54 4. TILTING VEHICLE CONTROL
min
U
r
;S
ro
W J D
N
P
kD1
U
.
k
/
r
R
ro
2
C
N
P
kD1
S
.k/
ro
Q
ro
2
s.t. W X
.
kC1
/
r
D A
rd
X
.k/
r
C B
rd
U
.k/
r
C E
rd
W
.0/
r
M
ro
X
.k/
C N
ro
U
.k/
L
ro
C S
.k/
ro
0 S
.
k
/
ro
S
ro;max
ˇ
ˇ
ˇ
U
.k/
r
ˇ
ˇ
ˇ
U
max
ˇ
ˇ
ˇ
U
.k/
r
U
.
k1
/
r
ˇ
ˇ
ˇ
U
slew;max
;
(4.20)
where S
.k/
ro
2 R
4
denotes slack variables for the roll envelope at the k th prediction step. e
disturbance is assumed to remain unchanged during the control horizon as W
.0/
. A
rd
; B
rd
; E
rd
are
the discretized matrices of A
r
; B
r
; E
r
derived from the roll model. e total cost to be minimized
is the l
2
norm of the control effort and slack variables, which are weighted by R
ro
and Q
ro
,
respectively.
e choice of the prediction horizon, as detailed in [70], should be long enough to see
the steady-state performance enhancement along with the transient degradation for an optimal
decision making. A ramp acceleration disturbance of 0:5g is applied to the system, and the
roll envelope is set as LTR
lim
D 0:5. MPC results with various prediction horizon settings are
shown in Figure 4.11. For a small prediction horizon as 0.5 s (i.e., N D 10; T D 50 ms), the
resultant tilting moment T
x
is in the wrong direction and makes LTR even worse. Results for the
optimized torque start to converge after the prediction horizon reaches 1 s. For computational
efficiency, a prediction horizon of 1 s is adopted for the control implementations.
It should also be noted that the LTR index overshoot is still noticeable as shown in Fig-
ure 4.11 with the active tilting control only. To further improve the vehicle stability, other on-
board control efforts should be incorporated, and an integrated approach is introduced in the
next section.
4.7.3 INTEGRATED STABILITY ENVELOPE
For lateral stability, a handling envelope has been proposed by researchers [81, 82]. e thresh-
olds for yaw rate .r
max
/ along with side-slip angles at rear wheels .˛
r;sat
/ are used to formulate
the envelope for lateral stability control. At each discrete control step k, this leads to
M
sh
X
.
k
/
L
sh
; (4.21)
where