chapter 4 dynamicsalonzo/books/dyn3.pdf · chapter 4 dynamics part 3 4.4 aspects of linear systems...
TRANSCRIPT
![Page 1: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/1.jpg)
Chapter 4 Dynamics
Part 3 4.4 Aspects of Linear Systems Theory 4.5 Predictive Modelling and System Identification
Mobile Robotics - Prof Alonzo Kelly, CMU RI 1
![Page 2: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/2.jpg)
Introduction • Nonlinear dynamical systems are the closest thing
to the engineering “theory of everything.” • Applies to:
– growth of bacteria – chemical reactions – financial markets – motion of the planets
• Most important and general model of a mobile robot.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 2
![Page 3: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/3.jpg)
Outline • 4.3 Aspects of Linear Systems Theory
– 4.4.1 Linear Time Invariant Systems – 4.3.2 State Space Representation of Linear Systems – 4.3.3 Nonlinear Dynamical Systems – 4.3.4 Perturbative Dynamics of Linear Systems – Summary
• 4.5 Predictive Modelling and System Identification
Mobile Robotics - Prof Alonzo Kelly, CMU RI 3
![Page 4: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/4.jpg)
Linear Time Invariant ODE s • These are of the form:
• Establishes a relationship between system state x(t) and its derivatives. – Implies that such a system will move (even when u(t)
is not present) – Called a dynamical system
Mobile Robotics - Prof Alonzo Kelly, CMU RI 7
“Forcing Function” “Input”
“Control”
![Page 5: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/5.jpg)
First Order System • Behavior governed by:
• Consider the discrete time equivalent:
• Hence output changes by an amount proportional to the distance-to-go.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 8
“Time Constant”
![Page 6: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/6.jpg)
Step Response • Useful to describe behavior of a
few special inputs. • Step response is response to
constant input applied for t >= 0. • Unforced response. Assume • Substitute into ODE: • Characteristic equation: • The roots of this equation play a
crucial role in determining system behavior.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 9
τs 1+ 0=
![Page 7: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/7.jpg)
Solution • Unforced solution: • Forced solution: • Complete solution: • For we must have • Total Solution:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 10
![Page 8: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/8.jpg)
Solution
• When t=τ, the system has moved …
• … of the total distance to the goal.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 11
![Page 9: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/9.jpg)
Laplace Transform • Extraordinarily powerful for manipulating
compounded ODEs intuitively. • Definition:
• s is a “complex frequency”
• The kernel is a damped sinusoid:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 12
![Page 10: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/10.jpg)
Laplace Transform • For a particular value of s:
• is a (function) dot product with a damped sinusoid.
• y(s) encodes the projections for every value of s. • Its just like a Fourier transform but for complex
frequency s.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 13
![Page 11: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/11.jpg)
Derivatives • Most important property for our purpose:
• Good news!
– Differentiation in the time domain is equivalent to multiplication by s.
• Bad news! – This is why differentiation amplifies noise.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 14
![Page 12: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/12.jpg)
Transforming ODEs • Recall the first order system ODE
• Transform the ODE itself:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 15
ODEs in the time domain become
algebraic eqns in the Laplace domain
![Page 13: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/13.jpg)
Transfer Function • Defined as the ratio of output to input:
• The roots of the characteristic polynomial always appear in the denominator of the transfer function.
• Known as the poles of the system. • An n-th order ODE has n poles.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 16
Characteristic polynomial!
![Page 14: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/14.jpg)
Block Diagrams • ODEs can be represented
graphically as block diagrams.
• Top is time domain, bottom is Laplace domain.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 17
uy+
- ∫ dtτ/1
y
u+
- s/1τ/1
y
y
y
![Page 15: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/15.jpg)
Special Block Diagram • This diagram:
• Is equivalent to this diagram • Derivation:
• So…
• For the 1st order system:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 18
u +
- )(sG y
)(sH
u)()(1
)(sHsG
sG+
y
ysτ+1
1u
![Page 16: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/16.jpg)
Frequency Response • Expresses the gain of the transfer function as
function of frequency: • Substitute into T(s):
• For 1st order system:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 19
![Page 17: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/17.jpg)
Frequency Response
• Huh? Decibels? – dB = 20 log10(amplitude) = 10 log10(power)
Mobile Robotics - Prof Alonzo Kelly, CMU RI 20
![Page 18: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/18.jpg)
Second Order System • One physical manifestation
is a damped oscillator:
• Newton’s second law:
• Rewrite:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 21
Physicists form
Mathematicians form
m f
yc
yk
![Page 19: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/19.jpg)
Simulation • Simulate with:
• Truthoid: You can teach yourself controls if you can write a dynamic simulator like the above.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 22
![Page 20: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/20.jpg)
2nd Order Step Response
Mobile Robotics - Prof Alonzo Kelly, CMU RI 23
![Page 21: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/21.jpg)
2nd Order Step Response • Take Laplace transform of 2nd order ODE:
• Transfer function:
• Behavior depends on the roots of the characteristic equation.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 24
![Page 22: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/22.jpg)
General 1st Order Solution • For the more general 1st order time-varying
system: • The following integrating function exists:
• The general solution therefore is: • When f(t) is constant:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 25
![Page 23: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/23.jpg)
Outline • 4.3 Aspects of Linear Systems Theory
– 4.4.1 Linear Time Invariant Systems – 4.3.2 State Space Representation of Linear Systems – 4.3.3 Nonlinear Dynamical Systems – 4.3.4 Perturbative Dynamics of Linear Systems – Summary
• 4.5 Predictive Modelling and System Identification
Mobile Robotics - Prof Alonzo Kelly, CMU RI 26
![Page 24: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/24.jpg)
State Space • Remember the special representation used for
Runge Kutta?
• State space = a minimal set of variables which can be used to predict future state given inputs: – Number of initial conditions in a differential equation.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 27
![Page 25: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/25.jpg)
Conversion of an LTI ODE • Consider the second order LTI ODE:
• Choose the state variables to be:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 28
![Page 26: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/26.jpg)
Conversion of an LTI ODE • Rewrite the second and the original ODE as:
• This is of the form:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 29
![Page 27: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/27.jpg)
Example: Damped Oscillator • By inspection:
• Hence, the system is of the form:
• Where x1 is the position and x2 is the velocity.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 30
![Page 28: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/28.jpg)
General Linear Dynamical Systems • State Equations:
• Visualize with:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 31
∫ dtu
dux y
G
F
H+
+
M
+
+
x· t( ) Fx t( ) Gu t( )+=y t( ) Hx t( ) Mu t( )+=
x
![Page 29: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/29.jpg)
Mobile Robotics - Prof Alonzo Kelly, CMU RI 32
∫ dtu
dux y
G
F
H+
+
M
+
+ x
![Page 30: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/30.jpg)
Vector Case – Constant Coefficient • When the system dynamics matrix F(t) is constant
wrt time:
• Recall: by definition (for any matrix A):
Mobile Robotics - Prof Alonzo Kelly, CMU RI 33
Φ t τ,( ) eF t τ–( )=Matrix
Exponential
eA A( )exp I A A 2
2!------ A3
3!------ …+ + + += =
![Page 31: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/31.jpg)
Solution – Vector Case • Knowing the transition matrix is equivalent to
knowing the solution because:
• This is the general solution to:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 34
x t( ) Φ t t0,( )x t0( ) Φ t τ,( )G τ( )u τ( ) τdt0
t
∫+=
x· t( ) F t( )x t( ) G t( )u t( )+=
Vector Convolution
Integral
![Page 32: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/32.jpg)
Outline • 4.3 Aspects of Linear Systems Theory
– 4.4.1 Linear Time Invariant Systems – 4.3.2 State Space Representation of Linear Systems – 4.3.3 Nonlinear Dynamical Systems – 4.3.4 Perturbative Dynamics of Linear Systems – Summary
• 4.5 Predictive Modelling and System Identification
Mobile Robotics - Prof Alonzo Kelly, CMU RI 35
![Page 33: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/33.jpg)
Nonlinear Dynamical System • Takes the form:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 36
x· t( ) f x t( ) u t( ) t, ,( )=
z t( ) h x t( ) u t( ) t, ,( )=
Nonlinear differential equation
Nonlinear algebraic equation
State Equations System Model Process Model
State Inputs
Forcing Function
Measurement Model Observer
![Page 34: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/34.jpg)
Solutions • Closed form solutions need not exist at all for
nonlinear equations. • With computers though, we can always integrate
like so:
• This case subsumes the linear case so anything true of nonlinear systems is true of a linear one. – Including the next few slides…..
Mobile Robotics - Prof Alonzo Kelly, CMU RI 37
x t( ) x 0( ) f x τ( ) u τ( ) τ, ,( ) τd0
t
∫+=
![Page 35: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/35.jpg)
Relevant Properties • Homogeneity (for some constant k):
• We say system is “homogeneous to degree n wrt u(t)”.
• u(t) must occur in f() as a factor like so:
• As a result, all terms of the Taylor series of f() over u(t) of order less than n vanish.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 38
f x t( ) k u× t( ),[ ] kn f x t( ) u t( ),[ ]×=
f x t( ) u t( ),[ ] un t( )g x t( )( )=
![Page 36: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/36.jpg)
Drift Free • All homogeneous systems are drift free.Their zero
input response is zero.
• Such systems can be stopped instantly by nulling the inputs.
• Similar to “drift-free” designation in control theory.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 39
u t( ) 0 x· t( )⇒ 0= =
![Page 37: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/37.jpg)
Reversibility & Monotonicity • Odd degree homogeneity implies a reversible
system.
• Even degree homogeneity implies monotonicity. Sign of derivative irrelevant to sign of u().
Mobile Robotics - Prof Alonzo Kelly, CMU RI 40
u2 t( ) u1 τ t–( )–= f2 t( )⇒ f1 τ t–( )–=
u2 t( ) u1 t( )–= f2 t( )⇒ f1 t( )=
![Page 38: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/38.jpg)
Outline • 4.3 Aspects of Linear Systems Theory
– 4.4.1 Linear Time Invariant Systems – 4.3.2 State Space Representation of Linear Systems – 4.3.3 Nonlinear Dynamical Systems – 4.3.4 Perturbative Dynamics of Linear Systems – Summary
• 4.5 Predictive Modelling and System Identification
Mobile Robotics - Prof Alonzo Kelly, CMU RI 41
![Page 39: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/39.jpg)
Linearizing a Nonlinear Diff Eq • Consider again:
• Suppose some u(t) generates some solution x(t). This is
the “reference trajectory”. • Suppose we want a solution for:
• The solution can be written as:
• Defines the state perturbation dx(t) as the difference in
solutions.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 42
x· t( ) f x t( ) u t( ) t, ,( )=
u' t( ) u t( ) δu t( )+=
x' t( ) x t( ) δx t( )+=
input perturbation
state perturbation
perturbed input
perturbed state
![Page 40: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/40.jpg)
Linearizing a Nonlinear Diff Eq • If the perturbed solution is a solution, then:
• Write a truncated Taylor Series at each point in
time for the derivative f():
• Where the two new matrices are the Jacobians:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 43
x'· t( ) x· t( ) δx· t( )+ f x t( ) δx t( )+ u t( ) δu t( )+ t, ,[ ]= =
f x t( ) δx t( )+ u t( ) δu t( )+ t, ,[ ] f x t( ) u t( ) t, ,[ ] F t( )δx t( ) G t( )δu t( )+ +≈
F t( )x∂
∂ fx u,
= G t( )u∂∂ f
x u,
=
![Page 41: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/41.jpg)
Linearizing a Nonlinear Diff Eq • At this point we have:
• Recall the original differential equation:
• Cancel it from the top one:
• If you know the Jacobians, you know the perturbative dynamics – the dynamics of error.
• If you know the transition matrix of that, you know the closed form solution to error dynamics. Mobile Robotics - Prof Alonzo Kelly, CMU RI 44
x· t( ) δx· t( )+ f x t( ) u t( ) t, ,( ) F t( )δx t( ) G t( )δu t( )+ +=
x· t( ) f x t( ) u t( ) t, ,( )=
δx· t( ) F t( )δx t( ) G t( )δu t( )+=Linear
Perturbation Equation
![Page 42: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/42.jpg)
Next year • Move slide 60 (or so) on perturbative dynamics od
State Est 1 here. The example will be used later in State Est1 to derive Integrated Heading error dynamics in dead reckoning.
• Also move slide 61, 62 on transition matrix.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 45
![Page 43: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/43.jpg)
Outline • 4.3 Aspects of Linear Systems Theory
– 4.4.1 Linear Time Invariant Systems – 4.3.2 State Space Representation of Linear Systems – 4.3.3 Nonlinear Dynamical Systems – 4.3.4 Perturbative Dynamics of Linear Systems – Summary
• 4.5 Predictive Modelling and System Identification
Mobile Robotics - Prof Alonzo Kelly, CMU RI 46
![Page 44: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/44.jpg)
Summary • Nonlinear dynamical systems cannot be solved in
closed form in general. • The general solution for linear, even time-varying,
dynamical systems exists. – Solution rests on Transition matrix
• Perturbative techniques linearize nonlinear differential equations – makes them solveable.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 47
![Page 45: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/45.jpg)
Outline • 4.3 Aspects of Linear Systems Theory • 4.5 Predictive Modelling and System Identification
Mobile Robotics - Prof Alonzo Kelly, CMU RI 48
![Page 46: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/46.jpg)
Introduction – the “ives” • Mobile robots must often be:
– Deliberative – decide among options – Perceptive – aware of the surroundings – Reactive – capable of fast action
• They must be both – smart and – fast
• … doing that involves tradeoffs.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 49
![Page 47: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/47.jpg)
Role of Dynamics • In support of the above, need to be …
– Predictive – able to project consequences – Active – able to execute a plan of action
• You need dynamics models for both of these.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 50
![Page 48: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/48.jpg)
Predictive Modeling • Must model …
– Information processing and propagation. – Physical vehicle / environment interaction.
• Often need to map …
– what you can do (exert forces) – what you care about (trajectory through space).
• Latter requires integrating the dynamics.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 51
![Page 49: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/49.jpg)
Outline • 4.3 Aspects of Linear Systems Theory • 4.5 Predictive Modelling and System Identification
– 4.5.1 Braking – 4.5.2 Turning – 4.5.3 Vehicle Rollover – 4.5.4 Wheel Slip and Yaw Stability – 4.5.5 Parameterization and Linearization of Dynamic
Models – 4.5.6 System Identification – Summary
Mobile Robotics - Prof Alonzo Kelly, CMU RI 52
![Page 50: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/50.jpg)
Reasons for Braking • A) Last resort response to problems.
– Collision is imminent due to • no solution or • inadequate planning or control.
• B) Deliberately slow down. – On slopes – The motion is finished. – In order to turn around.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 53
![Page 51: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/51.jpg)
Avoiding Collision • Requires precise knowledge of the time and space
required to react.
• These depend heavily on: – Speed (initial KE) – Friction (work done by friction) – Slope (change in PE)
Mobile Robotics - Prof Alonzo Kelly, CMU RI 54
Why Care about time?
![Page 52: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/52.jpg)
Braking Model • Assume brakes are applied instantly: • Free body diagram:
– Friction and Weight are coupled.
• Do heavier vehicles take longer to stop?
55 Mobile Robotics - Prof Alonzo Kelly, CMU RI
![Page 53: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/53.jpg)
Simple Model • Equate work done by external forces to initial
Kinetic energy (assume it is all used up).
• Solve for braking distance:
• Do heavier vehicles take more distance to stop?
Mobile Robotics - Prof Alonzo Kelly, CMU RI 56
12---mv2 µsmgsbrake=
s brakev2
2µ sg------------=
![Page 54: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/54.jpg)
Tangent: Falling • Do heavier objects fall faster?
Mobile Robotics - Prof Alonzo Kelly, CMU RI 57
Leaning Tower
small object
large object
![Page 55: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/55.jpg)
Impact of Slopes • Again equate work
done to initial KE:
• Solve for distance: • Effective coefficient
of friction: • Then, simply:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 58
12---mv2 µsmgcθ mgsθ–( )sbrake=
sbrakev2
2g µscθ sθ–( )-----------------------------------=
µef f µscθ sθ–( )=
sb rakev2
2µef fg---------------=
![Page 56: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/56.jpg)
Simple Model on Slopes • Critical angle exists
beyond which gravity overcomes friction….
• Atan() is highly nonlinear.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 59
µs cθ sθ– 0 θtan⇒ µ s= =
![Page 57: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/57.jpg)
General Case • More generally:
• Robots can compute this. – The terrain shape is known. – Keep integrating until KE exhausted. – Final value of s is stopping distance.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 60
F sd•
0
s
∫ 12---mv2=
![Page 58: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/58.jpg)
Rough Heuristic for Slopes • Make small angle assumptions:
• Change in effective coefficient:
• Ratio of sloped to level stopping distance:
• Stopping distance increases or decreases by the factor
Mobile Robotics - Prof Alonzo Kelly, CMU RI 61
cθ 1= sθ θ=
µef f θ( ) µs cθ sθ–( ) µs θ–≈=
10% slope reduces µ by 0.1
sθ
s0---- 1
1 θµ s-----–
-------------- 1 θµs-----+≈=
θ µ s⁄
![Page 59: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/59.jpg)
Outline • 4.3 Aspects of Linear Systems Theory • 4.5 Predictive Modelling and System Identification
– 4.5.1 Braking – 4.5.2 Turning – 4.5.3 Vehicle Rollover – 4.5.4 Wheel Slip and Yaw Stability – 4.5.5 Parameterization and Linearization of Dynamic
Models – 4.5.6 System Identification – Summary
Mobile Robotics - Prof Alonzo Kelly, CMU RI 62
![Page 60: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/60.jpg)
Turning • Goal is to cause terrain to exert a moment on the
vehicle – By 3rd law, vehicle must exert a moment on the
terrain.
• May actuate: – Wheel steering (Ackerman) – Wheel speeds (Differential, skid)
Mobile Robotics - Prof Alonzo Kelly, CMU RI 63
![Page 61: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/61.jpg)
Simple Motion Prediction • For small steer angles:
• Integrate the differential equations using “back
substitution”:
• Errors in steering are integrated twice to determine errors in predicted position.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 64
κ t( ) α t( )=
θ t( ) θ0 V t( )α t( )dt0
t
∫+=
x t( ) x0 V t( ) θ t( )( )dtcos0
t
∫+=
y t( ) y0 V t( ) θ t( )( )sin dt0
t
∫+=
Note mapping from inputs to outputs are
integrals.
The mapping from steer angle and
velocity onto the path the robot
follows. Assumes flat terrain.
![Page 62: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/62.jpg)
Reverse Turn @ Multiple Speeds • A curvature step is the
most ambitious maneuver.
• Not modeling steering response leads to collisions with obstacles above 3.5 m/sec speed.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 65
• One Curvature • Various Speeds
“Reverse Turn”
![Page 63: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/63.jpg)
Recall: Speed Coupling • Due to vehicle
dynamics…….
• The path followed is generally a function of speed.
• Therefore, they must be estimated together.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 66
![Page 64: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/64.jpg)
Reverse Turn @ Multiple Curvatures • Different steering
commands. Same speed (5 m/s).
• It takes a long distance to cross the forward (y) axis. – Its longer the faster
you are going.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 67
• One Speed • Various Curvatures
“Reverse Turn”
![Page 65: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/65.jpg)
Swerving • Recall our typical 2D equations of motion:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 68
![Page 66: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/66.jpg)
Swerving • Assuming velocity is constant, and
curvature rate is limited and constant, the yaw is given by:
• This gives the position coordinates as:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 69
“Clothoids”
![Page 67: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/67.jpg)
Swerving • Two limits on curvature (slipping and rollover) can
be computed from:
• Given all this, the equations for (x,y) can be integrated numerically to get….
Mobile Robotics - Prof Alonzo Kelly, CMU RI 70
![Page 68: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/68.jpg)
Swerving (Urmson)
Mobile Robotics - Prof Alonzo Kelly, CMU RI 71
1) Roughly linear!
2) Lower than stopping distance (v^2/10) at 10 m/s and beyond
Sometimes you can swerve in
time even when you cannot stop.
![Page 69: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/69.jpg)
Outline • 4.3 Aspects of Linear Systems Theory • 4.5 Predictive Modelling and System Identification
– 4.5.1 Braking – 4.5.2 Turning – 4.5.3 Vehicle Rollover – 4.5.4 Wheel Slip and Yaw Stability – 4.5.5 Parameterization and Linearization of Dynamic
Models – 4.5.6 System Identification – Summary
Mobile Robotics - Prof Alonzo Kelly, CMU RI 72
![Page 70: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/70.jpg)
Note • There is plenty of content on rollver in dyn1 too.
Check it all/
Mobile Robotics - Prof Alonzo Kelly, CMU RI 73
![Page 71: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/71.jpg)
Field Robots Motivation • Contemporary
mining, forestry, agriculture, and military vehicles, operate – on slopes and/or – at high speeds
• Field robots do rollover! – They at least need a
reactive system if predictive elements fail.
74 Mobile Robotics - Prof Alonzo Kelly, CMU RI
![Page 72: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/72.jpg)
Industrial Robots Motivation • Market forces
reward manufacturers of industrial truck that: – Are narrower, – Lift heavier loads, – Lift them higher.
• Automated industrial trucks face the same challenges.
75 Mobile Robotics - Prof Alonzo Kelly, CMU RI
Center of
Gravity
![Page 73: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/73.jpg)
PerceptOR - Yuma
Mobile Robotics - Prof Alonzo Kelly, CMU RI 76
• Some Robots live dangerously. • Listen for the distinctive “Crunch” of a ladar
sensor.
![Page 74: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/74.jpg)
UGCV – Roll Test
Mobile Robotics - Prof Alonzo Kelly, CMU RI 77
![Page 75: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/75.jpg)
Lift Truck Simulations
Mobile Robotics - Prof Alonzo Kelly, CMU RI 78
Ungoverned Governed
![Page 76: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/76.jpg)
Rollover • More likely in factory and field robots. • Happens due to combinations of:
– narrow wheel spacing, – high centers of gravity – high inertial forces (speeds and curvatures) – steep slopes
• Incidents may be: – Terrain induced (slide sideways into a curb) – Maneuver induced (turn too sharp on a hill)
Mobile Robotics - Prof Alonzo Kelly, CMU RI 79
![Page 77: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/77.jpg)
Examples • Tipover when stopping on
a downslope.
• Rollover when turning sharply.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 80
![Page 78: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/78.jpg)
Forms of Instability • Must distinguish two events:
– Point of wheel liftoff (still recoverable) – Point were cg passes over wheels (irrecoverable)
• The first occurs first and is easier to detect – Does not require knowledge of inertia.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 81
![Page 79: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/79.jpg)
NOTE • The book was updated to use a singel figure and
to not reverse the direction of the reactions as the figures do here.
• That changed the signs in a few places so the figures and the math need to be updated here to be consistent with the book.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 82
![Page 80: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/80.jpg)
Static Case • For translational equilibrium:
• For rotational equilibrium:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 83
fyufyl
+ mg φsin=
fzufzl
+ mg φcos=
fzut mg φhsin+ mg φ t
2---cos=
Sum moments about lower wheel. Do cross products (r x f) in body coordinates where it is easy
t
![Page 81: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/81.jpg)
Static Liftoff • Imagine raising the slope:
– fzu decreases – fzl increases
• At some point fzu=0 and the moment balance becomes:
• Can solve for the slope at which tipping occurs:
• Using this, can compute cg height using a tilt table.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 87
mg φhsin mg φ t2---cos=
φtan t2h------=
An important/famous vehicle design parameter affecting stability.
Gravity is the only force involved.
mg
mgsφ
mgcφ
![Page 82: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/82.jpg)
Static Liftoff • Since we are talking about
a moment of a single force… – Result can be understood
in terms of the direction of gravity.
• Liftoff criterion is first satisfied when gravity vector: – emanating from the center
of gravity (cg) – points at the lower wheel
contact point. Mobile Robotics - Prof Alonzo Kelly, CMU RI 88
![Page 83: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/83.jpg)
Dynamic Case • Use D’Alemberts principle:
– I.E. treat – ma like a real force.
• Moment balance:
• Solve for lateral acceleration in g’s:
• Set to get lateral acceleration threshold.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 89
fz it– mayh– mgsφh mgc φ t
2---+ + 0=
ayg----- t
2---cφ hsφ
tfzi
mg--------–+ h⁄=
fzi0= ay
g----- t
2---cφ hsφ+ h⁄=
Vehicle is turning left Ma is reversed in sense
per D’Alembert
![Page 84: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/84.jpg)
Dynamic Case • Rewrite last result:
• Liftoff when net noncontact specific force:
• Points at the outside wheel contact point. • A pendulum mounted at the cg aligns with this
vector.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 90
ay gsφ–gcφ
-------------------- t2h------=
f g a–=
![Page 85: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/85.jpg)
Interpretations • Static case is just special case of dynamic (ay=0) • Stability increases with:
– Lower cg h – Wider tread t – Lowering slope – Decreasing acceleration
• Slowing down • Reducing curvature
Mobile Robotics - Prof Alonzo Kelly, CMU RI 91
ay gsφ–gcφ
-------------------- t2h------=
![Page 86: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/86.jpg)
Stability Pyramid • Theory generalizes to vehicles of any shape. • Stability pyramid = the pyramid formed with the
wheel contact points with the cg at the apex. • Each edge is a potential tipover axis.
– Moment is: – Unbalanced when:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 92
Wheels need not be in the same
plane.
M r f ×=
M a• 0>
![Page 87: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/87.jpg)
Implementation • Some vehicles articulate mass so the cg would
have to be (re-)calculated in real time. • An accelerometer or inclinometer works like a
pendulum, but: – It probably cannot be placed at the cg. – So, acceleration transforms are necessary.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 93
![Page 88: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/88.jpg)
Outline • 4.3 Aspects of Linear Systems Theory • 4.5 Predictive Modelling and System Identification
– 4.5.1 Braking – 4.5.2 Turning – 4.5.3 Vehicle Rollover – 4.5.4 Wheel Slip and Yaw Stability – 4.5.5 Parameterization and Linearization of Dynamic
Models – 4.5.6 System Identification – Summary
Mobile Robotics - Prof Alonzo Kelly, CMU RI 94
![Page 89: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/89.jpg)
Slip Angle • Defined for a car as:
• Alternatively using body frame velocity
components:
• Can be defined for wheels too.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 95
ζ
ψ β ψ ζ–=
β 2 Vy Vx,( )atan=
![Page 90: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/90.jpg)
Generalized Slip Angle • Define the angle between the actual and intended
velocity:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 96
ζ
ψ
actual reference
β V V⋅( ) V V( )⁄[ ]acos=
![Page 91: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/91.jpg)
Generalized Slip Equation • The velocity may be incorrect in all 3 degrees of
freedom. • Express errors in body coordinates:
Mobile Robotics - Prof Alonzo Kelly, CMU RI 97
ζ
ψ
x·
y·
θ·
cθ sθ– 0sθ cθ 00 0 1
Vx
Vy
ω
δV x
δV y
δω
+
=
Vw
R θ( ) V δV+( )=
actual reference
![Page 92: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/92.jpg)
Wheel Slip Graphs
98 Mobile Robotics - Prof Alonzo Kelly, CMU RI
0.00
-0.05
-0.10
-0.15
-0.20
-0.25 0.2 0.4 0.6 0.8 1.0 1.2
δω (r
ads/
sec)
V (m/sec)
κ=0.4
κ=0.5
κ=0.6
0.25
0.20
0.15
0.10
0.05
0.0 0.2 0.4 0.6 0.8 1.0 1.2
V (m/sec)
κ=0.4 κ=0.5 κ=0.6
δVy (
m/s
ec)
0.00
-0.05
-0.10
-0.15
-0.20
-0.25 0.2 0.4 0.6 0.8 1.0 1.2
V (m/sec)
κ=0.4 κ=0.5 κ=0.6
δVx (
m/s
ec)
![Page 93: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/93.jpg)
Removing Slip with Prediction • Slip can be expressed as a function of actual or
reference velocity (and other things): • Compensate in body coordinates.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 99
ζ
ψ
V R 1– θ( ) Vw
δV–=
actual reference
![Page 94: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/94.jpg)
Outline • Introduction • Wheel Slip • Braking • Turning & Swerving • Rollover • System Identification • Summary
Mobile Robotics - Prof Alonzo Kelly, CMU RI 100
![Page 95: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/95.jpg)
Mobile Robotics - Prof Alonzo Kelly, CMU RI 101
System (dynamics)
System Model
Input
u
Output
y
Noise υ
Measurements z
Parameter Estimation
Prediction Error
z - y
Output prediction
y
Parameter Updates
![Page 96: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/96.jpg)
Mobile Robotics - Prof Alonzo Kelly, CMU RI 102
Side Slip
![Page 97: Chapter 4 Dynamicsalonzo/books/dyn3.pdf · Chapter 4 Dynamics Part 3 4.4 Aspects of Linear Systems Theory . 4.5 Predictive Modelling and System Identification . 1. Mobile Robotics](https://reader030.vdocuments.net/reader030/viewer/2022041012/5ebf5bd94367ab303a09ab3f/html5/thumbnails/97.jpg)
Summary • Braking distance:
– increases quadratically with initial speed – depends heavily on slope
• Turning and Swerving: – predicting steering maneuvers requires calibrated
dynamic models.
• Rollover stability can be measured with a pendulum at the cg.
Mobile Robotics - Prof Alonzo Kelly, CMU RI 103