Datasets:
tensorkelechi
/
arxiv-cstext

Dataset card Data Studio Files
xet
Community
1
text
stringlengths
0
89.3k
conclude that 32 also outperforms these methods Ta
ble 1 shows the computation times of the two methods
for the RoA computations where we notice that 32 is
slightly slower than the method in 25 due to the addi
tional constraints needed for recursive feasibility
158642 0 2
u1D4651864202468u1D4652
Chen et al 25
Ours 32
0 5 10 15 20 25
Timestepu1D45842024u1D462Fig 3 Closedloop trajectories of length 25 for 25 and 32
The average costs are 2143 7 for 25 and 1951 1 for 32
Even though 32 is slightly more conservative than the
method in 25 the latter has significant limitations
ie it requires a shrinking horizon implementation and
a RPI set for dynamics 1 both of which we high
light in the following First we compare the closed
loop performance of the two methods Figure 3 shows
200 closedloop state and input trajectories starting in
x0 70for both methods The average cost of
the closedloop trajectories are 2143 7 for 25 and 1951 1
for 32 which is a 9 improvement This behavior is
expected since 32 is applied in receding horizon while
the method in 25 is applied for a shrinking horizon
This is most evident in the first 5 timesteps where the
method in 25 exhibits jerky behavior in both the state
and input trajectories
62 Vertical Takeoff and Landing VTOL vehicle
Next we showcase the proposed filterbased SLTMPC
on a VTOL vehicle model We use 36 to derive the
dynamics of a VTOL vehicle inspired by the Rock
ETH 37 moving in a twodimensional plane as de
picted in Figure 4 We obtain the discretetime dynam
ics
px
i1px
itvx
i pz
i1pz
itvz
i 53a
vx
i1vx
itk1θi vz
i1vz
ituz
i 53b
θi1θitωi ω i1 tk2θiωit
Iuθ
i53c
RockETH30 m
15 mwind 2 m s
xz
Fig 4 Visualization of the setup for the VTOL vehicle
where pxzandvxzdenote the position and velocity of
the VTOL vehicle in xandzdirection respectively θ
andωare the angle and angular velocity of the vehicle
with respect to the upright position uzuθdenote the
control inputs the parameters k1k2define the interac
tion between the translational and rotational dynamics
Iis the inertia and tis the discretization time We
choose all parameters such that they mimic the Rock
ETH 37 and constrain its position to 15 m15 m and
0 m15 m in the xandzdirection respectively as de
picted in Figure 4 The control inputs are constrained
to 5 N5 N and 05 Nm 05 Nm for uzanduθ re
spectively For all details of the experimental setup see
Appendix B
The interaction parameters k1 k2are uncertain and we
assume a 5 modelling error on each of these parame
ters which we model as the parametric uncertainty A
Additionally we assume wind gusts of 2 ms inxdirec
tion which we model as the additive disturbance w We
assume no uncertainty in the input matrix Bnor any un
certainty affecting the zdirection For the given uncer
tainties it is not possible to compute a polytopic RPI set
for dynamics 1 therefore rendering both 23 and the
method in 25 inapplicable However we can compute
an RPI set for auxiliary dynamics 8 with only the ad
ditive disturbance set WwR6 w0075
In the following we apply 32 and the asynchronous
computation scheme 46 42 with memory size M 4
to the VTOL vehicle for a horizon of N 10 We show
the closedloop trajectories in the xzplane for 32 in
Figure 5 and for the primary process 46 in Figure 6
We notice that both the SLTMPC 32 and the pri
mary process with asynchronously computed tubes 46
are able to steer the vehicle to the origin Due to the
secondary process 42 only updating the tubes every
10thtimestep the SLTMPC with asynchronous compu
tation has a higher average closedloop cost of 5163 6
compared to 4799 4 for 32 However the asynchronous
computation is significantly faster than 32 as shown
in Table 2 Figure 6 also shows the convex combina
tion parameters λmfor the closedloop simulations The
memory slots M0and M1are both initialized with the
1615105 0 5 10 15
u1D465051015u1D467SLTMPC 32Fig 5 Closedloop trajectories in the xzplane of the VTOL
vehicle with SLTMPC 32 starting in x z 10 125
tubes computed by 42 for initial condition x z