Professional Documents
Culture Documents
Authorized licensed use limited to: Singapore University of Technology & Design. Downloaded on December 16,2021 at 05:13:06 UTC from IEEE Xplore. Restrictions apply.
constant slope angle ( ). In this case 45 , hence the 0
0 r/ 2 a 0 r/ 2 a
wheel moves freely at an angle 45 with the driven motion. 0
J1 R r/ 2 b , J 2 R r/ 2 b ,
A. FMWMR kinematics 0 0 1 0 0 1
Fig. 1 shows the schematic of a FMWMR. Let
0 r/ 2 a 0 r/ 2 a
O q , O r , and O wi ( i 1, 2 ,3, 4 ) are world frame, mobile robot
moving frame and wheel coordinate frame respectively. If J 3 R r/ 2 b , J 4 R r/ 2 b . Using
Pwi x wi y wi wi is the position vector of wheel in O wi , then 0 0 1 0 0 1
y w1
(3) and Jacobian matrices for each wheel, inverse kinematics
yr
x w1
model of the mobile robot is obtained as [8]
yw2 O w1 1 45
1
2b x r 1 1 1 1
R 2
Ow2 xw2 z xr
y r 1 1 1 1 (4)
4
45
r 1 / a b 1/ a b 1/ a b 1 / a b 3
4
2 Or yw4 xw4
yq
Ow4
y w3
where i ( i 1, 2 ,3, 4 ) is the angular velocity of each wheel.
4 45
3
45
2a given as Pq x q y q
R ( ) Pr , where
Oq
xq cos( ) sin( ) 0
Figure 1. Schematic of FMWMR
R ( ) sin( ) cos( ) 0 is the rotation matrix of
velocity vector is given as
0 0 1
x wi 0 ri sin( i ) 0 ix
(1) moving frame O r with respect to frame O q .
Pwi y wi R i ri cos( i ) 0 ir
wi 0 0 1 iz
B. FMWMR dynamics
where ix , ir , and iz are wheel angular
velocity around the For the dynamic modeling of the FMWMR it is assumed
hub, angular velocity of roller and wheel angular velocity that coordinate frame O r lies on the center of gravity of the
about the contact point, R i is the wheel radius and ri is the robot(Fig. 2) . F y1
sin( ) cos( )
5607
Authorized licensed use limited to: Singapore University of Technology & Design. Downloaded on December 16,2021 at 05:13:06 UTC from IEEE Xplore. Restrictions apply.
such that S q q R r ( ) S r and Fq R r ( ) Fr . Thus (5) , can be b1 x sin( 2 ), b 2 y sin( 2 ) , b3 2 cos ( )( x y ) ,
q 2
5608
Authorized licensed use limited to: Singapore University of Technology & Design. Downloaded on December 16,2021 at 05:13:06 UTC from IEEE Xplore. Restrictions apply.
where G is the switching gain. Hence, total control input 1 3
1 ~2
V (t ) (t )
T
a i (24)
u t ( t ) u r ( t ) u sw ( t ) 2 i 1 2i
g ( x ) ( c ( x d x ) xd f ( x ) ( ( t , u ( t ))
1 (20) First time derivative of Lyapunov function is obtained as
3
g ( x ) ( G . sign ( ( t )))
1
1 ~
V ( t ) ( t )
T
a i aˆ i (25)
Due to use of sign function, as the value of G increases, i 1 i
chattering increases considerably, therefore boundary layer Since, ( t ) aˆ p sat ( ( t )) ( t , u ( t ))
technique [10] is implemented where sign function is ~
and a i ( i 1, 2 ,3 ) aˆ i a i , (25) yields
replaced by saturation function to eliminate the chattering
and (20) becomes V ( t ) ( a p sat ( ( t )) ( t , u ( t )))
T
u t ( t ) u r ( t ) u sw ( t )
(26)
~ ( t ) sat ( ( t ) ) aˆ i
3
g ( x ) ( c ( x d x ) xd f ( x ) ( ( t , u ( t )) (21) a
1
i
i i
i
g ( x ) ( G . sat ( ( t )))
1 i 1
sign ( i ), i 0 3
~
aˆ i
The second term a i ( t ) i sat ( ( t ) i ) in (26)
where sat ( i ) i , and is a small i
, i i 1
tends to zero for t 0 . Moreover, ( t ) sat ( ( t )) ( t ) and
positive constant. However, it should be noted that use of
saturation function may increase the control input. ( t ) ( t , u ( t )) ( t ) ( t , u ( t )) , (26) is finally obtained as
B. Adaptive control law 3
V i ( t ) sat ( i ( t )) a i ( t , u ( t )) (27)
The control law in (21) removes the chattering of the i 1
u t ( t ) u r ( t ) u sw ( t )
uncertainties used in the simulation is taken as
g ( x ) ( c ( x d x ) xd f ( x ) ( ( t , u ( t ))
1
(23) 0 . 5 20 s t 26 s
g ( x ) ( aˆ p .sat ( ( t )))
1
( t , u ( t )) 0 . 9 sin( t ) t 50 s . Range of c is selected
0
The stability of the proposed adaptive robust controller is
guaranteed only if it satisfies by Lyapunov Theory [10]. For such that error e ( t ) 0 . Higher value of c minimizes the
this aspect, following procedure is adopted:
error but increases the control input. Hence, optimal value of
Theorem: The trajectory error for FMWMR converges
c is calculated by pattern search optimization method and
asymptotically to zero if control input is designed as in (23),
provided the control design parameters c , and (small is obtained as 22.7. In order to keep the sliding surface
positive constants) are chosen appropriately. proper selection of is necessary. Increasing the value of
brings the sliding function to zero. Accordingly, the value
Proof: Let a~ i ( i 1, 2 ,3 ) aˆ i a i is the estimated error
calculated as per optimization technique is obtained as 4.7.
where a p i is the nominal value of â p i . Lyapunov function is taken as 0.1.
is selected as Fig. 3 shows simulation results for the desired trajectory.
As can be seen from Fig. 3(a), the proposed controller is
5609
Authorized licensed use limited to: Singapore University of Technology & Design. Downloaded on December 16,2021 at 05:13:06 UTC from IEEE Xplore. Restrictions apply.
capable of tracking the desired trajectory in a more efficient control energy calculated using 2-norm method. The rise in
way as compared to PID controller. Fig. 3(b) shows control input voltage between 20 s to 26 s is because of
chattering free control input voltage at each wheel with uncertainties. The time history of estimated adaptive gain
U C . E 664 . 49 651 . 54 664 . 49 651 . 54 Nm which is the
T
PID 3 . 9 1 . 4 2 .7
T
88 71 64
T
100 104 .1 97 . 08
T
Kinect position
Figure 6: Overlay of several snapshots while performing trajectory I Figure 7: Image processing through Kinect for trajectory I
Drift due to
high torque
Figure 8: Overlay of several snapshots while performing trajectory II Figure 9: Image processing through Kinect for trajectory II
IMU sensor
xq ,yq Kinect
5610
Authorized licensed use limited to: Singapore University of Technology & Design. Downloaded on December 16,2021 at 05:13:06 UTC from IEEE Xplore. Restrictions apply.
in Fig. 10. For the simulations uncertainty is selected as integral absolute error (IAE), and integral time weighted
0 . 5 50 s t 53 s error (ITAE) for trajectory I has been presented in Table I
which proves the superiority of the adaptive robust
( t , u ( t )) 0 . 1 sin( t ) t 80 s
controller over PID controller. Results obtained shows that
0 while tracking the trajectory there is an error due to drift
Design parameters for proposed controller is based on which is associated with high torque requirement during the
criterion as discussed in section A. The values calculated turns. Another reasons for error in real time implementation
from pattern search optimization method are 30.6, 7.8, 0.5 can be related to delay in position and orientation data
for c , and respectively. The simulation results for the transfer and slippage of rollers mounted on the periphery of
the wheels.
desired trajectory is presented in Fig. 4. It is evident from
Fig. 4(a) that compared to PID controller the proposed
VI. CONCLUSION
controller is efficient enough to track the trajectory even
when the angular displacement of FMWMR is non zero. In this paper, we have proposed a robust adaptive
Moreover, the control input voltage at each wheel is shown controller for a FMWMR in presence of external disturbance
in Fig. 4(b) and has calculated control energy, and uncertainties which allows the robot to track any
complicated trajectory. First, equation of motion is derived.
U C . E 1435 1435 1413 1413 Nm. The high value of
T
5611
Authorized licensed use limited to: Singapore University of Technology & Design. Downloaded on December 16,2021 at 05:13:06 UTC from IEEE Xplore. Restrictions apply.