Majkut 2017

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Journal of Mechanical Science and Technology 31 (2) (2017) 893~901

www.springerlink.com/content/1738-494x(Print)/1976-3824(Online)
DOI 10.1007/s12206-017-0142-0

Crawler robot kinematics modeling by using a two-wheeled approach†


Konrad Majkut, Mariusz Giergiel and Piotr Kohut*
Department of Robotics and Mechatronics, AGH University of Science and Technology, 30-059 Cracow, Poland

(Manuscript Received October 20, 2015; Revised July 6, 2016; Accepted September 21, 2016)

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Abstract

The aim of this paper is to analyze the kinematics of a small crawler robot. A mathematical model of kinematics based on a two-
wheeled approach is proposed. This model is experimentally verified using vision-based motion measurements of a crawler vehicle
equipped with encoders and a remote control system. It is assumed that the vehicle moves along a few curves with different angular
speeds of its wheels. Based on the model of motion and the values of these speeds, numerical simulations are investigated. The results
obtained from numerical and experimental validation are presented and discussed. The comparison delivered some important conclusions.
Keywords: Crawler robot kinematics; Model verification; Vision-based measurements; Two-wheeled model
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

1. Introduction
The basis of the theory presented in this paper is the as-
sumption that the kinematics of a crawler robot (vehicle) can
be properly described using a two-wheeled parametrical
model. The parameters of this model can be found during
experimental motion of the vehicle. Studying the area of prob-
lems associated with kinematics, crawler-and wheeled-mobile
robots were examined in detail. The aim was to use a
wheeled-mobile robot model to describe crawler robot kine-
matics.
Therefore, different mathematical models of wheeled robots
in Refs. [1-4] were analyzed. Models of crawler robot kine-
Fig. 1. An example of a tracked vehicle model [8].
matics in Refs. [3, 5-26] were analyzed. The literature review
provided key information about models of kinematic crawler
robots previously applied. The example described in Ref. [6] iL/ iR are skid coefficients of the left/right track and r denotes
is presented in Fig. 1. Based on the picture in the article, a the wheels’ radius.
kinematic model was built and described with Eq. (1). One
rwL -VL
can see, that kinematic parameters depend, among others, on iL =
two skid coefficients – iL and iR. max(rwL ,VL )
(2)
rwR -VR
r iR = .
x& = [wL (1-iL )+wR (1-iR )] max(rwR ,VR )
2
y& = - x&tga (1) Skid coefficients are dimensionless. We can find this kind
r of model in the works mentioned earlier.
y& = [wR (1-iL )+wL (1-iR )]
2b In this work, we try to describe a model of kinematics of a
tracked vehicle looking from another perspective. There is a
where wL/wR denotes angular velocity of the left/right wheel, lack of kinematic models for tracked vehicles using substitute
two-wheeled models.
*
Corresponding author. Tel.: +48 12 6173396, Fax.: +48 12 6343505 The main difference between the existing kinematic models
E-mail address: pko@agh.edu.pl

Recommended by Associate Editor Kyoungchul Kong
and the model presented in this paper is the fact that its
© KSME & Springer 2017 mathematical formula does not take into account skid coeffi-
894 K. Majkut et al. / Journal of Mechanical Science and Technology 31 (2) (2017) 893~901

Fig. 3. Two-wheeled model.


Fig. 2. Dual-tracked vehicle with ICRs; example from Ref. [18].

cients. Instead of skid coefficients, the presented model in-


cludes a span between the so-called “virtual wheels”. These
wheels are placed in the tracks‘ ICRs (Instantaneous centers of
rotation).
Currently, the existing papers, for example Refs. [5, 13, 14,
18], describe kinematic models with ICRs and the main prob-
lem there is to find their location. The model presented in this
paper is original because the “virtual wheels” location is found. Fig. 4. Identification of the two-wheeled model.
It has been assumed that the wheels of our model should be
placed at such points where there is no slip between the wheel so-called V-axis (virtual axis) of virtual wheels and the second
and ground. It turns out that, during the vehicle’s motion in a goal is to find their radius and the span between them. This is
curve, these points are the tracks’ ICRs (Fig. 2). shown in Fig. 4.
The second essential difference between the presented Some reasons for using the two-wheeled model are men-
model and other solutions is the existence of a so-called “V- tioned below. These reasons are also its advantages:
axis” in our model. As explained in Sec. 2.1, it can be found - Simplicity of the mathematical model - the mathematical
experimentally. Its location changes when there are changes in formalism of the movement of the tracked vehicle is as simple
load balance. These two important differences make our as possible, which affects the computing speed;
model a new approach to the problem of a tracked robot’s - The two-wheeled model concisely describes the model of
kinematics. Such a description has a few benefits, such as the a tracked vehicle. It uses the pivot of wheels with regard to the
simplicity of the mathematical model, its concise form, versa- body of the real vehicle. Moreover, slides of tracks relative to
tility, the possibility of automatic calibration and intuitiveness. the base are interpreted as the wheels’ span of the model;
- Versatility - on the basis of the two-wheeled model, differ-
ent types of control systems of the tracked vehicle can be ap-
2. Theory
plied: with or without feedback, based on fuzzy logic, neural
At the beginning, certain assumptions have to be made. networks etc.;
First of all, the sizes of the robot are 0.3 m-1 m in length and - Possibility of automatic calibration - (identification of the
the mass is a maximum of 40 kg. Furthermore, a lack of inde- vehicle model's parameters), i.e. the control system based on
pendent shock absorption of wheels was assumed - the sus- the model alone would be able to determine the position of the
pension is stiff. A move was considered on a flat surface (the pivot of virtual wheels and the space between them, depend-
area of adjacency between the track and base is a rectangle). A ing on the surface the vehicle moves on. It also provides the
crawler robot is understood as a vehicle with two tracks. It ability to create databases of the model parameters. Therefore,
was assumed that the vehicle has stiff wheels which do not the control system of a robot based on this model would be
change their shape under the influence of ground pressure. A able to equip itself with adaptive features, i.e. the selection of
two-wheeled model of kinematics is also assumed to have stiff significant parameters influencing the kinematics, depending
wheels. In order to understand the reasons for using a two- on the surface;
wheeled model for a crawler robot’s kinematics, this model - Intuitiveness- the model and its behavior are easy to un-
has to firstly be defined. In Fig. 3 a two-wheeled vehicle mov- derstand, which is important when describing the movements
ing with the speed vP relative to the ground can be seen. The of a real vehicle
wheels have the radius rW and turn with the angular speed wW.
The fact that the vehicle has only two points in common with
2.1 First goal – finding the V-axis
its base is its characteristic feature (points L’ and P’ in Fig. 3).
Moreover, if non-slide movement is assumed, the momentary As mentioned earlier, when studying the kinematics of ob-
speed of these two points is equal to zero. To properly define jects we can only analyze their locations, speeds and accelera-
the model, two goals are required. The first goal is to find the tions without considering their physical properties and forces.
K. Majkut et al. / Journal of Mechanical Science and Technology 31 (2) (2017) 893~901 895

Fig. 6. Model relative to the vehicle’s body in the x0y0z0 system.

(a) (b)
performed. Fig. 6 shows the two-wheeled model put on the
Fig. 5. Different location of the V-axis depending on load position. model of the tracked robot in the x0y0z0 coordinate system
associated with the basement.
This means that the mathematical model of the object's As can be seen, the local coordinate system x1y1z1 is also
movement contains neither its physical properties nor the implemented, associated with the model of the robot in which
powers acting on it. Nevertheless, without violating this essen- the origin has been placed at the O1 point. The origins of the
tial condition of the modeling of kinematics, it is possible to next two frames of reference (x2y2z2 and x3y3z3) are placed in
present the situation in Fig. 5. The wheels are turning with the centers of the wheels (O2 and O3). Based on Fig. 6, it is
identical values of angular speed, however, their directions are possible to build the transformation matrix x1y1z1 into x0y0z0.
opposite. The result is the turnover of the vehicle around a
certain point of the rotation O. Thanks to the acquaintance of é cos b - sin b 0 xO1 ù
the trajectory of markers on the body, it is possible to deter- ê ú
sin b cos b 0 yO1 ú
mine the middle position of rotation. The V-axis goes through T10 = ê . (3)
ê 0 0 1 zO1 ú
this point. ê ú
One can see that the vehicle behaves in a totally different ëê 0 0 0 1 ûú
way in (a) and (b) cases (shown in Fig. 5). A conclusion
comes out of the above dissertations that, in order to correctly The transformation matrix x2y2z2into x1y1z1:
describe the kinematics with the two-wheeled model, it is
necessary to find the position of the V-axis relative to a real é0 1 0 xO2 ù
vehicle’s body. It is important to emphasize that research into ê ú
ê0 B
the relation between the load balancing and location of the V- 0 1 -( + e2 ) ú
T21 = ê 2 ú. (4)
axis is qualitative research. The only aim of this research is to ê1 0 0 zO2 ú
prove that a relation between them exists during the turn. ê ú
There is no mathematical relation considered between them. ëê0 0 0 1 úû

Otherwise, the condition of the modeling of the kinematics


would remain disturbed. If the relation exists, it would con- And the transformation matrix x3y3z3into x1y1z1:
firm that we need to seek the V-axis. If so, quantitative re-
search is needed. Based on a couple of 360-degree rotations, it é0 1 0 xO3 ù
is possible to specify the value of the location of the V-axis ê ú
ê0 B
relative to the vehicle’s body. 0 1 ( + e3 ) ú
T31 = ê 2 ú (5)
ê1 0 0 zO3 ú
ê ú
2.2 Second goal – finding wheel radii and span between ëê0 0 0 1 ûú
them

It is convenient for calculation purposes that the radii of the where B denotes the distance between tracks (between their
substitute model are the same as the radii of a real vehicle's symmetry axes), b is the orientation angle relative to the
propulsion wheels. In order to find the space between virtual ground (x0y0z0), e2 and e3 are distances depending on tracks’
wheels of the model, the V-axis location relative to the vehi- slip (depending also on the trajectory of the turn).
cle’s body should first be found. So, it is assumed that this is Hence, the matrix of the transformation x2y2z2 into x1y1z1:
896 K. Majkut et al. / Journal of Mechanical Science and Technology 31 (2) (2017) 893~901

T20 = T10T21 (6) Differentiating both of Eqs. (12) and (13), velocities relative
T = to the ground are yielded:
20
é B ù ρ& Ox'OBW
0 y0 z0
= vO' =
ê 0 cos b - sin b xO2 cos b + (
2
+ e2 )sin b + xO1 ú 2 2OBW

ê ú é & B ù
(7) &
ê-r2 b sin a 2 sin b + a& 2 r2 cos a 2 cos b + ( 2 + e2 ) b cos b + x&O1 ú
ê 0 sin b B
ê cos b xO2 sin b - ( + e2 )cos b + xO1 úú .
2 ê ú
ê ú ê r b& sin a cos b + a& r cos a sin b + ( B + e ) b& sin b + y& ú
ê1 0 0 zO1 + zO2 ú ê 2 2 2 2 2 2 O1 ú
ê ú 2
ë0 0 0 1 ê ú
û
ê -a& r
2 2 sin a 2 ú
ê 0 ú
And the matrix of the transformation x3y3z3into x0y0z0: ë û
(14)
T30 = T10T31 (8) ρ& Ox'OBW
0 y0 z0
= vO' =
3 3OBW

T = é & B ù
30 &
B ê-r3 b sin a 3 sin b + a& 3r3 cos a 3 cos b - ( 2 + e3 ) b cos b + x&O1 ú
é ù ê ú
ê 0 cos b - sin b xO3 cos b + ( + e3 )sin b + xO1 ú
2 ê r b& sin a cos b + a& r cos a sin b - ( B + e ) b& sin b + y& ú
ê ú ê 3 3 3 3 3 3 O1 ú .
ê B (9) 2
ê 0 sin b cos b xO3 sin b - ( + e3 )cos b + xO1 úú . ê ú
2 ê -a& 3r3 sin a 3 ú
ê ú ê ú
ê1 0 0 zO1 + zO3 ú ë 0 û
ê ú (15)
ë0 0 0 1 û

Points O’2OBW and O’3OBW lie on the virtual wheels’ circum- In particular, when there is a lack of skids, we are interested
ferences. Their locations in x2y2z2 and x3y3z3 are: in the form of Eqs. (14) and (15), when the points O’2OBW and
O’3OBW contact the ground (such a situation was also shown in
é r2 cos a 2 ù é r3 cos a 3 ù Fig. 3; point P’ contacts with the ground and its velocity is
ê ú ê ú
r sin a r sin a 3 ú equal to zero). In such a case α2 = α3 = π. Additionally, their
ρOx'OBW
2 y2 z 2
=ê 2 2 ú
, ρOx'OBW
3 y3 z 3
=ê 3 (10)
2 ê 0 ú 3 ê 0 ú speeds relative to the ground (x0y0z0) are zero:
ê ú ê ú
ëê 1 ûú êë 1 ûú
vO' =
2OBW

where r2 = r3 denote radii of the model’s wheels, a2 and a3


é & B & ù
denote angular position of O’2OBW and O’3OBW in the local ê -a 2 r2 cos b + ( 2 + e2 ) b cos b + x&O1 ú
coordinate system. ê ú
ê -a& r sin b + ( B + e ) b& sin b + y& ú (16)
Location of O’2OBW and O’3OBW in x0y0z0: ê 22 2 O1 ú = 0
2
ê ú
ρOx'OBW
0 y0 z0
= T20 ρOx'OBW
2 y2 z 2
, ρOx'OBW
0 y0 z0
= T20 ρOx'OBW
3 y3 z 3
. (11) ê 0 ú
2 2 3 3
ê 0 ú
ë û
Hence vO' =
3OBW

é & B & ù
ρOx'OBW
0 y0 z0
= ê -a 3r3 cos b - ( 2 + e3 ) b cos b + x&O1 ú
2 ê ú
ê -a& r sin b - ( B + e ) b& sin b + y& ú (17)
é B ù O1 ú = 0 .
ê r2 sin a 2 cos b + xO2 cos b + ( 2 + e2 )sin b + xO1 ú ê 33 2
3

ê ú ê ú
(12) ê 0 ú
ê r sin a sin b + x sin b - ( B + e )cos b + y ú
ê2 2 O2 2 O1 ú ê 0 ú
2 ë û
ê ú
ê r2 cos a 2 + zO1 + zO2 ú
ê ú On the basis of Fig. 6, another equation can be built:
ë 1 û
x0 y0 z0
ρO'OBW =
a& 2 r2 - a& 3 r3
3

é B ù b& = . (18)
ê r3 sin a 3 cos b + xO3 cos b - ( 2 + e3 )sin b + xO1 ú B + e2 + e3
ê ú
ê r sin a sin b + x sin b + ( B + e )cos b + y ú (13)
ê3 O1 ú .
3 O3
2
3
Selecting the first equation from Eq. (16) and the second
ê ú
ê r3 cos a 3 + zO1 + zO3 ú from Eq. (17) together with Eq. (18), a set of three equations
ê ú depending on O1 is obtained:
ë 1 û
K. Majkut et al. / Journal of Mechanical Science and Technology 31 (2) (2017) 893~901 897

ì a& 2 r2 2 2 B
ïe2 = b& - RS - lSx - lSy - 2
ï
í (25)
ïe = R 2 - l 2 - l - a& 3r3 - B .
ïî 3 b&
S Sx Sy
2

In order to properly identify the span between virtual


wheels, some experiments are needed. During a robot’s
movement along a curve, the following parameters are meas-
ured: – α'2 and α'3 – angular velocities of wheels,
– RS – point S trajectory radius.
– b'– angular velocity of robot orientation.
Fig. 7. A two-wheeled model with the chosen point S.
After calculating e2 and e3, a substitute two-wheeled model
ì B & is identified. It is then possible to proceed to the simulation of
ï x&O1 (t) = a& 2 r2 cos b - ( + e2 ) b cos b the movement of the tracked vehicle. Let’s simulate the mo-
ï 2
ï B tion of point S – the same one used to identify the values of e2
í y& O1 (t) = a& 3 r3 sin b + ( + e3 ) b& sin b (19)
and e3. Its real trajectory is already well-known. In order to
ï 2
ï& a& 2 r2 - a& 3 r3 simulate the motion of this point, a set of equations analogous
ï b (t) = B + e + e . to Eq. (19), but depending on S, should be built. On the basis
î 2 3
of Fig. 7, it is possible to build the transformation matrix
In order to use Eq. (19), e2 and e3 are needed at first: xSySzS into x0y0z0.

écos b - sin b 0 xS ù
ì x&O1 B a& r ê ú
ïe2 = - & - + 2& 2 sin b cos b 0 yS ú
ï b cos b 2 b TS0 = ê . (26)
í (20) ê 0 0 1 zS ú
&
ïe = xO1 - B - a& 3r3 ê ú
ï 3 b& cos b 2 ëê 0 0 0 1 ûú
î b&
ì a& 2 r2 - vO1 B The transformation matrix x2y2z2 into xSySzS:
ïe2 = -
ï b& 2
í (21)
v
ïe = O1 - a& r
3 3 B é0 1 0 lSx ù
- . ê ú
ï 3
î b& 2 ê0 B
0 1 -( + e2 ) + lSy ú
T2S = ê 2 ú (27)
ê1 0 0 zO2 ú
Eq. (19) can be generalized so that they can depend on any ê ú
point lying on the x1y1 surface, not only O1. This is shown in ëê0 0 0 1 ûú
Fig. 7. A new characteristic point lying anywhere on the body
of the vehicle was assumed. This is point S. According to Fig. and the transformation matrix x3y3z3 into xSySzS:
7, it is possible to write the equation as below:
é0 1 0 lSx ù
ê ú
ê0 B
vO1 vS 0 1 ( + e3 ) + lSy ú
b& = = . (22) T3S = ê 2 ú. (28)
R RS ê1 ú
0 0 zO3
ê ú
ëê0 0 0 1 ûú
Hence
Based on TS0, T2S, T3S and making the sequence of calcula-
v tions shown in Eqs. (3)-(17), a new set of equations depending
vO1 = S R . (23)
RS on S can be built.

ì B & & &


And ï x& S (t) = (a& 2 r2 - ( + e2 ) b + lSy b )cos b + lSx b sin b
ï 2
ï B
í y& S (t) = (a& 3 r3 + ( + e3 ) b& + lSy b& )sin b + lSx b& cos b (29)
R = RS2 - lSx2 - lSy . (24) ï 2
ï& a& 2 r2 - a& 3 r3
ï b (t) = B + e + e .
A set of equations depending on point S can be built: î 2 3
898 K. Majkut et al. / Journal of Mechanical Science and Technology 31 (2) (2017) 893~901

Table 1. Measurement results.

Diameters of chosen
h [m]
markers’ trajectories [m]
1) dM4 0.42479 dM5 0.42699 hCP0.19266
Load
2) dM4 0.42531 dM5 0.42916 hCP0.19338
front
3) dM4 0.42456 dM5 0.42721 hCP0.19293
1) dM2 0.44043 dM3 0.42531 hCT0.19634
Load
2) dM2 0.43695 dM3 0.42731 hCT0.19608
rear
3) dM2 0.43839 dM3 0.42661 hCT0.19618

Fig. 8. Scheme of the experimental setup.

After putting e2 and e3 into Eq. (29), this set of equations is


the ultimate form of description of a crawler robot’s motion.
Using it, the motion of any point lying on a robot’s body can
be stimulated. If the model Eq. (29) is correct, it should gener-
ate the trajectory of point S, theoretically identical (very simi-
lar in practice) to the real trajectory of it when the α'2 and α'3
angular velocities of wheels are known.

2.3 Conclusions from the theory

The basis of the theory is the assumption that the kinematics


of a crawler robot can be described using a two-wheeled
model. In order to identify this model, its V-axis has to be
found firstly. This is done by a sequence of 360-degree turns,
as shown in Fig. 5. When the location of the V-axis is known,
the span between the wheels must be specified. This span
depends on skids, so a sequence of measurements of α'2 , α'3,
RS and b' have to be provided and the values of e2 and e3 cal-
culated from Eq. (25). Knowing the set of e2(α'2 ,α'3) and
e3(α'2 ,α'3) these values can be entered into the model of kine-
matics expressed with Eq. (29). Using the model, it is possible
to simulate the trajectory of any point S on the robot’s body.

3. Experiment
In order to research a crawler robot’s kinematics, vision- Fig. 9. Examples of chosen markers’ trajectories.
based measurements have been made. In Fig. 8 the scheme of
the measuring stand is depicted. ments in this case are accurate and reliable. Furthermore, such
There are five markers (M1-M5) put on the body of the ve- a method for researching a crawler robot’s kinematics using
hicle shown in Fig. 8. Their position is registered by a camera TEMA Automotive® has not been reported in the literature
during movement. Using the TEMA Automotive® program, about this subject matter. Therefore, this is an original ap-
photo/video material can be analyzed to obtain the trajectories, proach to the problem.
velocities and accelerations of these markers. TEMA Automo-
tive® is professional motion analysis software used for crash
3.1 Finding the V-axis - experiment
and sled tests, measurements of deformations, airbag volume
etc. In order to achieve high measurement resolution, the ar- According to the theory, there should be a qualitative rela-
chitecture and configuration of the vision system were devel- tion between the load balancing and location of the V-axis.
oped and its configuration was arranged. A high resolution This can be proved by providing quantitative research – a
Canon 5D MarkII camera (21.1 megapixels ) constituted the sequence of 360-degree turns as described earlier.
main component of the vision system and was used to register This research consists of two stages: a) Movement with
the tracked vehicle’s motion. The vision system was calibrated. load put on the front of the vehicle (three measurements), b)
All measurements were conducted according to the applica- movement with load put on the rear of the vehicle (three
ble rules. The resolution of measurements is 0.023 mm. measurements). The results are shown in Table 1 and Fig. 9.
Therefore, it can be concluded that vision-based measure- In the first case, the markers M4 and M5 were observed. In
K. Majkut et al. / Journal of Mechanical Science and Technology 31 (2) (2017) 893~901 899

Table 2. Measurement results.

α'(t) [rpm] RS [m] b' [rad/s] e [m]


α'2 = 19.4 e2 = 0.0968
1) 0.1195 0.2227
α'3 = 0 e3 = - 0.004
α'2 = 19.57 e2 = 0.0815
SET1 2) 0.1216 0.2351
α'3 = 0 e3 = - 0.0013
α'2 = 19.57 e2 = 0.1002
3) 0.1168 0.2248
α'3 = 0 e3 = - 0.0076
α'2 = 31.62 e2 = 0.0344
1) 0.2522 0.2779
α'3 = 12.32 e3 = 0.0011
α'2 = 31.68 e2 = 0.0348
SET2 2) 0.2503 0.2796
α'3 = 12.32 e3 = -0.00004
Fig. 10. Location of the V-axis relative to the chosen markers. α'2 = 31.15 e2 = 0.0360
3) 0.2549 0.2705
α'3 = 11.97 e3 = 0.0042
the second case – the markers M2 and M3 were seen. The α'2 = 30.88 e2 = 0.0385
1) 0.1701 0.3518
values of the diameters of trajectories are obtained from the α'3 = 5.046 e3 = 0.0099
least square method. The exact location of the V-axis relative SET3 2)
α'2 = 31.14
0.1677 0.3581
e2 = 0.0386
to chosen markers as shown in Fig. 10, can be found knowing α'3 = 5.11 e3 = 0.0074
hCP and hCT. hCP is the distance between the V-axis and M4M5 α'2 = 30.68 e2 = 0.0387
3) 0.1709 0.3483
(when the load is put on the front side of the vehicle), hCT is α'3 = 4.99 e3 = 0.0107
the distance between the V-axis and M2M3 (when the load is α'2 = 38.39 e2 = 0.0435
1) 0.1496 0.4677
α'3 = 5.01 e3 = - 0.0018
put on the rear of the vehicle). The values of h depend on the
markers’ trajectory diameters. α'2 = 38.59 e2 = 0.0364
SET4 2) 0.1511 0.4798
α'3 = 4.95 e3 = 0.0012
The experiment proved clearly that there is a qualitative re-
lation between the load balancing and location of the V-axis. α2 = 38.79 e2 = 0.0429
3) 0.1511 0.4707
α'3 = 4.98 e3 = 0.0004
The difference between the locations is significant, and this
α2 = 45.69 e2 = 0.0491
proves that finding the V-axis is an essential part of the two- 1) 0.1502 0.5437
α'3 = 5.02 e3 = 0.0037
wheeled model identification.
α'2 = 46.48 e2 = 0.0404
SET5 2) 0.1467 0.5805
α'3 = 6.26 e3 = - 0.0055
3.2 Finding the span between wheels - experiment
α'2 = 45.67 e2 = 0.0407
3) 0.1444 0.5756
As stated earlier, in order to properly identify the span be- α'3 = 5.38 e3 = - 0.0035
tween virtual wheels, a sequence of experiments must be per-
formed. Measurements were conducted for five different pairs
of α'2 and α'3 (SET1-SET5) as shown in Table 2. In addition,
the location of the V-axis should be known. It is assumed that
the additional load is put in the front of the vehicle (location of
the V-axis as in case a)). To obtain e2(α'2 ,α'3) and e3(α'2 ,α'3)
from Eq. (25), the following parameters have to be measured
during the robot’s movement along curves:
– α'2 and α'3 – angular velocities of wheels,
– RS – point S trajectory radius Fig. 11. Example of measurement results for SET4 1) - α'2 and α'3.
– b'– angular velocity of robot orientation.

The measurement results are shown in Figs. 11-13 and Ta-


ble 2. RS is the point S trajectory radius obtained from the least
square method.

4. Model verification
It turns out that for all pairs of α'2 and α'3, e2 and e3 have dif-
ferent values. This can be explained by the non-linear nature
of slides between the tracks and ground (Fig. 14).
In order to fully identify the substitute model, the empirical
characteristics e2 = f(α'2, α'3) and e3=f(α'2, α'3) are needed.
However, seeking these characteristics is not the purpose of Fig. 12. Example of measurement results for SET4 1) - RS.
900 K. Majkut et al. / Journal of Mechanical Science and Technology 31 (2) (2017) 893~901

Table 3. Comparison results. The maximum Euclidean norm means the largest distance
(deviation) of the simulated trajectory from the real trajectory.
Max. euclidean RS_MEAN Max. Euclidean
norm [m] [m] norm/RS_MEAN [%] RS_MEAN is the mean value of RS from Table 2 for the proper
SET1 0.0007 0.1195 0.58
SET. It can be clearly seen that the values of the last column
do not exceed 2.5 %. On that basis, it can concluded that the
SET2 0.0026 0.2522 1.03
simulations are precise enough because the mentioned devia-
SET3 0.0022 0.1701 1.3
tions of the trajectory are small enough relative to the trajec-
SET4 0.0028 0.1496 1.87 tory’s radius.
SET5 0.0037 0.1502 2.46
5. Final conclusions
In this paper the new approach to the kinematics of a
crawler robot modeling by using a two-wheeled substitute
model have been described. The two essential differences
between the presented model and other solutions have been
distinguished, namely the tracks' ICRs and the existence of the
axis of virtual wheels.
In order to analyze a crawler robot’s kinematics, analysis of
its trajectory was performed using vision-based methods.
Considering the software-hardware part of the vision system,
it is worth noting that employed TEMA Automotive® pro-
gram with the high resolution DSLR camera for investigating
robots’ kinematics is an original approach to these types of
Fig. 13. Example of b' measurement results for SET4 1). issues. Application of the developed vision system allowed
easy and highly accurate measurements of kinematic charac-
teristics to be conducted. Measurement resolution amounted to
0.02 mm. A movement was examined with the additional load
installed at the front and back of the vehicle. It was proven
that there is a correlation between the load location and the V-
axis location. It was also proven that there is an existence be-
tween the slide of tracks and the span between the wheels of
the substitute model. This was confirmed by measurements
during the movement along the curve of the circle.
Finally, the model was verified by simulating the motion of
the vehicle along the curve with different sets of the speed of
Fig. 14. Identification of two-wheeled model parameters.
drive wheel pairs and comparing the trajectories of this motion
with the real trajectory.

References
[1] T. Buratowski, Mobile robots - selected issues, AGH Press,
Krakow (2013).
[2] L. Zhijun and S. G. Shuzhi, Fundamentals in Modeling and
Fig. 15. Comparison of real and simulated trajectory SET4 – example. Control of Mobile Manipulators, CRC Press (2013).
[3] P. F. Muir, Modeling and control of mobile robots, Ph.D.
this work. The purpose is to prove the thesis that with a substi- Thesis, Carnegie Mellon University, Pittsburgh (1988).
tute two-wheeled model it is possible to correctly describe the [4] J. Madsen, T. Heyn and D. Negrut, Methods of tracked vehi-
motion of the tracked vehicle. Without losing the generality of cle system modeling and simulation, Based Engineering Re-
the thesis, it is possible to show that, for a specific pair of α'2 ports, TR-2010-01 (2010).
and α'3 (SET of α'2 and α'3), the mathematical substitute model [5] F. Zhejun, Modeling and control of autonomous tracked
Eq. (29) correctly describes the motion of the real vehicle. vehicles, University of Michigan (1995).
Trajectory simulations should be performed based on this [6] M. D. Tehmoor and R. G. Longoria, Slip Estimation for
model for specific α'2 and α'3 values and then the simulated Small-Scale Robotic Tracked Vehicles, American Control
and real trajectories compared. The initial conditions have Conference, Marriott Waterfront, Baltimore, MD, USA
been taken from the measurements. The results of comparison (2010) 6816-6821.
are shown in Fig. 15 and Table 3. [7] D. Endo, Y. Okada, K. Nagatani and K. Yoshida, Path fol-
K. Majkut et al. / Journal of Mechanical Science and Technology 31 (2) (2017) 893~901 901

lowing control for tracked vehicles based on slip- [20] Z. Shiller, W. Serate and M. Hua, Trajectory planning of
compensating odometry, Proc. of the IEEE/RSJ Interna- tracked vehicles, Proc. of the IEEE Int. Conference on Ro-
tional Conference on Intelligent Robots and Systems, San botics and Automation, Atlanta, GA, 3 (1993) 796-801.
Diego, CA, USA (2007) 2871-2876. [21] S. Shoval and A. Shapiro, Dual-tracked mobile robot for
[8] Z. Fan, Y. Koren and D. Wehe, Tracked mobile control: motion in challenging terrains, Journal of Field Robotics, 28
hybrid approach, Control Eng. Practice, 3 (3) (1995) 329- (5) (2011) 769-791.
336. [22] T.-K. Yeu et al., Study on underwater navigation of crawler
[9] M. Gianni, F. Ferri, M. Menna and F. Pirri, Adaptive robust type mining robot, Proc. of the IEEE Int. Conference on Ro-
Three-dimensional trajectory tracking for actively articulated botics and Automation, OCEANS'11 MTS/IEEE KONA,
tracked vehicles, Journal of Field Robotics (2015) DOI: ISSN 0197-7385 (2011) 1-6.
10.1002/rob.21584. [23] G. G. Wang, S. H. Wang and C. W. Chen, Design of turn-
[10] R. Gonzalez, F. Rodriguez, J. L. Guzman and M. Beren- ing control for a tracked vehicle, IEEE Control Systems
guel, Localization and control of tracked mobile robots un- Magazine, 10 (3) (1990) 122-125.
der slip conditions, Proc. of the IEEE International Confer- [24] J. Y. Wong, Theory of Ground Vehicles, 3rd Edition, Wiley,
ence on Mechatronics, Malaga, Spain (2009) DOI: 10.1109/ New York (2001).
ICMECH.2009.4957141. [25] L. Yugang and L. Guangjun, Modeling of tracked mobile
[11] M. Kitano and M. Kuma, An analysis of horizontal plane manipulators with consideration of track-terrain and vehicle-
motion of tracked vehicles, J. Terramechanics, 14 (1977) manipulator interactions, Robotics and Autonomous Systems,
211-225. 57 (11) (2009) 1065-1074.
[12] T. Le Anh, C. R. David and H. F. Durrant-Whyte, Estima- [26] Y. Zhang and T. Huang, Research on a tracked omnidirec-
tion of tack-soil interactions for autonomous tracked vehicles, tional and cross-country vehicle, Mechanism and Machine
Proc. of the IEEE Int. Conference on Robotics and Automa- Theory, 87 (2015) 18-44.
tion, Albuquerque, New Mexico, 2 (1997) 1388-1393.
[13] J. Kim, F. C. Park and Y. Park, Design, analysis and con-
trol of a wheeled mobile robot with a nonholonomic spheri- Konrad Majkut received Ph.D. degree
cal CVT, The Int. Journal of Robotics Research, 21 (5) in field of Automatics and Robotics
(2002) 409-426. from Faculty of Mechanical Engineering
[14] J. L. Martínez, A. Mandow, J. Morales, S. Pedraza and A. and Robotics AGH University of Sci-
García-Cerezo, Approximating kinematics for tracked ence and Technology, Poland, in 2016.
mobile robots, The Int. Journal of Robotics Research, 24 His scientific interests focus on mecha-
(10) (2005) 867-878. tronics, automatics and robotics, applied
[15] J. L. Martínez, A. Mandow, J. Morales, S. Pedraza and A. mechanics, mobile robots and robot
García-Cerezo, Kinematic modelling of tracked vehicles by control.
experimental identification, Proc. of 2004 IEEORSJ
International Conference on Intelligent Robots and Systems, Mariusz Giergiel received Ph.D. de-
Sendal, Japan, 2 (2004) 1487-1492. gree from Faculty of Mechanical Engi-
[16] S. Moosavian, A. Ali and A. Kalantari, Experimental slip neering and Robotics, AGH University
estimation for exact kinematics modeling and control of a of Science and Technology, Poland, in
tracked mobile robot, IEEE/RSJ Int. Conference on Intelli- 1992. Since 2005 he is Professor at
gent Robots and Systems, Acropolis Convention Center Nice, AGH UST. His scientific interests focus
France (2008) 95-100. on automatics and robotics, applied
[17] H. Murakami, K. Watanabe and M. Kitano, A mechanics and mechatronics, robot
mathematical model for spatial motion of tracked vehicles control, mobile robots, underwater robots.
on soft ground, J. Temmechanics, 29 (1) (1992) 71-81.
[18] K. Nagatani, D. Endo and K. Yoshida, Improvement of the Piotr Kohut received Ph.D. degree in
odometry accuracy of a crawler vehicle with consideration Automatics and Robotics from Faculty
of slippage, Proc. of the IEEE Int. Conference on Robotics of Mechanical Engineering and Robot-
and Automation, Roma, Italy (2007) 2752-2757, DOI:10.1109/ ics, AGH University of Science and
ROBOT.2007.363881. Technology, Poland, in 2002. His scien-
[19] J. Pentzer and S. Brennan, Model-based prediction of tific interests focus on mechatronics,
skeed-steer robot kinematics using online estimation of track vision systems, methods of image proc-
instantaneous centers of rotation, Journal of Field Robotics, essing and analysis, 3D measurement
31 (3) (2014) 455-476. techniques.

You might also like