YouSub
Untitled

en English

Untitled English subtitles

0 downloads · 663 lines · added Aug 2, 2015

Report a problem
What's wrong?
Advertisement

English transcript

Alright, hello everybody; let us continue our lecture series for this course, optimal
control guidance and estimation.
Talking about lecture number six here, where we will review some calculus of variations
concept and this a two part lecture series and the first lecture will be taken this time
and we will generalize some of the concepts in the next lecture. So, why do we do that?
Because, one very strong of backbone of optimal control theory relies on this calculus of
variation approach actually. In fact, the entire problem can be viewed
out or viewed as nothing but calculus of variation problem actually. That is what we will do
but, typically the calculus of variation is a vast subject also in its own right. So,
we will not talk of too much detail about everything that is involved; what will just
see are the concepts that are relevant and then we will proceed further for optimal control
ideas and of that actually next. So, this is the motivation; why we want to study calculus
of variation actually here. So, first thing is before going there little bit of motivation
we can, we have seen some of these things before; what will have is a re-look at that.
What is an optimal control formulation that is what we are, our ultimate aim is actually.
So, in a way it can be summarized something like that which I think we have all discussed
about it well before. But our objective an optimal control problem is to find an admissible
time history of control variable u of t in the domain t0 to t f; t will also be in that
domain. We have to find out all u of t and rather, u of t trajectory within this t0 to
t f which causes the system governed by the system dynamics to follow an admissible trajectory
ok. So, it is a control trajectory we are interested
to find out; so that the state trajectory becomes an admissible. If you view it as a
solution to this I mean, this equation or whichever. But, on the way has to optimize;
that means minimize or maximize a meaningful performance index which can be given like
that in a kind of its very generic form. But, that does not mean it is the only form basically;
it can be having here one way of defining it, but it is a very generic way; this can
be defined something like that and it also is to set - force the system to satisfy certain
boundary conditions ok. So, what is our objective? All that our objective is, to find the some
sort of control I mean, control trajectory in this domain.
So that it will satisfy all these nice things actually; it will give a solution of the state
which is admissible; it will minimize or maximize a cos function and it also satisfy a certain
boundary condition with both initial condition as well as the final condition actually. That
is our objective and obviously, it falls in the control variable calculus variation because
this is a path dependent on optimization. Optimization is what we are talking about
here and then there is a dynamic system involved in the process actually. So, at a point it
does not make too much sense but, it is an evaluation in time, it makes a lot of sense
actually.
So, looking at this particular cross function you can talk about, again discuss this a little
bit before as well. So, if you talk about minimum time then I can talk about minimizing
this fellow and hence I can select phi 0 and l one this thing 0 it is one then it is nothing,
but t f minus t 0. So, that is want I want to minimize here actually ok.
Now, if you want about minimum control effort and this is this is what I will do this is
I will put it as 0 and this one I will take it as something like you transpose r u then
if I if; that means, I minimizing the control effort in a way actually. So, this means phi
is 0 and l is the nothing more like that. Then, if you talk about minimum deviation
of state about c with minimum control effort is this helical probably in certain ideas
that I discussed before as well. So, then you can select something like this;
in other words, your l is coming like that and phi is 0 actually. So, like that you can
variety of things you can talk about and this particular thing talks about something like
a terminal time dependent of a component of the j and this talks about some path dependent
t 0 to t f what is happening on the way actually. So, this is what happens at t f and this is
happens at what happens from t0 to t f these 2 components are built around that kind of
ideas actually.
So, this all can do that and again if you talk about minimum deviation of state from
origin and think like that then it is it gives a standard kind of quadratic regulators sort
of ideas actually. Then, you can take values as something like this and if you want to
minimize the control while the final state reaches close to a constant value when this
is, this must in a error quantity want to minimize that then you can talk about a phi
component now at equal to t f this is to minimum. On the way, you want to control effort to
be minimum; so, l has to be like that actually. So, various things can be defined they are
just taken I have kind of taken examples out of actually. You can think of your own problem
and construct a cos function your own way actually. So, if you just see that this particular
problem is directly falling on the calculus a various problem because we are interested
in finding a state trajectory finding a control trajectory which will ultimately also give
you some sort of admissible state trajectory. So, it is a trajectory finding problem sort
of thing and with respect to those kind of trajectories everything is optimal to each
other. That means, the cos function is either maximum or minimum that we want to do as well
as it satisfies the boundary condition actually. So, directly it falls on the calculus of variation;
that is why we were interested to see some concepts of this calculus of variations actually
– ideas, alright.
So, before that now let us know a little bit fundamental theorem are calculus as well as
usual every talk about something like derivatives or integrals and all that it just good to
know that actually; that is what is happening here.
Now that the limits are constant number actually and the integrals do not contain any function
of state and thing I mean this function of these x variable then it comes out to be 0,
but what happens if the limit itself is a function of state? Actually, the function
of the pre variable then what happens, d by dx of this quantity happens to be f of x provided
f of x is continuous actually. These are something directly from calculus
not really calculus of variation actually advance calculation actually. Now, what happens
of these 2 limits are constant, but inside the function inside the integral nearby function
which is a function of the dependent variable then what happens actually. It is integrated
over y, but the result is a function of x anymore this integral is integrated over y.
So, y is gone actually because that will be evaluated eventually, but it ultimately results
in a function of x. So, that is you can talk about derivative
of x actually. So, this result is given something like that provided of this f of x has continuous
partial derivative of del f by del x actually then you have something like little bit extension
of that and you tell now what if in addition to that conditions my limits are also function
of. So, pre variable as psi 1 and psi 2 instead of constants a and b we have this functions
not coming there actually. Then, it satisfies something very close to that what you know
here plus additional components actually. This will be something; this is what you want
to be here plus these 2 components will come here actually.
Some of these things will be used in our calculus of variation analysis concepts also gives
a some transposal eddy conditions ideas and all that actually; we see that there anyway.
So, this is a quick greens of these theorems 1, 2, 3 which we do not want to prove here
anything, but with this back ground in mind about the calculus of variation concepts actually.
So, what is the difference between calculus and calculus of variations? On one end we
talk about functions which is part of the calculus and the other end we talk about functional
actually ok. What is functional? You can think of something like a function of function sort
of things. So, j is a function of x, but x itself is a function t. So, that kind of thing
we talk about a functional basically and typically, the functional are also kind of I mean scalar
values and thing like that. Ultimately, it will be integrated; once we integrate it will
ultimately result in some sort of scalar value actually. Whereas anyway, that is coming back
to that function is something defined like this to its value of independent variable.
There is a corresponding value of the dependent variable and we can know these functions from
very basic ideas already. Actually, I do not describe too much on that. But, as a functional
to which function there is a corresponding value of the dependent variable actually.
That is I mean, in a way you can it feel similar basically; but in I mean, if you just look
at a kind of an activity idea sense then this is a function one is a independent variable
and we got a function and it is a function of a function basically. So, that kind of
ideas you can actually... So, here the differences is to each value of the independent variable
there is a corresponding value of the dependent variable and here the reference to each function
for this is just the difference actually. To each function, there is a corresponding
value of the dependent variable. Here it is to each value of the independent variable
there is a corresponding value of a were here is a to each function there is a corresponding
value actually . So, that is the difference actually ok.
Now, we have talk about what is increment of a function; obviously, it is defined something
like that x of t plus delta t minus x t and what is this here increment of a functional
we need to talk about j of x t plus delta x t and this delta is denoted something like
this and this itself is a function of t you know that ok. So, and then this will give
you some sort of increment of a functional in the sense this minus that will give you
some delta j basically ok.
Pictorially also that is possible to denote, we will see that in a second lecture. Now,
delta j if we talk about that what about some sort of a small kind of a example talk about
j is something like this. The delta j by definition cannot be like that just from this the definition
actually. So, we have j of x plus delta x both are one sense of time again minus j of
x of t and j of x of t is defined something like that. So, I can put it somewhere like
that, but j of x plus delta x I can evaluate wherever x is there I will put x plus delta
x and then make it square and thing like that then I will integrate I mean even only a kind
of combined the 2 integrals and cancelled out some prompts and think later that way
I will this 2 x square and 2 x square and think like that will go well. So, what we
will what will be left out with this kind of a quantity actually ok.
As long as you know this delta x of t as a function it is on wait then I can talk about
evaluating these integral along with this x of t actually I too I too know both x of
t as well as the corresponding variations throughout the time domain then I can evaluate
this corresponding delta j basically.
Now, this is j pictorially representing something like this and this is a function and it is
an increment you have a function and independent variable in the x axis. You have a functional
and the independent function in the yx I mean, x axis. So, you that is what the instead of
a value we talk about some set of a value of the function itself basically in x of t.
Alright, so, we will in other things will remain same I mean, if you whatever happens
here as a as a independent variable sense will happen everything will happen here. In
this functions project in the function sense basically, but remember this itself is the
function x of t is in be a function actually. We are just evaluating that function in a
particular value of time and then interpreting what is happening there actually. So, we talk
about let us say increment delta t we talk about variation delta x and then talk about
some there this is the d f and this is delta j inclined that actually. And, typically will
even though we know that this is a kind of our I mean suppose you should change the function
from here to here this independent variable then the function goes from here to there;
but if you put a first order approximation these are just like a kind of elegant approximation
of sort of thing and then go up to that point then you will end up there actually will not
to the end of there and also that is the defect thing and all that is the error quantity basically.
Similarly, think it will happen also here nearly both there, but this is the error quantity
actually ok. So, most of our analysis will begin limited to the first order set of things
actually. So, that is why this pictorial representation has on each other and more that you can say
see complete from this book actually.
How do you calculate? Whatever you define as a differential of a function it terms to
the variation of a functional layer ok. So, what in the differential of a function, delta
f star is nothing, but f of t star plus delta t minus f of t star and then you can evaluate
that way. So, it turns out to be d f and it’s kind of defined d square of anything like
that. So, this is how t is and then if we talk about
delta t goes to0 then we talk about the divide these both sides by delta t and then take
delta t going to0 then it is represents that the d f by d I is all that is the first rid
of. So, you will end up with some approximations with we talk about actually. d f is nothing
f dot into delta t in general that is what I told here; d f is nothing but f dot into
delta t that is what related that actually. F dot is the slope and delta t and then d
f be back actually. So, this is how it is what we are coming to the other side of the
thing variation of a functional you can again talk about very similar and then first variation,
second variation like that you can draw and then you will end up with something like this
actually. So, most of the time we will be worried about first variation in that will
be define something like that way and second variations are important only for sufficiency
check and all that later, ok. So, I mean, in one class later we will talk about that.
So, what we are talking here is a result 1; what is derivative variation is variation
of derivative and result 2 which talks about integration of variation is variation of integration
strong trivial and in the proof applies to very trivial also. But, it is a very powerful
implication actually. So, what we are talking here? Derivative of a variation: this is by
definition like that; this is the variation quantity and this is d y. Now, what we are
talking here? All these, let me see whether I have a picture somewhere.
This is what it here talking about. So, it is actually a x of t what we are dealing here
and some x star of t and the difference between that is also a function right I mean if you
talk about various values of this x axis you will get various values of the difference
actually. The difference is nothing, but delta x actually
that is what we are talking. So, if you have a path we are talking about if variation around
the path that will, I will take some of the other alternative path around actually that
is the concept of variation actually. So, this is what I am talking here actually. So,
you can see that this that variation and then derivative of that and then simply use the
definition of this variation x minus x star and then it will lies as to evaluate the derivative
and these 2 difference is nothing, but the variation of x dot by definition again is
that the sum derivative minus sum derivative of star value actually. So, by definition
it is nothing but delta of x dot actually. Again, very same algebra; similar algebra
you take the integration and then put it by the definition and then separate it out. It
can be done because both is derivative and linear operators are actually linear operator.
That is why it happens; actually talk about this; separate it out and by definition from
here is nothing but variation of this function; this functional actually. We consider this
integration itself is a functional; then it is nothing but variation of that function
actually. So, what we are talking about it is integration
of variation by definition; like that is variation of integration which is that by definition
like that. So, very simple to line proof at very powerful result as we will see little
later actually.
First thing, some exercise problem creation of this particular functional; how do you
do that? Also, some small kind of comment that tells you that just evaluate the variation;
that means, first variation not many certain power variation and all that actually. Default
variation means first variation actually. We talk about this problem; we want if we
evaluate the first variation or just the variation of the j then method one you can just go out
to definition that then put the j. Then try to expand it; put the definitions and then
end up with some expressions like that and remember second order, third order of the
variations are neglected.
So, you can neglect all these terms and you will end up with something like this actually.
What about method 2? This is you can directly go to the result; this is the result that
we discussed; this one then you can simply put it back here then, evaluate the del j
by del x directly. Then you can the except this area of variation of integral is integral
of variation here; that means, this is what we are talking about. It will be like this;
this one can be pushed inside, now this integral actually.
Then you can end up with something like this. Directly we are taking help of del j by del
x evaluation actually. That is the difference and that simplifies value of algebra and you
will end up with something like this actually. This derivative can be pushed inside; this
limits are constant; now can be pushed inside and then you can have a simplification term
4 x plus 3 here. So, that is how the reference there actually.
Now, what are this some boundary condition ideas and all that? I remember this optimal
control problem also talks about boundary conditions and here you can talk about a variety
of boundary conditions. It can be a fixed end point, free end point, partially fixed,
partially free, something like that actually. So, when you talk about fixed end points what
we are telling is t0 my x of t0 is a fixed value and t f my x of t f is also a fixed
value. So, this is the like you too have a solution as to satisfy these kind of end points.
Actually, t0 x of t0 is a number I mean, specified number and t f and x f is also a specified
number actually. So, that is a kind of fixed end point problems
of and free end points problem is can be both these; that means, it is either completely
free. We do not really know about where it lies and things like that or it may be required
to lie on a curve actually. Many of these things practical examples in probably demand
that for example, as if you really want to launch a satellite. When you go and launch
a satellite anywhere in the orbit is fine actually. That is, once you have, once you
satisfy the orbital conditions, it will keep on evolving in the orbit anywhere. So, that
in your job actually does not matter with may which point with the orbital condition
said the point; obviously, if you consider that they we nothing but a function.
So, orbital equations will happen to be an elliptic and then that elipse as a equation,
governing equation sort of things. So, if your variables happen to merge
…

Need subtitles from another video?