0 1.0]
b = [1.0, 2.0, 3.0, 4.0, 1.0]
origin = [0.0, 0.0]
polygon = HPolygon(A,b)
ball = Ball1(origin, 1.3)
ball2 = Ball2(origin, 1.0)
A = [2.0 0.0; 0.0 0.5]
# Scaling
ball_scaled = 2.0 *ᶜ ball
# Linear Map
ball_lm = A *ᶜ ball
# Translation of ball
ball_trans = [2.0,2.0] +ᶜ ball
# Minkowski Sum
sum = polygon +ᶜ ball
# Pontryagin Difference
diff = polygon -ᶜ ball2
# polygon_after_diff = diff + ball2
# Intersection
intersection = polygon ∩ᶜ ball
# Reflection
polygon_reflected = reflect(polygon)
# Minkowski Difference
diff_minsk = ball_trans +ᶜ reflect(polygon)
# chebyshev center
c, r = InvariantSets.chebyshev_center(polygon,  get_radius=true)
chebyball = Ball2(c, r)

Invariant Sets & Control Systems

Name	Constructor	Description
preset	preset(A, X)
state_constraints	state_constraints(X, U, K)
maximum_invariant_set	maximum_invariant_set(A, X, N)
maximum_control_invariant_set	maximum_control_invariant_set(A, B, X, U, N)
terminal_set	terminal_set(A, B, X, U, K)
feasible_set	feasible_set(A, B, X, U, Xf, N)
maximal_RPI_set
minimal_RPI_set
tightened_constraints

Integration with MathematicalSystems.jl

A = [0.9 0.5; 0 0.9]
B = [1., 0]
X = BallInf(zeros(2), 10.)
U = BallInf(zeros(1), 1.)

K = [0.1 0.1]

autSys = @system x⁺ = A*x  x∈X
ctrlSys = @system x⁺ = A*x + B*u x∈X u∈U

IS = maximum_invariant_set(autSys)
MIL = maximum_invariant_set(ctrlSys, K)
MCI = maximum_control_invariant_set(ctrlSys)

Integration with JuMP.jl

using InvariantSets
using JuMP
m = Model()
@variable(m, X[1:2, 1:5])
@variable(m, U[1:1, 1:2]) # U[1,1:5] does not work
constru= BallInf(zeros(size(U,1)), 2.0)
constrx = HPolyhedron([1 -2.], [1.])
InvariantSets.add_constraint!(m, X, constrx)
InvariantSets.add_constraint!(m, U, constru)
# Feasibility
# Subject to
# X[1,1] - 2 X[2,1] <= 1.0
# X[1,2] - 2 X[2,2] <= 1.0
# X[1,3] - 2 X[2,3] <= 1.0
# X[1,4] - 2 X[2,4] <= 1.0
# X[1,5] - 2 X[2,5] <= 1.0
# U[1,1] <= 2.0
# -U[1,1] <= 2.0
# U[1,2] <= 2.0
# -U[1,2] <= 2.0

Comparison to MATLAB and MPT

Geometric operations with polyhedra

Matlab and MPT

P1 = Polyhedron( 'A', [1 -2.1 -0.5; 0.8 -3.1 0.9; -1.2 0.4 -0.8], 'b', [1; 4.9; -1.8])
P1.isEmptySet()
P1.isBounded()
P1.plot()

P2 = Polyhedron('lb', [-1; -2], 'ub', [3; 4])
P2.computeVRep
P2.V

P3 = Polyhedron([4, -1; 4, 5; 8, 3])
P3.computeHRep

V = [ -1.7 -0.4; -0.4  0.7; 1.2 -0.8; 0 0.8; 1.3 0.9; -0.3 0.6];
P4 = Polyhedron(V);
x0 = [0; 0];
P4.contains( x0 )
x1 = [3; 0];
P4.contains( x1 )

P5 = Polyhedron([ 1.8  -4.8; -7.2 -3.4; -4.2 1.2; 5.8  2.7]);
data = P5.chebyCenter()

Julia and InvariantSets

P1 = Polyhedron([1 -2.1 -0.5; 0.8 -3.1 0.9; -1.2 0.4 -0.8],  [1; 4.9; -1.8])
isempty(P1) # or P1 |> isempty
isbounded(P1)
plot(P1)

P2 = Hyperrectangle(low=[-1, -2], high=[3,4])
P2hrep = convert(HPolytope, P2)
P2vrep =  tovrep(P2hrep)
P2vrep.vertices

P3vrep = VPolytope([4 -1; 4 5; 8 3.]);
P3hrep = HPolytope(tosimplehrep(P3vrep)...)

V = [ -1.7 -0.4; -0.4  0.7; 1.2 -0.8; 0 0.8; 1.3 0.9; -0.3 0.6];
P4 = VPolytope(V');
[0., 0.] ∈ P4
[3., 0.] ∈ P4

P5 = VPolytope([ 1.8  -4.8; -7.2 -3.4; -4.2 1.2; 5.8  2.7]');
chebyshev_center(P5)

Maximum Control Invariant Sets

Matlab and MPT:

% computes a control invariant set for LTI system x^+ = A*x+B*u
system = LTISystem('A', [1 1; 0 0.9], 'B', [1; 0.5]);
system.x.min = [-5; -5];
system.x.max = [5; 5];
system.u.min = -1;
system.u.max = 1;
InvSet = system.invariantSet()
InvSet.plot()

Julia:

using InvariantSets, Polyhedra, CDDLib
using MathematicalSystems
using Plots
A = [1 1; 0 0.9]
B = [1; 0.5]
X = Hyperrectangle(low=[-5, -5], high=[5, 5])
U = Hyperrectangle(low=[-1], high=[1])
system = @system x⁺ = A*x + B*u x∈X u∈U
InvSet = maximum_control_invariant_set(system)
plot(InvSet)