FLATNESS-BASED CONTROL OF A

PARALLEL ROBOT ACTUATED BY

PNEUMATIC MUSCLES

Harald Aschemann ∗ and Eberhard P. Hofer ∗

Dept. of Measurement, Control and Microtechnology,

∗

University of Ulm, 89081 Ulm, Germany

e-mail:{harald.aschemann,ep.hofer}@e-technik.uni-ulm.de

Abstract: This paper presents a flatness-based control for a two-degree-of-freedom

parallel robot driven by two pairs of pneumatic muscle actuators. The robot
consists of a light-weight closed-chain structure with four moving links connected
by revolute joints. The two base joints are active and driven by pairs of pneumatic
muscles by means of a toothed belt and pulley. Exploiting differential flatness with
end-effector position and mean muscle pressures as flat outputs, a cascaded trajec-
tory control is designed. Simulation results demonstrate an excellent control per-
formance and point out the potential of this novel robot.Copyright c 2005 IFAC

Keywords: Model-based control, robotics, nonlinearity, mechanisms, pneumatic

systems

1. INTRODUCTION of Ulm involves the use of pneumatic muscles as

actuators for parallel robots, which are known
Pneumatic muscles are tension actuators, which for providing high stiffness, and especially for the
consist of a fibre-reinforced vulcanised rubber tub- capability of performing fast and highly accurate
ing with connection flanges at both ends. The motions of the end-effector.
working principle is based on the specially de-
signed fibre structure that leads to a muscle con- The parallel robot, which serves as a platform
traction in longitudinal direction when the pneu- for the development and the investigation of non-
matic muscle is filled with compressed air by linear control approaches, is depicted in fig. 1.
means of a proportional valve. Pneumatic muscles It is characterised by a closed-chain kinematic
offer several advantages in comparison to classi- structure formed by four moving links and the
cal cylinders: significantly less weight, no stick- robot base offering two degrees of freedom. All
slip effects, insensitivity to dirty working environ- joints are revolute joints, two of which - the cranks
ment, and a larger maximum force. The achievable - are actuated by a pair of pneumatic muscles,
closed-loop performance using such pneumatic respectively. Here, the coordinated contraction of
muscle actuators in combination with sophisti- a pair of pneumatic muscles is transformed into
cated non-linear control has already been investi- a rotation of the according crank by means of a
gated thoroughly by experiments at a one-degree- toothed belt and a pulley. The mass flow rate of
of-freedom test rig (Aschemann and Hofer, 2004). compressed air into and out of a pneumatic muscle
This experimental platform consists of a carriage is controlled by means of a proportional valve. The
with pneumatic muscles arranged at opposite incoming air is available at a maximum pressure
sides, which allow for rectilinear movements on of 7 bar, whereas the outlet air is discharged at
two guideways. Current research at the University atmospheric pressure, i.e. 1 bar. To avoid pressure
 é xE ù z
r(t ) = ê z ú
êë E úû k2 = - 1
mE k1 = 1
lP lP
m A , J A , s A , lA
q 1 (t ) q 2 (t ) k2 = 1
CA r
a a x k 1 = -1

é q1(t ) ù
q(t ) = êê ú
ú
q (t )
ëê 2 ûú
Fig. 2. Ambuigity of the inverse kinematics

Fig. 1. Two-degree-of-freedom parallel robot

the actuated axis JA , center of gravity distance
driven by pneumatic muscles
sA to the centre of gravity CA , length of the link
declines when large mass flow rates are required, a lA , pulley radius r) and the end-effector E (mass
compensator reservoir for each muscle is utilised. mE ), which can be modelled as lumped mass.
In this paper, the modelling of the mechatronic The inertia properties of the remaining two links
system is addressed first. Second, after proving with length lP , which are designed as light-weight
the flatness property for the proposed flat output construction, shall be neglected in comparison to
candidates, a cascade control based on differen- the other links.
tial flatness is envisaged for the resulting non- The inertial xz-coordinate system is chosen in
linear system model. The control design for the the middle of the straight line that connects
inner control loops involves decentralised pres- both base joints with distance 2a, as depicted in
sure controls for each pneumatic muscle with high (fig. 1). The motion of the parallel robot is com-
bandwidth, whereas the central outer control loop pletely described by introducing two generalised
design deals with decoupling control of the end- coordinates q1 (t) and q2 (t) that denote the two
effector position in the xz-plane and the two mean crank angles, which are combined in the vector
pressures of both pairs of pneumatic muscles. q = [q1 , q2 ]T . Analogously, the vector of the end-
Thus, desired trajectories for the end-effector po- effector coordinates is defined as r = [xE , zE ]T .
sition in the xz-plane and the two mean pressures
of the according pairs of pneumatic muscles can
be tracked independently with high accuracy. The For a given end-effector position r the correspond-
trajectory control of the mean pressure repre- ing crank angles follow from the inverse kinemat-
sents an effective means for increasing stiffness ics q = q(r, k1 , k2 ), which can be determined
concerning disturbance forces acting on the end- in symbolic form. The given ambuigity is taken
effector (Bindel et al., 1999). Simulation results into account by introducing the configuration pa-
of the closed-loop system show excellent tracking rameters k1 and k2 as shown in fig. 2. The re-
performance as well as steady-state accuracy. lationship between the corresponding velocities
is obtained by differentiation q̇ = J(r, k1 , k2 )ṙ,
where J(r, k1 , k2 ) denotes the Jacobian. Analo-
gously, the acceleration relationship is given by
2. NON-LINEAR SYSTEM MODELLING
q̈ = J(r, k1 , k2 ) r̈ + J̇(r, k1 , k2 ) ṙ.
As for modelling, the mechatronic system under The direct kinematics yields the vector of end-
consideration is divided in a mechanical system effector coordinates for given crank angles, i.e.
part and a pneumatic system part, which are r = r(q, k3 ). Similar to the inverse kinematics,
coupled by the torques resulting from the tension the configuration parameter k3 is introduced to
forces of the pair of pneumatic muscles, respec- cope with two possible configurations. The re-
tively. Here, as opposed to the model of (Carbonell lationships between velocities and between ac-
et al., 2001), the dynamics of the internal muscle celerations are derived by taking advantage of
pressures is also taken into account. the inverse dynamics. This leads directly to ṙ =
J−1 (q, k1 , k2 , k3 )q̇ and r̈ = J−1 (q, k1 , k2 , k3 )[q̈ −
J̇(q, k1 , k2 , k3 )ṙ]. At this, singularities in the Ja-
2.1 Modelling of the Parallel Robot cobian can be avoided by model-based trajectory
planning.
The chosen multi-body model of the parallel robot The according equations of motion for the ac-
part consists of three rigid bodies (fig. 1): the two tuated links follow directly from the free-body
cranks as actuated links with identical properties diagramm applying Euler’s Law
(mass mA , reduced mass moment of inertia w.r.t.
 F1E F2E b2 (x3, z3) fibres. This contraction effect can be exploited
b1 (x3, z3) to generate forces. The resulting force FM ij of a
mE g F2E
F1E g2 (x3, z3) pneumatic muscle depends non-linearly on the ac-
g1 (x3, z3) cording internal pressure pM ij as well as the con-
q1 mA g traction length ∆`M ij . The contraction lengths of
pulley
mA g (radius r) q2 the pneumatic muscles are related to the gener-
FM1l FM1r FM2l FM2r alised coordinates, i.e. the crank angles qi0 . The
position of the crank angle, where the correspond-
ing right pneumatic muscle is fully contracted, is
denoted by qi0 . Consequently, by considering the
transmission consisting of toothed belt and pulley,
the following constraints hold for the contraction
lengths of the muscles
Fig. 3. Free-body diagram of the parallel robot
JA · q̈1 = τ1 − mA · g · sA · cos q1
∆`M il (qi ) = r · (qi − qi0 ), (4a)
+F1E · lA · sin β1 + η1 , (1a)
∆`M ir (qi ) = ∆`M, max − r · (qi − qi0 ). (4b)
JA · q̈2 = τ2 − mA · g · sA · cos q2
−F2E · lA · sin β2 + η2 . (1b) Here, ∆`M, max is the maximum contraction given
by 25% of the uncontracted length. The force
Here, the drive torque τi of drive i depends on the characteristic FM ij (pM ij , ∆`M ij ) of the pneu-
corresponding muscle forces, i.e. τi = r · [FM il − matic muscle yields the resulting static tension
FM ir ]. The disturbance torque ηi accounts for force for given internal pressure pM ij as well as
friction effects as well as remaining uncertainties given contraction length ∆`M ij . This non-linear
in the muscle force characteristics (5) of drive i, force characteristic has been identified by static
respectively. The coupling forces F1E and F2E are measurements and, then, approximated by the
obtained from Newton’s Second Law applied to following polynomial description
the end-effector
FM ij = F̄M ij (∆`M ij ) · pM ij − fM ij (∆`M ij )
     3 4
mE · ẍE cos γ1 − cos γ2 F1E X X
am · ∆`m bn · ∆`nM ij .



= (2) = M ij · p Mi −
m3 · (g + z̈E ) sin γ1 sin γ2 F2E
m=0 n=0
The equations of motion in minimal form for the (5)
crank angles can be obtained in two steps. First, The dynamics of the internal muscle pressure fol-
the last equation has to be solved for the unknown lows directly from a mass flow balance in combi-
forces FiE , which then can be eliminated in (1a) nation with the pressure-density relationship. As
and (1b). Second, the substitution of the variables the maximum internal muscle pressure is limited
γi = γi (q), βi = β(q), and r̈ = r̈(q, q̇, q̈) resulting by a maximum value of pM ij,max = 7 bar, the
from direct kinematics leads to the envisaged ideal gas equation pM ij = ρM ij · R · TM ij can be
minimal form of the equations of motion utilised as accurate description of the thermody-
q̈ = q̈(q, q̇, τ1 , τ2 ). (3) namic behaviour of the compressed air. Here, the
density ρM ij , the gas constant R, and the ther-
modynamic temperature TM ij are introduced. For
the thermodynamic process a polytropic change of
2.2 Modelling of the Pneumatic Actuators
state is employed
The parallel robot is equipped with four pneu- pM ij p0
n = const. or pM ij = n · ρnM ij , (6)
matic muscle actuators. At this, the indices of all ρM ij ρ0
variables describing a particular pneumatic mus-
where n denotes the identified polytropic expo-
cle are chosen as follows: the first index i = 1, 2
nent. Thus, the relationship between the time
denotes the drive under consideration, described
derivative of the pressure and the time derivative
by the generalised coordinate qi (t), whereas the
of the density results in
second index j = l, r stands for the mounting
position, i.e. for the left or the right pneumatic ṗM ij = n · R · TM ij · ρ̇M ij . (7)
muscle. A mass flow ṁM ij into the pneumatic
muscle leads to an increase in internal pressure The mass flow balance for the pneumatic muscle
pM ij . In the case of a free movable muscle without yields
external forces, the increased internal pressure ρ̇M ij · VM ij = ṁM ij − ρM ij · V̇M ij . (8)
results in an enlarged tubing diameter associated
with a contraction ∆`M ij,f ree of the muscle in The volume characteristic of the pneumatic mus-
longitudinal direction due to specially arranged cle can be approximated with high accuracy by
the following non-linear function of both contrac- dim (uijp ) = 1 holds, the differential flatness prop-
tion length and muscle pressure, where the coeffi- erty is proven. The contraction length ∆`M ij as
cients in this polynomial approximation have been well as its time derivative ∆`˙M ij can be consid-
identified by measurements ered as scheduling parameters in a gain-scheduled
3
X 1
X adaptation of kuij and kpij .
VM ij (∆`M ij , pM ij ) = am · ∆`m
M ij · bn · pnM ij .
m=0 n=0
(9) With the internal pressure as flat output, its first
Finally, by inserting (7) and (9) into (8), the time derivative ṗM ij = vijp is introduced as new
pressure dynamics for the muscle i becomes control input. Consequently, the state variable of
the corresponding Brunovský form has to be pro-
n vided by means of measurements, i.e. zijp = pM ij .
ṗM ij = ∂VM ij
[ R · TM ij · ṁM ij Each pneumatic muscle is equipped with a pres-
VM ij + n · ∂pM ij · pM ij
 sure transducer mounted at the connection flange
∂VM ij ∂∆`M ij that connects the muscle with the toothed belt.
− · · pM ij · q̇i . (10)
∂∆`M ij ∂qi The scheduling parameter ∆`M ij results from the
measured crank angle qi , which is obtained by an
encoder providing high resolution. Furthermore,
3. FLATNESS-BASED FEEDBACK the second scheduling parameter ∆`˙M ij is derived
CONTROL DESIGN from the crank angle qi by means of real differ-
entiation using a DT 1-System with the transfer
3.1 Differential Flatness function GDT 1 (s) = s/(T1 · s + 1). The error
dynamics of each muscle pressure pM ij can be
Differential flatness is a prerequisite for flatness- asymptotically stabilised by the following control
based control of non-linear systems, which are law
usually given in state space representation, i.e.
vijp = ṗM ijd + αijp · (pM ijd − pM ij ), (13)
ẋ = f (x, u). A system is denoted as differentially
flat (Fliess et al., 1995)
if appropriate flat outputs where the constant αijp is determined by pole
y = y x, u, u̇, . . ., u(`) exist that placement. Here, the desired value ṗM ijd can
be obtained either by real differentiation of the
(i) allow for expressing all system states x and
corresponding control input uij in (20) or by
all system inputs u as a function of these flat
model-based calculation using only desired val-
outputs y as well
as their time derivatives, i.e.
x = ues, i.e. ṗM ijd = ṗM ijd (rd , ṙd , r̈d , pM id , ṗM id ). The
x y, ẏ, . . ., y(β) and u = u y, ẏ, . . ., y(β+1) ,
corresponding desired trajectories are obtained
(ii) are differentially independent, i.e. they are not from a trajectory planning module that provides
connected by differential equations. synchronous time optimal trajectories according
to given kinematic constraints. Defining epij =
If the first condition is fulfilled, the second condi-
pM ijd − pM ij as control error w.r.t. the internal
tion is equivalent to dim(u) = dim(y).
muscle pressure, the corresponding error dynam-
ics is governed by the first order differential equa-
tion
3.2 Flatness-Based Pressure Control
ėpij + αijp · epij = 0. (14)
The non-linear state equation (10) for the internal In each input channel, the non-linear valve char-
muscle pressure pM ij represents the basis for acteristic (VC) is compensated by pre-multiplying
the decentralized pressure control. It can be re- with its approximated inverse valve characteristic
formulated as (IVC). This inverse valve characteristic is imple-
mented as look-up-table and depends both on
ṗM ij = kuij (∆`M ij , pM ij ) · ṁM ij
  (11) the commanded mass flow and on the measured
− kpij ∆`M ij , ∆`˙M ij , pM ij · pM ij . internal pressure.
With the internal muscle pressure as flat output
candidate yijp = pM ij , (11) can be solved for the
3.3 Flatness-Based Decoupling Control
mass flow as control input uijp = ṁM ij and leads
to the inverse model for the pressure control
For the outer control loop, the following flat out-
1 put candidates are chosen: the end-effector po-
ṁM ij = · [ṗM ij
kuij (∆`M ij , pM ij ) (12) sition in the xz-plane, i.e. xE and zE , and the
 
+ kpij ∆`M ij , ∆`˙M ij , pM ij · pM ij ]. mean pressure for each pair of muscle, i.e. the
mean value pM i = (pM il + pM ir )/2 of the left
Since the internal pressure pM ij as state variable and the right muscle pressure. The trajectory
is identical to the flat output and dim (yijp ) = control of the mean pressure allows for increasing
stiffness concerning disturbance forces acting on é xE ù
zx = ê ú
é q1 ù
q = êq ú
ê x& ú
the carriage (Bindel et al., 1999). As the decen-
Stabilisation of
x-position
ë Eû Real differentation ëê 2 ûú
and
éx ù error dynamics direct kinematics
tralised pressure controls have been assigned a ê Ed
ê
ú
ú [pM 1l pM 1r ]T
xE
ê x& Ed ú u1 = x&&E
high bandwidth, these four subsidiary controlled p M 1ld m& M 1l

2-dof parallel robot

ê ú zE
êêë x&&Ed úúû pM 1rd m& M 1r
muscle pressures pM ij can be considered as ideal u 2 = z&&E é pM 1d ù
Inverse
dynamics pM 2ld
Decentralised
pressure m& M 2l pM 1
ê ú
control inputs of the outer control loop. Subse- éz ù
p
ëê M 2d úû pM 2rd controls
m& M 2r
ê Ed ú pM 2
quent differentiation of the first two flat output ê
ê z&Ed
ú
ú [pM 2l pM 2r ]T
ê ú
candidates until one of the control inputs appears ê z&&
ëê Ed
ú
ûú
Stabilisation of
z-position
Real differentation
and
leads to error dynamics é zE ù direct kinematics é q1 ù
zz = ê ú q = êq ú
ê z&E ú êë 2 úû
ë û

y1 = xE , ẏ1 = ẋE , Fig. 4. Control implementation

ÿ1 = ẍE (xE , zE , ẋE , żE , pM 1j , pM 2j ) , (15a)
v1 = ẍEd + α2x · (ẋEd − ẋE ) + α1x · (xEd − xE )
Z t
y2 = zE , ẏ2 = żE , +α0x · (xEd − xE )dτ, (21a)
0
ÿ2 = z̈E (xE , zE , ẋE , żE , pM 1j , pM 2j ) , (15b)
v2 = z̈Ed + α2z · (żEd − żE ) + α1z · (zEd − zE )
whereas the third and fourth flat output candi- Z t
dates directly depend on the control inputs +α0z · (zEd − zE )dτ. (21b)
0

y3 = pM 1 = 0.5 · (pM 1l + pM 1r ) , (16a) Here, the desired trajectories for the end-effector
y4 = pM 2 = 0.5 · (pM 2l + pM 2r ) , (16b) position in x-direction xEd and in z-direction zEd
as well as their first two time derivatives have to
The differential flatness can be proven as follows: be provided, whereas the desired trajectories for
all system states can be directly expressed by the the mean pressure of drive 1, i.e. pM 1d , and the
flat outputs and their time derivatives mean pressure of drive 2, i.e. pM 2d , are directly
xE = y1 , ẋE = ẏ1 , zE = y2 , żE = ẏ2 . (17) employed in a feedforward manner

Analogously, the internal muscle pressures as in- v3 = pM 1d , v4 = pM 2d . (22)

puts are given by the following function of the Due to the integral control parts, which are meant
flat outputs and a finite number of their time to counteract the disturbance torque ηi acting on
derivatives drive i, respectively, and guarantee a vanishing
  steady-state control error w.r.t. the end-effector
pM 1l (y1 , ẏ1 , ÿ1 , y2 , ẏ2 , ÿ2 , y3 , y4 ) position, the dynamics of the corresponding posi-
 pM 1r (y1 , ẏ1 , ÿ1 , y2 , ẏ2 , ÿ2 , y3 , y4 ) 
u= tion errors in x-direction ex = xEd − xE as well as
 pM 2l (y1 , ẏ1 , ÿ1 , y2 , ẏ2 , ÿ2 , y3 , y4 )  . (18)

in z-direction ez = zEd −zE are of third order. The
pM 2r (y1 , ẏ1 , ÿ1 , y2 , ẏ2 , ÿ2 , y3 , y4 ) coefficients α2x , α1x , and α0x as well as α2z , α1z ,
The non-linear state transformation that yields and α0z are specified by pole placement according
the measured state variables of the corresponding to a desired pole configuration. The implemen-
Brunovský form, i.e. the end-effector position and tation of the flatness-based control structure is
the end-effector velocity in x− and z-direction, is depicted in fig. 4.
given by the direct kinematics of the parallel robot

4. SIMULATION RESULTS
   
xE (q) zE (q)
zx = , zz = , (19)
ẋE (q, q̇) żE (q, q̇)
Tracking performance as well as steady-state ac-
Here, in contrast to the approach presented curacy w.r.t. end-effector position and mean pres-
in (Aschemann and Hofer, 2004), the end-effector sure have been investigated by simulation studies
acceleration has not to be determined neither by of a parallel robot with the following dimensions:
evaluation of the equation of motion (3) nor by a = 0.5 m, lA = 0.4 m, sA = 0.2 m, lP = 0.8 m.
a double real differentiation of the measured po- For this purpose, the tracking of a triangle-like
sition signal. By inserting the new defined inputs desired trajectory for the controlled variables as
v1 = ẍE , v2 = z̈E , v3 = pM 1 , and v4 = pM 2 shown in fig. 5 has been considered. At this, the
as well as the transformed states into (18), the desired value for the mean pressure is held con-
inverse dynamics becomes stant at 4.0 bar.
u = u (zx , zz , v1 , v2 , v3 , v4 ) . (20) The first part consists of a motion from the
The error dynamics of the end-effector positions starting point (x = 0 m, z = 1 m) to the
xE and zE can be asymptotically stabilized with point (x = −0.2 m, z = 0.6 m). The second
the control laws part comprises a movement in x-direction by
 1.5 pM1ld [bar]
x [m] 6 6
Ed
0.9 1 zEd [m] p [bar]
M1l
0.8
0.5 4 4
0.7 pM1rd [bar]
0 p [bar]
0.6 2 2 M1r
des. traj. in xz−plane [m]
0.5 −0.5 0 5 10 0 5 10
−0.3−0.2−0.1 0 0.1 0.2 0.3 0 5 10
t [s] t [s] t [s]
2 20
xpp [m/s2] pM2rd [bar]
Ed 6 6
1 zppEd [m/s2] p [bar]
10 M2r

0 4 4
0 pM2ld [bar]
−1 xp [m/s]
Ed p [bar]
zp [m/s] 2 M2l 2
Ed
−2 −10
0 5 10 0 5 10 0 5 10 0 5 10
t [s] t [s] t [s] t [s]

Fig. 5. Desired trajectories and according tracking Fig. 7. Desired values and actual values of the four
errors for carriage position and mean pressure muscle pressures
−3 −3
4
x 10
4
x 10 robot with two degrees of freedom driven by pneu-
ex [m] ez [m] matic muscles. The modelling of this mechatronic
2 2
system leads to a system of non-linear differential
0 0
equations of eigth order. For the characteristics of
−2 −2 the pneumatic muscles polynomials serve as good
−4
0 5 10
−4
0 5 10
approximations. The non-linearity of the valve is
t [s] t [s] linearised by means of a pre-multiplication with
0.1
epM1 [bar]
0.1
epM2 [bar]
its approximated inverse characteristic. The inner
0.05 0.05 control loops of the cascade involve a flatness-
0 0 based control of the internal muscle pressure with
−0.05 −0.05
high bandwidth. The outer control loop achieves
a decoupling of the end-effector position and the
−0.1 −0.1
0 5 10 0 5 10 mean pressures as controlled variables. Simulation
t [s] t [s]
results emphasise the excellent closed-loop perfor-
Fig. 6. Tracking errors w.r.t. the end-effector po- mance with maximum position errors of approx.
sition (top) and the mean pressures (bottom) 3 mm during the movements, vanishing steady-
state position error and steady-state pressure er-
0.4 m to the point (x = 0.2 m, z = 0.6 m). ror of less than 0.01 bar. Future research will
The third part involves the return motion to the address experiments at a prototype system.
starting point. These fast trajectories have been
generated considering and, consequently, avoiding
saturation effects due to the mass flow limitations. REFERENCES
The corresponding tracking errors are depicted in Aschemann, H. and E.P. Hofer (2004). Flatness-
fig. 6. As for the end-effector position, the maxi- based trajectory control of a pneumatically
mum tracking errors during the acceleration and driven carriage with uncertainties. Proceed-
deceleration intervals are approx. 3 mm, whereas ings of NOLCOS 2004, Stuttgart, Germany,
the steady-state error becomes zero due to the in- pp. 239–244.
tegral control part. Concerning the mean pressure, Bindel, R., R. Nitsche, R. Rothfuß and M. Zeitz
tracking errors of approx. 0.1 bar occur during the (1999). Flatness based control of two valve
acceleration and deceleration intervals, whereas hydraulic joint actuator of a large manip-
the steady-state error is less than 0.01 bar. ulator. CD-ROM-Proceedings of ECC 1999,
The corresponding desired values as well as ac- Karlsruhe, Germany.
tual values of the decentralised pressure controls Carbonell, P., Z. P. Jiang and D. Repperger
are depicted in figure 7. As can be seen, these (2001). Comparative study of three nonlin-
decentralised control loops achieve an excellent ear control strategies for a pneumatic mus-
tracking of the desired values provided by the cle actuator. Proceedings of NOLCOS 2001,
outer decoupling control. Saint-Petersburg, Russia, pp. 167–172.
Fliess, M., J. Levine, P. Martin and P. Rouchon
(1995). Flatness and defect of nonlinear sys-
5. CONCLUSIONS tems: Introductory theory and examples. Int.
J. Control 61, pp. 1327–1361.
In this paper, a cascaded trajectory control based
on differential flatness is presented for a parallel