2010 IEEE International Conference on Control Applications

Part of 2010 IEEE Multi-Conference on Systems and Control

Yokohama, Japan, September 8-10, 2010

Observer Based Output Feedback Tracking Control of Robot Manipulators

Erkan Zergeroglu and Enver Tatlicioglu

Abstract— In this paper, we developed a new observer based structure can be designed to compensate for the parametric
output feedback (OFB) tracking controller for rigid-link robot uncertainty of the robot dynamics.
manipulators. Specifically, a model independent variable struc- The paper is organized as follows. In Section II the
ture like observer structure in conjuction with the use of desired
system dynamics in the controller design have been utilized dynamical model of the robot manipulator with its properties
to remove the link velocity dependency of the controller and used in the analysis and design of the proposed observer-
the asymptotic stability of the observer-controller couple is controller couple are presented while Section III contains
then guaranteed via Lyapunov based arguments. Simulation the error system development and problem formulation. In
results are included to demonstrate the observer/controller Section IV the design and stability analysis of the controller
performance.
and observer are proposed. In Section V, we demonstrate
I. I NTRODUCTION the effectiveness of the proposed method through simulation
results obtained from a two link, direct drive planar robot
Nearly all commercially available robot manipulators do
manipulator. Finally, concluding remarks are presented in
not have link velocity sensors and the ones that have velocity
Section VI.
sensors the sensor outputs are, most of time, contaminated
with noise. Therefore the output feedback tracking control II. ROBOT M ODEL
of robot manipulators, where only link position information
The mathematical model for an n DOF, revolute joint, di-
is available, have received considerable interest in robotics
rect drive robot manipulator is assumed to have the following
literature over the past years. The existing solutions to the
form [13]
forementioned problem can be categorized as observer based
[1], [2] and filtered based [3], [4], [5], [6], [7] methods. In M (q)q̈ + Vm (q, q̇)q̇ + G(q) + Fd q̇ = τ (1)
observer based methods either a model based [2], [8] or a
model free observer is used to estimate the velocity signal, where q(t), q̇(t), q̈(t) ∈ Rn denote the link position, velocity,
where in filtered based approaches surrogate filters are used and acceleration, respectively, M (q) ∈ Rn×n represents the
to overcome the need of velocity measurements. positive-definite, symmetric inertia matrix, Vm (q, q̇) ∈ Rn×n
In this paper we present a new model free observer represents the centripetal-Coriolis matrix, G(q) ∈ Rn is the
based output feedback controller, some of the past research gravitational vector, Fd ∈ Rn×n denotes the constant, diag-
that applied a similar approach are as follows: In [9] a onal, positive-definite viscous friction matrix, and τ (t) ∈ Rn
variable structure output feedback controller was designed represents the torque input control vector. In the subsequent
to compensate for the lack of link velocity measurement. development, we will assume that the left-hand side of (1)
Similarly in [10] a discontinuous controller with a high gain is first-order differentiable.
observer was proposed for the stabilization of a class of The dynamic model given by (1) exhibits the following
nonlinear systems. Recently a Luenberger like observer with properties that will be utilized in the subsequent control
an extra switching term was proposed in [11] for the output development and the associated stability analysis.
feedback control of robot manipulators. Property 1: The inertia matrix can be bounded from
In this paper, inspired by the observer structure given in above and below by the following inequalities [13]
[17], we propose a new model free observer in conjunction
m1 In ≤ M (q) ≤ m2 In (2)
with a desired robot model based controller formulation
for the output feedback tracking control of robot manipu- where m1 and m2 are positive constants, and In is the
lators. The observer/controller structure proposed achieves standard n × n identity matrix. Likewise, the inverse of the
semi-global asymptotic tracking despite the lack of velocity inertia matrix can be bounded as follows [13]
measurements. Though in its current form the proposed 1 1
methodology require the exact knowledge of the system In ≤ M −1 (q) ≤ In · (3)
m2 m1
parameters, with considerably small effort adaptive and Property 2: The inertia and the centripetal-Coriolis matri-
repetitive learning versions of the same observer/controller ces satisfy the following relationship [14]
 
E. Zergeroglu is with the Department of Computer Engineering, Gebze T 1
Institute of Technology, 41400, Gebze, Kocaeli, Turkey. ξ Ṁ (q) − Vm (q, q̇) ξ = 0 ∀ ξ ∈ Rn (4)
ezerger@bilmuh.gyte.edu.tr
2
E. Tatlicioglu is with the Department of Electrical & Electronics Engineer-
ing, Izmir Institute of Technology, Urla, Izmir, 35430 Turkey. where Ṁ (q) represents the time derivative of the inertia
envertatlicioglu@iyte.edu.tr matrix.

978-1-4244-5363-4/10/$26.00 ©2010 IEEE 602

 Property 3: The centripetal-Coriolis matrix satisfies the IV. O BSERVER -C ONTROLLER D ESIGN
following relationship [8] Based on the subsequent error system development and the
stability analysis we propose the following velocity observer
Vm (q, ν)ξ = Vm (q, ξ)ν ∀ ξ, ν ∈ Rn . (5)
Property 4: The norm of the centripetal-Coriolis and fric- q̂˙ = p + K0 q̃ − Kc e
(12)
tion matrices can be upper bounded as follows [13] ṗ = K1 Sgn (q̃) + K2 q̃ − αKc e
where p (t) ∈ Rn is an auxiliary variable, Sgn (·) ∈ Rn is
kVm (q, ξ)k ≤ ζc1 kξk , kFd k ≤ ζf ∀ ξ ∈ Rn . (6)
defined as
T
where ζc1 and ζf are positive constants. ∀ζ ∈ Rn

Sgn (ζ) = sgn (ζ1 ) sgn (ζ2 ) ...sgn (ζn )
Property 5: The inertia, centripetal-Coriolis, and gravity (13)
terms in (1) can be upper bounded as follows [15] with sgn (·) being the standard signum function, K0 , Kc ,
K1 , K2 ∈ Rn×n are diagonal, positive define gain matrices
kM (ξ) − M (ν)ki∞ ≤ ζm1 kξ − νk and α was defined in (11). It is straightforward to show that
M −1 (ξ) − M −1 (ν) i∞ ≤ ζm2 kξ − νk the time derivative of (12) yields
(7)
kVm (ξ, η) − Vm (ν, η)ki∞ ≤ ζc2 kηk kξ − νk
kG(ξ) − G(ν)k ≤ ζg kξ − νk q̂¨ = K1 Sgn (q̃) + K2 q̃ + K0 q̃˙ − Kc r (14)

∀ ξ, ν, η ∈ Rn , where ζm1 , ζm2 , ζc2 , and ζg ∈ R are positive where the definition of r (t) given in (11) has been utilized.
bounding constants, and k·ki∞ denotes the induced infinity Similarly, assuming that exact knowledge of all the system
norm of a matrix. parameters are available, the control torque input signal τ (t)
The robot dynamics given in (1) can be written in terms is designed to have the following form
 
of the desired trajectory in the following manner τ = Wd + Kp e + Kc α (qd − q̂) + Kc q̇d − q̂˙ (15)
Wd = M (qd )q̈d + Vm (qd , q̇d )q̇d + G(qd ) + Fd q̇d (8) where the first term, Wd (·), defined in (8) is the desired robot
dynamics, Kp ∈ Rn×n is diagonal positive define control
where Wd (qd , q̇d , q̈d ) ∈ Rn is a function of the desired link gain matrix and Kc , α were previously defined. Note that
position, velocity, and acceleration vectors, denoted by qd (t), using the fact that
q̇d (t), q̈d (t) ∈ Rn , respectively.
qd − q̂ = e + q̃ (16)
III. P ROBLEM F ORMULATION the control torque input given in (15) can be rewritten in the
The control objective is to design a link position tracking following advantageous form
controller for the robot manipulator model given by (1) under τ = Wd + Kp e + Kc r + Kc s. (17)
the restrictive constraint that only the link position variable
q(t) is available for control development. We will quantify A. Observer Analysis
the control objective by defining the link position tracking After utilizing (1) and (14), the velocity observation error
error e(t) ∈ Rn as follows dynamics can be obtained as

e , qd − q (9) q̃¨ = q̈ − q̂¨

(18)
= N0 − K1 Sgn (q̃) − K2 q̃ − K0 q̃˙ + Kc r
where we assume that qd (t) and its first three time deriva-
where the auxiliary term N0 (t) ∈ Rn is defined as
tives are bounded functions of time. To account for the
unmeasurable link velocity constraint, we define q̂˙ (t) ∈ Rn N0 = M −1 (q) {τ − Vm (q, q̇) q̇ − G (q) − Fd q̇} . (19)
as the observed velocity signal. The corresponding velocity
and position observation error signals q̃˙ (t), q̃ (t) ∈ Rn are After inserting (17) and (8) in (19), we can write N0 (t) in
defined as the following form
˙
q̃˙ = q̇ − q̂, N0 = Nd + Nb (20)
(10)
q̃ = q − q̂.
where the auxiliary functions Nd (t) ∈ Rn and Nb (t) ∈ Rn
To ease the presentation of the analysis, we will use two are defined as
auxiliary variables, filtered tracking error, denoted by r (t) ∈ Nd (t) , q̈d (21)
Rn , and filtered observation error, denoted by s (t) ∈ Rn as
and
s , q̃˙ + αq̃ Nb (t) ,


r , ė + αe, and (11) M −1 (q) − M −1 (qd ) M (qd ) q̈d
+M −1 (q) {Vm (qd , q̇d ) q̇d − Vm (q, q̇) q̇
where α ∈ R is a positive control gain. It should be +G (qd ) − G (q) + Fd (q̇d − q̇)
noted that, from (11), regulating r (t) and s (t) ensures the +Kp e + Kc (r + s)} .
regulation of e (t) and q̃ (t), respectively. (22)

603
 Remark 1: Exploiting the boundedness properties of the and substituting (26), time derivative of (28) and (29), we
desired trajectory, we can show that Nd (t) and Ṅd (t) are can obtain
bounded. Furthermore, as illustrated in Appendix I, after
 
K2
using (5), (6), (7), and the mean value theorem [19], Nb (t) V̇0 = sT − s + K c r + Nb . (31)
α
can be upper bounded as follows
The first term in the brackets of (31) will be used for both
2
kNb (t)k ≤ ρo1 kek + ρo2 krk + ρo3 krk + ρo4 ksk (23) damping the unwanted effects of the term Nb (t) in the
composite stability analysis and to ensure the convergence of
where ρoi , i = 1, .., 4 are some positive known bounding the observation error. The second term is designed to cancel
functions that depend on the mechanical parameters and the out the interconnection term between the observer/controller
desired trajectory and k·k denotes the Euclidean norm. subsystem. At this point, we are ready to proceed to the error
Taking the time derivative of (11) and inserting (18), system development.
the dynamics for the filtered observation error s (t) can be
obtained as follows B. Error System Development
To obtain the dynamics of r (t), we take its time deriva-
ṡ = Nd + Nb − K1 Sgn (q̃) − K2 q̃ − (K0 − α) q̃˙ + Kc r (24) tive and premultiply the resulting equation by M (q), and
and when the observer gains are selected to satisfy after utilizing (1) and (9)), and performing some algebraic
manipulation, to obtain
α (K0 − α) = K2 (25)
M (q) ṙ = −Vm (q, q̇) r + Ws − τ (32)
the expression in (24) can be rearranged to have the following n
where the auxiliary term Ws (t) ∈ R is defined as
form
Ws = M (q) (q̈d + αė) + Vm (q, q̇) (q̇d + αe) + G (q) + Fd q̇.
K2
ṡ = Nd + Nb − K1 Sgn (q̃) − s + Kc r. (26) (33)
α After substituting the control law (17) into (32) we obtain
Based on the expression in (26), we can state the following the following closed-loop dynamics for r (t)
preliminary Lyapunov-like analysis for the observer design.
Specifically, we define the following non negative scalar M (q) ṙ = −Vm (q, q̇) r + χ − Kc r − Kc s − Kp e (34)
function, Vo (t), as follows where the disturbance term χ (e, r, t) ∈ Rn is defined as
1 T follows
V0 = s s + P0 (27) χ = Ws − Wd (35)
2
where the scalar auxiliary function P0 (t) ∈ R is defined as with Wd (·) term was defined in (8).
Remark 2: As illustrated in [13], and also shown in Ap-
Zt pendix I, we can exploit the boundedness properties of the
P0 = ζ0 − w0 (σ) dσ (28) desired trajectory and the properties of the robot dynamics
t0 in (5), (6), (7), to show that the norm of the variable χ (·)
defined in (35) can be upper bounded as
with the scalar function w0 (t) ∈ Rn and the non-negative
constant ζ0 ∈ R defined as kχ (·)k ≤ ρ1 (e) kek + ρ2 (e) krk (36)
w , sT [Nd − K1 Sgn (q̃)] where ρ1 (e) ∈ R and ρ2 (e) ∈ R are known positive
P0n (29)
ζ0 , i=0 K1i |q̃i (0)| − q̃ T (0) Nd (0) bounding functions. The above bound will be exploited to
obtain the stability result presented in the next section.
where the subscript i = 1, 2, ..., n denotes the ith element of
the vector or diagonal matrix. Following a similar analysis1 C. Stability Analysis
to that of [16], [17], it can be proven that when K1 satisfies The combination of error systems in (26) and (34) yields
the following sufficient condition the following stability result for the observation error and
the position tracking error.
1
K1i > kNdi (t)k∞ + Ṅdi (t) (30) Theorem 1: The velocity observer in (12) and the control
α ∞
law in (15) ensure that the closed-loop observer/controller is
where the subscript i = 1, 2, ..., n denotes the ith element semi-globally asymptotically stable in the sense that
of the diagonal matrix and k·k∞ denotes the L∞ norm, then
ke (t)k , q̃˙ (t) → 0 as t → ∞ (37)
P0 (t) of (28) is always non zero, that is P0 (t) ≥ 0 and
V0 (t) is a positive-definite
p Lyapunov function with respect provided that the controller and observer gains are selected to
to s (t) and P0 (t). After taking the time derivative of (27) satisfy (25), (30), and the controller gain Kc and the observer
gain K2 are designed as follows
1 Though the analysis very similar to that of the one given in the

references, for the completeness of the presentation we have included it Kc = 1 + ρ2 + kn ρ21 In

(38)
in Appendix II K2 = α 1 + ρo4 + kn ρ2o1 + ρ2o2 + ρ2o3 In

604
where ρ1 (e), ρ2 (e) were defined in (36), ρoi , i = 1, 2, 3, 4 From (43), a sufficient condition on (49) can be obtained as
were defined in (23) and kn is a nonlinear damping gain 1

V (t)

selected to satisfy the following condition 1− 1+ >0
  2kn λ1
λ2 2 and hence at this point the analysis can be reformulated as
kn > 1 + kz (0)k /2 (39)
λ1 
V (t)

2
V̇ ≤ −β kxk provided that 2kn > 1 + (50)
and z (t) ∈ R3n+1 defined as follows λ1
√ T where β ∈ R is some positive constant (0 < β ≤ 1). Due
z (t) , sT P0 rT eT

(40)
to the negative semi-defineteness of V̇ (t), the maximum
and the positive bounding constants λ1 ∈ R and λ2 ∈ R are value that V (t) can have is its initial value (i.e., V (0)),
defined as therefore, from (43), a more conservative condition on kn
λ1 = 12 min {1, m1 , λmin {Kp }} can be obtained to have the following form
(41)
λ2 = 12 max {1, m2 , λmax {Kp }} 2 λ2 2
Proof: We start our proof by introducing the following V̇ ≤ −β kxk provided that 2kn > 1 + kz (0)k (51)
λ1
non negative function in the form
that is when kn is selected to satisfy (39), we can ensure
1 1 that V (t) in (42) remains bounded therefore z (t) ∈ L∞ ,
V = V0 + rT M (q) r + eT Kp e. (42)
2 2 thus e (t), r (t), s (t), P0 (t) ∈ L∞ . Following standard
From (42) V (t) can be upper and lower bounded as signal chasing arguments we can show that all signal in
2 2 2 the closed loop system are bounded and e (t) and q̃˙ (t) are
λ1 kxk ≤ λ1 kzk ≤ V ≤ λ2 kzk (43)
uniformly continuous (from the boundedness of their time
3n derivatives), furthermore after integrating both sides of (51),
where x (t) ∈ R is defined as
T it is easy to see that x (t) ∈ L2 and therefore e (t), q̃ (t),
x (t) , sT rT eT

, (44) q̃˙ (t) ∈ L2 . Finally, after utilizing a direct application of
where z (t) was defined in (40) and the positive constants Barbalat’s Lemma [12], we can obtain the result given in
λ1 , λ2 were defined in (41). After differentiating (42) with (37) provided that the gain condition of (39) is satisfied.
respect to time, then substituting (31), (34), and cancelling V. S IMULATION R ESULTS
common terms results in The proposed, observer based OFB controller was sim-
 
T K2 2 ulated on a two-link, direct-drive, planar robot manipulator
V̇ ≤ s − s + Nb + rT [χ − Kc r] − αλ kek (45)
α having the following dynamics [18]
  
where Property 2 was utilized and λ ∈ R+ denotes the p1 + 2p3 c2 p2 + p3 c2 q̈1
minimum eigenvalue of Kp . After applying (23) and (35) p2 + p3 c2 p2 q̈2  
to (45), we can form an upper bound on V̇ (t) as follows −p3 s2 q̇2 −p3 s2 (q̇1 + q̇2 ) q̇1
+ (52)
V̇ ≤ −αλ 2 2 2  p3 s2 q̇1   0   q̇2
h kek − krk − ksk i fd1 0 q̇1 τ1
+ ρo1 kek ksk − kn ρ2o1 ksk
2 + =
h i 0 fd2 q̇2 τ2
2
+ ρo2 krk ksk − kn ρ2o2 ksk (46) where p1 = 3.473[kg-m2], p2 = 0.193 [kg-m2], p3 =
0.242 [kg-m2 ], fd1 = 5.3 [Nm-sec], fd2 = 1.1 [Nm-sec],
h i
2 2
+ ρo3 krk ksk − kn2 ρ2o3 ksk
h
2
i c2 , cos(q2 ) and s2 , sin(q2 ).
+ ρ1 kek krk − kn ρ21 krk . The simulations were performed using the following de-
sired position trajectory
Completing the squares of the terms in the brackets we obtain 


    57.30 sin(t) 1 − exp −0.3t3


1 2 1 1 2 2 2 qd (t) = [deg] (53)
V̇ ≤ − αλ − kek − 1 − − krk krk −ksk 45.84 sin(t) 1 − exp −0.3t3
2kn 4kn 4kn
(47) where the exponential... term was included to ensure that
which using the definition of x (t) of (44) can be further q̇d (0) = q̈d (0) = q d (0) = 02×1 and the observer/controller
upper bounded as follows gains were selected as
 
Ko = diag 8 6 K1 = diag  20 20
 
1 
2

2
V̇ ≤ − 1 − 1 + kxk kxk . (48) 
Kc = diag 0.012 0.08 Kp = diag 60 32
2kn
α = 1.2
The sign of the upper bound of V̇ (t) is determined by the (54)
term in the brackets of (48). This term has to be positive with the initial link positions selected as q (0) =
T
to ensure the negative semi-definiteness of V̇ (t), that is to

10 10 deg. The link position tracking error is de-
ensure the negative semi definiteness of V̇ (t) we must have picted in Figure 1, while the control torque is shown in
1  2
 Figure 2. From Figure 1, it is clear that the position tracking
1− 1 + kxk > 0. (49) objective was met.
2kn

605
 5 [5] J. Yuan and Y. Stepanenko, “Robust Control of Robotic Manipulators
Link 1 Tracking Error [deg] without Velocity Measurements”, Int. J. Robust and Nonlinear Control,
0
vol. 1, pp. 203–213, 1991.
[6] F. Zhang, D.M. Dawson, M.S. de Queiroz, and W. E. Dixon, “Global
−5
Adaptive Output Feedback Control of Robot Manipulators”, IEEE Tr.
−10
on Automatic Control,vol. 45, no. 6, pp. 1203–1208, 2000.
[7] E.Zergeroglu, D.M. Dawson, M.S. de Queiroz, and M. Krstic, ”On
0 2 4 6 8 10 12 14 16 18 20
time [sec] Global Output Feedback Control of Robot Manipulators”, IEEE Inter-
national Conference on Decision and Control, pp. 5073–5078, Sydney,
4 Australia, 2000.
[8] S. Nicosia and P. Tomei, “Robot Control by Using Only Position
Link 2 Tracking Error [deg]

0 Measurements”, IEEE Tr. on Automatic Control, vol. 35, no. 9, pp.

−2 1058–1061, 1990.
−4 [9] C. Canudas de Wit and J. Slotine, “Sliding Observers for Robot
−6 Manipulators”, Automatica, vol. 27, no. 5, pp. 859–864, 1991.
−8 [10] S. Oh and H.K. Khalil, “Output Feedback Stabilization using Variable
−10 Structure Control”, Int. J. of Control, vol. 62, no. 4, pp.831–848, 1995.
0 2 4 6 8 10 12 14 16 18 20
time [sec] [11] A. Abdessameud, M.F. Khelfi, “A Variable Structure Observer for
the Control of Robot Manipulators”, Int. J. Appl. Math. Compt. Sci.,
Fig. 1. Link Tracking Errors vol.16, no.2, pp. 189–196, 2006.
[12] M. Krstic, I. Kanellakopoulos, and P. Kokotovic, Nonlinear and
Adaptive Control Design, New York, NY, USA: John Wiley & Sons,
15
1995.
Link 1 Control Input [Nt m]

10 [13] F.L. Lewis, C.T. Abdallah, and D.M. Dawson, Control of Robot
5
Manipulators, New York, NY, USA: Macmillan Publishing Co., 1993.
[14] P. Tomei, “Adaptive PD Controller for Robot Manipulators”, IEEE Tr.
0
on Robotics and Automation, vol. 7, no. 4, pp. 565–570, 1991.
−5 [15] N. Sadegh, and R. Horowitz, “Stability and Robustness Analysis of a
−10
Class of Adaptive Controllers for Robot Manipulators”, Int. J. Robotics
0 2 4 6 8 10 12 14 16 18 20 Research, vol. 9, no. 3, pp. 74–92, 1990.
time [sec] [16] B. Xian, D.M. Dawson, M.S. de Queiroz, and J. Chen, “A Continuous
Asymptotic Tracking Control Strategy for Uncertain Nonlinear Sys-
4
tems”, IEEE Tr. on Automatic Control, vol. 47, no. 7, pp. 1206–1211,
Link 2 Control Input [Nt m]

3 2004.
2 [17] B. Xian, M.S. de Queiroz, D.M. Dawson and M.L. McIntyre, “Output
1 Feedback Variable Structure-like Control of Nonlinear MEchanical
0
Systems”, Proc. of the 42nd IEEE Conference on Decision and
Control, pp. 368–373, Maui, HA, USA, 2003.
−1
[18] Direct Drive Manipulator Research and Development Package Oper-
−2
0 2 4 6 8 10 12 14 16 18 20 ations Manual, Integrated Motion Inc., Berkeley, CA, USA, 1992.
time [sec]
[19] H. K. Khalil, Nonlinear Systems, 3rd Edition, New York, NY, USA:
Prentice Hall, 2002.
Fig. 2. Control Torque Inputs
A PPENDIX I
P ROOF OF B OUNDS
VI. C ONCLUSION In this appendix, we illustrate how the upper bounds of
In this paper, we have presented a new observer based Nb (t) in (23) and χ (t) in (36) are obtained. We start with
output feedback tracking controller for robot manipulators. A exploting the expression given in (22), which can be rewritten
novel observer-controller couple was introduced that ensures in the following form
semi-globally asymptotic the tracking despite the lack of link

Nb = M −1 (q) − M −1 (qd ) M (qd ) q̈d
velocity measurements. Simulation results are presented to +M −1 (q) {Vm (qd , q̇d ) q̇d − Vm (q, q̇d ) q̇d }
illustrate the tracking performance of the observer-controller +M −1 (q) {2Vm (q, ė) q̇d − Vm (q, ė) ė} (55)
couple. Future work will focus on extending the proposed +M −1 (q) {G (qd ) − G (q) + Fd (q̇d − q̇)}
result to adaptive and learning output feedback controllers +M −1 (q) {Kp e + Kc r + Kc s}
for robot manipulators.
where (5) was utilized. After applying (3), (6), and (7), we
R EFERENCES can obtain an upper bound for the right-hand-side of (55) as

[1] S. Arimoto, V. Parra-Vega, and T. Naniwa, “A Class of Linear Velocity

1
kNb (t)k ≤ {ζm1 m1 m2 kq̈d k + ζc2 kq̇d k (56)
Observers for Nonlinear Mechanical Systems”, Proc. Asian Control m1
Conf., pp. 633–636, Tokyo, Japan, 1994. +λmax {Kp } + ζg } kek
[2] H. Berghuis and H. Nijmeijer, “A Passivity Approach to Controller-
Observer Design for Robots”, IEEE Tr. on Robotics and Automation, 1
+ {2ζc1 kq̇d k + ζf + λmax {Kc }} krk
vol. 9, no. 6, pp. 740–754, 1993. m1
[3] T.C. Burg, D.M. Dawson, and P. Vedagarbha, “A Redesigned DCAL 1 1
2
Controller without Velocity Measurements: Theory and Demonstra- + ζc1 krk + λmax {Kc } ksk
tion”, Robotica, vol. 15, pp. 337–346, 1997. m1 m1
[4] E. Zergeroglu, W.E. Dixon, D. Haste, and D.M. Dawson, “A Com-
posite Adaptive Output Feedback Tracking Controller for Robotic where the fact that kr (t)k ≥ kė (t)k was utilized. From the
Manipulators”, Robotica, vol. 17, pp. 591–600, 1999. structure of (57), it is clear that the bounding functions in

606
(57) are valid and ρoi , i = 1, 2, 3, 4 are defined as After integrating the second integral on the right-hand side
of (61) by parts, the following expression is obtained
1 Z t Z t
ρ01 , ζm1 m2 kq̈d k + ζc2 kq̇d k (57) w0 (σ) dσ = q̃ T (σ) α [Nd (σ)
m1 t0 t0
1 1 t
+ λmax {Kp } + ζg , −K1 Sgn (q̃ (σ))] dσ + q̃ T (σ) Nd (σ)
m1 m1 Z t
t0
2 1 1 dNd (σ)
ρ02 , ζc1 kq̇d k + ζf + λmax {Kc } , − q̃ T (σ) dσ
m1 m1 m1 t0 dσ
1 m
ρ03 , ζc1 , X t
m1 − K1i |q̃i (σ)||t0
1 i=1
ρ04 , λmax {Kc } .
m1 t
Z
= q̃ T (σ) α [Nd (σ)
t0

For (36), we start with the previously found bound on the 1 dNd (σ)
− − K1 Sgn (q̃ (σ)) dσ
same term [13] (see Chapter 6 Eq: 6.2-9) as α dσ
+q̃ T (t) Nd (t) − q̃ T (t0 ) Nd (t0 )
2 n
kχk ≤ ζ1 kek + ζ2 kek + ζ3 krk + ζ4 krk kek (58) X
− K1i (|q̃i (t)| − |q̃i (t0 )|) . (62)
i=1
where ζi , i = 1, 2, 3, 4 are positive bounding constants that The right-hand side of (62) can be upper-bounded as follows
depend on the desired trajectory and physical parameters Z t Z tX n
(i.e., link mass, link length, friction coefficients, etc.). The w0 (σ) dσ ≤ |q̃i (σ)| α [|Ndi (σ)| (63)
right-hand-side of the expression in (58) can be rewritten in t0 t0 i=1
the following form 1 dNdi (σ)

+ − K1i dσ
α dσ
kχk ≤ (ζ1 + ζ2 kek) kek + (ζ3 + ζ4 kek) krk . (59) Xn
+ |q̃i (t)| (|Ndi (t)| − K1i ) + ζ0 .
i=1
When the bounding functions ρ1 (e) and ρ2 (e) are selected
If K1 is chosen to satisfy (30), then it is easy to obtain the
as
following expression from (63)
Z t
ρ1 (e) = ζ1 + ζ2 kek (60) w0 (σ) dσ ≤ ζ0 (64)
t0
ρ2 (e) = ζ3 + ζ4 kek
thus; from (28), it can be concluded that P0 (t) is non-
negative.
then the bound given in (36) is satisfied.

A PPENDIX II
T HE GAIN CONDITION OF K1

In this appendix, we will illustrate how the sufficient

condition of (30) is obtained. After substituting the definition
of s (t) in (11) into (29) and then integrating w0 (t) in time,
results in the following expression

Z t Z t
w0 (σ) dσ = q̃ T (σ) α [Nd (σ) (61)
t0 t0
−K1 Sgn (q̃ (σ))] dσ
Z t T
dq̃ (σ)
+ Nd (σ) dσ
t dσ
Z 0t T
dq̃ (σ)
− K1 Sgn (q̃ (σ)) dσ.
t0 dσ

607