Professional Documents
Culture Documents
Mayank Baranwal
University of Illinois at Urbana-Champaign
Rolling Bodies
Introduction
An old problem in the field of holonomy asks: Given a pair of orientations for a sphere
resting on a plane, is there a closed path along which one can roll the sphere (without slipping or
twisting), starting with the first orientation, and return to the origin with the sphere in the second
orientation? The answer is yes, and the goal of this article is to provide an elementary
proof of this fact. Moreover, this article also discusses the existence of optimal control for
the rolling problem of two rigid bodies (Riemannian manifolds).
Rolling is a rigid motion in the embedding space, subject to holonomic and nonholonomic constraints. The rolling motion is then described by the action of the Euclidean
group SEn = SOn Rn on Rn :
SEn := {X = (R, s) : R SOn , s Rn }
SEn Rn Rn
(X, p) 7 X p = Rp + s.
Even though the goal of this article is to discuss the rolling problem in full detail, a complete analysis would require several tools from differential geometry. Hence, it will be
useful to first develop an understaing of these mathematical preliminaries.
Section 2 of this article talks about Fiber bundles and Vector bundles. Section 3 essentially
defines what is meant by the directional derivative of vector fields on manifolds. Section
4 talks about describing connections on manifolds as a means to not only compare the
tangent spaces at different points, but to also allow to perform calculus on the manifolds.
This is followed by a brief introduction to Pontryagin maximum principle. The subsequent
sections then discuss the original rolling problem.
Fiber Bundle
2.1
Vector Bundle
A special class of fiber bundles, called vector bundles, are those whose fibers are vector
spaces (to qualify as a vector bundle the structure group of the bundle must be a linear
group). Important examples of vector bundles include the tangent bundle and cotangent
bundle of a smooth manifold.
2.2
Sections
A section (or cross section) of a fiber bundle is a continuous map f : B E such that
( f (x)) = x for all x in B. Since bundles do not in general have globally defined sections,
one of the purposes of the theory is to account for their existence. A section picks out an
element of each fiber, in a continuous way.
2.3
Vertical Bundle
In mathematics, the vertical bundle of a smooth fiber bundle is the subbundle of the
tangent bundle that consists of all vectors that are tangent to the fibers. Formally, let
2
2.4
(1)
3.1
3.2
3.3
3.4
Let X Vec(M), and let t : M M be the flow of X. This means that 0 is the identity
map
d
t (p)|t=0 = Xp p M
dt
4
t, t is a diffeomorphism. It can be shown that s t = s+t , s, t < for which the flow
is defined. The Lie derivative of a vector field Y Vec(M) in the direction X is the vector
field LX Y defined by
Yp (t Y)p
p M
(LX Y)(p) := lim
t0
t
Here t Y is the pushforward of the vector field Y by the diffeomorphism t , defined by
(dt )1
Y 1 , where dt is the derivative of t .
t (p) t (p)
If our manifold is Rn , then there actually is a natural choice of covariant directional derivative. We define : Vec(Rn ) Vec(Rn ) Vec(Rn ) by
X Y(p) := lim
Yp+tXp Yp
t0
Connection
4.1
i j = g
2
x j
xk
xl
=0
A curve (t) is called a geodesic if (t)
(t)
Hamiltonian Mechanics
L
q k
THe quantities pk are the components of the momentum vector, which is in fact an element
of T(t) Q . We can see this as follows, Let W be an n-dimensional vector space, and let
f : W R be any differentiable function. Then the differential d f~v at a point ~
v W is
a linear transformation d f~v : W R. Thus, by definition of the dual space, d f~v W .
Consequently, the differential d f gives a correspondence d f : W W via
n
X
f i
dx
~
v 7 d f~v =
i
x
~
v
i=1
Taking W as the vector space T(t) Q, the momentum at the point (t) be the vector
dL(q k ) T(t) Q . Hence, we can think of the momentum vector p as a covector field
along the curve given at each point by dL(q k ) .
Consider now the Hamiltonian function H defined by
H=
n
X
k=1
Since we can write the quantity q k in terms of the components pk , we can view the
Hamiltonian as a time-dependent function on TQ . Given any configuration space Q, we
define the cotangent bundle TQ as the phase space of the system. If Q is an n-dimensional
manifold, then TQ is a manifold of dimension 2n. Thus, the Hamiltons equations of motion
are
q k
H
pk
H
p k = q
k
5.1
additional independent variables, as shown below. Start by finding a chart that trivializes
the cotangent bundle TQ (i.e. a local trivialization):
TQ |U ' U Rn
Then introduce the Hamiltonian as
H(q, p) =
1 ab
g (q)pa pb
2
Here, gab (q) is the inverse of the metric tensor: gab (q)gbc (q) = ac . The geodesic equations
can then be written as
q a
H
pa
= gab (q)pb
H
1
p a = q
a = 2
gbc (q)
pb pc
qa
The second order geodesic equations are easily obtained by substitution of one into the
other. The flow determined by these equations is called the cogeodesic flow. The first of
the two equations gives the flow on the tangent bundle TQ, the geodesic flow. Thus, the
geodesic lines are the projections of integral curves of the geodesic flow onto the manifold
Q. This is a Hamiltonian flow, and that the Hamiltonian is constant along the geodesics:
dH H a H a
= a q +
p = p a q a + p a q a = 0
dt
q
pa
Thus the geodesic flow splits the cotangent bundle into level sets of constant energy
QE = {(q, p) TQ : H(q, p) = E}
for each energy E 0, so that
TQ =
QE
E0
5.2
Poisson Bracket
q
pi qi
i pi
i=1
Hamiltons equations of motion have an equivalent expression in terms of the Poisson
bracket. This may be most directly demonstrated in an explicit coordinate frame. Suppose
that f (p, q) is a function on the manifold. Then from the multivariable chain rule,
f
f
d
f (p, q) =
q +
p
dt
q
p
7
f H
q p
f H
p q
= { f, H}
Let u (t) be an optimal control and let x (t) be the corresponding trajectory. Then there
exists a function p (t) and a constant {0, 1}, satisfying (, p (t)) , (0, 0)t, such that:
1) x (t) and p (t) satisfy the canonical equations
x = Hp (x , u , p , )
p = Hx (x , u , p , )
with the boundary conditions x (t0 ) = x0 , x (t1 ) = x1 (t1 unspecified), where the Hamiltonian is defined as
H(x, u, p, ) := <p, f (x, u)> + L(x, u) u U , t [t0 , t1 ]
2) H(x (t), u (t), p (t), ) H(x (t), u(t), p (t), ) u U , t [t0 , t1 ]
3) H(x (t), u (t), p (t), ) 0, t [t0 , t1 ]
6.1
We use the Orbit Theorem and Pontryagin Maximum Principle to an intrinsic geometric
model of a pair of rolling rigid bodies. The main reult is: the system is completely
controllable iff the bodies are not isometric. We also state an optimal control problem
and study its extremals. Let us consider the following example of a sphere rolling on a plane,
7.1
Geometric Model
q : Tx M Tx M
An isometry q is a state of the system, and the state space is the connected 5-dimensional
manifold
n
o
| x M, x M,
q an isometry
Q = q : Tx M Tx M
q : Tx M Tx M,
(q) = x, (q)
= x,
q Q, x M, x M
<ei , e j >M = i j
i, j = 1, 2
7.2
For the 2-dimensional manifold, we describe Riemannian geodesics, Levi-Civita connection and parallel translation on TM TM
7.2.1
Riemannian geodesics
For any fixed point x0 , x1 M, we seek for the shortest curve in M connecting x0 and x1 :
x = u1 e1 (x) + u2 e2 (x), x M, (u1 , u2 ) R2
x(0) = x0 , x(t1 ) = x1 ,
Zt1
Zt1
1/
1
x>
2 dt = (u21 + u22 ) /2 dt min.
l=
<x,
0
Earlier, we had shown that arc-length parameterized extremal trajectories, i.e. unit speed
trajectories in this problem (Riemannian geodesics) are projections of trajectories of the
normal Hamiltonian field and that the geodesic splits cotnagent bundle into level sets of
constant energy:
~
h1 = cos
h2 = sin
Cotangent bundle of a Riemannian manifold can be identified with tangent bundle via
the Riemannian strucutre:
TM TM ,
v = <v, >
Then H is identified with the spherical bundle of unit tangent vectors to M.
S = {v TM | ||v|| = 1} TM
~
Levi-Civita Connection
Levi-Civita connection is the unique connection on the spherical bundle S M such that:
(1) velocity of Riemannian geodesic is parallel along the geodesic,
(2) parallel translation preserves angle, i.e., horizontal lifts of vector fields on the base M
~
~0 , H
~0 =
~ .
T H = span H,
,H
,H
We have,
~
H
~0
H
+ h2 e2 + c2
= h1 e1 + c1
= h2 e1 + c1
+ h1 e2 + c2
ci C (M)
11
7.3
Admissible Velocities
Let us return to the rolling manifolds problem and write down admissibility conditions
non-slipping and non-twisting for a curve q(t) Q as restricitons on velcity q(t).
Decompose
in the orthonormal frames:
velocities of the contact curves in M and M
x = a1 e1 (x) + a2 e2 (x),
+ a2 e2 (x).
x = a1 e1 (x)
Then the non-slipping condition reads:
a1 = a1 cos a2 sin ,
a2 = a1 sin + a2 cos .
(t)
In the cotangent bundle, the non-twisting condition means that if
(t) = x(t), (t) H
is a parallel covector field along a curve x(t) M, then
= x(t),
(t)
(t)
H
M
is a parallel covector field along the curve x(t)
Summing up, the admissibility conditions for rolling bodies determine constraints along
contact curves, i.e., a rank two distribution on Q spanned locally by the vector fields
,
X1 = e1 + cos e1 + sin e2 + (c1 + c1 cos + c2 sin )
Admissible motions of the rolling bodies are trajectories of the control system
q = u1 X1 (q) + u2 X2 (q),
7.4
q Q,
u1 , u2 R.
Controllability
by k and k
Denote the Gaussian curvatures of the Riemannian manifolds M and M
to Q:
respectively. We lift these curvatures from M and M
k(q) = k (q) ,
= k (q)
k(q)
,
q Q.
ii
h h
~ H,
~ = k , where k is the curvature of the manifold.
Corollary: H,
12
X4 = [X1 , X3 ]
"
#
X5 = [X2 , X3 ]
"
#
other hand, if k(q) = k(q) at all points q O, implies that the distribution is integrable,
thus dimO = 2.
7.5
7.5.1
x)
i.e., k k , 0 on Q. Then, by above
for any x M, x M,
Suppose that k(x) , k(
theorem, the rolling system is controllable. The length minimization problem is stated as
the following optimal control problem:
q = u1 X1 + u2 X2 , q Q u = (u1 , u2 ) R2 ,
q(0) = q0 , q(t1 ) = q1 , t1 f ixed,
Zt1
l = (u21 + u22 )1/2 dt min.
0
13
read respectively as
Notice that projections of ODE to M and M
x = u1 e1 + u2 e2 , x M,
x = u1 (cos e1 + sin e2 ) + u2 ( sin e1 + cos e2 ) ,
x M,
Zt1
(u21 + u22 )dt min,
0
Maximum Principle
As we showed in the proof of the Theorem, the vector fields X1 , . . . , X5 form a frame on
Q. Denote the corresponding Hamiltonians linear on fibers in TQ :
gi () = <, Xi >,
TQ ,
i = 1, . . . , 5.
H = u1 g1 () + u2 g2 () + (u21 + u22 ),
2
and the corresponding Hamiltonian system is
= u1 ~g1 () + u2 ~g2 (),
7.6
TQ
Results
For a sphere rolling on a plane, it turns out that the point of contact of the sphere rolling
optimally traces Euler elastica on the plane.
7.7
Conclusion
We saw how Orbit Theorem and Pontryagin Maximum Principle can be used to analyze the motion of a pair of rolling rigid bodies. We obtained an exact condition on the
curvatures of the two manifolds for the rolling system to be controllable. Summing up,
Euler Elastica
References
1. Agrachev, Andrei A., and Yuri Sachkov. Control theory from the geometric viewpoint.
Vol. 2. Springer, 2004.
2. Jurdjevic, Velimir. Geometric control theory. Cambridge university press, 1997.
3. Lovett, Stephen. Differential geometry of manifolds. AMC 10 (2010): 12.
4. Johnson, Brody Dylan. The nonholonomy of the rolling sphere. American Mathematical
Monthly 114, no. 6 (2007): 500-508.
5. Barbero Lin, Mara. A geometric study of abnormality in optimal control problems for
control and mechanical control systems. (2008).
6. Leite, Ftima Silva. Introduction to geometric control theory-controllability and Lie bracket.
(2010).
7. Goldberg, Timothy E., What is a Connection, and What is it good for?.
8. Barden, Dennis, and Charles Thomas. An introduction to differential manifolds. AMC
10 (2003): 12.
9. Fiber bundle. Wikipedia.
15