Advertisement
English transcript
Hello everyone, we will continue with our lecture series in this optimal control guidance
and estimation course. So, far you have seen many concepts including linear quadratic regulator
for about two lectures and we will continue the discussion on L Q R and this lecture as
well actually.
So, this is a what is coming up today I mean this lecture. This is a outline of the outline
of the lecture first is review of L Q R using state transition matrix approach; that means,
whatever we discuss last lecture we will have a quick review of that. Because we need to
use that next for tactical missile guidance application as well just an example of how
optimal control whom gives us a platform to design a advance guidance laws and all that
more on that we will see little later. But one approach using state transition matrix
we will discuss in this lecture actually. Then towards the end of this lecture we will
also talk about frequency domain interpretation for LQR as well as some robustness property
results. So, how frequency domain interpretation people
can do that but typically when you talk about linear systems then this laplace transformers
and other things will come into picture there. So, we will have some idea about what people
talk about a frequency domain representation and how it leads to some sort of a equation
using which you can also design the matrix actually we will also give example on that
alright.
So, first is the summary of what we discussed last time this is what we discussed must revise
standard L Q R problems. We have a quadratic cross function, penalty function at the end
and cross function for the path actually and the path constraint represents to be static
equation boundary condition are given like this. This is the standard form, standard
equation that we have been discussing under the frame work of L Q R. And here we will
talk about soft constraint problem; that means, X of t f is actually free it is close to 0
by this minimization this type minimization. So, this is a this the problem that we are
talking under the frame work of a soft constraint.
So, we went ahead and derived all these things first of all we noticed that this is phi of
X and l of X and all that. And using this Hamiltonian and like that this necessary conditions
state constraint optimal that cannot be like this. And then the boundary conditions happens
to be lamda f equal to S f X f with after that we went ahead and substituted this control
expression in this third equation.
And then noted I mean we could notice that this can be written as something like X dot
lamda dot is a coupled side of equation like this where this is a nothing but the Augmental
system I think. And because it is a linear homogenous linear system we can also write
the solution in the form of equation and matrix. And hence you can write lamda X and lamda
at t is nothing but phi of t t f and X and lamda at time t f. By the way just a comment
some people intent to write it in something like a initial condition also. And other words
it will be phi of t t 0 and then X lamda of t 0 again.
It depends on your situation and then which is the advantage something like that sometimes
in the literature you may see something like that both are correct actually any way. So,
writing this way then we could partition this matrix as some phi 1 1 and phi 1 2 phi 2 1
phi 2 2 sort of thing and then.
You could notice that using this boundary condition lambda f equal to S f X f. You can
also write this kind of equation where X of t is dictated by X of f through the state
transition matrix where X of t t f this algebra we discussed last time already. And similarly,
you can write also write lambda of t in the second half of the equation and the through
a fairly similar algebra leading towards that actually.
So, when you do that when we write it something like this then I expect X of t to be a X t
t f times X f and lamda of t f to be matrix times X of f this is not lamda this is X of
f. So, t is equal to t f it must satisfy the boundary condition X equal to X f and lamda
f equal to S f X f. So, we can get this final boundary condition as well actually. So, using
this boundary condition this turns out to be the boundary condition for this.
And also not that the this state transition matrix will satisfy the same differential
equation. So, we have a differential equation and we have a boundary condition. So, essentially
we can integrated by chords and store the solution and think like that actually. The
problem; however, is X f is not known basically because ultimately after getting the solution
for the state transverse matrix still the solution for X of t lamda of t requires the
information about X of f, X of f is not know. So, how do you handle that now it is easy
for because the state transition matrix representation allows us to write it this as well. X of t
0 equal to X of t 0 t f into X f and that is valid because this expression is valid
for any time including initial time. So, you put initial time condition here then
you will get X of t 0 is equal to X of t X t f times t. So, that is what you write it
here and typically state transition matrices are never singular. So, you can always invert
it and get it in this way. So, X of f information is available now. So, you can go back to that
and then write X of t and lamda of t something like this because of X of t and lamda of t
is already there as an expression. Now, you got X f so, using that expression you can
substitute here you get expression for X of t and lamda of t like this it is.
So, finally we have the control expression equal to minus R inverse lamda. So, you can
substitute this lamda of t you got lamda of t here like that. And then you can always
define this matrix what you see here as some sort of a gain matrix way of t and hence you
can write minus K t times X 0. And again it gives some sort of a sample data feedback
law and if you consider wherever you are I mean wherever the condition is that mean initial
condition then you can represent this control law or something like U of t is equal to minus
K t into X t that is how you get the gain matrix for this actually.
This is all about soft constraint problem how about hard constraint problem we also
discuss about that. And it tells out that this class of problems demands that there
are part of the state vectors is equal to 0 at t equal to A. And hence we have this
sort of a formulation which talks about I mean the regular standard L Q R problem over
the boundary conditions demands that part of the states is equal to 0. And that dimension
of that subside can be as n actually in other words entire state vector can be constraint
if it is necessary do that. Then the regular I mean the analysis we discuss
about that in last class tells out the because of the hard constraint we have to augmented
I mean cross turns tells out to be something like this where I happens to be additional
logarithm multiply sort of constraint itself.
Then we carried out the similar algebra but the boundary conditions turn out to be different.
And in other words we have this first 1 to q except t f equal to 1 but after that q plus
1 n except t f equal to 0. I mean lamda of t f is equal to 0 now account for all that
in the solution procedure then you now, but at it is we collected the terms of these quantities
which are non 0 and then define this mu vector or something like this actually. Then using
these mu vector we could write this X of t and lamda of t something like this and then
the analysis were very similar to what we had before in other words the state transition
matrix will satisfy the same differential equations.
However, the boundary conditions turn out to be something like this actually. The retail
analysis and derivation and all we have already done in the last class actually. So, because
of the boundary condition difference if they very different solution actually. So, now
it terms out that we cannot always claim that the this X of t 0 t f remains a singular non
singular, but if it remains non singular then this expression is valid. And in that case
we can always right this actually; that means, the formula in the solution is fairly similar
to what we had before. But the real actual values numbers will be different because the
boundary conditions are different.
Again similar to previous analysis the continuous data in case of a continuous data; that means,
we have this t 0 verses t and all that the small type probably alright. So, that is the
small one this for again if you take wherever the states I mean initial condition then you
can write the control or something like this. Now, here the problem happens that s t goes
to t f then this approach is that but X of t f t f you see the boundary condition it
tells a one type of a row actually ones of rows which are 0. So, essentially X of t f
t f is nothing but a singular matrix. And hence there is a problem at the end actually
but that is expected because the ambition was little very high in other words we insisted
on 0 terminal error. So, I mean if you have a soft constraint this
problem typically does not arise, but you have a hard problem this problem does arise.
Then the I mean the usual approach is after some towards the very end when t approach
is very close to t f then you do not update your gain matrix. So, use the previous gain
matrix and all that actually and so, that can be some sort of a what I called a linear
actually. So, in practical problems typically you can do it and recognize that way. So,
anyway this was the summary of discussion last time and this class like to use this
concept and then apply this through an example. A good example which talks about optimal missile
guidance through state transition matrix solution of L Q R.
So, let us see how the formulation is and how and where it leads to and think like that.
Now, what is the fundamental problem of tactical missile guidance situation in a two d plane
in the plane of engagement assuming that it is some sort of a two d engagement towards
the end we can write it this way. So, there is a reference line typically parallel to
I mean to this origins sort of thing horizontal line and then there is a missile changing
the target it is a this is missile and then there is a target. So, it is changing the
target, target is running away with the velocity V T with an angle theta t with respect to
what is as called line of sight this yellow is stands for line of sight.
And similarly, the velocity of the missile is something like V M which is a which is
different angle theta M and the way it corrects the direction of V M is through applying this
lateral resolution a M. You can think of something like a centrifugal force and think like that
way. So, the movement we have some sort of a acceleration that is perpendicular to V
M then it will try to rotate the V M vector. So, remember we do not have a thrust vector
control and like that I mean do not have a thrust manipulation control. This V M with
typically is not controlled because there are several reasons for that the thing it
is not a very good control manipulation thing you cannot manipulate it as you wish in a
fast derive itself. So, V M is typical left as is it sort of thing,
but assuming that V M is larger V T; that means, the V M is moving fast compare to V
T then there is a only the direction correction of V M is will typically lead to engagement.
And this concept relation something like rotation of line of sight vector I mean this line of
sight vector what you see rotation of line of sight vector. That means, if the line of
sight vector does not rotate then ultimately it leads to something called a collision case.
And it leads to collision triangles sort of thing this is what happens actually. If it
does not I mean if you start rotating the same vector properly we manipulate properly.
Then this is what will happen in other words if this is your initial L O S after sometime
this is your L O S that is your L O S. Like that this all these after sometime they remains
parallel and a target appears larger and larger; that means, you are approaching the target
anyway. So, ultimately it leads to collision at some point of time actually that is the
whole idea. The whole idea tells us that somehow it applies some sort of a lateral resolution
this a M is called lateral resolution to V M. So, that it the turning will take place
V M turning take place and ultimately leads to a collision actually, that is the whole
idea of typical missile guidance there are several variants that of course.
And then one of the variants will talk about that A M should be applied perpendicular to
L O S instead of perpendicular to V M, that is geometrically more correct and then more
precise also. But\ applying a perpendicular vector to V M is practically easier than compare
to I mean let us compare to applying thing which is perpendicular to L O S actually.
So, there are again mechanization issues and all that and there are also results which
show the does not really matter. What will matter is the magnitude of a M will vary in
such a way that ultimately it both will lead to the similar sort of results. Now we will
compare in our something like this. And the result tells us something called proportional
navigation guidance which is there for a long time really tells us that a M is nothing but
N V M lamda dot. So, that means, if I apply a lateral resolution perpendicular to V M
something like these where it is proportional to V M as well as proportional to line of
sight rate actually. I mean this angle rate the rate at which it changes I will make it
proportional to V M and proportional to lamda dotmas well. And ultimately it leads to this
kind of a expression where N is known as navigation concern actually. And traditionally it has
been shown before that if N is equal to 3 then it tells some sort of a optimal performance.
That is not the major observations before in a classical since by many researchers.
Then it turns out that formally it can be shown that N equal to three nothing but an
optimal guidance actually. That is what the main motivation of this particular lecture
I am going to take you through this example and prove that N equal to 3 is nothing but
a linear quadratic optimal missile guidance actually. Anyway so, this is the problem and
we want to solve this problem without knowing this we want to come up with some sort of
a lateral resolution trajectory. Lateral resolution history rather which will leads to this collision
actually. Then we will correlate this expression with that and so, that this can be represented
and something like this actually. Anyway the system dynamics that is accounted
for is quite simple very simple rather it tell tells us the missile and the target here
this situation is lightly reversed. I have actually taken this example from so, that
book uses some sort of a notation like this I thought I will and that is why. So, anyway
missile is here it moves with a velocity V and there is target here. So, this is L O
S actually line of sight what we are applying is lateral resolution perpendicular to V which
direction a and this is the velocity vector capital V and this is a velocity vector along
the line of resolution basically. So, and if you have this one this is V and t f minus
t is something called t go. So, V into t f minus t will take you there actually assuming
V is constant. So, the velocity missile velocity assume to
be constant and also on the way we will assume that this angle that sigma what you see is
line of sight sort of things that remains small actually. So, the whole idea here is
to somehow close this y t, y t as to be nullified and if t f computation is proper; that means
this fellow is no other in mean this missile it cannot escape it as to go to the target
actually. X equal to t f is going there the question is by the time the what this is called
something called actually, by the time becomes zero your heights also becomes zero then the
point lies on the target itself. So, how do you do that we will consider the system dynamics
or something very simple. Secondly, we have been this two sort of kinematic
equations rather v dot equal to a in this direction and y dot equal to V; obviously,
the cost function what you are interested in is to minimize y f as much as possible.
So, this outside the integral and we are interested in some sort of a lateral resolution minimization
also primarily because lateral resolution leads to turning. In the movement some something
is moving on turning the projection through L Q R, is larger and it essentially leads
to something called induced stall. Induced stall is primary factor for reducing the velocity
and all that you want to enter too much of lateral resolution.
For that you can preserve the velocity; that means, you can preserve the energy actually,
essentially it leads to the high impact velocity as well as larger range actually. Think like
that those are the benefits by having a minimum lateral resolution. And additional benefit
as additional benefit is as a for a minimal if a lateral resolution demand is minimum.
Then obviously, through the autopilot loop and then the control loop and all you are
will also turn out to be minimum actually remember are typically bounded by certain
values and all that actually there rates and there values and all that. So, having lateral
resolution as minimum as possible helps us in several other things actually.
So, we account for this a square minimization on the way but the primary motivation is to
minimize y f actually. So, y f should be x los actually that is the reason why this cost
function is selected. So, the problem is the premium this it is not a hard constraint problem
it is a it is actually a soft constraint problem. And then this soft constraint problem tells
us that y f square possible at the end while on the way lateral resolution should be a
square actually. This optimum L Q R formulation as this cost function usually the system dynamics.
Now, to precede further first we have to formulate exactly put it into the l q r frame work and
for doing that we define a state vector v. And I mean v and y that is a state vector,
U that lateral resolution which is nothing but a that is our control vector or control
scalar rather here. And once you define that then you can represent the system dynamics
as X dot equal to a X plus v u where a and v turns out to be like that. And similarly,
because there is nothing in the state side here the Q as happens to be all zeros 0 0
0 where as R happens to be 1 and this and S f happens to be like this 0 0 0 c remember
the waiting is c times y f and y f is a second component of a state vector.
So, it tells out that c should not appear here, but it should appear here and that because
the definition of state vector is v y and not y v like that. And also this time to go
definition is what I told t f minus t that is the time to go definition actually; that
means, how much time is left from here to go and the target actually. So, then I mean
following our results that is we discussed I mean the few slides earlier we first we
go back to this augmentation matrix x dot lamda dot which is given something like that.
Now, we have all these matrixes right a we know b know q we know r we know s f we know.
So, using all these this augmented state matrix I can write it something like this and then
A I mean if I substitute all that it turns out to be like that.
And solution happens to be x of t lamda of t is nothing but phi of t t f into x t f and
lamda t f actually. So, essentially you remember this is a linear time in variant systems.
So, phi of t t f is nothing but phi of t minus t f that is all results in linear systems
theory you can do that as long as the system matrix is time in variant actually. So, what
is the thing I mean we want to compute some sort of a because the time in variant case
we can also compute this phi
…
and estimation course. So, far you have seen many concepts including linear quadratic regulator
for about two lectures and we will continue the discussion on L Q R and this lecture as
well actually.
So, this is a what is coming up today I mean this lecture. This is a outline of the outline
of the lecture first is review of L Q R using state transition matrix approach; that means,
whatever we discuss last lecture we will have a quick review of that. Because we need to
use that next for tactical missile guidance application as well just an example of how
optimal control whom gives us a platform to design a advance guidance laws and all that
more on that we will see little later. But one approach using state transition matrix
we will discuss in this lecture actually. Then towards the end of this lecture we will
also talk about frequency domain interpretation for LQR as well as some robustness property
results. So, how frequency domain interpretation people
can do that but typically when you talk about linear systems then this laplace transformers
and other things will come into picture there. So, we will have some idea about what people
talk about a frequency domain representation and how it leads to some sort of a equation
using which you can also design the matrix actually we will also give example on that
alright.
So, first is the summary of what we discussed last time this is what we discussed must revise
standard L Q R problems. We have a quadratic cross function, penalty function at the end
and cross function for the path actually and the path constraint represents to be static
equation boundary condition are given like this. This is the standard form, standard
equation that we have been discussing under the frame work of L Q R. And here we will
talk about soft constraint problem; that means, X of t f is actually free it is close to 0
by this minimization this type minimization. So, this is a this the problem that we are
talking under the frame work of a soft constraint.
So, we went ahead and derived all these things first of all we noticed that this is phi of
X and l of X and all that. And using this Hamiltonian and like that this necessary conditions
state constraint optimal that cannot be like this. And then the boundary conditions happens
to be lamda f equal to S f X f with after that we went ahead and substituted this control
expression in this third equation.
And then noted I mean we could notice that this can be written as something like X dot
lamda dot is a coupled side of equation like this where this is a nothing but the Augmental
system I think. And because it is a linear homogenous linear system we can also write
the solution in the form of equation and matrix. And hence you can write lamda X and lamda
at t is nothing but phi of t t f and X and lamda at time t f. By the way just a comment
some people intent to write it in something like a initial condition also. And other words
it will be phi of t t 0 and then X lamda of t 0 again.
It depends on your situation and then which is the advantage something like that sometimes
in the literature you may see something like that both are correct actually any way. So,
writing this way then we could partition this matrix as some phi 1 1 and phi 1 2 phi 2 1
phi 2 2 sort of thing and then.
You could notice that using this boundary condition lambda f equal to S f X f. You can
also write this kind of equation where X of t is dictated by X of f through the state
transition matrix where X of t t f this algebra we discussed last time already. And similarly,
you can write also write lambda of t in the second half of the equation and the through
a fairly similar algebra leading towards that actually.
So, when you do that when we write it something like this then I expect X of t to be a X t
t f times X f and lamda of t f to be matrix times X of f this is not lamda this is X of
f. So, t is equal to t f it must satisfy the boundary condition X equal to X f and lamda
f equal to S f X f. So, we can get this final boundary condition as well actually. So, using
this boundary condition this turns out to be the boundary condition for this.
And also not that the this state transition matrix will satisfy the same differential
equation. So, we have a differential equation and we have a boundary condition. So, essentially
we can integrated by chords and store the solution and think like that actually. The
problem; however, is X f is not known basically because ultimately after getting the solution
for the state transverse matrix still the solution for X of t lamda of t requires the
information about X of f, X of f is not know. So, how do you handle that now it is easy
for because the state transition matrix representation allows us to write it this as well. X of t
0 equal to X of t 0 t f into X f and that is valid because this expression is valid
for any time including initial time. So, you put initial time condition here then
you will get X of t 0 is equal to X of t X t f times t. So, that is what you write it
here and typically state transition matrices are never singular. So, you can always invert
it and get it in this way. So, X of f information is available now. So, you can go back to that
and then write X of t and lamda of t something like this because of X of t and lamda of t
is already there as an expression. Now, you got X f so, using that expression you can
substitute here you get expression for X of t and lamda of t like this it is.
So, finally we have the control expression equal to minus R inverse lamda. So, you can
substitute this lamda of t you got lamda of t here like that. And then you can always
define this matrix what you see here as some sort of a gain matrix way of t and hence you
can write minus K t times X 0. And again it gives some sort of a sample data feedback
law and if you consider wherever you are I mean wherever the condition is that mean initial
condition then you can represent this control law or something like U of t is equal to minus
K t into X t that is how you get the gain matrix for this actually.
This is all about soft constraint problem how about hard constraint problem we also
discuss about that. And it tells out that this class of problems demands that there
are part of the state vectors is equal to 0 at t equal to A. And hence we have this
sort of a formulation which talks about I mean the regular standard L Q R problem over
the boundary conditions demands that part of the states is equal to 0. And that dimension
of that subside can be as n actually in other words entire state vector can be constraint
if it is necessary do that. Then the regular I mean the analysis we discuss
about that in last class tells out the because of the hard constraint we have to augmented
I mean cross turns tells out to be something like this where I happens to be additional
logarithm multiply sort of constraint itself.
Then we carried out the similar algebra but the boundary conditions turn out to be different.
And in other words we have this first 1 to q except t f equal to 1 but after that q plus
1 n except t f equal to 0. I mean lamda of t f is equal to 0 now account for all that
in the solution procedure then you now, but at it is we collected the terms of these quantities
which are non 0 and then define this mu vector or something like this actually. Then using
these mu vector we could write this X of t and lamda of t something like this and then
the analysis were very similar to what we had before in other words the state transition
matrix will satisfy the same differential equations.
However, the boundary conditions turn out to be something like this actually. The retail
analysis and derivation and all we have already done in the last class actually. So, because
of the boundary condition difference if they very different solution actually. So, now
it terms out that we cannot always claim that the this X of t 0 t f remains a singular non
singular, but if it remains non singular then this expression is valid. And in that case
we can always right this actually; that means, the formula in the solution is fairly similar
to what we had before. But the real actual values numbers will be different because the
boundary conditions are different.
Again similar to previous analysis the continuous data in case of a continuous data; that means,
we have this t 0 verses t and all that the small type probably alright. So, that is the
small one this for again if you take wherever the states I mean initial condition then you
can write the control or something like this. Now, here the problem happens that s t goes
to t f then this approach is that but X of t f t f you see the boundary condition it
tells a one type of a row actually ones of rows which are 0. So, essentially X of t f
t f is nothing but a singular matrix. And hence there is a problem at the end actually
but that is expected because the ambition was little very high in other words we insisted
on 0 terminal error. So, I mean if you have a soft constraint this
problem typically does not arise, but you have a hard problem this problem does arise.
Then the I mean the usual approach is after some towards the very end when t approach
is very close to t f then you do not update your gain matrix. So, use the previous gain
matrix and all that actually and so, that can be some sort of a what I called a linear
actually. So, in practical problems typically you can do it and recognize that way. So,
anyway this was the summary of discussion last time and this class like to use this
concept and then apply this through an example. A good example which talks about optimal missile
guidance through state transition matrix solution of L Q R.
So, let us see how the formulation is and how and where it leads to and think like that.
Now, what is the fundamental problem of tactical missile guidance situation in a two d plane
in the plane of engagement assuming that it is some sort of a two d engagement towards
the end we can write it this way. So, there is a reference line typically parallel to
I mean to this origins sort of thing horizontal line and then there is a missile changing
the target it is a this is missile and then there is a target. So, it is changing the
target, target is running away with the velocity V T with an angle theta t with respect to
what is as called line of sight this yellow is stands for line of sight.
And similarly, the velocity of the missile is something like V M which is a which is
different angle theta M and the way it corrects the direction of V M is through applying this
lateral resolution a M. You can think of something like a centrifugal force and think like that
way. So, the movement we have some sort of a acceleration that is perpendicular to V
M then it will try to rotate the V M vector. So, remember we do not have a thrust vector
control and like that I mean do not have a thrust manipulation control. This V M with
typically is not controlled because there are several reasons for that the thing it
is not a very good control manipulation thing you cannot manipulate it as you wish in a
fast derive itself. So, V M is typical left as is it sort of thing,
but assuming that V M is larger V T; that means, the V M is moving fast compare to V
T then there is a only the direction correction of V M is will typically lead to engagement.
And this concept relation something like rotation of line of sight vector I mean this line of
sight vector what you see rotation of line of sight vector. That means, if the line of
sight vector does not rotate then ultimately it leads to something called a collision case.
And it leads to collision triangles sort of thing this is what happens actually. If it
does not I mean if you start rotating the same vector properly we manipulate properly.
Then this is what will happen in other words if this is your initial L O S after sometime
this is your L O S that is your L O S. Like that this all these after sometime they remains
parallel and a target appears larger and larger; that means, you are approaching the target
anyway. So, ultimately it leads to collision at some point of time actually that is the
whole idea. The whole idea tells us that somehow it applies some sort of a lateral resolution
this a M is called lateral resolution to V M. So, that it the turning will take place
V M turning take place and ultimately leads to a collision actually, that is the whole
idea of typical missile guidance there are several variants that of course.
And then one of the variants will talk about that A M should be applied perpendicular to
L O S instead of perpendicular to V M, that is geometrically more correct and then more
precise also. But\ applying a perpendicular vector to V M is practically easier than compare
to I mean let us compare to applying thing which is perpendicular to L O S actually.
So, there are again mechanization issues and all that and there are also results which
show the does not really matter. What will matter is the magnitude of a M will vary in
such a way that ultimately it both will lead to the similar sort of results. Now we will
compare in our something like this. And the result tells us something called proportional
navigation guidance which is there for a long time really tells us that a M is nothing but
N V M lamda dot. So, that means, if I apply a lateral resolution perpendicular to V M
something like these where it is proportional to V M as well as proportional to line of
sight rate actually. I mean this angle rate the rate at which it changes I will make it
proportional to V M and proportional to lamda dotmas well. And ultimately it leads to this
kind of a expression where N is known as navigation concern actually. And traditionally it has
been shown before that if N is equal to 3 then it tells some sort of a optimal performance.
That is not the major observations before in a classical since by many researchers.
Then it turns out that formally it can be shown that N equal to three nothing but an
optimal guidance actually. That is what the main motivation of this particular lecture
I am going to take you through this example and prove that N equal to 3 is nothing but
a linear quadratic optimal missile guidance actually. Anyway so, this is the problem and
we want to solve this problem without knowing this we want to come up with some sort of
a lateral resolution trajectory. Lateral resolution history rather which will leads to this collision
actually. Then we will correlate this expression with that and so, that this can be represented
and something like this actually. Anyway the system dynamics that is accounted
for is quite simple very simple rather it tell tells us the missile and the target here
this situation is lightly reversed. I have actually taken this example from so, that
book uses some sort of a notation like this I thought I will and that is why. So, anyway
missile is here it moves with a velocity V and there is target here. So, this is L O
S actually line of sight what we are applying is lateral resolution perpendicular to V which
direction a and this is the velocity vector capital V and this is a velocity vector along
the line of resolution basically. So, and if you have this one this is V and t f minus
t is something called t go. So, V into t f minus t will take you there actually assuming
V is constant. So, the velocity missile velocity assume to
be constant and also on the way we will assume that this angle that sigma what you see is
line of sight sort of things that remains small actually. So, the whole idea here is
to somehow close this y t, y t as to be nullified and if t f computation is proper; that means
this fellow is no other in mean this missile it cannot escape it as to go to the target
actually. X equal to t f is going there the question is by the time the what this is called
something called actually, by the time becomes zero your heights also becomes zero then the
point lies on the target itself. So, how do you do that we will consider the system dynamics
or something very simple. Secondly, we have been this two sort of kinematic
equations rather v dot equal to a in this direction and y dot equal to V; obviously,
the cost function what you are interested in is to minimize y f as much as possible.
So, this outside the integral and we are interested in some sort of a lateral resolution minimization
also primarily because lateral resolution leads to turning. In the movement some something
is moving on turning the projection through L Q R, is larger and it essentially leads
to something called induced stall. Induced stall is primary factor for reducing the velocity
and all that you want to enter too much of lateral resolution.
For that you can preserve the velocity; that means, you can preserve the energy actually,
essentially it leads to the high impact velocity as well as larger range actually. Think like
that those are the benefits by having a minimum lateral resolution. And additional benefit
as additional benefit is as a for a minimal if a lateral resolution demand is minimum.
Then obviously, through the autopilot loop and then the control loop and all you are
will also turn out to be minimum actually remember are typically bounded by certain
values and all that actually there rates and there values and all that. So, having lateral
resolution as minimum as possible helps us in several other things actually.
So, we account for this a square minimization on the way but the primary motivation is to
minimize y f actually. So, y f should be x los actually that is the reason why this cost
function is selected. So, the problem is the premium this it is not a hard constraint problem
it is a it is actually a soft constraint problem. And then this soft constraint problem tells
us that y f square possible at the end while on the way lateral resolution should be a
square actually. This optimum L Q R formulation as this cost function usually the system dynamics.
Now, to precede further first we have to formulate exactly put it into the l q r frame work and
for doing that we define a state vector v. And I mean v and y that is a state vector,
U that lateral resolution which is nothing but a that is our control vector or control
scalar rather here. And once you define that then you can represent the system dynamics
as X dot equal to a X plus v u where a and v turns out to be like that. And similarly,
because there is nothing in the state side here the Q as happens to be all zeros 0 0
0 where as R happens to be 1 and this and S f happens to be like this 0 0 0 c remember
the waiting is c times y f and y f is a second component of a state vector.
So, it tells out that c should not appear here, but it should appear here and that because
the definition of state vector is v y and not y v like that. And also this time to go
definition is what I told t f minus t that is the time to go definition actually; that
means, how much time is left from here to go and the target actually. So, then I mean
following our results that is we discussed I mean the few slides earlier we first we
go back to this augmentation matrix x dot lamda dot which is given something like that.
Now, we have all these matrixes right a we know b know q we know r we know s f we know.
So, using all these this augmented state matrix I can write it something like this and then
A I mean if I substitute all that it turns out to be like that.
And solution happens to be x of t lamda of t is nothing but phi of t t f into x t f and
lamda t f actually. So, essentially you remember this is a linear time in variant systems.
So, phi of t t f is nothing but phi of t minus t f that is all results in linear systems
theory you can do that as long as the system matrix is time in variant actually. So, what
is the thing I mean we want to compute some sort of a because the time in variant case
we can also compute this phi
…