8/3/2019 4 Odometry Final
1/22
DOMETRY MOTION MODELMOTION AND MAPS
Nguyn Th L Linh
Nguyn ng KhoaNguyn Thi Ha
Nguyn Vn Tun
on Xun TrngV Th T ThanhTrn Vit Quc
Slide 1
8/3/2019 4 Odometry Final
2/22
Odometry Motion Model
Closed-form Calculation
Sampling Algorithm
Maps
Outline
2
Slide 2
8/3/2019 4 Odometry Final
3/22
ODOMETRY MOTION MODEL
Odometry is the estimation of distance and direction
from a previous location using wheel encoder
information of robot instead of commanded velocity
(Robots with wheel encoders)
3
higher accuracy than velocity
good for short term, relative position estimation
measurements only available retrospectively
unusable for accurate motion planning and control
Slide 3
8/3/2019 4 Odometry Final
4/22
Example Wheel Encoders
These modules require
+5V and GND to powerthem, and provide a 0 to5V output. They provide+5V output when they"see" white, and a 0V
output when they "see"black.
These disks aremanufactured out of highquality laminated colorplastic to offer a very crispblack to white transition.This enables a wheelencoder sensor to easilysee the transitions.
Source: http://www.active-robots.com/ 4
Slide 4
8/3/2019 4 Odometry Final
5/22
Closed-form calculation
In the time interval ( t-1, t ] :
Robot moves from to
Odometrymotion information:
yxxt 1 ''' yxxt
5
T
tttxxu
1
T
rottransrottu 21
Transform into
Slide 5
8/3/2019 4 Odometry Final
6/22
6
22 )'()'( yyxxtrans
)','(atan21 xxyyrot
12
'rotrot
Closed-form calculation
Slide 6
8/3/2019 4 Odometry Final
7/22
Here are the true values of the rotationand translation from the measured ones
with independent noise (with zero
mean) and variance b:
||||11 211
transrotrotrot
||||22 221
transrotrotrot
|||| 2143
rotrottranstranstrans
Closed-form calculation
b
Slide 7
8/3/2019 4 Odometry Final
8/22
22 )'()'( yyxxtrans
)','(atan21 xxyyrot
12' rotrot
22 )'()'( yyxxtrans
)','(atan2 1 xxyyrot
12' rotrot
)
|
|,
(prob trans21rot11rot1rot1 p|))||(|,(prob rot2rot14trans3transtrans2 p
)||,(prob trans22rot12rot2rot3 p
odometry values (u)
values of interest (xt-1, xt)
T
t yxx )(1 T
t yxx )(
),( baprob
),( 1ttt xuxp 8
Closed-form calculationAlgorithm
: Tttt xxu 1
Slide 8
8/3/2019 4 Odometry Final
9/22
9
Closed-form calculationGraphs :
Slide 9
8/3/2019 4 Odometry Final
10/22
)||sample( 21111 transrotrotrot
|))||(|sample( 2143 rotrottranstranstrans
)||sample( 22122 transrotrotrot
)cos(' 1rottransxx
)sin(' 1rottransyy
21' rotrot
Tt yxx ',',' sample_normal_distribution
10
Sampling algorithm
22 )'()'( yyxxtrans
)','(atan21 xxyyrot
12 ' rotrot
Slide 10
8/3/2019 4 Odometry Final
11/22
11
Sampling algorithmGraphs :
Slide 11
8/3/2019 4 Odometry Final
12/22
Start
Sampling from the motion model
12Slide 12
8/3/2019 4 Odometry Final
13/22
13
Maps
If robot has a map of the environment, ....
Such an opportunity gives us more information,
and also makes the model more complex. Model is
represented as p(x(t)|u(t),x(t-1),m) instead of
p(x(t)|u(t),x(t-1))
This model is called as map-based motion model
Most important problem is that there is a
probability of an unoccupied path between x(t)
and x(t-1)
This model can be calculated by:
)',|()|(),',|( xuxpmxpmxuxp 13Slide 13
8/3/2019 4 Odometry Final
14/22
14
Maps (Algorithms)
Algorithm motion_model_with_map ( x(t), u(t), x(t-1), m );
return p(x(t)|u(t), x(t-1))*p(x(t)|m)
Algorithm sample_motion_model_with_map ( u(t), x(t-1), m );
do
x(t) = sample_motion_model(u(t), x(t-1))
= p(x(t)|m)
Until >0 Return
14Slide 14
8/3/2019 4 Odometry Final
15/22
Map-Consistent Motion Model
)',|()|(),',|( xuxpmxpmxuxp
Approximation:
),',|( mxuxp)',|( xuxp
Map free estimate of
motion modelconsistency of
pose in the map
=0 when placed
in an occupied cell
Obstacle grown
by robot radius
Slide 15
8/3/2019 4 Odometry Final
16/22
5.4.3
8/3/2019 4 Odometry Final
17/22
5.4.3
8/3/2019 4 Odometry Final
18/22
MOTION AND MAPS
Slide 16
8/3/2019 4 Odometry Final
19/22
MOTION AND MAPS
Slide 17
8/3/2019 4 Odometry Final
20/22
y l chun ha thng thng
Thut ton cho m mnh di chuyn da trn bn
Slide 18
8/3/2019 4 Odometry Final
21/22
Slide 19
8/3/2019 4 Odometry Final
22/22
Nhm 3 v 4
Slide 20