Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
8Activity
0 of .
Results for:
No results containing your search query
P. 1
Adaptive Neuro-Fuzzy Extended Kalman Filtering for Robot Localization

Adaptive Neuro-Fuzzy Extended Kalman Filtering for Robot Localization

Ratings: (0)|Views: 49|Likes:
Published by IJCSI Editor
Volume 7, Computer Science Issues, International Journal, Computer Science, IJCSI, www.IJCSI.org, Open Access, refereed journal, March 2010
Volume 7, Computer Science Issues, International Journal, Computer Science, IJCSI, www.IJCSI.org, Open Access, refereed journal, March 2010

More info:

Published by: IJCSI Editor on Apr 23, 2010
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

08/06/2013

pdf

text

original

 
IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 2, No 2, March 2010ISSN (Online): 1694-0784ISSN (Print): 1694-0814
 15
Adaptive Neuro-Fuzzy Extended KalmanFiltering for Robot Localization
Ramazan Havangi, Mohammad Ali Nekoui and Mohammad TeshnehlabFaculty of Electrical Engineering, K.N. Toosi University of TechnologyTehran, Iran
Abstract
Extended Kalman Filter (EKF) has been a popular approach tolocalization a mobile robot. However, the performance of theEKF and the quality of the estimation depends on the correct apriori knowledge of process and measurement noise covariancematrices (
Q
and
 R
, respectively). Imprecise knowledge of these statistics can cause significant degradation in performance.This paper proposed the development of an Adaptive Neuro-Fuzzy Extended Kalman Filtering (ANFEKF) for localization of robot. The Adaptive Neuro-Fuzzy attempts to estimate theelements of 
Q
and
 R
matrices of the EKF algorithm, at eachsampling instant when measurement update step is carried out.The ANFIS supervises the performance of the EKF with the aimof reducing the mismatch between the theoretical and actualcovariance of the innovation sequences. The free parameters of ANFIS are trained using the steepest gradient descent (SD) tominimize the differences of the actual value of the covariance of the residual with its theoretical value as much possible. Thesimulation results show the effectiveness of the proposedalgorithm.
 Keywords:
 Extended Kalman Filter, Localization, Fuzzy Inference System and Neuro-Fuzzy
1. Introduction
Mobile robot localization is the problem of estimating arobot pose (position, orientation) relative to itsenvironments. Two different kinds of localization exist:relative and absolute. Relative localization is also knownas dead-reckoning (DR). It is realized through themeasures provided by sensors measuring the dynamics of variables internal to the vehicle. Typical internal sensorsare encoders which are fixed to the axis of the drivingwheels. The basic drawback of this method is that theerror of robot’s position and orientation generally growsunbounded with time. Absolute localization is performedprocessing the data provided by a proper set of sensorsmeasuring some parameters of the environment in whichthe vehicle is operating. The methods to obtain absolutemeasurements can be divided into methods based on theuse of landmarks and methods based on the use of maps.The main drawback of absolute measures is theirdependence on the characteristics of the environment.Possible changes to environmental parameters may giverise to erroneous interpretation of the measures providedby the localization algorithm. In this paper, we integratethe advantages of "the relative localization" and "theabsolute localization" and make them complementary,which will enable the mobile robot to localize itself moreaccurately. To this purpose, data provided from odometric,laser range finder and MAP are combined togetherthrough EKF. The localization based on EKF proposed inthe literatures [1-9] for the estimation of robot pose.However, a significant difficulty in designing an EKF canoften be traced to incomplete a priori knowledge of theprocess covariance matrix
Q
and measurement noisecovariance matrix
 R
[10-13]. In most robot localizationapplication these matrices are unknown. On the otherhand, it is well known how poor estimates of noisestatistics may seriously degrade the Kalman filterperformance [12], [16]. One of the efficient ways toovercome the above weakness is to use an adaptivealgorithm for localization. There have been manyinvestigations in the area of adaptive algorithm for robotlocalization [5], [8], [9], [18], [19], [20]. In [15] a Neuro-fuzzy assisted extended kalman filter-based approach forsimultaneous localization and mapping (SLAM) problemis presented. In this algorithm, a Neuro-fuzzy approach isemployed to adapt the matrix
 R
only while
Q
iscompletely known.Also, as the computational load of this algorithm is veryhigh, it cannot be implemented in real-time application. Inthis paper the EKF coupled with adaptive Neuro-FuzzyInference System has been presented to adjust the matrices
Q
and
 R
. Main advantage of this algorithm, comparedto prior ones, is its fast and efficient approach in terms of 
 
IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 2, No 2, March 2010www.IJCSI.org16the computational cost and therefore its suitability for real-time applications.
2. Kinematics Modeling Robot and itsOdometery
The state of robot can be modeled as (
,,
 x y
 
) that(
,
 x y
) are the Cartesian coordinates and
 
is theorientation respectively to global environment. Thekinematics equations for the mobile robot are in thefollowing form [1-2] and [4]:()cos([])()()sin([])()sin()
vvv
 x V v v y f X V v vV vv B
   
    
(1)Where
B
is the base line of the vehicle and
u
 
is the control input consisting of a velocityinput
and a steer input
 
,
as
shown in Fig.1.
Fig.1 The robot and Feature
The process noise
v
v v v
 
is assumed to be appliedto the control input,
v
v
to velocity input, and
v
 
to thesteer angle input. The vehicle is assumed to be equippedwith a sensor (range-laser finder) that provides ameasurement of range
i
and bearing
i
 
to an observedfeature
i
  
relative to the vehicle as following:
221
()()()tan
i i iiii
 x x y yh  y y x x
 
   
(2)where(,)
i i
 x y
is the position landmark in map and
 
 
related to observation noise.
3. Extended Kalman Filter (EKF)
Kalman filter (KF) is widely used in studies of dynamicsystems, analysis, estimation, prediction, processing andcontrol. Kalman filter is an optimal solution for thediscrete data linear filtering problem. KF is a set of mathematical equations which provide an efficientcomputational solution to sequential systems. The filter isvery powerful in several aspects: It supports estimation of past, present, and future states (prediction), and it can doso even when the precise nature of the modeled system isunknown. The filter is derived by finding the estimator fora linear system, subject to additive white Gaussian noise.However, the real system is non-linear; Linearizationusing the approximation technique has been used tohandle the non-linear system. This extension of thenonlinear system is called the Extended Kalman Filter
 
(EKF). The general non-linear system and measurementform is as given by equations (3) and (4) as follows:
1
(,)
k k k k 
 x f x u w
(3)
()
k k
 z h x v
(4)The system, measurement noises are assumed to beGaussian with zero mean and are represented by theircovariance matrices
Q
and
 R
:{}0kj[]0 kj
k j
 E wQ E w w
(5){}0kj[]0 kj
k j
 E v R E v v
(6)The Extended kalman filter algorithm has two groups of equations [14]:1) The prediction equations:The extended Kalman filter predicts the future state of system
1
ˆ
 x
based on the available system model
(.)
 f 
 and projects ahead the state error covariance matrix
1
P
 using the time update equations:
1
ˆ(,)
k k
 x f x u
(7)
1
T k k k k u k u
P f P f G Q G
(8)Where
 
IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 2, No 2, March 2010www.IJCSI.org17
111222333
 f f  x y f f  f  f  X x y f f  x y
   
(9)
111222123312
u
 f u u f  f Gu u u f u u
(10)2) Measurement updates equationsOnce measurements
 z
become available the Kalmangain matrix
is computed and used to incorporate themeasurement into the state estimate
ˆ
 x
. The state errorcovariance for the updated state estimate
P
is alsocomputed using the following measurement updateequations:
1
()
T k k k k k k
K P H H P H R
(11)
ˆˆˆ(())
k k k k
 x x K z h x
(12)
()
k k k
P I K H P
(13)Where
 I 
is an identity matrix and
 H 
is following:
h H  X 
(14)In the above equations,
ˆ
 x
is an estimation of the systemsate vector
 x
and
P
is the covariance matrixcorresponding to the state estimation error defined by
ˆˆ{()())
k k k k
P E x x x x
(15)The difference between the prediction and observedmeasurements is called the measurement innovation, orresidual, generally denoted as
:
ˆ()
k k
r z h x
(16)The innovation represents the additional informationvariable to the filter in consequence to the observation
 z
. For an optimal filter the innovation sequence is asequence of independent Gaussian random variables.
4. Localization Based on EKF
We assume that robot knows map of environment. TheEKF estimates a robot pose given a map of theenvironment and range-bearing of landmarksmeasurements. For this purpose, the data provided byodometric, map and laser range-finder are fused togetherby means of an EKF. To design EKF, The continuous timemodel formulated must be reformulated in the discretetime. The discrete time kinematics model is:
(1)()(()())cos(()[()()])(1)()(()())sin(()[()()])(1)(()())()sin(()())
vvv
 x k x k T V k v k k k v  y k y k T V k v k t t v k V k v k T k v  B
   
    
 
 (17)And the discrete time observation model is:
221
(()())(()())()()()()()tan()()()()
i i iiii
 x k x k y k y k r  y k y k k  x k x k 
 
   
(18)Fig.2 briefly describes the cyclic localization procedureduring the localization procedure using EKF.
Fig.2 Localization algorithm based on EKF
The algorithm consists of such steps as positionprediction, observation, measurement prediction, matchingand estimation. The detailed discussion of these steps willbe followed.Step1) InitializationInitialize the state vector
0
ˆ
 x
and covariance matrix
0
P
of the mobile robotStep 2) Robot Position PredictionThe robot position at time step k+1 is predicted based onits old localization (at time step k) and its movement duethe control input
u
:

Activity (8)

You've already reviewed this. Edit your review.
1 hundred reads
engineer86 liked this
sarojxyz liked this
sarojxyz liked this
suganyaggn liked this
vladimir_zhidkov liked this
ei.ballz liked this
transducer liked this

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->