Probabilistics Methods in Robotics (KOM 613E) : Hakan Temeltas, Prof - DR

İTÜ-EEF Dept.
Of Control Engineering
Probabilistics Methods in Robotics

(KOM 613E)
Autum Somester
Hakan Temeltas , Prof.Dr.
KOM 613E : Probabilistics Methods in Robotics 1

İTÜ-EEF Dept. Of Control Engineering
Lecture Plan
Topic Reading
Introduction and useful concepts Chap. 1,2
Bayes filter Chap. 2
The Kalman filter Chap. 3
The Extended Kalman filter Chap. 3
The Unscented Kalman filter Chap. 3
Information Filters Chap. 3
Motion models Chap. 4
Measurement models Chap. 6
Lecture Plan
Topic Reading
EKF Localization Chap. 7
Occupancy maps Chap. 9
Particle filters Chap. 4
Particle filters: localization Chap. 8
SLAM and Scan matching Chap. 10
EKF-SLAM Chap. 10
Graph-SLAM Chap. 11
Fast-SLAM Chap. 13
Course Evaluation Criteria

Method Quantity Percentage(%)
Midterm Exams __ __
Homeworks __ __
Final Exam __ __
(or Final Project)

References
1. Thrun, S., Burgard, W., & Fox, D. (2005). Probabilistic
robotics. Intelligent robotics and autonomous agents.
Cambridge, Mass: MIT Press.
2. Montemerlo, M., & Thrun, S. (2007). FastSLAM: A
scalable method for the simultaneous localization and
mapping problem in robotics. Springer tracts in advanced
robotics, 27. Berlin: Springer.
3. Siegwart, R., & Nourbakhsh, I. R. (2004). Introduction to
autonomous mobile robots. Intelligent robotics and
autonomous agents. Cambridge, Mass: MIT Press.
4. Thrun, S., Fox, D., Burgard, W., & Dellaert, F. (2000).
Robust Monte Carlo Localization for Mobile Robots.
Artificial Intelligence , 99-141.

Lecture 1: Introduction

“When you understand something, then

you can find the math to express that
understanding. The math doesn’t provide
the understanding.”— Lamport

Introduction: Why Probabilistic Robotics?

• Examples of Robotics Systems :
– Mobile Platforms,
– Industrial Robotics manipulators,
– driverless cars
– …
• Applications :
– Planetary explorations
– Assembly lines
– Robot assistive surgery
– ……
 Robotics systems accumulates enormous level
of UNCERTAINTY


• Examples of Robotics Systems :
Motion
Perception
Control
Cognition Localization
Operator Commands Prior Knowledge

Trying to navigate in unknown environment Usually a robot
would create an internal model of this environment calculate
control signals based on this model

In navigation problem, each component has own
uncertainty !! Due to :
1) Unpredictable environments
a) they could be dynamic, unstructured, …
2) Errors in the sensors used to detect environments
a) sensitive to environment,
b) discretization errors, …
3) Computer models themselves have errors
a) low resolution,
b) limited set of features,
c) linear approximations, …


4) Algorithmic approximations
a) For real-time systems,
b) approximate algorithms are used which reduce
accuracy
5) Robot Actuation is unpredictable
a) control noise,
b) mechanical failure,
c) low power…

Introduction: Revisit Probability Theory

• X : a random variable and x denote a specific
event that X might take.
• i.e. : a coin flip, where X can take on the values
head or tail.
 p(X = x) : the probability that the
random variable X has value x
 p(X = head) = p(X = tail) = 1/2

x
 p(X = x)  1: Discrete probabilities sum to
one
 probabilities are always non-negative ( p(X = x)  0 )


• To simplify the notation, Use p(x) instead of
writing p(X = x)
• Continuous random variables: Take on a
continuum of values.
• All continuous random variables have probability
density functions (PDFs)
• A common density function:
“ one-dimensional normal distribution

with mean  and variance 2 ”

• Normal distribution is given by a Gaussian
function :  2
 1 ( x ) 
1  
p(x)=  2 
  2
2 
 2 

2 e
• This will be mostly shown : N( x;  ,  2 )
• Normal distributions over vectors are called
multivariate.
1
 1 
p( x)  det  2   exp  ( x   )  ( x   ) 
 T 1
2
 2 
 : mean vector ;  : covariance matrix

• PDF always integrates to 1 :
 p ( x ) dx  1
• However , a PDF is not bounded above by 1
• The following terms are identical:
“ probabiltiy ≡ probablity density ≡ pdf ”
• joint distribution of two random variables X and Y is

given by
p( x, y)  p( X  x and Y  y )


Continuous Random Variables :
• X takes on values in the continuum.
• p(X=x), or p(x), is a probability density function(pdf).
a function which can be integrated to obtain the
probability that the random variable takes a value in a
given interval.
b 
Pr( x  (a, b))   p( x)dx Pr( x)   p( x)dx  1
a 
• E.g. p(x)
x

Some Probability Density Functions :
• Uniform:
• Triangular:


Exponential pdf :


Normal (Gaussian) pdf :


PDF in 2 dimensions:
• p(x, y) = probability density of random
variables (X, Y ) at location (x, y) :


• If X and Y are independent :
p( x, y)  p( x) p( y)
• If we know that Y ’s value is y ; Then the probability,
that X’s value is x, defined conditionaly
p ( x, y )
p( x y ) 
p( y )
• If X and Y are independent :
p ( x) p ( y )
p( x y )   p ( x)
p( y )

• theorem of total probability:
p ( x)   p( x y ) p ( y ) : disrete case
y
p( x)   p ( x y ) p ( y )dy : continious case
• Bayes rule : Reletes contidional probability

p( x y ) and it’s inverse p( y x)


• Bayes Rule:
p ( y x) p ( x) p ( y x) p ( x)
p( x y )   (disrete)
p( y)  x p( y x) p( x)
p ( y x) p ( x) p ( y x) p ( x)
p( x y )   (continious)
p( y)  p( y x) p( x)


• Bayes rule plays an important role in
probabilistic robotics.
• If x is a quantity that we would like to infer
from y:
– probability p(x) : “prior probability distribution”
– y : “data (e.g., a sensor easurement)”
– p( x y ) : “posterior probability distribution”
• Denominator of bayes rule (p(y)) does not
depend on x
 p(y)-1 is written as a normalizer : 

• Bayes rule with normalizer :
p ( x y )   p ( y x) p ( x)
• The expectation of a random variable X:
E  X    xp( x) (discrete)
x
E  X    p( x)dx (continious)
• Entropy of a random variable:
H ( P)  E   log 2 p( x)    p( x) log 2 p( x)
x
(The entropy is the expected information that the value
of x carries: -log2p(x) is the number of bits required)

• An additional random variable (i.e. z
variable). p ( y x, z ) p ( x z )
p( x y , z ) 
p( y z )
• x and y are independed on z:
p( x, y z )  p( x z ) p( y z ) :conditional independence
• This is equavalent:
p( x z )  p( x z , y )
p( y z )  p( y z , x)
Introduction: Statistical Tools

• Mean (Expectation) :
– Average value
– Is the expected value of some random variable
(also known 1st moment)
– μ  EX
– μ   xp(x)dx (continuous case)
– μ  i 0 x i p(xi )
n
(discrete case)


• Function Expectation :
– E[f(X)] is the expected value of f(x) where x is
drawn from X’s distribution
– Note that in general E[f(X)]  f(E[X])

– If f(x) is a linear function:
E[aX + b] = aE[X] + b


• Variance :
– Is a measure of the variability of a random variable
– 
Var X   E X  μ 
2

–  2   x -  2 p(x)dx (continuous case)
σ  i 0 x i - μ  p(xi )
n
– 2 2
(discrete case)
– Note the square root of variance is the standard deviation


• Covariance :
– is a measure of how much two random variables
change together.
– Cov X, Y   E X  μ Y  ν   E XY   μν
– If two variables tend to vary together (that is, when

one of them is above its expected value, then the
other variable tends to be above its expected value
too), then the covariance between the two variables
will be positive.
– when one of them is above its expected value the
other variable tends to be below its expected value,
then the covariance between the two variables will
be negative.

• Covariance and correlation:
– The correlation coefficient ρxy is defined by the
ratio:
– Two variables are called uncorrelated if their

covariance is 0,
– if X and Y are normal random variables and their
correlation coefficient is 0, then they are
independent.
– Cov[X, Y ] = 0 , ρxy = 0, E[XY ] = E[X]E[Y ]


• Covariance Matrix:
– Given n random variables {X1, . . . ,Xn} their
covariance matrix is defined by:
 C11  C1n 
      
C n1  C nn 

Cij  Var X i , X j 


• Normal or Gaussian Distribution :
– A particularly useful pdf is the normal distribution.
Normalized Gaussian curves with expected value μ and variance σ2. The corresponding parameters are
a = 1/(σ√(2π)), b = μ, c = σ
- p(X) = (2πσ)-0.5 exp{-0.5 (x- μ)2/σ2}
- p(X) = det(2πΣ)-0.5 exp{-0.5 (x- μ)TΣ-1 (x- μ)} : where x, μ are vectors and Σ is a matrix


• Multivariate Normal distributions :
– If x is a multi-dimensional vector (column vector):
x=(x1, x2, …xn}T
– the density function of the multivariate Normal
distribution s:


• Bayes Formula :
P ( x, y )  P ( x | y ) P ( y )  P ( y | x ) P ( x )

P( y | x) P( x) likelihood  prior
P( x y )   : posterior
P( y ) evidence

Statistical Notation in Prob. Robotics

State : vector x
• A collection of all aspects of the robot and its environment
that can impact the future.
• These aspects can be static or dynamic
• we denote the state as x which is comprised of many
variables.
• the state at time t is xt
Measurements : vector z (the state at time t is zt)
• provides information about a momentary state of the
environment (camera images, range scans,…)
Control Inputs : vector u (the control at time t is ut)
• carry information about the change of state in the
environment (the velocity of a robot.)
Introduction:
Statistical Notation in Prob. Robotics
Typical state variables include:
• Pose – location and orientation relative to a global
coordinate frame. Rigid mobile robots have up to 6
variables max. 3 for planar.
• Configuration – more often used in manipulators, includes
the joint angles of revolute joints
• Velocities - Rigid mobile robots have up to 6 variables max.
3 for planar. (one for each pose variable)
• Feature locations – E.g. tree, wall, etc. They can be just x,y
or full pose. There can from 1 to several billion features.
• Landmarks – features that are static, distince and
recognized reliably.
Introduction:
State Estimation Example
• Problem: detect if a door open or closed?
• Have some measurement z
• Rephrase problem: What is the probability
that the door is open given z?
What is P(open|z)?

Introduction:
• Method: We note that Problem:
– P(open|z) is diagnostic knowledge
– P(z|open) is causal knowledge
– Often causal knowledge is easier to obtain
– Bayes rule allows us to use one to calculate the other:
P( z | open) P(open)
P(open | z ) 
P( z )

Introduction:
Sample Data :
• P(z|open) = 0.6 P(z|open) = 0.4
• P(open) = P(open) = 0.5
P( z | open) P(open)
P(open | z ) 
P( z | open) p(open)  P( z | open) p(open)
0.6  0.5 3
P(open | z )    0.6
0.6  0.5  0.4  0.5 5
• z raises the probability that the door is open

Introduction:
Combining Evidence :
• Suppose our robot obtains another observation
z2.
• How can we integrate this new information?
• More generally, how can we estimate
P(x| z1...zn )?
Recursive Bayesian Updating
P( zn | x, z1,, zn  1) P( x | z1,, zn  1)
P( x | z1,, zn) 
P( zn | z1,, zn  1)
Introduction:
Markov assumption:
• zn is independent of z1,...,zn-1 if we know x.
P( zn | x)P( x | z1,  , zn  1)
P ( x | z1, , zn) 
• x P( zn | z1, , zn  1)
  P( zn | x)P( x | z1,  , zn  1)
 1...n[  P( zi | x)]P( x)
i 1...n
An equivalent statement to the above is:

“The future is independent of the past if the
present is known. ”
Introduction:
State Estimation Example 3
• Consider closing the door problem
• pdf is depicted as STATE TRANSITION:
P(x|u,x’) for u = “close door”:
0. 9
0. 1 open cl osed 1
0
• If the door is open, the action “close door”
succeeds in 90% of all cases.
Introduction:
• Given probability of current state x’, and a control
input u, we can calculate the probability of new
state x by integrating over all possible paths to get
to x.
• Continuous case:
P( x | u)   P( x | u, x' ) P( x' )dx'

• Discrete case:
P( x | u )   P( x | u , x' ) P( x' )
Introduction:
Bayes Filters: Framework
Given :
• Stream of observaions z and action data u:
dt={u1, z1,…., ut, zt}
• Sensor model : P(z  x)
• Motion model : P(x  u,x’)
• Prior probability of system state P(x).
Desired :
• Estimate of state x of a dynamical system
• Posterior of the state also called Belief :
Bel(xt)=P(xt  u1, z1,…., ut, zt )
Introduction: Bayes Filters

Markov Assumption :
• Using the Markov Assumption, we can view the estimation
process as:
p( zt | x0:t , z1:t , u1:t )  p( zt | xt ) p( xt | x1:t 1 , z1:t , u1:t )  p( xt | xt 1 , ut )

Underlying Assumptions
• Static world
• Independent noise
• Perfect model, no approximation errors

Bel ( xt )   P( zt | xt )  P( xt | ut , xt 1 ) Bel ( xt 1 ) dxt 1
1. Algorithm Bayes_filter( Bel(x),d ):
2. 0
3. If d is a perceptual data item z then
4. For all x do
5. Bel ' ( x)  P( z | x) Bel ( x)
6.     Bel ' ( x)
7. For all x do
8. Bel ' ( x)   1Bel ' ( x)
9. Else if d is an action data item u then
10. For all x do
11. Bel ' ( x)   P( x | u, x' ) Bel ( x' ) dx'
12. Return Bel’(x)


Bayes Filters are Familiar! :
Bel ( xt )   P( zt | xt )  P( xt | ut , xt 1 ) Bel ( xt 1 ) dxt 1
• Kalman filters
• Particle filters
• Hidden Markov models
• Dynamic Bayesian networks
• Partially Observable Markov Decision
Processes (POMDPs)

Summary
• Bayes rule allows us to compute probabilities
that are hard to assess otherwise.
• Under the Markov assumption, recursive
Bayesian updating can be used to efficiently
combine evidence.
• Bayes filters are a probabilistic tool for
estimating the state of dynamic systems.

Next class…
• State Estimation by Gaussian Filter : Kalman

Filters(KF) and Extended Kalman Filters (EKF).
have any questions or concerns!

Probabilistics Methods in Robotics (KOM 613E) : Hakan Temeltas, Prof - DR

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Probabilistics Methods in Robotics (KOM 613E) : Hakan Temeltas, Prof - DR

Uploaded by

Copyright:

Available Formats

İTÜ-EEF Dept.

Probabilistics Methods in Robotics

Hakan Temeltas , Prof.Dr.

KOM 613E : Probabilistics Methods in Robotics 1

Course Evaluation Criteria

KOM 613E : Probabilistics Methods in Robotics 6

KOM 613E : Probabilistics Methods in Robotics 7

KOM 613E : Probabilistics Methods in Robotics 9

“When you understand something, then

KOM 613E : Probabilistics Methods in Robotics 10

Introduction: Why Probabilistic Robotics?

KOM 613E : Probabilistics Methods in Robotics 11

Introduction: Why Probabilistic Robotics?

Operator Commands Prior Knowledge

Introduction: Why Probabilistic Robotics?

KOM 613E : Probabilistics Methods in Robotics 13

Introduction: Why Probabilistic Robotics?

KOM 613E : Probabilistics Methods in Robotics 14

Introduction: Revisit Probability Theory

KOM 613E : Probabilistics Methods in Robotics 15

Introduction: Revisit Probability Theory

“ one-dimensional normal distribution

Introduction: Revisit Probability Theory

Introduction: Revisit Probability Theory

“ probabiltiy ≡ probablity density ≡ pdf ”

• joint distribution of two random variables X and Y is

KOM 613E : Probabilistics Methods in Robotics 18

Introduction: Revisit Probability Theory

Introduction: Revisit Probability Theory

KOM 613E : Probabilistics Methods in Robotics 20

Introduction: Revisit Probability Theory

KOM 613E : Probabilistics Methods in Robotics 21

Introduction: Revisit Probability Theory

KOM 613E : Probabilistics Methods in Robotics 22

Introduction: Revisit Probability Theory

KOM 613E : Probabilistics Methods in Robotics 23

Introduction: Revisit Probability Theory

Introduction: Revisit Probability Theory

p( x)   p ( x y ) p ( y )dy : continious case

• Bayes rule : Reletes contidional probability

KOM 613E : Probabilistics Methods in Robotics 25

Introduction: Revisit Probability Theory

KOM 613E : Probabilistics Methods in Robotics 26

Introduction: Revisit Probability Theory

Introduction: Revisit Probability Theory

Introduction: Revisit Probability Theory

Introduction: Statistical Tools

KOM 613E : Probabilistics Methods in Robotics 31

Introduction: Statistical Tools

– Note that in general E[f(X)]  f(E[X])

KOM 613E : Probabilistics Methods in Robotics 32

Introduction: Statistical Tools

– Note the square root of variance is the standard deviation

KOM 613E : Probabilistics Methods in Robotics 33

Introduction: Statistical Tools

– If two variables tend to vary together (that is, when

Introduction: Statistical Tools

– Two variables are called uncorrelated if their

KOM 613E : Probabilistics Methods in Robotics 35

Introduction: Statistical Tools

KOM 613E : Probabilistics Methods in Robotics 36

Introduction: Statistical Tools

KOM 613E : Probabilistics Methods in Robotics 37