This document appears to be lecture slides for a course on controlling mobile robots. It introduces linear systems and how they can be used to model robotic systems. Specific examples are provided, including modeling a point mass, pendulum, two robots on a line meeting at a point, and the process of linearizing nonlinear systems around an operating point to obtain linear approximations. The document discusses writing systems in state-space form and computing Jacobian matrices to obtain the A, B, C matrices for linear time-invariant models. The overall content focuses on modeling robotic and other physical systems using linear differential equations and state-space representations.
This document appears to be lecture slides for a course on controlling mobile robots. It introduces linear systems and how they can be used to model robotic systems. Specific examples are provided, including modeling a point mass, pendulum, two robots on a line meeting at a point, and the process of linearizing nonlinear systems around an operating point to obtain linear approximations. The document discusses writing systems in state-space form and computing Jacobian matrices to obtain the A, B, C matrices for linear time-invariant models. The overall content focuses on modeling robotic and other physical systems using linear differential equations and state-space representations.
This document appears to be lecture slides for a course on controlling mobile robots. It introduces linear systems and how they can be used to model robotic systems. Specific examples are provided, including modeling a point mass, pendulum, two robots on a line meeting at a point, and the process of linearizing nonlinear systems around an operating point to obtain linear approximations. The document discusses writing systems in state-space form and computing Jacobian matrices to obtain the A, B, C matrices for linear time-invariant models. The overall content focuses on modeling robotic and other physical systems using linear differential equations and state-space representations.
This document appears to be lecture slides for a course on controlling mobile robots. It introduces linear systems and how they can be used to model robotic systems. Specific examples are provided, including modeling a point mass, pendulum, two robots on a line meeting at a point, and the process of linearizing nonlinear systems around an operating point to obtain linear approximations. The document discusses writing systems in state-space form and computing Jacobian matrices to obtain the A, B, C matrices for linear time-invariant models. The overall content focuses on modeling robotic and other physical systems using linear differential equations and state-space representations.
Mobile Robots Dr. Magnus Egerstedt Professor School of Electrical and Computer Engineering
Module 3 Linear Systems
How make mobile robots move in effective, safe,
predictable, and collaborative ways using modern control theory?
School
of
Electrical
and
Computer
Engineering
Lecture 3.1 A Simple Robot
So far we have seen Controls Basics Mobile Robots Need to be a bit more systematic in our discussion Rich class of models: LINEAR SYSTEMS!! General enough Simple enough Expressive enough Relevant enough
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.1.1
Controlling a Point Mass
Given a point mass on a line whose acceleration is directly controlled:
u p = u p Want to write this on a compact/general form
x1 = p x2 = p
x 1 = x2
x 2 = u
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.1.2
On State Space Form
x 1 = x2 x 2 = u
x=
x1 x2
Store these variables in a single, 2-dimensional state
variable
x =
x 1 x 2
y = p = x1 =
x2 u
1 0
0 1 0 0
x1 x2
0 1
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.1.3
On State Space Form
x 1 x 2
x2 u
= x = =
y = p = x1 = 1 0 x
Or, even more generally
A=
0 0
1 0
, B=
0 1 0 0
0 1
, C=
x = Ax + Bu y = Cx
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
x1 x2
1 0
0 1
3.1.4
A 2D Point-Mass px = ux py = uy
u = (ux , uy )
p = (px , py ) x 1 = px x2 = px x 3 = py x4 = py u1 = ux u2 = uy y 1 = px y 2 = py
A=
B=
C=
0 0 0 0 0 1 0 0 1 0
1 0 0 0 0 0 0 1 0 0
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
0 0 0 0
0 0
1 0
0 0 1 0
x = Ax + Bu y = Cx
3.1.5
LTI Systems x = Ax + Bu y = Cx This is a so-called LTI (Linear Time-Invariant) System on State-Space form!
x n u m
y p
A: nn B : nm C : pn
x =
A x + B u
y =
C x
n 1 (n n)(n 1) (n m)(m 1) n1 n1
p1
(p n)(n 1) p1
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.1.6
Lecture 3.2 State-Space Models
The general LTI model is
x y u
state output input
x = Ax + Bu y = Cx
Main Question: How should the input be selected?
But First: How can such systems be understood? And where do they come from? Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.2.1
Example 1: The Car Model
Recall our old friend, the car model
c v = u v m If we care about/can measure the velocity:
x = Ax + Bu y = Cx,
c A = , B = , C = 1 m
If we care about/can measure the position we have the same
general equation with different matrices:
A=
0 0
, B=
0 c/m
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
, C=
3.2.2
Example 2: Pendulum Newtons 2nd tells us that
Not linear!
= sin() + cu
1 sin For small angles we get x = Ax + Bu, y = Cx
A=
0 g/
1 0
, B=
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
0 c
, C=
3.2.3
Example 3: Two Simple Robots
Consider two robots on a line
x 2 = u2
x 1 = u1 x1
x =
x2
1 0 0 1
(A = 0)
The Rendezvous Problem: Have them meet at the same
location Idea: Have them aim towards each other:
u1 = x2 x1 u2 = x1 x2
x =
A closed-loop feedback law!
1 1
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
1 1
x 3.2.4
Rendezvous
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.2.5
Example 4: Unicycle Robot
(x, y) x = v y = v =
1 x = v cos y = v sin =
Not linear!
We need to be more systematic/
clever when it comes to generating LTI models from nonlinear Still not linear! systems!
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.2.6
Lecture 3.3 Linearizations
Classifying systems as linear and non-linear is like
classifying objects in the Universe as bananas and non-bananas. -unknown
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.3.1
Linearizations Given a non-linear model
x = f (x, u), y = h(x)
We want to find a local, linear model around an operating point
(xo , uo ) (x = xo + x, u = uo + u) The new equations of motion become
= x x o = x = f (xo + x, uo + u) x Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.3.2
Linearizations = f (xo + x, uo + u) x Taylor expansion
f f = f (xo , uo ) + (xo , uo )x + (xo , uo )u + H.O.T x u A
h y = h(xo + x) = h(xo ) + (xo )x + H.O.T x Assumptions:
f (xo , uo ) = 0 h(xo ) = 0 Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.3.3
Linearizations
x = f (x, u)
y = h(x)
x = xo + x u = uo + u
f (xo , uo ) = 0
h(xo ) = 0
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
= Ax + Bu x
y = Cx
A = f x (xo , uo )
B =
u (xo , uo )
C = h x (xo )
3.3.4
Computing the Jacobians
x n , u m , y p , f =
f = x
f1 x1 f2 x1
f1 x2
.. .
fn x1
nn
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
.. .
f1 f2 .. . fn f1 xn f2 xn
.. .
fn xn
h1
.. , h =
hm
3.3.5
Computing the Jacobians
x n , u m , y p , f =
f = x
f1 x1 f2 x1
.. .
fn x1
f1 x2
.. .
f1 xn f2 xn
.. .
fn xn
f = u
f1 u1 f2 u1
.. .
fn u1
.. .
nm
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
f1 um f2 um
.. .
fn um
f1 f2 .. . fn
h1
.. , h =
hm
3.3.5
Computing the Jacobians
x n , u m , y p , f =
f = x
f = u
f1 x1 f2 x1
f1 x2
.. .
fn x1
f1 u1 f2 u1
.. .
fn u1
.. .
f1 xn f2 xn
.. .
.. .
f1 um f2 um
.. .
fn um
fn x n
h = x
h1 x1
.. .
hp x1
.. .
pn
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
f1 f2 .. . fn
h1
.. , h =
hm
h1 xn
.. .
hp xn
3.3.5
Example: Inverted Pendulum
= sin + u cos
x1 = x2 f (x, u) =
g/ sin(x1 ) + u cos(x1 ) x2 = , h(x) = x1 y = x1
(xo , uo ) = (0, 0) A=
f1 x1 f2 x1
f1 x2 f2 x2
= (0,0)
0 g/ cos(x1 )
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
1 0
= (0,0)
0 g/
1 0
3.3.6
Example: Inverted Pendulum
= sin + u cos
x1 = x2 f (x, u) =
g/ sin(x1 ) + u cos(x1 ) x2 = , h(x) = x1 y = x1
(xo , uo ) = (0, 0) B=
f1 u f2 u
= (0,0)
0 cos(x1 )
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
= (0,0)
0 1
3.3.6
Example: Inverted Pendulum
= sin + u cos
x1 = x2 f (x, u) =
g/ sin(x1 ) + u cos(x1 ) x2 = , h(x) = x1 y = x1
(xo , uo ) = (0, 0) C=
h x1
h x2
(0,0)
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.3.6
Example: Unicycle
(x, y)
x = v cos x1 = x, x2 = y, x3 =
y = v sin y1 = x1 , y2 = x2 , y3 = x3 = u1 = v, u2 =
(xo , uo ) = (0, 0)
1 0 1 A = 0, B = 0 0 , C = 0 0 1 0
0 0 1 0 0 1
x 2 = 0 ??? Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.3.7
Punchlines Sometimes the linearizations give reasonable models and sometimes they do not Despite the fact that they are only local approximations, they are remarkably useful (when they work)
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.3.8
Lecture 3.4 LTI Systems
u
x = Ax + Bu y = Cx
Lets figure out how such systems behave.
Start by ignoring the input term:
x = Ax x(t0 ) = x0 What is the solution to this system?
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
For higher-order systems, we just get a matrix version of this
x = Ax, x(t0 ) = x0 x(t) = eA(tt0 ) x0
matrix exponential Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.4.2
Matrix Exponentials The definition is just like for scalar exponentials
eat =
a k tk
k=0
Derivative:
k!
eAt =
Ak t k
k=0
k!
k=0
k=1
k=1
k=0
kAk tk1 Ak1 tk1 Ak t k d Ak t k =0+ =A =A dt k! k! (k 1)! k!
d At e = AeAt dt Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.4.3
Solving the Controlled Equation
The matrix exponential plays such an important role that it has its own name: The State Transition Matrix
eA(tt0 ) = (t, t0 ) d dt (t, t0 ) = A(t, t0 ) x = Ax x(t) = (t, )x( ) (t, t) = I But what if we have the controlled system: x = Ax + Bu Claim:
x(t) = (t, t0 )x(t0 ) +
(t, )Bu( )d
t0
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.4.4
Solving the Controlled Equation
Claim:
x(t) = (t, t0 )x(t0 ) +
x(t0 ) = (t0 , t0 )x(t0 ) + I
(t, )Bu( )d t0 t0
(t0 , )Bu( )d t0
x(t0 ) = x(t0 )
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.4.5
Solving the Controlled Equation
Claim:
x(t) = (t, t0 )x(t0 ) +
(t, )Bu( )d t0
d d x(t) = A(t, t0 )x(t0 ) + dt dt d dt
(t, )Bu( )d t0
d f (t, )d dt
f (t, )d = f (t, t) + t0 t0 t (t, t)Bu(t) + A(t, )Bu( )d t0
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.4.6
Solving the Controlled Equation
Claim:
x(t) = (t, t0 )x(t0 ) +
(t, )Bu( )d t0
d d t x(t) = A(t, t0 )x(t0 ) + (t, )Bu( )d dt dt t0
t d x(t) = A (t, t0 )x(t0 ) + (t, )Bu( )d + Bu(t) dt t0 d x = Ax + Bu dt Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.4.6
In Summary x = Ax + Bu, y = Cx
y(t) = C(t, t0 )x(t0 ) + C
(t, )Bu( )d t0
(t, ) = eA(t )
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.4.7
Lecture 3.5 Stability
First order of business is always trying to figure out if the system blows up or not Recall the control design objectives: Stability Tracking Robustness Other objectives
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.5.1
Scalar Systems It is useful to start with scalar systems to get some intuition about what is going on
x = ax x(t) = eat x(0)
150
a>0: 100
50
0.5
1.5
2.5
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.5
4.5
3.5.2
Scalar Systems It is useful to start with scalar systems to get some intuition about what is going on
x = ax x(t) = eat x(0)
1
a<0:
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.5
1.5
2.5
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.5
4.5
3.5.2
Scalar Systems It is useful to start with scalar systems to get some intuition about what is going on
x = ax x(t) = eat x(0)
2
a=0:
1.8
1.6
1.4
1.2
0.8
0.6
0.4
0.2
0.5
1.5
2.5
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.5
4.5
3.5.2
Three Cases Asymptotically Stable: x(t) 0, x(0) Unstable: x(0) : x(t) Critically Stable: in-between (doesnt blow up but doesnt go to zero either)
x = ax x(t) = eat x(0)
a > 0 : unstable a < 0 : asymptotically stable
a = 0 : critically stable
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.5.3
From Scalars to Matrices?
x = Ax x(t) = eAt x(0)
We cannot say that A>0, but we can do the next best
thing eigenvalues! eigenvalue C
Av = v
eigenvector n
The eigenvalues tell us how the matrix A acts in
different directions (eigenvectors) In MATLAB:
>> eig(A)
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.5.4
Example A=
1 0
0 1
1 = 1, 2 = 1, v1 =
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
1 0
, v2 =
0 1
3.5.5
Stability x = Ax x(t) = eAt x(0) Asymptotically Stable (if and only if): Unstable (if):
Re() < 0, eig(A)
We will design for this!
eig(A) : Re() > 0
Critically Stable (only if):
Re() 0, eig(A)
Critically Stable (if): one eigenvalue is 0 and the rest have
negative real part OR two purely imaginary eigenvalues and the rest have negative real part Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.5.6
A Tale of Two Pendula
A=
A=
0 1 1 0
0 1 1 0
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
1 = j, 2 = j critically stable! oscillates!
1 = 1, 2 = 1 unstable!
3.5.7
Lecture 3.6 Swarm Robotics
Lets use the stability results to solve the so-called Rendezvous Problem in swarm robotics The setup: Given a collection of mobile agents who can only measure the relative displacement of their neighbors (no global coordinates) This is what agent i can measure
Problem: Have all the agents meet at the same (unspecified) position
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.6.1
The Two-Robot Case
We have already seen the two-robot case (scalar does not matter)
x 1 = u1
x 2 = u2
x1
x2
If the agents simply aim towards each other:
u1 = x2 x1 u2 = x1 x2
x =
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
1 1
1 1
3.6.2
The Two-Robot Case
A=
1 1
1 1
1 = 0, 2 = 2 critically stable
Fact: If one eigenvalue is 0 and all others have negative real
part, then the state will end up in the so-called null-space of A
null(A) = {x : Ax = 0}
For this particular A, the null-space is
null(A) = {x : x =
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
, }
3.6.3
The Two-Robot Case
We have that
x1 , x2 (x1 x2 ) 0
Rendezvous is achieved! If there are more than two agents, they should probably aim towards the centroid of their neighbors (or something similar)
x i =
jNi
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
(xj xi )
3.6.4
The Multi-Robot Case
x i =
jNi
(xj xi )
x1 .. x= . xN
x = Lx
Fact: If the underlying graph is
connected then the graph Laplacian L has one zero eigenvalues and the rest are positive Critically Stable!
.. null(L) = {x : x = . }
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.6.5
The Multi-Robot Case
x i =
jNi
(xj xi )
x = Lx
xi , i (xi xj ) 0, i, j Rendezvous is achieved!
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.6.6
Rendezvous
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.6.7
Beyond Rendezvous
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.6.8
And Now For Real
2 Kheperas Same PID go-to-goal controller as before The proposed high-level controllers are used to pick a intermediary goal-points Robots keep track of position using odometry Positions are communicated rather than sensed (later)
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.6.9
Lecture 3.7 Output Feedback
So now we know that the first order of business is to stabilize the system such that all the eigenvalues have negative real part
x = Ax + Bu y = Cx
? Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.7.1
Back to the Worlds Simplest Robot
u
x = y=
p 0 1 0 0 1 0
x+
u > 0 if y < 0 u < 0 if y > 0
p = u
0 1
Idea: Move towards the origin!
u = y
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.7.2
Or, In General u = Ky = KCx x = Ax + Bu = Ax BKCx = (A BKC)x Pick, if possible, K such that
Re() < 0 eig(A BKC)
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.7.3
Back to the Robot
x = x =
0 1 0 0
0 1 1 0
0 1
eig(A BKC) = j Critically stable!
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.7.4
Back to the Robot
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.7.5
Whats The Problem?
The problem is that we do not take the velocity into account! We need to use the full state information in order to stabilize this system! Problem 1 (How to do that?) Problem 2 (But we do not know x, we know y?) Next Module! (and some next lecture)
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.7.6
Lecture 3.8 State Feedback
Need to stabilize the system such that all the eigenvalues have negative real part Using state feedback!
x = Ax + Bu y = Cx
? Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.8.1
Closing the Loop
x = Ax + Bu u = Kx
closed-loop dynamics
x = Ax + Bu = Ax BKx = (A BK)x Pick, if possible, K such that the closed-loop system is stabilized, i.e.,
Re(eig(A BK)) < 0
Module 4!
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.8.2
Back to the Robot
u , x 2 , K : 1 2
K = k1 k2
0 1 0 k1 x =
0 0 1
0 1 x = x k1 k2
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
k2
3.8.3
Picking the Gains
In the next module, we will pick gains in a systematic manner, but for now, lets try
k1 = k2 = 1
0 1 A BK = 1 1 eig(A BK) = 0.5 0.866j Asymptotically stable! Damped oscillations
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.8.4
Attempt 1
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.8.5
Another Attempt k1 = 0.1, k2 = 1
0 1 A BK = 0.1 1 eig(A BK) = 0.1127, 0.8873
Asymptotically stable! No oscillations
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.8.6
Attempt 2
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
3.8.7
Eigenvalues Matter It is clear that some eigenvalues are better than others. Some cause oscillations, some make the system respond too slowly, and so forth In the next module we will see how to select eigenvalues and how to pick control laws based on the output rather than the state.
Magnus
Egerstedt,
Control
of
Mobile
Robots,
Georgia
Ins<tute
of
Technology
