Geometric

control
.
Introduction to geometric control theory
Dynamical
systems - controllability and Lie bracket -
control
systems

Reachable
sets and Ftima Silva Leite
controllability

Affine control Department of Mathematics Institute of Systems and Robotics

systems
INSTITUTO DE SISTEMAS
Departamento de Engenharia Electrotcnica e
Distributions Plo II * 3030 290 Coimbra * Portugal
Telef. +351 239 796 200/1 Fax. +351 239 4
and N. Cont. (VAT) 502 854 227
integrability

Main results

Optimal University of Coimbra, Portugal

control

Seminar of the Mathematics PhD Program

UCoimbra-UPorto
Coimbra, 10 November, 2010
 Nonlinear Geometric Control

Geometric
control
.

Dynamical Control theory is a theory that deals with influencing the

systems

control
behavior (controlling) of dynamical systems.
systems

Reachable Many processes in industries like robotics and

sets and
controllability aerospace industry have strong nonlinear dynamics.
Affine control
systems
The configuration spaces of many mechanical systems
Distributions
and are smooth manifolds (Lie groups, symmetric
integrability
spaces,...). Techniques from differential and
Main results

Optimal
Riemannian geometry are fundamental in modern
control control theory.
 The evolution ...

Geometric
control Beginning - 60s
.

Dynamical
systems

control
systems Roger Brockett
Reachable - the father -
sets and
controllability

Affine control
systems

Distributions
and Adultness - mid 70s
integrability

Main results Agrachev, Bloch, Crouch, Nijmeijer, Jurdjevic, Krener,

Optimal
control
Sachkov, Sontag, Sussmann, Van der Schaft, ...

The steam of publications has grown sharply in recent

years and gives every indication of continuing to grow...
 What is a dynamical system?

Geometric
control A dynamical system is a differential equation
.

Dynamical
x = f (x), x M.
systems

control
systems M is a smooth manifold.
Reachable
sets and
controllability f is a (smooth) vector field on M:
Affine control
systems
x M 7 f (x) Tx M (tangent space to M at x).
Distributions
and
integrability

Main results

Optimal is a solution of x = f (x) is an integral curve of f

control

A dynamical system is a vector field

 Vector fields and flows

Geometric Assume that the vector field f is complete (i.e., for all x0 M the
control
. solution x(t, x0 ) of the Cauchy problem x = f (x), x(0) = x0 , is
defined for all t R).
Dynamical
systems
Flow generated by the vector field f :
control
systems
t exp(t f ) : M M ,t R
Reachable
sets and x0 7 x(t, x0 )
controllability

Affine control
systems

Distributions
and
If x = f (x) describes the dynamics of a moving fluid in M, then
integrability exp(t f ) takes any particle of the fluid from a position x0 and
Main results moves it for a time t R to the position exp(t f )(x0 ) = x(t, x0 ).
Optimal
control

(If f not complete, flow is local)

 Dynamical systems

Geometric
control A dynamical system evolving on a smooth manifold M is a vector
.
field f (.) on M
Dynamical x(t) = f (x(t)), x(t) M.
systems

control
systems

Reachable
sets and
controllability

Affine control
systems

Distributions The dynamics of this system is determined by the flow of one

and
integrability vector field only. The future x(t, x0 ) is completely determined by
Main results the present state x0 .
Optimal
control
In order to affect (control) the dynamics we must consider
a family of vector fields.
 What is a control system?

Geometric
control A control system evolving on M is a family of vector fields f (., u)
.
on M, parameterized by the controls u.
Dynamical
systems x(t) = f (x(t), u(t)), x(t) M, u(t) U Rm
control
systems

Reachable
sets and
controllability

Affine control
systems

Distributions
and
integrability x is the state of the system, M is the state space
Main results
u is the input or control of the system
Optimal
control

In control theory we can change the dynamics of the control

system at any moment of time by changing the control u.
 Technical assumptions

Geometric
control x = f (x(t), u(t)), x(t) M, u(t) U Rm
.
The controls belong to a class U of admissible controls. Its choice
Dynamical depends on what the control system is modeling.
systems

control On the set of admissible controls:

systems
U contains all the piecewise constant functions with values in U,
Reachable
sets and which are piecewise continuous from the right.
controllability

Affine control Results on the continuity of solutions guarantee that

systems if a more general control function is approximated by
Distributions
and
piecewise constant functions, the solution of the control
integrability system for this class of admissible controls
Main results approximates the solution of the original system.
Optimal
control
On the vector fields: For each x0 M and u U, the ODE

x(t) = f (x(t), u(t)), x(0) = x0

has a solution for all t [0, [.
 Lie brackets

Geometric
control
. The set of all smooth vector fields on M forms a Lie algebra.
Dynamical
systems

control We can perform linear combinations and Lie brackets.

systems

Reachable
sets and In terms of coordinates, vector fields may be identified with
controllability
column matrices.
Affine control
systems

Distributions Lie bracket of two vector fields

and
integrability g f
[f , g](x) = (x)f (x) (x)g(x).
Main results
x x
Optimal
control
( g
x is the Jacobian matrix of g).
 Lie brackets (cont.)

Geometric
control
. The Lie bracket of 2 vector fields measures non-commutativity of
Dynamical
the corresponding flows
systems

control
systems

Reachable [f , g] 0 exp(t f ) exp(sg) = exp(sg) exp(t f ), s, t R.

sets and
controllability
exp(-tf)
Affine control
systems exp(-tf)
exp(sg)
Distributions
and
exp(-tg) exp(tg)
integrability
exp(-sg)
x
Main results
exp(tf)
Optimal
control x
2 exp(tf)
exp(t [f,g])

f and g commute f and g dont commute

 The power of Lie brackets

Geometric
control
.
The Lie bracket allows studying interconnections between
different dynamical systems.
Dynamical
systems

control
systems

Reachable
sets and For control theory it is particularly important that
controllability

Affine control
systems
[f , g] 6 span{f , g}.
Distributions
and
integrability

Main results

Optimal
control
 Reachable sets

Geometric
control
fu (.) := f (., u), F = {fu }uU
.

Dynamical Reachable set of F from a point x0 M

systems

control
systems R(x0 ) = {exp(tk fuk ) exp(t1 fu1 )(x0 ) | k N, fui F, ti 0}
Reachable
sets and
controllability The reachable set characterize the states that can be
Affine control
systems
reached from a given initial state x0 M in positive time, by
Distributions choosing various controls and switching from one to another
and
integrability from time to time.
Main results

Optimal x0
control

Only forward-in-time motions allowed!

Reachable setset
Reachable
 Controllability

Geometric
control
. Controllability
Dynamical Controllability is the ability to steer a system from a given
systems
initial state to any final state, in finite time, using the
control
systems available controls.
Reachable
sets and
controllability

Affine control
systems
A system is said to be controllable if R(x) = M, x M.
Distributions
and
integrability

Main results

Optimal
control
 Affine control systems

Geometric m
X
control
.
x = g0 (x) + ui gi (x)
i=1
Dynamical
| {z }
systems f (x,u)
control
systems g0 is the drift vector field - specifies the dynamics in the absence
Reachable of controls.
sets and
controllability
gi , i = 1, m, are called the control vector fields
Affine control
systems

Distributions Assumptions:
and
integrability
On the set of admissible controls:
Main results
U consists of all the piecewise constant functions with values in
Optimal
control U, which are piecewise continuous from the right.
On the vector fields:
g0 , g1 , , gm are smooth (of class C ). m n = dim(M).
 The car

Geometric f
control
. q
q2
Dynamical
systems x2 q1

control
systems
x1
Reachable
sets and
controllability
The state of the car:
Affine control
systems
position of its center of mass (x1 , x2 ) R2
Distributions orientation angle S 1 (relative to the positive direction of
and
integrability the axis x1 )
Main results

Optimal The state space:

control

M = {x = (x1 , x2 , ) | x1 , x2 R, S 1 } = R2 S 1 .
 The car (cont.)

f
Geometric
control Possible kinds of motion: q
q2
.
x2 q1

Dynamical
systems x1

control
systems
Linear motion: drive the car
q forward and backward with some
Reachable fixed linear velocity u1 = x12 + x22
sets and
controllability x1 = u1 cos
Affine control x2 = u1 sin (dynamical system for linear motion)
systems
= 0

Distributions
and
integrability
Rotational motion: turn the car around its center of mass with with
Main results
some fixed angular velocity u2 =
Optimal
control x1 = 0
x2 = 0 (dynamical system for rotational motion)
= u2

 The car (cont.)

Geometric
control In vector form:
.

x1 cos 0
Dynamical
systems x = x2 R2 S 1 , g1 (x) = sin , g2 (x) = 0 .
control 0 1
systems

Reachable
sets and
controllability
Combining both kinds of motion in an admissible way:
Affine control
systems

Distributions
and
x = u1 g1 (x) + u2 g2 (x)
| {z } | {z }
integrability
linear motion rotational motion
Main results

Optimal
control Affine control system, underactuated.

The control u = (u1 , u2 ) can take any value in U R2

 Typical maneuver in parking a car

Geometric

2
control
.

2
2
2
forbidden
Dynamical
systems

control
systems
Four motions with the same amplitude perform forbidden motion:
Reachable
sets and

2
2

controllability

2
2

2
2

2
Affine control
systems

Distributions 1. motion 2. rotation

and
integrability forward counterclockwise
Main results

Optimal
2

2
2
2

2
2

2
control 2

3. motion 4. rotation
backward clockwise
 The Lie bracket does your job!

Geometric
control
cos 0
. g1 (x) = sin , g2 (x) = 0 ,
0 1
Dynamical
systems

control

systems
0 0 sin 0 sin
Reachable
[g1 , g2 ](x) = 0 0 cos 0 = cos .
sets and 0 0 0 1 0
controllability

Affine control
systems
The vector field g1 generates the forward/backward motion.
Distributions
and
integrability
The vector field g2 generates the clockwise/counterclockwise
Main results rotation.
Optimal
control
The vector field [g1 , g2 ] generates the motion in the direction
perpendicular to the orientation of the car.
 Distributions and integrability

Geometric
control A distribution on the manifold M is a map which assigns to
.
each point in M a subspace of the tangent space at this point:
Dynamical
systems M 3 x 7 (x) Tx M.
control
systems
dim() = k if dim((x)) = k , x M.
Reachable
sets and
controllability
Example: M = R3 \{0}, (x) = {v R3 | v > x = 0}
Affine control
systems (x) is the tangent space at x to the sphere centered at 0
Distributions passing through x.
and
integrability This distribution has a special property: for every x M, there
Main results exists a smooth 2-dim submanifold Nx of M (the sphere centered
Optimal at 0 passing through x) which is everywhere tangent to .
control

This property is called integrability

 Distributions and integrability (cont.)

Geometric
control An integral manifold of a distribution is a submanifold N of M
.
satisfying
Dynamical Tx N = (x), x N.
systems

control
A distribution is integrable if, for every x M, there exists a
systems (maximal) integral manifold, N(x), of through every point
Reachable x M, or equivalently, there exists a (integral) foliation on M
sets and In[12]:= SphericalPlot3D @81, 2, 3<, 8, 0, Pi<, 8, 0, 3 Pi 2<, Axes False, Boxed FalseD
controllability whose tangent bundle is .
Affine control
systems

Distributions
and
integrability

Main results
Out[12]=

Optimal
control

An integral foliation on M = R3 \{0}

 Involutive distribution

Geometric
control The integrability of a distribution depends on its involutivity.
.
A distribution on M is said to be involutive if, x M,
Dynamical
systems
f (x), g(x) (x) [f , g](x) (x).
control
systems

Reachable
sets and
controllability

Affine control Frobenius theorem

systems
Suppose a distribution has constant dimension. Then,
Distributions
and
integrability
is integrable if and only if is involutive.
Main results

Optimal
control
 Control systems without drift

Geometric
control
. m
X
x = ui gi (x), x M, (unconstrained inputs).
Dynamical
systems i=1

control
systems

Reachable
sets and Control distribution
controllability
(x) = span{g1 (x), , gm (x)} Tx M.
Affine control
systems

Distributions
and
integrability

Main results

Optimal
control
 Control distribution

Geometric Control distribution: (x) = span{g1 (x), , gm (x)}

control
.
Example: M = R2 , m = 1, g1 (x) 6= 0, for all x.
Dynamical
systems The control distribution is 1-dimensional. Through each point
control x0 R2 passes a curve (x0 ) = x(t, x0 ) which is everywhere
systems
tangent to . is integrable.
Reachable
sets and
controllability
What is the reachable set from x0 ?
Affine control x(t,x0)
systems
x0
Distributions R(x0 ) = (x0 ) 6= R2
and
integrability

Main results The system is not controllable!

Optimal
control
 Control distribution (example)

Geometric
control x2 0 x3
.
x = u1 x1 + u2 x3 + u3 0 , x R3 \{0}.
Dynamical 0 x2 x1
systems

control The control distribution is 2-dimensional.

systems

Reachable [g1 , g2 ] = g3 [g2 , g3 ] = g1 [g3 , g1 ] = g2 .

sets and
controllability

Affine control
is involutive, so is integrable.
systems

Distributions d 
2x(t)> x(t) = x(t)> x(t) = 0.

and
integrability dt
Main results
Consequently, the maximal integral manifold of , at a given
Optimal
control
x M, is the sphere centered at the origin, passing through x.
The reachable set from x M is contained in a 2-dimensional
sphere. The system is not controllable.
Integrability of control distribution rules out controllability!?
 Control distribution (example)

Geometric
control Back to the car model
.

cos 0
Dynamical
systems g1 (x) = sin , g2 (x) = 0
control 0 1
systems

Reachable The control distribution is 2-dimensional.

sets and
controllability
Affine control
sin
systems [g1 , g2 ](x) = cos 6 (x).
Distributions 0
and
integrability
The control distribution is not involutive. So, is not integrable.
Main results

Optimal
The reachable sets are not restricted to 2-dimensional
control submanifolds.
Is the system controllable? Our experience says yes!
 Bracket generating property

Geometric
control If the system
. m
X
x = ui gi (x), x M
Dynamical
systems i=1
control
systems
is controllable, its control distribution
Reachable
sets and = span{g1 , , gm }
controllability

Affine control should satisfy a property that is intuitively opposite to integrability.

systems

Distributions
and
integrability The distribution = span{g1 , , gm } on M is said to be bracket
Main results generating if the iterated Lie brackets
Optimal
control gi , [gi , gj ], [gi , [gj , gk ]], , 1 i, j, k m,

span the tangent space of M at every point.

 Rashevsky-Chow Theorem

Geometric In other words:

control
.
= span{g1 , , gm } is bracket generating iff
Dynamical
systems

control Liex (F) = Tx M, for every x M.

systems

Reachable
sets and
controllability

Affine control Theorem (Rashevsky-Chow)

systems

Distributions
Assume that M is connected.
and
integrability
If the control distribution = span{g1 , , gm } is bracket
Main results
generating, then the (drift free) system
Optimal
control
m
X
x = ui gi (x), x M
i=1

is controllable.
 m
Geometric X
control x = g0 (x) + ui gi (x), x M
. | {z }
i=1
drift
Dynamical
systems The presence of drift significantly complicates the question of
control controllability!
systems

Reachable
sets and It is possible that the trajectories of a system cant be
controllability
restricted to a lower dimensional sub-manifold, and yet the
Affine control
systems system is uncontrollable, as this example shows.
Distributions
and
x22
     
x1 0
integrability
= +u , M = R2 .
Main results
x2 0 1
Optimal
control The reachable set from a point z R2 is:

R(z) = w R2 | w1 > z1 {z}.



Thus, the system is accessible but not controllable!

 The rolling sphere

Geometric The rolling sphere consists of a sphere in 3-space, rolling without

control
.
slip or twist over the tangent space at a point.

Dynamical M1

systems

control
systems

Reachable dev
sets and M0

controllability

Affine control
systems
This rigid motion is described by the action of SE(3) (the special
Distributions
and
Euclidean group), but has 2 types of constraints:
integrability

Main results Holonomic constraints (sphere keeps tangent to the plane

Optimal during motion)
control
Nonholonomic constraints (sphere cant twist or slip)
No twist (performing spins not allowed!)
No slip (performing slidding not allowed!)
 Kinematic equations for the rolling sphere

Geometric The motion of S 2 , when rolling over the south tangent plane is
control
.
described by the following right-invariant control system evolving
on SE(3) = R3 n SO(3):
Dynamical
systems

u1
control
systems
s = u2 (translational velocity)
Reachable
0
sets and
controllability
0 0 u1
Affine control
systems R = 0 0 u2 R (rotational velocity)
Distributions u1 u2 0
and
integrability

Main results
This system is controllable.
Optimal
control
The rolling sphere is a complete nonholonomic system.
 Forbidden motions

Geometric
control Question:
.
How to steer the rolling sphere from one initial configuration to
Dynamical any other admissible configuration, without violating the
systems
nonholonomic constraints (i.e, avoiding forbidden motions)?
control
systems

Reachable
sets and
Forbidden motions
controllability

Affine control
systems Twists Slips
Distributions
and
l l

integrability

Main results

Optimal
control

Answer: Realizing the forbidden motions by rolling the

sphere without slip or twist!
 Realizing a twist

Geometric Realizing a rotation of an angle around the z-axis:

control
.

Dynamical S2 = M1
systems

control . . . M2
.
systems
M4 .
.

Reachable
sets and p j/2
controllability .
Affine control M3
systems

Distributions
and M. 5
.
M6 = z(j) S2
integrability

Main results
. . .
Optimal . j/2
control M4
p
 Realizing a twist

Geometric A twist is a rotation around the z-axis.

control
.

cos sin 0
Dynamical z() = eA12 = sin cos 0
systems
0 0 1
control
systems Eulers theorem guarantees that any rotation around the z-axis
Reachable decomposes as rotations around the x-axis and the y -axis.
sets and
controllability

Affine control
But to perform a twist, such decomposition has to be carefully
systems chosen, so that the angles of rotation around these 2 orthogonal
Distributions axis add up to zero.
and
integrability
Decomposition corresponding to the previous picture:
Main results

Optimal
control z() = x( ) y ( ) x() y ( ) x( )
2 2 2 2
 Realizing a slip

Geometric A slip is a pure translation.

control
.
d(p0 , p1 ) = multiple of 2
Dynamical
systems Roll along the segment p0 p1 .
control
systems d(p0 , p1 ) 6= multiple of 2
Reachable Roll along the sides of the isosceles triangle in the picture.
sets and
controllability

Affine control
systems l is the smallest integer satisfying 2l > d(p0 , p1 ).
Distributions
and
integrability
p
Main results
0
Optimal
control
p
2pl 1
q
 Optimal control

Geometric
control Controllability doesnt care about the quality of the trajectory
.
between two states, neither the amount of control effort!
Dynamical
systems

control
systems Optimal control
Reachable
sets and
What is the optimal way to control the system? We may
controllability require smooth trajectories, minimizing costs, ...
Affine control
systems

Distributions
and
integrability

Main results

Optimal
control
 Rolling optimally

Geometric
control Given 2 admissible configurations, roll the sphere upon the
.
tangent plane from the first configuration to the second, so that
Dynamical the curve traced in the plane by the contact point be the shortest
systems
possible.
control
systems

Reachable 1
R t1
sets and J(u) = 2 0
(u12 + u22 ) dt min (cost functional)
controllability

Affine control subject to:

systems
Distributions
u1
and
integrability
s(t) = u2
0
Main results
(control system)
Optimal
control 0 0 u1
R(t) = R(t) 0 0 u2
u1 u2 0

X (0) = X0 = (s0 , R0 ) X (t1 ) = X1 = (s1 , R1 ) (boundary cond.)

 The rolling sphere meets Euler

Geometric
control
.
To solve optimal control problems one needs to understand the
Dynamical Pontryagin Maximum Principal (Hamiltonian equations,
systems
symplectic geometry,...).
control
systems

Reachable
sets and
controllability

Affine control The rolling sphere meets Euler

systems

Distributions The point of contact of the sphere rolling optimally traces Euler
and
integrability
elastica on the plane!
Main results

Optimal
control
 Elastic curves of Euler (1744)

Geometric An elastic rod is a 1-dimensional object which is flexible

control
.
(bendable but not stretchable), which looks like a portion of a
of the curva elastica and its ramifications. Leibniz had proposed: F
straight line in its natural state.
substantiated, that the extensions are proportional to the stretching force
Dynamical
systems
as Hookes law of the spring. Bernoulli questioned this relationship, and,
the elastica
Which form takes an generalized evenwhen
elastic rod to nonlinear
subjectdisplacement.
to external forces
control
systems applied to its ends?
Reachable
sets and
5 James Bernoulli poses the elastica pro
Jacob Bernoulli posed this problem in 1691 and showed that the
controllability

Affine control
elastic
R 2 energy of a deformed
James Bernoulli elastic
posed the preciserod is proportional
problem toin 1691:
of the elastica
systems (t) dt, where (t) is the geodesic curvature.
Distributions
and
integrability

Main results

Optimal
control
 Elastic curves of Euler (1744)

Geometric
control Euler (1744) also studied the variational principle
.
Z
Dynamical 2 dt min
systems

control
systems that gives rise to the shape that minimizes the elastic energy.
Reachable
sets and
controllability

Affine control
systems

Distributions
and
integrability

Main results

Optimal
control
 Eulers Elastica

Geometric
control Euler also sketched these beautiful curves even before the
. discover of the elliptic functions (Carl Jacobi) was borne!
Dynamical
systems

control
systems

Reachable
sets and
controllability

Affine control
systems

Distributions
and
integrability

Main results

Optimal
control
 References

1 A. Agrachev and Y. Sachkov, Control Theory from the Geometric Viewpoint. Springer, Berlin, 2004.
Geometric
2 V. Jurdjevic, Geometric Control Theory. Cambridge Univ. Press, Cambridge, 1997.
control
. 3 V.Jurdjevic and H. Sussmann, Control Systems on Lie Groups. Journal of Differential Equations, 12
(1972), 313-329.

Dynamical 4 M. Kleinsteuber, K. Hper and F. Silva Leite, Complete Controllability of the N-sphere - a constructive
systems proof. Proc. 3rd IFAC Workshop on Lagrangian and Hamiltonian Methods for Nonlinear Control
(LHMNLC06). Nagoya, Japan (19-21 July, 2006).
control 5 Y. Sachkov, Control Theory on Lie Groups. Journal of Mathematical Sciences, Vol. 156, No. 3, 2009.
systems
6 J. Zimmerman, Optimal control of the Sphere S n Rolling on E n . Math. Control Signals Systems, 17
Reachable (2005), 14-37.
sets and
controllability

Affine control
systems

Distributions
and
integrability

Main results

Optimal
control