Scribd
Upload
142 views

Optimal Control Minimum Time Hamiltonian

This document discusses optimal control for continuous time dynamical systems with partially fixed or free terminal states and times. It covers the Euler-Lagrange equations and transversality conditions for these cases, as well as examples like minimum time control problems and Zermelo's problem.

Uploaded by

darin koblick
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
142 views

Optimal Control Minimum Time Hamiltonian

This document discusses optimal control for continuous time dynamical systems with partially fixed or free terminal states and times. It covers the Euler-Lagrange equations and transversality conditions for these cases, as well as examples like minimum time control problems and Zermelo's problem.

Uploaded by

darin koblick
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 11

Optimal Control

Lecture 9

Optimal Control for Continuous Time Dynamical Systems: Partially fixed terminal state,
free terminal time cases

John T. Wen

February 12, 2002


JTW-OC9 RPI ECSE 6440 Optimal Control.

Outline

 time invariant case

 partially fixed terminal state problem

 free terminal time problem (including minimum time)

 a static optimization perspective

 examples

Ref: Bryson & Ho, Sec. 2.7-2.8


Last Time

 unicycle examples: simplify the TPBVP as much as possible (reducible to an alge-


braic nonlinear least square problem).

February 12, 2002Copyrighted by John T. Wen Page 1


JTW-OC9 RPI ECSE 6440 Optimal Control.

Time Invariant Case

Along the optimal trajectory,


∂H ∂H ∂H ∂H
Ḣ = + u̇ + λ̇ + ẋ
∂t ∂u ∂λ ∂x
∂H ∂H
= + u̇ + f T λ̇ λ̇T f
∂t ∂u
∂H
= :
∂t
If f and L do not depend on t explicitly, then ∂H
∂t = 0, or H is a constant along the optimal
trajectory. This is used in the minimum drag nose shape problem.

February 12, 2002Copyrighted by John T. Wen Page 2


JTW-OC9 RPI ECSE 6440 Optimal Control.

Fixed or Partially Fixed Terminal State

Consider the minimization problem


Z tf
J = φ(x(t f ); t f ) + L(x(t ); u(t ); t ) dt
t0

subject to the dynamical constraint

ẋ(t ) = f (x(t ); u(t ); t ); x(t0 ) = x0 :

Suppose, in addition, a function of the final state is specified (e.g., certain component).
For example, for a point-to-point motion control problem, final velocity may be specified
to be zero. Let the constraint be ψ(x(t f ); t f ) = 0 where ψ 2 R q . We can include the
constraint through an additional Lagrange multiplier ν:
Z tf  
J = φ(x(t f ); t f ) + ν ψ(x(t f ); t f ) +
T
L(x(t ); u(t ); t ) + λ (t ) ( f (x(t ); u(t ); t ) ẋ(t )) dt :
T
t0

February 12, 2002Copyrighted by John T. Wen Page 3


JTW-OC9 RPI ECSE 6440 Optimal Control.

Fixed or Partially Fixed Terminal State (cont.)

The first order condition yields the same Euler-Lagrange Equations as before, except with
respect to x(t f ):
 T " T #
∂φ ∂ψ
λ(t f ) = + ν :
∂x(t f ) ∂x(t f )
The TPBVP now involves 2n ODE’s and q algebraic equations ψ(x(t f ); t f ) = 0 (and q
additional variables in ν).
Special case: ψi (x(t f )) = xi (t f ) and φ = φ(xq+1 ; : : : ; xn ). Then
8
< ν i = 1; : : : ; q
i
λi (t f ) = ∂φ
: i = q + 1; : : : ; n:
∂xi (t f )

February 12, 2002Copyrighted by John T. Wen Page 4


JTW-OC9 RPI ECSE 6440 Optimal Control.

Free terminal time problem

When t f is free, we need to include δt f in the first optimality condition for the addition
equation (called Transversality Condition):
 
∂φ
Free terminal state case: + H =0
∂t f tf
Fixed terminal state case: H jt f =0 (H = 0 in the time invariant case.)
 
∂φ
T ∂ψ
Partially fixed terminal state: +ν + H = 0:
∂t f ∂t tf

February 12, 2002Copyrighted by John T. Wen Page 5


JTW-OC9 RPI ECSE 6440 Optimal Control.

Minimum time control

Consider the minimum time control problem with


Z tf
J= dt ; ẋ = f (x; u); x(0) = x0 ; x(t f ) = x f :
0

Hamiltonian: H = 1 + λT f (x; u).


 T
∂f
Costate: λ̇ = ∂x λ
 T
∂f
Control: ∂u λ=0
Transversality: H (t f ) = 0
Time invariance: H (t ) = constant = 0.

February 12, 2002Copyrighted by John T. Wen Page 6


JTW-OC9 RPI ECSE 6440 Optimal Control.

Examples

1. Zermelo’s problem (Example 1 in Sec.2.7 in B& H)


2. Brachistochrone problem (Problems 2 and 6 in Sec 2.7 in B& H)

February 12, 2002Copyrighted by John T. Wen Page 7


JTW-OC9 RPI ECSE 6440 Optimal Control.

Another Approach to Free Terminal Time Problem

Let t f be a new state variable, i.e., x n+1 = t f . Then

ẋn+1 = 0:

Define τ = t t
t f = xn+1 and xτ (τ) = x(xn+1 τ). State dynamics in terms of τ:
2 3 2 3 2 3 2 3
dxτ
xn+1 f (xτ (τ); u(xn+1τ); xn+1 τ) xτ (0) x0
4 dτ 5 = 4 5; 4 5=4 5:
dxn+1
dτ 0 xn+1 (0) xn+1 (1)

The optimal control problem is now a fixed terminal time problem:


Z1
J = φ(xτ (1); xn+1 ) + xn+1 L(xτ (τ); u(xn+1 τ); xn+1 τ) dτ:
0

Note that if the original system is linear, the transformed system is bilinear.

February 12, 2002Copyrighted by John T. Wen Page 8


JTW-OC9 RPI ECSE 6440 Optimal Control.

Static Optimization Perspective

We have seen the optimal constrol problem can be converted to an equivalent TPBVP
which can in turn be posed as a nonlinear least square problem involving n variables (e.g.,
final or inital costates).
Alternatively, we can take a static perspective (as in HW#2). For a given input u =

u(t ) : 0  t  t f . Let the solution of ẋ = f (x; u; t ), x(t 0 ) = x0 , be x(t ) = Φ(t; u; x0 ; t0 ).
Then, for given t0 and x0 ,
Z tf
J = φ(x(t f ); t f ) + L(x(t ); u(t ); t ) dt
0

can be written as a function of u:


J = J1 (u)
which is a static functional optimization problem. To solve this problem, some finite
dimensional approximation of u is needed.

February 12, 2002Copyrighted by John T. Wen Page 9


JTW-OC9 RPI ECSE 6440 Optimal Control.

Static Optimization (cont.)

Typically, u is written as a linear combination of basis functions:


`
u(t ) = ∑ λi ψi (t ) = ΨT (t )λ:
i=1

The problem then reduces to a finite dimensional static optimization problem:

J = J1 (ΨT λ) = J2 (λ):

Typical choices of ψi : Fourier basis, Legurre polynomials, wavelet functions, Walsh func-
tions, radial basis functions (RBF), etc.

February 12, 2002Copyrighted by John T. Wen Page 10

You might also like