Professional Documents
Culture Documents
20 Slam
20 Slam
2
Introduction to Autonomous Mobile Robots (2 nd Ed), Seigwart
Instructor: Dr. Latif Anjum
The SLAM Problem
4
Introduction to Autonomous Mobile Robots (2 nd Ed), Seigwart
Instructor: Dr. Latif Anjum
The SLAM Problem
Given:
• The robot’s controls
• Relative observations
Wanted:
• Map of features
5
Introduction to Autonomous Mobile Robots (2 nd Ed), Seigwart
Instructor: Dr. Latif Anjum
SLAM Applications
Indoors Undersea
Space Underground
• Online SLAM:
p(xt , m | z1:t ,u1:t ) … p(x 1:t , m | z1:t ,u1:t ) dx1dx2 ...dxt 1
Integrations (marginalization) typically
done one at a time
yt [ xt , m1 ,..., mn 1 ]T
yt [ xt , yt , t , 0 , r0 ,..., n 1 , rn 1 ]T
As the robot moves and takes measurements, the state vector and
covariance matrix are updated using the standard equations of the
extended Kalman filter
S r S l S r S l
cos( )
xˆt xt 2
t 1
2b
yˆ y S r S l S r S l
Pˆt Fy Pt 1 Fy Fu Qt Fu
T T
t t sin( t 1 )
ˆ
t t 2 2b
S r S l
ˆ
0 0 b
rˆ0 r0 0
... ... 0
ˆ ˆ ...
n 1
n 1
rˆn 1 rn 1 0
Introduction to Autonomous Mobile Robots
0 (2 Ed), Seigwart
nd
Instructor: Dr. Latif Anjum
EKF based SLAM: Comparison with EKF localization
LOCALIZATION SLAM
The state X is ONLY the robot configuration, that The state Y is the robot configuration X plus the
is: features mi=(ri,φi) that is:
xt [ xt , yt , t ]T yt [ xt , yt , t , 0 , r0 ,..., n 1 , rn 1 ]T
The prediction function is: The prediction function is:
xˆt f ( xt 1 , ut ) yˆ t f ( yt 1 , ut )
S r S l S r S l
S r S l S r S l cos( )
cos( t 1 ) xˆt xt 2
t 1
2b
xˆt xt 1 2 2b yˆ y S r S l S r S l
sin( )
yˆ y S r S l sin( S r S l )
ˆ
t t
2
t 1
2b
t t 1 t 1 t t S S
2 2b r l
ˆt t 1 S r S l ˆ 0 0 b
rˆ0 r0 0
b
... ... 0
ˆ ˆ ...
n 1
n 1
ˆ
rn 1 rn 1 0
Pˆt Fx Pt 1 Fx Fu Qt Fu
T T
0
Pˆt Fy Pt 1 Fy Fu Qt Fu
Introduction to Autonomous Mobile Robots (2 nd Ed), Seigwart T T
Instructor: Dr. Latif Anjum
EKF based SLAM: estimation phase
yt yˆ t K t ( z h( x))
ˆ
P P K K
T
t t t IN t
where
IN HPH T R
K t PH ( IN ) 1
Introduction to Autonomous Mobile Robots (2 nd Ed), Seigwart
Instructor: Dr. Latif Anjum
EKF SLAM: considerations
At start, when the robot takes the first measurements, the covariance matrix is populated
by assuming that these (initial) features are uncorrelated, which implies that the off
diagonal elements are set to zero.
Px 0 0 ... 0 0
0 Pm0 0 ... 0 0
0 0 Pm1 ... 0 0
P0
... ... ... ... ... ...
0 0 0 ... Pmn2 0
0 0 0 ... 0 Pmn1
Pˆt Fy Pt 1 Fy Fu Qt Fu
T T