You are on page 1of 6

2008 IEEE/RSJ International Conference on Intelligent Robots and Systems

Acropolis Convention Center


Nice, France, Sept, 22-26, 2008

Experimental Slip Estimation for Exact Kinematics Modeling and


Control of a Tracked Mobile Robot
S. Ali .A. Moosavian1 Arash Kalantari2
Advanced Robotics & Automated Systems (ARAS) Laboratory
Department of Mechanical Engineering, K. N. Toosi Univ. of Technology
Tehran, Iran, P.O. Box 19395-1999, Fax: (+98) 21 8867-4748
Email: moosavian@kntu.ac.ir, Kalantari@ee.kntu.ac.ir

Abstract—Tracked Mobile Robots (TMRs) can be considered risk of longitudinal and lateral slippage. So, slippage
as the most important type of mobile robots. Large contact area identification and including that in the kinematics model is
of tracks with the ground provides superior advantages for required to enhance the path tracking control of TMRs.
TMRs such as better mobility in unstructured environments, Much of the work done on tracked vehicles has been well
though it may cause a higher risk of slippage. In this paper, an
summarized in [1]. An extended Kalman filter has been used
experimental slip model is proposed for exact kinematics
modeling, and the parameters of this model will be determined in [2] to identify soil parameters, and slip of the tracks. A
based on experimental analysis of ResQuake. This is a tele- method for identifying soil parameters during traverse of a
operative rescue mobile robot with great capabilities in tracked vehicle on unknown terrain has been presented in
climbing obstacles in destructed areas, and its performance was [3]. A kinematics approach has been proposed for TMRs in
demonstrated in Rescue robot league of RoboCup 2005 in [4], in order to improve their motion control and pose
Osaka (Japan), achieving the 2nd best design award, and
estimation. A method for computing track forces and speeds
RoboCup 2006 in Bremen (Germany) achieving the best
operator interface award. Therefore, ResQuake is used here as of a planar tracked vehicle has been presented in [5].
an experimental platform to study the relationship between This paper presents an experimental model for slip
slippage of tracks and two main physically meaningful factors,
i.e. radius of the tracking path and speed of the robot. The slip coefficients of tracks of a TMR. ResQuake as depicted in
coefficients will be obtained as an exponential function of Fig. 1 is a tracked robot which an illustrative description of
radius of curvature of the path. To validate the obtained its main characteristics has been presented in [4].
results, the proposed model will be used along with two path
tracking controllers, and it is empirically demonstrated that the
developed model drastically improves the system performance
in terms of lower path tracking errors.

Index terms: Tracked Mobile Robot, Kinematics, Slippage, Slip


Coefficients, Path Tracking Control

I. INTRODUCTION
Application of mobile robotic systems has been increasing
in different areas during recent years. Many types of such
systems have been designed and manufactured that can be
categorized based on the locomotion system. Wheeled,
legged, and tracked robots are three major types of mobile Fig. 1. ResQuake, the experimental platform of this research
robots. In Tracked Mobile Robots (TMRs), the locomotion
is based on tracks which offer a large contact area with the Two exponential functions will be presented to describe
ground and consequently high mobility in unstructured slip coefficients of inner and outer tracks in terms of path
environments. Skid steering is most widely used to drive curvature. The efficiency of obtained functions is evaluated
TMRs. In this type of steering the heading angle of robot is empirically in path tracking using two different control
controlled in the same way as differentially driven wheeled schemes. It is shown that utilizing the slip functions along
mobile robots (WMRs) i.e. by imposing different speeds to with controller improves the performance of path tracking
the left and right tracks. controller. It is also concluded that for limited speeds the
Although tracks offer TMRs many real advantages, great slippage in tracks is almost independent from speed of
maneuverability for instance, but they usually increase the tracks. A thorough localization method is also presented in
this paper. This method which uses scans of a Laser Range
Finder to localize the robot is implemented with a minimum
1- Associate Professor
2- Graduate student
978-1-4244-2058-2/08/$25.00 ©2008 IEEE. 95
frequency of 10 Hz. To this end, kinematics model of the φ& = c[ω r i r − ω l il ] (4c)
robot will be developed in Section II. The localization
where b is equal to the half width of the robot, c is a constant
method and the relationship between slippage of tracks and
equal to r/2b and α is the slip angle of robot which has a
radius of path and robot speed are studied in Section III by
performing various experiments. Obtained functions are nonzero value in the presence of side slippage.
validated in Section IV where the model is used along with Lateral slippage happens mainly due to centrifugal force
exerted to the robot when moving on a curved path with a
two different path tracking control schemes. It will be
experimentally shown that the developed model drastically relatively high speed. But maximum longitudinal speed of
the chosen platform does not exceed 0.3 m/s which will
improves the system performance in terms of reduced path
result in a negligible centrifugal force. Besides, the design
tracking errors.
of tracks treads as explained in [6] results in large lateral
friction force which in turn helps the robot not to slide
II. KINEMATICS MODEL
laterally. Therefore, the lateral slippage is neglected, and
A mobile rigid platform has three degrees of freedom Eq. (4) is rewritten as:
(DOF) in a horizontal plane, which can be defined either in
x& = cb[ω r i r + ω l il ] (5a)
the body coordinate frame f : { x, y , φ } , fixed to the body of
robot with the x axis along with the tracks, or in the inertial y& = 0 (5b)
coordinate frame F : { X , Y , ϕ } fixed to the plane of motion, φ& = c[ω r i r − ω l il ] (5c)
as shown in Fig. 2. It should be mentioned that Eq.(5c) gives a positive φ& for
C counterclockwise rotations. These components can be
α transferred into the inertial frame F as:
y R X& = x& cos ϕ (6a)
Y& = x& sin ϕ (6b)
Left Track (Outer Track)
x
ϕ& = φ& (6c)
Substituting Eqs. (5) into Eqs. (6), direct kinematics will be
V recapitulated in matrix form as:
α
Y f
2b
 X  cbir cos ϕ cbil cos ϕ 
d     ω 
Y = cbir sin ϕ cbil sin ϕ   r  (7)
dt   
Right Track (Outer Track)
ω
− cil   l 
F X
Fig. 2. Coordinate frames and geometric parameters  ϕ   cir
The direct kinematics is defined in the main frame as: III. IDENTIFYING SLIP COEFFICIENTS
( X& , Y& , ϕ& ) = Ω(ω r , ω l ) (1)
A. Proposed Algorithm
which relates the velocity components to the angular speeds
Physical considerations of a TMR motion on a curved
of the right and left tracks. The speed of the right track is
path imply the dependence of longitudinal slippage to terms
calculated by:
like radius of curvature of the path and speed of the robot.
Vr = rω r (1 − i r ) = rω r ir (2) The type of surface may also affects the intensity of
where r is the driving wheel radius, ω r is the angular speed longitudinal slippage but in this research it is assumed that
and i r is the slip coefficient of this track. The longitudinal the surface remains the same type, and so its effect is not
distinguished. In order to find the relationship between the
slip of right track is also defined as following:
foregoing factors and coefficients of longitudinal slippage, a
Vt − Vrr Vrj set of experiments has been performed. ResQuake is planned
ir = = 1− (3)
Vt Vt to follow five different paths with curvatures of radius:
where Vt is the theoretical speed, and Vrr is the real speed R = 0.5q1 (m) q1 = 1,2,...,5 (8)
of the right track. Eq. (2) and (3) can be simply rewritten for and on each path, five different speeds:
the left track. v = 0.05q 2 (m / s ) q 2 = 1,2,...,5 (9)
On the other hand, the velocity components in the body Therefore, 25 experiments have been performed. In all of
frame can be obtained as: the experiments the field is paved with stone. The slip
x& = cb[ω r ir + ω l il ] (4a) coefficients are calculated in each experiment by using
Eq. (2) which requires measuring the real speed of each
y& = cb[ω r i r + ω l i l ] tan(α ) (4b) track during the experiment.

96
B. Localization Method and Experiment Setup 4 0.2

The real speed of each track is calculated by means of 3


0.15

differentiating posture (position and heading) of the robot 2


0.1

Velocity
Posture
with respect to time. This needs a very smooth and accurate 1
0.05
localization to decrease the noise resulted by differentiation 0
X(m) dX/dt
as much as possible. An overview of the proposed -1 Y(m) 0 dY/dt
φ (rad) dφ /dt
localization method and the robot control system is -2 -0.05
0 5 10 15 20 25 0 5 10 15 20 25
demonstrated in Fig. 3. time(s) time(s)
Fig. 4. Position and velocity components of the robot calculated by the
proposed localization algorithm

0 0

-0.02 -0.02

-0.04 -0.04

-0.06 -0.06

ii

ii
-0.08 -0.08
R=0.62m V=5cm/s
-0.1 R=1.34m -0.1 V=10cm/s
R=2.05m V=15cm/s
-0.12 R=2.68m -0.12 V=20cm/s
R=3.12m V=25m/s
-0.14 -0.14
5 10 15 20 25 0.5 1 1.5 2 2.5 3 3.5
V(cm/s) R(m)

0.07 0.07
R=0.62m V=5cm/s
0.06 0.06
R=1.34m V=10cm/s
Fig. 3. Localization and control system overview 0.05 R=2.05m
R=2.68m
0.05 V=15cm/s
V=20cm/s
0.04 R=3.12m 0.04 V=25m/s
0.03 0.03
io

io
In order to localize the robot a long thick plate is mounted 0.02 0.02
on top of the robot’s axis of symmetry. A Laser Range 0.01 0.01

Finder (LRF) is installed in the same height the data of 0 0

which is used to localize the robot. Localization and high -0.01


5 10 15 20 25
-0.01
0.5 1 1.5 2 2.5 3 3.5
V(cm/s) R(m)
level control are implemented on the same PC, and resulting
commands are then sent to the low level controller via a Fig. 5. Slip coefficients of the inner and outer tracks in terms of: (left)
serial connection; the robot and the LRF are connected to the Observed path curvature, (right) speed of the robot
same PC. Low level controller consists of an embedded
servomechanism that regulates the angular speed set points As illustrated in Fig. 6, by increasing the path curvature,
of driving sprockets which are the outputs of high level the slippage of tracks decreases and tends to zero. This
controller. The data of LRF regarding the long plate are implies a specific behavior for slippage in terms of path
extracted using clustering method. A line is fitted to the curvature, while it is not the case for the speed changes. It is
clustered points, the slope of which is equal to the heading also noticeable that obtained slip coefficients for the outer
angle of robot, and the coordinates of its center are equal to track are positive which means the speed of this track is less
the coordinates of robot. than the expected value whilst speed of the inner track is
An inevitable noise still exists in computed posture due to even more than what expected which results in a negative
the locomotion type of robot and intrinsic noise of LRF. A slip coefficient.
0.05 0
low pass filter is applied to the computed coordinates of
-0.02
robot to smooth the data as much as possible. The calculated 0.04
-0.04
heading angle is also filtered by means of a second order 0.03

Butterworth filter with a normalized cutoff frequency of 0.5. -0.06


io

ii

0.02
-0.08
C. Obtained Results 0.01
-0.1

Fig. 4 shows the position and velocity components of 0 -0.12


0 1 2 3 4 0 1 2 3 4
robot calculated by the localization algorithm during the R(m) R(m)

experiment for speed of 0.15 m/s and radius of curvature of Fig. 6. Slip coefficients of the inner and outer tracks in terms of path
curvature
1.5 m. The acquired data are then applied to calculate the
In order to obtain a relationship between slip coefficients
slip coefficients of the tracks in the planned experiments.
and path curvature, the speed data were averaged. An
Fig. 5 shows the calculated slip coefficients for inner and exponential function is then used to define slip coefficient as
outer tracks, for various observed path curvatures and robot a function of the radius of path:
speeds. It should be noted that in a counterclockwise turn the i (R ) = α e β R (10)
left track is the inner track and the right track is the outer where α and β can be obtained, and substituted as:
one.
io ( R ) = 0.07.e −0.68 R (11a)

97
ii ( R ) = −0.15.e −0.63 R (11b) and then it decelerates for 5 seconds until complete stop:
As mentioned before, switching between the inner and  X d = −0.0075t 2 + 0.075t

outer tracks should be regarded with a change in sign of ϕ& . 2
Yd = 0.013t − 0.13t 40.8 < t < 45.8 (14e)
Therefore, Eq. (11) can be rewritten for the left and right ϕ = 4π / 3
tracks as follows:  d
The desired trajectories are depicted in Fig. 7.
ir ( R ) = 0.07.e −0.68 R 5
 ϕ& ≤ 0 (12) X
il ( R ) = −0.15.e −0.63 R
4 d
Y
3 d

Planned Trajectories
φd
ir ( R) = −0.15.e −0.63 R 2

 ϕ& > 0 (13) 1

il ( R) = 0.07.e −0.68 R 0

-1

-2
IV. MODEL VERIFICATION AND CONTROL
-3
To verify the developed model for slip compensation two -4
0 10 20 30 40
sets of experiments are conducted, in which the robot is time(s)

planned to follow a desired circular path. Two feedforward Fig. 7. Planned trajectories for the verification experiments
and kinematic based controllers are implemented to control B. Feedforward control
the motion of robot in these experiments. In each case, the
In feedforward control schema, angular speed of driving
system performance is studied with and without slip
sprockets should be calculated for every time step and sent
compensation.
to the low level controller as a set point. This can be done
A. Trajectory planning using the following equation:
Consider that the center of TMR must follow a desired −1
ω r  cbir cbil   x& 
Cartesian trajectory (Xd(t),Yd(t)) where t ∈[0,T ] The ω  =  ci  φ& (15)
planned path consists of three major parts, two straight lines  l   r − cil   
connected with a circular arc with a radius of 1.25m. It is where x& and φ& are the linear and angular speeds of robot.
assumed that robot starts from the initial posture Therefore, the corresponding set points can be calculated by
of (0,0, π / 2) . The robot moves along the y axis and speed differentiating desired trajectories:
increases by time up to 0.15 m/s during 5 seconds:
x& = X& d2 + Y&d2 (16)
X d = 0
 2 φ& = ϕ& (17)
Yd = 0.015t t <5 (14a) d

ϕ = π / 2 Two experiments have been carried out to show the


 d efficiency of slip functions. In the first experiment the
The robot continues moving along y axis for the next feedforward commands are calculated assuming that no slip
7 seconds with a constant speed of 0.15 m/s: happens, i.e. ir = 0, il = 0 , while in the second test
X d = 0 feedforward commands are obtained by calculating slip
 coefficients using Eq. (11), and replacing them into Eq. (15).
Yd = 0.15t 5 < t < 12 (14b)
ϕ = π / 2 Fig. 8 shows the result of the both experiments. It is clear
 d that robot follows the planned trajectories much better when
Then, it moves with constant angular speed of 0.12 rad/s slippage is compensated.
and constant linear speed of 0.15 m/s until its heading angle
reaches 4π / 3 rad: C. Kinematic Based Control
 X d = − cos( 0.12t + π / 2) As a second assessment of the developed exact kinematics
 model exploiting obtained slip coefficients, the robot motion
Yd = sin(0.12t + π / 2) 12 < t < 33.8 (14c)
ϕ = 0.12t + π / 2 is controlled by means of a kinematic based controller
 d presented in [7]. A combination of a nominal feedforward
In the last portion of the path, the robot moves with command with a feedback action on the error at each time
constant speed of 0.15(m/s) for another 7 seconds: step is used in this tracking control schema. To this end,
 X d = −0.06t defined position and orientation errors in the inertial frame
 as ( X d − X , Yd − Y , ϕ d − ϕ ) can be expressed in the body
Yd = 0.10t 3 3.8 < t < 40.8 (14d)
ϕ = 4π / 3 frame using a rotation matrix as follows:
 d

98
e x   cos ϕ sin ϕ 0  X d − X 
 
e = e y  = − sin ϕ cos ϕ 0  Yd − Y  (18)
eφ   0 0 1  ϕ d − ϕ 
 
Multiplying the matrices of right hand side of Eq. (18) and
then time differentiation yields the error dynamics. Using
Eq. (5) the error dynamics is simplified to:
e& x   0 φ&d 0 e x  cos eφ   x& 
   &       (19)
e& y  = − φ d 0 0 e y  +  sin eφ  x& d −  0 
e&φ   0 0 0 eφ   0  φ& − φ&d 
  
Fig. 9. Position errors in the body frame
3

2.5 A linear feedback law is defined as:


2
u1 = − k1e x (24a)
u 2 = − k 2 sign ( x& d (t ))e y − k 3 eφ (24b)
Y(m)

1.5

Then, assuming a desired closed-loop characteristic equation


1
as:
0.5 Desired Path ( s + 2ξa )(s 2 + 2ξas + a 2 ) = 0 (25)
Observed Path With Slip Compensation
0 Observed Path Without Slip Compensation It can be obtained by choosing the following gains:
-3.5 -3 -2.5 -2
X(m)
-1.5 -1 -0.5 0
k1 = k 3 = 2ξa (26)
Fig. 8. Feedforward control of robot with and without slip compensation
a − φ&d (t )
2 2
Linearization of Eq. (19) around the desired trajectory, i.e. k2 = (27)
sin(e ) = e , cos(e ) = 1 , yields: x& d (t )
To prevent k2 from growing to infinity, parameter “a” in
e&x   0 φ&d 0  e x  1 0 
&   &   u  Eq. (28) is defined as:
e y  =  −φd 0 x&d  e y  +  0 0   1  (20)
u a = φ&d2 (t ) + bx& d2 (t ) (28)
e&φ   0
   0 0  eφ   0 1   2 
Therefore, the gains are obtained as:
where:
k1 = k 3 = 2ξ φ&d2 (t ) + bx& d2 (t ) (29)
 u1   x& d cos eφ − x& 
u  =  φ& − φ&  (21) k 2 = b x& d (30)
 2  d 
This leads to the following nonlinear time-varying
Assuming the feedforward inputs x& d and φ&d to be controller if written in terms of the original control inputs:
constant (linear and circular paths) makes the system an LTI x& = x& d cos(φd − φ )
(31a)
system, the controllability matrix of which can be introduced + k1 [( xd − x) cos φ + ( y d − y) sin φ ]
as:
φ& = φ&d + k 2 sign ( x& d )[( y d − y ) cos φ − ( x d − x) sin φ ] (31b)
Γd = [ B AB A 2 B ] (22) + k 3 (φ d − φ )
Substituting Eq. (20) into (22) yields: Initial posture of the robot is assumed to be
1 0 0 0 − φ&d φ&d x& d 
2
(0.3,−0.2, π / 3) and control specifications are chosen as
 
ΓC = 0 0 − φ&d x& d 0 0  (23) ξ = 0.7 and b = 10 . Then, two experiments are conducted.
0 1 0 1 0 1  In the first one, the high level control commands are chosen

as angular speeds of sprockets by assuming that no slip
This matrix has full rank if either x& d or φ&d is nonzero,
happens, i.e. ir = 0, il = 0 . In the second test, the radius of
which means in this condition it is possible to stabilize the
system with static feedback. If the time-varying desired path is calculated in every time step, R = x& / φ& . This radius is
then substituted into Eq. (11) to obtain the slip coefficients
inputs x& d and φ&d are used then the nonsingularity of the
for that step. The angular speeds of sprockets are then
controllability gramian should be checked, [8]. calculated by replacing the obtained coefficients along with
x& d and φ&d in Eq (15).

99
0
2.5

ex
-0.2 With Slip Compensation
2 Without Slip Compensation
-0.4
0 10 20 30 40
1.5
time(s)
Y(m)

0.2 With Slip Compensation


1 Without Slip Compensation
0.1

ey
0
0.5
Desired Path -0.1
Observed Path With Slip Compensation
0 10 20 30 40
Observed Path Without Slip Compensation time(s)
0
0.4
-3 -2.5 -2 -1.5 -1 -0.5 0 With Slip Compensation
X(m) Without Slip Compensation

||e||
Fig. 10. Path tracking performance with and without slip compensation 0.2

Fig. 10 shows the trajectory tracking performance of the 0


0 10 20 30 40
controller in both experiments. Cartesian error and its norm time(s)
are also depicted in Fig. 11. It can be seen in Fig. 10 that this Fig. 11. Cartesian tracking: error and its norm with and without slip
control schema has a satisfactory performance in both compensation
0.1
experiments, but identifying slippage and including that in ir
the exact kinematics model improves its performance. Fig. 0.05 il
11 reveals that the norm of Cartesian error in second run is Slip Coefficients 0
less than the first run, especially where the robot starts to
-0.05
turn which is exactly due to the compensation of slippage.
Estimated slip coefficients during the trajectory tracking test -0.1

are presented in Fig. 12. -0.15


0 10 20 30 40
time(s)
V. CONCLUSIONS Fig. 12. Slip coefficients of right and left tracks obtained from slip
functions during trajectory tracking
Performing various sets of experiments, the effects of the
speed changes and the path curvature on slippage of a TMR [4] Martínez, J. L., Mandow, A., Morales, J., Pedraza, S., and
were studied in this paper. Slip coefficients of tracks were García-Cerezo, A., “Approximating Kinematics for Tracked
modeled as an exponential function of path curvature using Mobile Robots”, International Journal of Robotics Research,
the results of implemented experiments. The obtained Volume 24 , Issue 10 (October 2005) ,Pages: 867 - 878
functions were evaluated by two sets of experiments. It was [5] Shiller, Z., and Serate, W., “Trajectory Planning of Tracked
Vehicles”, ASME Journal of Dynamic Systems, Measurement
concluded that for limited speeds the slippage in tracks is and Control, December 1995, Vol. 117, No. 4, pp. 619-624.
almost independent from speed of tracks. A systematic [6] Moosavian, S. Ali A., Semsarilar, Hesam, and Kalantari,
localization method was also presented based on scans of a Arash, “Design and Manufacturing of a Mobile Rescue
Laser Range Finder which was implemented with a Robot” , IEEE Int. Conference on Intelligent Robots and
minimum frequency of 10 Hz. It was shown that identifying Systems(IROS), Beijing, China, October 2006.
the slippage effect to develop an exact kinematics model for [7] Luca, A., Oriolo, G., and Vendittelli, M., “Control of wheeled
mobile robots: An experimental overview,” in: S. Nicosia, B.
a TMR considerably improves the system performance in Siciliano, A. Bicchi, P. Valigi, (Eds.) RAMSETE —
terms of reduced path tracking errors. Articulated and Mobile Robotics for Services and
Technologies, Springer-Verlag, 2001.
REFERENCES [8] Gregor Klancar, and Igor Skrjanc, “Tracking-error model-
based predictive control for mobile robots in real time”,
[1] Wong, J.Y., Theory of Ground Vehicles, Third Edition, John
Journal of Robotics and Autonomous Systems, 55 , 460–469
Wiley & Sons Inc., 2001.
(2007).
[2] Tuan Le, A., Rye, D.C., and Durrant-Whyte, H.F., “Estimation
[9] Yun, X., and Yamamoto, Yoshio, “Stability Analysis of the
of track-soil interactions for autonomous tracked vehicles”,
Internal Dynamics of a Wheeled Mobile Robot”, Journal of
Proceedings of IEEE International Conference on Robotics
Robotic Systems, 14(10), 697–709 (1997).
and Automation, 20-25 April 1997, Page(s):1388 – 1393,
[10] Papadopoulos, E. and Poulakakis, J., “Planning and Model-
[3] Song, Z.; Hutangkabodee, S.; Zweiri, Y.H.; Seneviratne, L.D.;
Based Control for Mobile Manipulators,” Proc. Of IEEE/RSJ.
Althoefer, K., “Identification of soil parameters for unmanned
Int. Conf. on Intelligent Robots and Systems, Takamatsu,
ground vehicles track-terrain interaction dynamics”, SICE
Japan, Oct., 30- Nov.5, 2000.
2004 Annual Conference, Volume 3, 4-6 Aug. 2004
[11] Craig, John J., Introduction to Robotics, Mechanics and
Page(s):2255 – 2260.
Control, Second edition, Addison Wesley,1989.

100

You might also like