Experimental validation of a new adaptive control scheme for quadrotors MAVs Gianluca Antonelli † , Elisabetta Cataldi † , Paolo Robuffo Giordano ⊕ , Stefano Chiaverini † , Antonio Franchi ≀ † University of Cassino and Southern Lazio, Italy http://webuser.unicas.it/lai/robotica ⊕ CNRS at IRISA and Inria Bretagne Atlantique, France http://www.irisa.fr/lagadic ≀ Max Planck Institute for Biol. Cybernetics, Germany http://www.kyb.mpg.de/research/dep/bu/hri/ IROS 2013 Antonelli Cataldi Robuffo Chiaverini Franchi Tokyo, 5 November 2013
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Experimental validation of a new adaptive
control scheme for quadrotors MAVs
Gianluca Antonelli†, Elisabetta Cataldi†,Paolo Robuffo Giordano⊕, Stefano Chiaverini†, Antonio Franchi≀
†University of Cassino and Southern Lazio, Italyhttp://webuser.unicas.it/lai/robotica
⊕CNRS at IRISA and Inria Bretagne Atlantique, Francehttp://www.irisa.fr/lagadic
≀Max Planck Institute for Biol. Cybernetics, Germanyhttp://www.kyb.mpg.de/research/dep/bu/hri/
IROS 2013
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Trajectory tracking control for quadrotor
Adaptive with respect to
uncertainty in total massuncertainty in Center Of Mass (CoM)presence of 6-DOF external disturbances
Assumption: closed-loop orientation dynamics faster thantranslational one
Stability analysis & numerical simulations1
Experimental results
1Antonelli, Arrichiello, Chiaverini, Robuffo Giordano, “Adaptivetrajectory tracking for quadrotor MAVs in presence of parameteruncertainties and external disturbances”, AIM 2013
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Kinematics
earth-fixed
O x
z
y
η1
body-fixed
Ob
xb
zb
yb
u, surge
w, heave
v, sway
p, roll
r, yaw
q, pitch η1 =[
x y z]T
η2 =[
φ θ ψ]T
ν1 = RBI η1
ν2 = T (η2)η2
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Dynamics
Mathematical model expressed in body-fixed frame
Mν +C(ν)ν + τW + g(RBI ) = τ ,
beyond the common terms, we model
τW =
[
RBI O3×3
O3×3 RBI
]
γW
γW ∈R6 external disturbance constant in the inertial frame (wind)
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Dynamics -2-
Exploiting the linearity in the parameters
Φ(ν,ν,RBI )γ = τ
and rewriting with respect to the inertial frame while separating thexy dynamics from z:
[
Φxy(η, η, η)φz(η, η, η)
]
γ = RIBτ 1
with γ ∈ R16:
mass (1 parameter)
first moment of inertia (3 p.)
inertia tensor (6 p.)
external disturbance (6 p.)
τ =
[
τ 1
τ 2
]
=
00Z
K
M
N
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Thrust
Assuming CoM coincident with Ob
xy
z
O
xbyb zb
Ob
f1
f2
f3
f41
2
3
4
ωt,1
ωt,2
ωt,3
ωt,4
τt,1
τt,2
τt,3
τt,4
l
fi = bω2t,i
τt,i = dω2t,i
τ 1 =
00
4∑
i=1
fi
τ 2 =
l(f2 − f4)l(f1 − f3)
−τt,1 + τt,2 − τt,3 + τt,4
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Mapping from the angular velocities to the force-torques
CoM influences the mapping from thrust generated from the
motors to the vehicle forces/moments
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Inverse mapping
Any controller determines a control action[
Zc Kc Mc Nc
]Tfurther
projected onto the motor input u ∈ R4
u = B−1v
Zc
Kc
Mc
Nc
where B−1v ∈ R
4×4 is
B−1v =
l − rC,x
4bl0
1
2bl−l − rC,x
4dll − rC,y
4bl
1
2bl0
l − rC,y
4dll + rC,x
4bl0 −
1
2bl−l + rC,x
4dll + rC,y
4bl−
1
2bl0
l + rC,y
4dl
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Current inverse mapping
When the CoM position estimate rC is affected by an error, the real
mapping becomes
Z
K
M
N
= Bv|rC B−1v
∣
∣
rC
Zc
Kc
Mc
Nc
=
1 0 0 0rC,y
21 0
brC,y
2drC,x
20 1 −
brC,x
2d0 0 0 1
Zc
Kc
Mc
Nc
wrong CoM estimate ⇒ a coupling from altitude and yaw control
actions onto roll and pitch dynamics
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Controller block diagram
η1d
ψdφd, θd
Zc
posor
Kc
Mc
Nc
B−1
v
umotors
w2
t,iBv
Z
K
M
N
τW
η
plant
Classical MAV control architecture with adaptation wrt parametersand compensation of the CoM position
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Altitude controller
error
z = zd − z ∈ R
sz = ˙z + λz z ∈ R
full version
Z =1
cosφ cos θ(φzγ + kvzsz)
˙γ = K−1γ,zφ
Tz sz
with γ ∈ R16
reduced version
Z =1
cosφ cos θ(γz + kvzsz)
˙γz = k−1γ,zsz
with γz ∈ R1
the reduced version designed to compensate only for persistent
terms ⇒ null steady state error wrt a minimal set of parameters!
(λz > 0, kvz > 0,Kγ,z > O, kγ,z > 0)
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Horizontal controller
error
ηxy =[
xd − x yd − y]T
∈ R2
sxy = ˙ηxy + λxyηxy ∈ R2
full version
virtual input solutions of:
[
cφsθ−sφ
]
=1
ZRz (Φxyγ + kv,xysxy)
˙γ = K−1γ,xyΦ
Txysxy
with γ ∈ R16
reduced version
virtual input solutions of:
[
cφsθ−sφ
]
=1
ZRz
(
γxy + kv,xysxy)
˙γxy = k−1γ,xysxy
with γxy ∈ R2
again: the reduced version compensates only for persistent terms
⇒ null steady state error wrt a minimal set of parameters!
(λxy > 0, kv,xy > 0,Kγ,xy > O, kγ,xy > 0)
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Orientation controller
The inputs are the desired roll, pitch and yawThe commanded forces map onto the real ones according to
K = Kc +rC,y
2Zc +
brC,y
2dNc
M = Mc +rC,x
2Zc −
brC,x
2dNc
N = Nc
Neither the altitude nor the yaw control loop are affected by rC , thus both Zc
and Nc convergence to a steady state valueRoll and pitch control can be designed by considering the estimation error asan external, constant, disturbance:
K = Kc +1
2
(
Zc +b
dNc
)
rC,y
M = Mc +1
2
(
Zc −b
dNc
)
rC,x
The disturbance value is unknown and its effect may be compensated by
resorting to several adaptive control laws well known in the literature
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
CoM estimation
PD control for roll and pitch =⇒ steady-state error because of thewrong CoM estimateA simple integral action can counteract this effect resulting a zerosteady-state error
[
˙rC,x
˙rC,y
]
= −krC
[
θd − θ
φd − φ
]
, krC > 0
As a byproduct, in absence of moment disturbance, the estimates(rC,x, rC,y) are driven towards the real CoM offsets (rC,x, rC,y)
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Stability analysis
Altitude controller: let γ = γ − γ and consider the Lyapunov function
V (sz, γ) =m
2s2z +
1
2γTKγ,zγ
Along the system trajectories
V (sz, γ) = sz(
mzd −mz +mλz ˙z)
− γTKγ,z˙γ
= sz (φzγ − cosφ cos θZ)− γTKγ,z˙γ = −kvzs
2z ≤ 0
State trajectories are boundedAsymptotic stability can be further proven by resorting to Barbalat’sLemma as in classical adaptive control schemesSimilar machinery for the horizontal controller case
Antonelli CataldiRobuffoChiaverini Franchi Tokyo, 5 November 2013
Experimental results
Experiments run at the Max Planck Institute of Tubingen, Germany