Robot Navigation in Outdoor Environments Using Markov Models

1
Robot Navigation inOutdoor EnvironmentsUsing
Markov Models

• Alberto Vale - PhD Student

vale_at_isr.ist.utl.pt
http//www.isr.ist.utl.pt http//lrm.isr.ist.utl.p
t
Instituto Superior Técnico Instituto de
Sistemas e Robótica Av.Rovisco Pais, 1 1049-001
Lisboa - Portugal
2
Objective
Robot Navigation in Outdoors Environment

• Highly non-structured environments
• Large amount of available information
• Physical area with large dimensions

3
Problem Relevance

• Safety concerns are leading to an increase in
  the use of robots. Mainly in outdoors
  environments where a communication channel might
  not be available and the robot may have to
  operate autonomously rather than being remotely
  operated by a central station
• Outdoors environments mean large and
• unstructured physical area, which can
• change in time and where scarce
• a priori information is usually
• available

4
Navigation Uncertainty
Uncertainty
Uncertainty
T1
T2
T
Mobile platform navigation along time
Uncertainty

T3
Uncertainty
Impossible to work with
5
Navigation Uncertainty Bounding
T
T1
T2
Uncertainty
Uncertainty
Uncertainty
...
Probabilistic Approach
Probabilistic Approach
...
Probabilistic Approach
6
Navigation Block Diagram
7
Environment Model
Topological Map
Summer School, EURON - EPFL, Lausanne -
Switzerland
8
Markov Models (to support robot navigation)
qt is the robot state in time instant t, qt?
s1,s2, ... ,si, ... ,sN ot is the observation
in time instant t QT q1,q2,...,qT is a
sequence of states from t1 to tT OT
o1,o2,...,oT is a sequence of observations
from t1 to tT
states of the topological map
9
Set of parameters of the model

• Initial State Distribution
• State Transition Probability Distribution
• Observation Probability Distribution

a priori information
dependent of distances between states
10
Localization
How to identify the state qt (or sequence of
states) based on observations obtained until time
instant T ?
11
Navigation
Being in state si at instant t, which are the
best observations to reach the state sj at
instant t? ?
12
Simulation Results
Experimental results of Robot Localization with 6
states (s1, s2, s3, s4, s5, s6)
Each state is identified with 3 different
attributes
(example)
13
Simulation Results
Localization probability as result of a path
execution
Pj - via points
14
Future Development

• Development of new techniques to adjust the
  model parameters aij (state transition
  probability distribution)
• Adjust the parameters kil , uil and Ril of the
  observation probability distribution according to
  attributes
• Identify new attributes (if necessary) which
  adds more information to each state
• Identify and remove useless attributes

15
Future Development
As a challenging application, this will be
applied in the Rescue Project.
This project will endow a team of two outdoors
robots with cooperative navigation capabilities
in search and rescue-like operation under
large-scale catastrophe scenarios.
The outdoor navigation will be applied on the
wheeled robot using all the sensors information
from the team.
16
Bibliography
L. Rabiner, "A Tutorial on Hidden Markov Models
and Selected Applications", Proceeding of the
IEEE, February 1989 S. Thrun, W. Burgard and D.
Fox, "A Probabilistic Approach to Concurrent
Mapping and Localization for Mobile Robots",
Machine Learning 31, pages 29-53,1998 E. Altman,
"Constrained Markov Decision Processes", Chapman
Hall/CRC, 1999 M. Kijima, "Markov
Processes for Stochastic Modeling", Chapman
Hall, 1997 A. Papoulis, "Probability Random
Variables and Stochastic Processes",
McGraw-Hill,1991