Robotics 2

Adaptive Trajectory Control

Prof. Alessandro De Luca

 Motivation and approach

n need of adaptation in robot motion control laws

n large uncertainty on the robot dynamic parameters
n poor knowledge of the inertial payload
n characteristics of direct adaptive control
n direct aim is to bring to zero the state trajectory error, i.e.,
position and velocity errors
n no need to estimate on line the true values of the dynamic
coefficients of the robot (as opposed to indirect adaptive control)
n main tool and methodology
n linear parametrization of robot dynamics
n nonlinear control law of the dynamic type (the controller has its
own ‘states’)

Robotics 2 2
 Summary of robot parameters
n parameters assumed to be known
n kinematic description based, e.g., on Denavit-Hartenberg parameters
({𝛼" , 𝑑" , 𝑎" , 𝑖 = 1, … , 𝑁} in case of all revolute joints), including link
lengths (kinematic calibration)
n uncertain parameters that can be identified off line
n masses 𝑚" , positions 𝑟-" of CoMs, and inertia matrices 𝐼" of each link,
appearing in combinations (dynamic coefficients) ⇒ 𝑝 ≪ 10×𝑁
n parameters that are (slowly) varying during operation
n viscous 𝐹5" , dry 𝐹6" , and stiction 𝐹7" friction at each joint ⇒ 1 ÷ 3×𝑁
n unknown and abruptly changing parameters
n mass, CoM, inertia matrix of the payload w.r.t. the tool center point

when a payload is firmly attached to the robot E-E, only the 10 parameters of the
last link are modified, influencing however most part of the robot dynamics

Robotics 2 3
 Goal of adaptive control
n given a twice-differentiable desired joint trajectory 𝑞; (𝑡)
n with known desired velocity 𝑞̇ ; (𝑡) and acceleration 𝑞̈ ; (𝑡)
n possibly obtained by kinematic inversion + joint interpolation
n execute this trajectory under large dynamic uncertainties
n with a trajectory tracking error vanishing asymptotically
𝑒 = 𝑞; − 𝑞 ⟶ 0 𝑒̇ = 𝑞̇ ; − 𝑞̇ ⟶ 0
n guaranteeing global stability, no matter how far are the initial
estimates of the unknown/uncertain parameters from their true
values and how large is the initial trajectory error

n identification is not of particular concern: in general, the estimates of

dynamic coefficients will not to converge to the true ones!
n if this convergence is a specific extra requirement, then one should
use (more complex) indirect adaptive schemes

Robotics 2 4
 Linear parameterization

𝑀 𝑞 𝑞̈ + 𝑆 𝑞, 𝑞̇ 𝑞̇ + 𝑔 𝑞 + 𝐹5 𝑞̇ = 𝑢
n there exists always a (𝑝-dimensional) vector 𝑎 of dynamic
coefficients, so that the robot model takes the linear form
𝑌 𝑞, 𝑞,̇ 𝑞̈ 𝑎 = 𝑢
n vector 𝑎 contains only unknown or uncertain coefficients
n each component of 𝑎 is in general a combination of the
robot physical parameters (not necessarily all of them)
n the model regression matrix 𝑌 depends linearly on 𝑞̈ ,
quadratically on 𝑞̇ (for the terms related to kinetic energy),
and nonlinearly (trigonometrically) on 𝑞

Robotics 2 5
 Trajectory controllers
based on model estimates
n inverse dynamics feedforward (FFW) + PD (linear) control
J 𝑞; 𝑞̈ ; + 𝑆O 𝑞; , 𝑞̇ ; 𝑞̇ ; + 𝑔N 𝑞; + 𝐹P5 𝑞̇ ; + 𝐾L 𝑒 + 𝐾7 𝑒̇
𝑢=𝑀
𝑢N ;
n (nonlinear) control based on feedback linearization (FBL)
J 𝑞 (𝑞̈ ; + 𝐾L 𝑒 + 𝐾7 𝑒)̇ + 𝑆O 𝑞, 𝑞̇ 𝑞̇ + 𝑔N 𝑞 + 𝐹P5 𝑞̇
𝑢=𝑀
J 𝑆,
𝑀, O 𝑔,
N 𝐹P5 ⟺ estimate 𝑎N
n approximate estimates of dynamic coefficients may lead to
instability with FBL due to temporary ’non-positive’ PD gains
J 𝑞 𝐾L < 0!)
(e.g., 𝑀
n not easy to turn these laws in adaptive schemes: inertia
inversion/use of acceleration (FBL); bounds on PD gains (FFW)
Robotics 2 6
 A control law easily made ‘adaptive’
n nonlinear trajectory tracking control (without cancellations)
having global asymptotic stabilization properties
J 𝑞 𝑞̈ ; + 𝑆O 𝑞, 𝑞̇ 𝑞̇ ; + 𝑔N 𝑞 + 𝐹P5 𝑞̇ ; + 𝐾L 𝑒 + 𝐾7 𝑒̇
𝑢=𝑀
n a natural adaptive version would require ...
𝑎Ṅ = designing a suitable update law
(in continuous time)
n without extra assumptions, it can be shown only that joint velocities
become eventually “clamped” to those of the desired trajectory
(zero velocity error), but a permanent residual position error is left
n idea: on-line modification with a reference velocity
𝑞̇ ; ⟶ 𝑞̇ R = 𝑞̇ ; + Λ(𝑞; − 𝑞) Λ>0
typically Λ = 𝐾7UV𝐾L (all matrices will be chosen diagonal)
Robotics 2 7
 Intuitive interpretation of 𝑞̇ R
n elementary case
n a mass ‘lagging behind its mobile reference (𝑒 > 0) on a linear rail
controlled mobile
mass reference 𝑞̇ R = 𝑞̇ ; + Λ 𝑒
𝑢 𝑞̇
𝑞̇ ;

𝑞 𝑒>0 𝑞; (𝑡)
‘enhanced’ velocity error 𝑠 = 𝑞̇ R − 𝑞̇ > 𝑞̇ ; − 𝑞̇ = 𝑒̇

𝑢 = 𝐾7 𝑠 = 𝐾7 𝑞̇ R − 𝑞̇ = 𝐾7 𝑞̇ ; + Λ 𝑒 − 𝑞̇ = 𝐾7 𝑒̇ + 𝐾7 Λ 𝑒
𝐾L
n a mass ‘leading in front’ of its mobile reference (𝑒 < 0)

in a symmetric way, a ‘reduced’ velocity error will appear (𝑠 < 𝑒̇ )

Robotics 2 8
 Adaptive control law design
n substituting 𝑞̇ R = 𝑞̇ ; + Λ𝑒, 𝑞̈ R = 𝑞̈ ; + Λ𝑒̇ in the previous nonlinear
controller for trajectory tracking
𝑢=𝑀J 𝑞 𝑞̈ R + 𝑆O 𝑞, 𝑞̇ 𝑞̇ R + 𝑔N 𝑞 + 𝐹P5 𝑞̇ R + 𝐾L 𝑒 + 𝐾7 𝑒̇
= 𝑌 𝑞, 𝑞,̇ 𝑞̇ R , 𝑞̈ R 𝑎N + 𝐾L 𝑒 + 𝐾7 𝑒̇
dynamic parameterization of PD stabilization
the control law using current estimates (diagonal matrices, >0)
(note here the 4 arguments in 𝑌(X) !)
n update law for the estimates of the dynamic coefficients (𝑎N becomes
the 𝑝-dimensional state of the dynamic controller)

𝑎Ṅ = Γ𝑌 Z (𝑞, 𝑞,̇ 𝑞̇ R , 𝑞̈ R )(𝑞̇ R − 𝑞)̇ Γ>0 (diagonal)

estimation gains
𝑠 (variation rate of estimates)
‘modified’ velocity error
Robotics 2 9
 Asymptotic stability of trajectory error
Theorem
The introduced adaptive controller makes the tracking error along
the desired trajectory globally asymptotically stable
𝑒 = 𝑞; − 𝑞 ⟶ 0, 𝑒̇ = 𝑞̇ ; − 𝑞̇ ⟶ 0
Proof
n a Lyapunov candidate for the closed-loop system (robot +
dynamic controller) is given by
1 Z 1 Z 1 Z UV
𝑉 = 𝑠 𝑀 𝑞 𝑠 + 𝑒 𝑅𝑒 + 𝑎[ Γ 𝑎[ ≥ 0
2 2 2
𝑠 = 𝑞̇ R − 𝑞̇ (= 𝑒̇ + Λ𝑒) 𝑅>0 𝑎[ = 𝑎 − 𝑎N
constant matrix error in parametric
modified velocity error (to be specified later) estimation

𝑉=0 ⟺ 𝑎N = 𝑎, 𝑞 = 𝑞; , 𝑠 = 0 (⇒ 𝑞̇ = 𝑞̇ ; )
Robotics 2 10
 Proof (cont)
n the time derivative of V is
1 Z
𝑉̇ = 𝑠 𝑀̇ 𝑞 𝑠 + 𝑠 Z 𝑀 𝑞 𝑠̇ + 𝑒 Z 𝑅𝑒̇ − 𝑎[ Z Γ UV 𝑎Ṅ
2
since 𝑎[̇ = −𝑎Ṅ (𝑎̇ = 0)
n the closed-loop dynamics is given by
𝑀 𝑞 𝑞̈ + 𝑆 𝑞, 𝑞̇ 𝑞̇ + 𝑔 𝑞 + 𝐹5 𝑞̇ =
=𝑀 J 𝑞 𝑞̈ R + 𝑆O 𝑞, 𝑞̇ 𝑞̇ R + 𝑔N 𝑞 + 𝐹P5 𝑞̇ R + 𝐾L 𝑒 + 𝐾7 𝑒̇
subtracting the two sides from 𝑀 𝑞 𝑞̈ R + 𝑆 𝑞, 𝑞̇ 𝑞̇ R + 𝑔 𝑞 + 𝐹5 𝑞̇ R
leads to
𝑀(𝑞)𝑠̇ + 𝑆 𝑞, 𝑞̇ + 𝐹5 𝑠 =
=𝑀 ` 𝑞 𝑞̈ R + 𝑆a 𝑞, 𝑞̇ 𝑞̇ R + 𝑔[ 𝑞 + 𝐹b5 𝑞̇ R − 𝐾L 𝑒 − 𝐾7 𝑒̇
` = 𝑀 − 𝑀,
with 𝑀 J 𝑆a = 𝑆 − 𝑆,
O 𝑔[ = 𝑔 − 𝑔,
N 𝐹b5 = 𝐹5 − 𝐹P5
Robotics 2 11
 Proof (cont)
n from the property of linearity in the dynamic coefficients, it follows
𝑀(𝑞)𝑠̇ + 𝑆 𝑞, 𝑞̇ + 𝐹5 𝑠 = 𝑌 𝑞, 𝑞,̇ 𝑞̇ R , 𝑞̈ R 𝑎[ − 𝐾L 𝑒 − 𝐾7 𝑒̇
n substituting in 𝑉,̇ together with 𝑎Ṅ = Γ𝑌 Z 𝑠, and using the skew-
symmetry of matrix 𝑀̇ − 2𝑆 we obtain
1
̇𝑉 = 𝑠 Z 𝑀̇ 𝑞 − 2𝑆 𝑞, 𝑞̇ 𝑠 − 𝑠 Z 𝐹5 𝑠 + 𝑠 Z 𝑌𝑎[
2
−𝑠 Z (𝐾L 𝑒 + 𝐾7 𝑒)̇ + 𝑒 Z 𝑅𝑒̇ − 𝑎[ Z 𝑌 Z 𝑠
= −𝑠 Z 𝐹5 𝑠 − 𝑠 Z (𝐾L 𝑒 + 𝐾7 𝑒)̇ + 𝑒 Z 𝑅𝑒̇
n replacing 𝑠 = 𝑒̇ + Λ 𝑒 and being 𝐹5 = 𝐹5Z (diagonal)

𝑉̇ = −𝑒 Z (ΛZ 𝐹5 Λ + ΛZ 𝐾L )𝑒
a complete −𝑒 Z 2ΛZ 𝐹5 + ΛZ 𝐾7 + 𝐾L − 𝑅 𝑒̇ − 𝑒̇ Z (𝐹5 + 𝐾7 )𝑒̇
quadratic form
in 𝑒, 𝑒̇ !
Robotics 2 12
 Proof (end)
n defining now (all matrices are diagonal!)

Λ = 𝐾7UV 𝐾L > 0 𝑅 = 2𝐾L (𝐼 + 𝐾7UV 𝐹5 ) > 0

leads to
𝑉̇ = −𝑒 Z ΛZ (𝐹5 + 𝐾7 )Λ𝑒 − 𝑒̇ Z (𝐹5 + 𝐾7 )𝑒̇
= −𝑒 Z 𝐾L 𝐾7UV (𝐹5 + 𝐾7 )𝐾7UV 𝐾L 𝑒 − 𝑒̇ Z (𝐹5 + 𝐾7 )𝑒̇ ≤ 0
and thus
𝑉̇ = 0 ⟺ 𝑒 = 𝑒̇ = 0
the thesis follows from Barbalat lemma + LaSalle theorem

the set of states of convergence has zero trajectory error and

a constant value for 𝑎N , not necessarily the true one (𝑎[ ≠ 0)

Robotics 2 13
 Remarks
n if the desired trajectory 𝑞; (𝑡) is persistently exciting, then also the
estimates of the dynamic coefficients converge to their true values
n condition of persistent excitation
n for linear systems: # of frequency components in the desired trajectory
should be at least twice as large as # of unknown coefficients
n for nonlinear systems: the condition can be checked only a posteriori
(a certain motion integral should be permanently lower bounded)
n in case of known absence of (viscous) friction (𝐹5 ≡ 0), the same
proof applies (a bit easier in the final part)
n the adaptive controller does not require the inverse of the inertia
matrix (true or estimated), nor the actual robot acceleration (only the
desired acceleration), nor further lower bounds on 𝐾L > 0, 𝐾7 > 0
n adaptation can be also used only for a subset of dynamic coefficients,
the remaining being known (𝑌𝑎 = 𝑌f;fgh 𝑎Nf;fgh + 𝑌ijklj 𝑎ijklj )
n the non-adaptive version (using accurate estimates) is a static
tracking controller based on the passivity property of robot dynamics
Robotics 2 14
 Case study: Single-link under gravity
model 𝐼 𝜃̈ + 𝑚𝑔𝑑 sin 𝜃 + 𝑓5 𝜃̇ = 𝑢 (with friction)
𝑢 𝑑
linear parameterization
𝐼
𝑌 𝜃, 𝜃,̇ 𝜃̈ 𝑎 = 𝜃̈
𝑚
sin 𝜃 𝜃̇ 𝑚𝑔𝑑 = 𝑢
𝑓5
𝜃
adaptive controller

𝑒 = 𝜃; − 𝜃 Λ > 0 𝑢 = J𝐼 𝜃̈ + 𝑚𝑔𝑑 t sin 𝜃 + 𝑓J5 𝜃̇ + 𝑘L 𝑒 + 𝑘7 𝑒̇

𝑘L J
𝐼 ̇ ̈R
̇𝜃R = 𝜃̇; + 𝑒 𝛾 V 𝜃
𝑘7 𝑎Ṅ = 𝑚𝑔𝑑
t ̇ = 𝛾u sin 𝜃 (𝜃̇R − 𝜃) ̇
𝛾" > 0, 𝑖 = 1,2,3 J ̇ 𝛾 𝜃 ̇R
𝑓5 v

Robotics 2 15
 Simulation data
n real dynamic coefficients
𝐼 = 7.5, 𝑚𝑔𝑑 = 6, 𝑓5 = 1
n initial estimates
J𝐼 = 5, t = 5,
𝑚𝑔𝑑 𝑓J5 = 2
n control parameters
𝑘L = 25, 𝑘7 = 10, 𝛾" = 5, 𝑖 = 1,2,3
n test trajectories (starting with 𝜃 0 = 0, 𝜃̇ 0 = 0)
n first
𝜃; 𝑡 = − sin 𝑡 Note: same test trajectories
n second used also for robust control

𝜃̈; 𝑡 = (periodic) bang-bang acceleration profile with

𝐴 = 1 rad/s2, 𝜔 = 1 rad/s
Robotics 2 16
 Results
first trajectory
note the nonlinear system dynamics
(no sinusoidal regime at steady state!)
𝑒

𝑒̇

position and velocity errors control torque

𝜃; 𝑡 = − sin 𝑡

Robotics 2 17
 Results
second trajectory
note the torque discontinuities
𝑒 (due to those of the desired acceleration)

𝑒̇

position and velocity errors control torque

𝜃̈; 𝑡 = (periodic) bang-bang acceleration profile

Robotics 2 18
 Estimates of dynamic coefficients

t
𝑚𝑔𝑑 − 𝑚𝑔𝑑 𝐼 − 𝐼J

t
𝑚𝑔𝑑 − 𝑚𝑔𝑑
𝐼 − 𝐼J

𝑓5 − 𝑓J5 𝑓5 − 𝑓J5

errors 𝑎[ = 𝑎 − 𝑎N
first trajectory second trajectory
only the estimate of the viscous all three estimates of
friction coefficient converges dynamic coefficients converge
to the true value to their true values
Robotics 2 19
 A special case: Adaptive regulation
n adaptation in case 𝑞; is constant
n no special simplifications for the presented adaptive control
law (designed for the general tracking case…)
𝑢=𝑀 J 𝑞 𝑞̈ R + 𝑆O 𝑞, 𝑞̇ 𝑞̇ R + 𝑔N 𝑞 + 𝐹P} 𝑞̇ R + 𝐾L 𝑒 + 𝐾7 𝑒̇

𝑎Ṅ = Γ𝑌Z (𝑞, 𝑞,̇ 𝑞̇ R , 𝑞̈ R )(𝑞̇ R − 𝑞)̇

since 𝑞̇ R = Λ(𝑞; − 𝑞) and 𝑞̈ R = −Λ𝑞̇ do not vanish!
n a different case would be the availability of an adaptive
version of the trajectory tracking controller
𝑢=𝑀 J 𝑞 𝑞̈ ; + 𝑆O 𝑞, 𝑞̇ 𝑞̇ ; + 𝑔N 𝑞 + 𝐹P} 𝑞̇ ; + 𝐾L 𝑒 + 𝐾7 𝑒̇
since, when 𝑞; collapses to a constant, only the adaptation
of the gravity term would be left over (which is what one
would naturally expect…)
Robotics 2 20
 An efficient adaptive regulator
n use a linear parameterization of the gravity term only
𝑔 𝑞 = 𝐺(𝑞)𝑎‚
with a 𝑝𝑔-dimensional vector 𝑎𝑔
n an adaptive regulator yielding global asymptotic stability
of the equilibrium state (𝑞𝑑, 0) is provided by
𝑢 = 𝐺 𝑞 𝑎N‚ + 𝐾L 𝑞; − 𝑞 − 𝐾7 𝑞̇
2𝑒
𝑎Ṅ ‚ = 𝛾𝐺 Z 𝑞 u − 𝛽𝑞̇ , 𝛾>0
1+2 𝑒
where 𝑒 = 𝑞𝑑 − 𝑞, 𝐾𝑃 > 0, 𝐾𝐷 > 0 (symmetric), and
𝛽 > 0 is chosen sufficiently large
(see paper by P. Tomei, IEEE TRA, 1991; available as extra material on the course web)
Robotics 2 21
 An adaptive regulator
Sketch of asymptotic stability analysis

n use the function

Z𝑀 𝑞 𝑒
𝛽 Z Z 2𝑞̇ 1 Z
𝑉 = 𝑞̇ 𝑀 𝑞 𝑞̇ + 𝑒 𝐾L 𝑒 − u + 𝑎N‚ − 𝑎‚ 𝑎N‚ − 𝑎‚
2 1+2 𝑒 2
n a sufficient condition for 𝑉 to be a Lyapunov candidate is that
2𝑀„
𝛽>
𝑀ƒ 𝐾L,ƒ
n a sufficient condition which guarantees also that
𝑉̇ = ⋯ ≤ −𝑎 𝑒 u − 𝑏 𝑞̇ u ≤ 0, 𝑎 > 0, 𝑏 > 0
is 𝑆(𝑞, 𝑞)
̇ ≤ 𝛼6 𝑞̇
u
2𝑀„ 1 𝐾7,ƒ 𝛼6
𝛽 > max , + 4𝑀„ +
𝑀ƒ 𝐾L,ƒ 𝐾7,ƒ 2𝐾L,ƒ 2
Note: for any symmetric, positive definite matrix 𝐴
𝐴„ = 𝜆Œ•Ž 𝐴 = 𝜆Œ•Ž 𝐴Z 𝐴 = 𝐴 and thus, e.g.,
V V
𝑞̇ Z 𝑀 𝑞 𝑞̇ ≥ 𝑀ƒ 𝑞̇ u
𝐴ƒ = 𝜆Œ•• 𝐴 u u
Robotics 2 22