1

Global Positioning of Robot Manipulators with Mixed Revolute and Prismatic Joints
Josip Kasac, Branko Novakovic, Dubravko Majetic and Danko Brezak Faculty of Mechanical Engineering and Naval Architecture University of Zagreb, I. Lucica 5, HR-10000 Zagreb, Croatia e-mail: {josip.kasac, branko.novakovic, dubravko.majetic, danko.brezak}@fsb.hr Phone: +385-1-6168-{375, 354, 348, 357} Fax: +385-1-6168-351 Corresponding author: Josip Kasac

Abstract The existing controllers for robot manipulators with uncertain gravitational force can globally stabilize only robot manipulators with revolute joints. The main obstacles to the global stabilization of robot manipulators with mixed revolute and prismatic joints are unboundedness of the inertia matrix and the Jacobian of the gravity vector. In this paper a class of globally stable controllers for robot manipulators with mixed revolute and prismatic joints is proposed. The global asymptotic stabilization is achieved by adding a nonlinear proportional and derivative term to the linear PID controller. By using Lyapunovs direct method, the explicit conditions on the controller parameters to ensure global asymptotic stability are obtained. Index Terms Position control, manipulators, robot dynamics, stability.

I. I NTRODUCTION It is well known that a PD plus gravity compensation controller can globally asymptotically stabilize a rigid-joints manipulator with both revolute and prismatic joints [1]. This approach has drawbacks since the gravitational torque vector, which depends on some usually uncertain parameters, is assumed to be known accurately. To overcome the parametric uncertainties on the gravitational torque vector, an adaptive version of PD controller has been introduced in [2], guaranteeing global asymptotic stability. The main drawback of this approach is that the gravity regressor matrix has to be known. On the other hand, most industrial robots are controlled by linear PID controllers which do not require any component of robot dynamics into its control law. A simple linear and decoupled PID feedback controller with appropriate control gains achieves the desired position without any steady-state error. This is the main reason why PID controllers are still used in industrial robots. The local asymptotic stability of the linear PID controller in a closed loop with robot manipulator is proved in [3],[4] and semiglobal asymptotic stability is shown in [5], [6]. By looking at the proof, it can be seen that the cubic term in the derivative of the Lyapunovs function hampers the global asymptotic stability. This is the reason to believe that linear PID control is inadequate to cope with highly nonlinear systems like robot manipulators, since the design of the linear PID control law is based solely on local arguments. The rst nonlinear PID controller which ensures global asymptotic stability (GAS) is proposed in [7]. In this work,
This work was supported by the National Scientic Foundation of Republic of Croatia under Grant No. 0120025, Application of Articial Intelligence in Robotics and Manufacturing Systems.

which was inspired by the results of [2], it is proven that global convergence is still preserved if the regressor matrix is replaced by the constant matrix. Since the regressor matrix is constant, the control law can be interpreted as a nonlinear PID controller which achieves GAS by normalized nonlinearity in the integrator term of the control law. The second approach to achieving GAS is the scheme of Arimoto [8] that uses a saturation function in the integrator to render the system globally asymptotically stable, just as the normalization did in [7]. A unied approach to both above mentioned controllers, which have a linear derivative term, linear or saturated proportional term, and a class of nonlinear integral action is given in [9]. An alternative approach to global asymptotic stabilization of robot manipulator is delayed PID (PId D) [5]. PId D can be understood as a simple PD controller to which an integral action is added after some transient of time. The idea of this approach consists of patching a global and a local controller. The rst drives the solutions to an arbitrarily small domain, while the second yields local asymptotic stability. None of the proposed controllers takes into account the effects of actuators saturation. Recently, several saturated PID controllers for robot manipulators with uncertain gravitational force have been reported: a semiglobal saturated linear PID controller [10] and two global saturated nonlinear PID controllers [11], [12]. Also, progress is made in the adaptive set point control of robot manipulators. In [13] an adaptive set point controller with uncertain gravity regressor is proposed and global asymptotic convergence is proved. In [14] a semiglobal adaptive controller with amplitude-limited torque and uncertainty in the kinematic and dynamic models is developed. All of the mentioned nonlinear PID controllers can globally or semiglobally stabilize robot manipulators with revolute joints only. In this paper an approach to GAS of robot manipulators with mixed revolute and prismatic joints is presented. In this approach GAS is achieved by adding a nonlinear proportional and derivative term to the linear PID controller. Explicit conditions on controller parameters which guarantee GAS are given. Throughout the paper we use the notation: x for the Euclidean norm of the vector x Rn , M {A} and m {A} for the maximal and minimal eigenvalues, respectively, of the symmetric positive denite matrix A, and I for the identity matrix of the appropriate dimension. This paper is organized as follows. Robot dynamics and its

main properties are presented in Section II. The main results are presented in Section III, where a class of PID controllers with nonlinear proportional and derivative gains is introduced and conditions for global asymptotic stability are established. The simulation results are presented in Section IV. Finally, the concluding remarks are emphasized in Section V. II. ROBOT DYNAMICS The model of n-link rigid-body robotic manipulator, in the absence of friction and disturbances, is represented by M (q) + C(q, q)q + g(q) = u, q (1)

where q is the n 1 vector of robot joint coordinates, q is the n 1 vector of joint velocities, u is the n 1 vector of applied joint torques and forces, M (q) is n n inertia matrix, C(q, q)q is the n1 vector of centrifugal and Coriolis torques and g(q) is the n1 vector of gravitational torques and forces, obtained as the gradient of the robot potential energy U (q) g(q) = U (q) . q (2)

The assumptions (4)-(6) are valid for all practically used robot manipulators. If the robot has no prismatic joints, then d1 , d2 , c2 , kg2 = 0, and the inertia matrix and the Jacobian of the gravity vector are uniformly bounded. In that case we get well known properties of the robot manipulators with revolute joints only, where a1 = m {M }, a2 = M {M }, c1 = kc and kg1 = kg (see e.g. [18], [19], [20]). From a practical point of view, the prismatic joints of robot manipulators are always limited by physical limits and the uniform bounds exist in (4)-(6) within the robot workspace. However, as pointed in [16], with the uniform bounds within the robot workspace the entire stability analysis becomes local, and it is necessary to know the domain of attraction. It is often very difcult to explicitly characterize the domain of attraction that could be much smaller then the robot workspace. So, a global result is always more useful and exact characterization of upper bounds in (4)-(6) for all q Rn is crucial in establishing global stability. III. A C LASS OF N ONLINEAR PID C ONTROLLERS A. Main Result Proposition 1: Consider the robot dynamics (1) in closed loop with nonlinear PID controller u = KP q KD q KI P () D ()q, (10) q q q = q, (11) where KP , KD and KI are n n constant positive denite symmetric matrices, P () and D () are n n positive q q denite diagonal matrix functions which can be written in the following form P () = kP q D () = q
(1) kP , (1) kD (2) kD (1)

We assume that the matrix C(q, q) is dened using the Christoffel symbols. The following properties of the robot dynamics with mixed revolute and prismatic joints are important for stability analysis (see e.g. [6], [15], [16], [17]). Property 1. The matrix M (q)2C(q, q) is skew-symmetric, i.e., z T (M (q) 2C(q, q))z = 0, z Rn . (3) Property 2. The inertia matrix M (q) is a positive denite symmetric matrix which satises a1 z
2

q I, q I+
(2) kD

(12) q I,
2

z T M (q)z (a2 + c2 q + d2 q 2 ) z 2 ,

(1) kD

(13)

(4)

for all z, q Rn , where a1 , a2 > 0, c2 , d2 0. Property 3. The Coriolis and centrifugal terms C(q, q)q satisfy C(q, q)q (c1 + d1 q ) q 2 , (5) for all q, q Rn , where c1 , d1 0. Property 4. There exist such positive constants kg1 and kg2 that the Jacobian of the gravity vector satises g(q) kg1 + kg2 q , q Rn . q (6)

where and are positive constants. If the following conditions are satised kP > kg2 , k1 = m {KP } kg > 0, k1 > max M {KI } m1 + kc d1 + d2 m , , (1) (2) m {KD } kD kD
(1)

(14) (15) , (16)

where kg is dened by (9) and m = a2 + c2 qd + d2 qd 2 , m1 = c2 + 2d2 qd , kc = c1 + d1 qd , (17) (18) (19)

The property (6) follows from the fact that the potential energy depends linearly on translational coordinates, while rotational coordinates appear only in trigonometric form with period 2. Property 5. (See the Appendix A) For any constant vector of desired joint positions qd Rn , the functions 1 1 G() = kg q 2 + kg2 q 3 + U (q) U (qd ) q T g(qd ), (7) q 2 3 2 3 T D() = kg q + kg2 q + q (g(q) g(qd )), q (8) are globally positive denite with respect to q Rn , where q = q qd denotes the joint position error vector, and kg = kg1 + kg2 qd . (9)

1 then the equilibrium [T q T ( + KI g(qd ))T ]T = 0 is q globally asymptotically stable.

B. Proof of Main Result The stability analysis is based on Lyapunovs direct method, and can be divided in four parts. First, error equations for the closed-loop system (1), (10), and (11) are determined. Second, the Lyapunov function (LF) candidate is proposed. Then, a global stability criterion on system parameters is established. Finally, the LaSalle invariance principle is invoked to guarantee the asymptotic stability.

1) Error Equations: The stationary state of the system (1), (10), (11) is q = 0, q = 0, = , and satises g(qd ) = KI . If a new variable z = is introduced, then the system (1), (10), (11) becomes M (q) + C(q, q)q + g(q) g(qd ) = u, q u = KP q KD q KI z P () D ()q, q q q z = q, (20) (21) (22)

and using property (4) we get V1 1 (1) (2) (m {KD } + kD q + kD q 2 ) q 2 1 2 (a2 + c2 q + d2 q 2 ) q 2 + 2 2 + a1 q + 0. q

2

(29)

Using the triangle inequality q q + qd , and rearrang ing the previous expression we get V1 1 1 (1) (m {KD } m) q 2 + (kD m1 ) q 3 + 2 2 1 (2) + (kD d2 ) q 4 + a1 q + 2 , q (30) 2

whose origin [T q T z T ]T = 0 R3n is the unique q equilibrium. 2) Lyapunov Function Candidate: We will now construct a Lyapunov function for the system (20)-(22). First, an output variable y = q + with some > 0 is introduced, and q the inner product between (20) and y is made, resulting in a nonlinear differential form which can be separated in the following way dV (, q, z) q = W (, q), q (23) dt where V (, q, z) is the Lyapunov function candidate (see the q Appendix B). For easier determination of conditions for positive deniteness of function V and W , the following decompositions are made: V (, q, z) = V1 (, q) + V2 (, z) and W (, q) = q q q q W1 (, q) + W2 (), where q q 1 1 q q V1 (, q) = q T M (q)q + T M (q)q + T KD q + q 2 2 1 (1) 1 (2) + kD q 3 + kD q 4 , (24) 3 4 1 1 V2 (, z) = z T KI z + q T KI z + q T KP q + q 2 2 1 (1) + kP q 3 + U (q) U (qd ) q T g(qd ), (25) 3 and W1 (, q) = q T M (q)q + q T (KD + D ())q + q q + T (M (q) C(q, q))q, q (26) T W2 () = q (KP KI ) + q q + kP
(1)

where m and m1 are dened by (17) and (18), respectively. The function V1 is positive denite if the following conditions are satised m {KD } > , m kD > , m1
(1)

kD > . d2

(2)

(31)

Further, we consider function V2 , which can be rearranged in the following form V2 =

T 1 1 1 z + q KI z + q + 2 1 (1) 1 + kP q 3 + q T KP q + U (q) U (qd ) q T g(qd ) 3 2 1 q T KI q . (32)

Applying property (7) we get V2 1 1 k1 M {KI } 2 + G() 0, q q

2

1 (1) k kg2 3 P

(33)

which is positive denite if the conditions > M {KI } , k1 (34)

+ T (g(q) g(qd )). q

(27)

In this way, the problem of determining conditions for positive deniteness of function V (, q, z), which contains three q variables, is transformed into two simpler problems of determination of conditions for positive deniteness of functions V1 (, q) and V2 (, z), which contain only two variables. The q q second advantage of the above mentioned decomposition of functions V and W is the elimination of unspecied constant from the nal stability conditions. 3) Stability criterion determination: The following step is determination of conditions for positive deniteness of the function V and positive semi-deniteness of the function W . First, we consider function V1 , which can be rearranged in the following form 1 1 T q q q V1 = (q + ) M (q) (q + ) 2 q T M (q) + 2 2 1 1 (1) 1 (2) + T KD q + kD q 3 + kD q 4 , q (28) 2 3 4

and (14) are satised, where k1 is dened by (15). Comparing (34) with (31), the following conditions for positive deniteness are obtained k1 m {KD } > M {KI }m,
(1) k1 kD (2) k1 kD

(35) (36) (37)

> M {KI }m1 , > M {KI }d2 ,

including condition (14). Note that, in the above-stated conditions, the unspecied positive constant is eliminated. The following step is the condition which ensures that the time derivative of LF is a negative semi-denite function, i.e. W 0. First, we consider function W1 . Applying properties (4) and (5) we get W1 (m {KD } + kD q + kD q 2 ) q (a2 + c2 q + d2 q 2 ) q 2 (c1 + d1 q ) q q 2 0.
(1) (2) 2

(38)

Using triangle inequality q q + qd , and rearranging the previous expression we get W1 [m {KD } m] q 2 + (1) + [kD (m1 + kc )] q q +
(2) [kD 2 2

x0

q2

y1

+ (39)

(d1 + d2 )] q

q ,

where kc is dened by (19). The function W1 is positive denite if the following conditions are satised m {KD } > , > , (d1 + d2 ) > . (40) m (m1 + kc ) Further, we consider function W2 . Using property (8) we get W2 (k1 M {KI }) q + D() 0, q
2 (1) kD (2) kD

lc1 x1

m2 g q1

(1) kP

kg2

+ (41)
Fig. 1.

which is positive denite if the conditions (34) and (14) are satised. Comparing (34) with (40) the following conditions for positive deniteness are obtained k1 m {KD } > M {KI }m,
(1) k1 kD > M {KI }(m1 + kc ), (2) k1 kD

The robot manipulator with revolute and prismatic joints.

;;;
M (q) =

m1 g y0

(42) (43) (44)

So, the control law (45) provides an alternative approach to the set-point control of robot manipulators with all revolute joints without using saturation functions. IV. S IMULATION E XAMPLE We consider a 2-DOF manipulator with revolute and prismatic joints [22], [16], as shown in Fig. 1. The matrices M (q), C(q, q) and vector g(q) in (1) are given by
2 2 m1 lc1 + m2 q2 + I1 + I2 0

> M {KI }(d1 + d2 ),

including condition (14)-(15). Also, in the above-stated conditions, the unspecied positive constant is eliminated. Notice that the conditions (35)-(37) are trivially implied by the conditions (42)-(44). So, the conditions (42)-(44), including (14)-(15), are the nal stability conditions which guaranty global stability. Finally, by invoking the LaSalle invariance principle, we conclude asymptotic stability. The conditions (42)-(44) can be represented in a more compact form by the expression (16). Remark 1. From the stability conditions it follows that the choice of the controller gains depends on the norm of the desired position qd . In other words, larger gains should be chosen if the desired position qd is larger. This property of the stability conditions is consequence of the fact that the inertia matrix and the Jacobian of the gravity vector are not uniformly bounded. The larger values of the robot position q increase the upper bounds of the mentioned matrices, implying also larger values of the controller gains in the stability conditions. Replacing qd from expressions (9), (17)-(19) with the norm of the maximal desired position of robot manipulators qd M , we get the stability conditions which are valid for any qd in the robot workspace. Remark 2. The control law (10)-(13) can also be applied to robot manipulators with all revolute joints. In that case, parameters kg2 , d1 , c2 , d2 vanish and the following simplied control law u = KP q KD q + kD q q KI
(1) t

0 , m2

(47) (48) (49)

C(q, q) = g(q) =

m2 q 2 q 2 m2 q1 q2

m2 q 1 q 2 , 0

m1 lc1 g cos q1 + m2 gq2 cos q1 , m2 g sin q1

q ( )d,
0

(45)

provides global asymptotic stability [21], if the following condition is satised m {KP } kg > max M {KI } M {M } kc , m {KD } k (1) D . (46)

where q = [q1 q2 ]T , q1 is the revolute joint variable and q2 is the prismatic joint variable. Inserting (47), (48) and (49) in expressions (4), (5) and (6) and using denitions and properties of matrix norms, we get c1 = 0, d1 = 2m2 , c2 = 0, d2 = m2 , 2 a2 = max{m1 lc1 + I1 + I2 , m2 }, kg1 = max{2m1 lc1 g, 2m2 g}, kg2 = 2m2 g. The numerical values of the model parameters are: m1 = 1 kg, m2 = 1 kg, lc1 = 0.8 m, I1 + I2 = 0.1 kg m2 . Initial conditions are q1 (0) = q1 (0) = q2 (0) = 0, and q2 (0) = 1 m. The nal desired positions are q1d = /2 rad and q2d = 2 m. The controller gains are chosen in accordance with stability conditions (14)-(16) as KP = diag{200, 200}, KD = diag{20, 20}, (1) (1) (2) KI = diag{300, 300}, kP = 20, kD = 20, kD = 10. Fig. 2. shows the position errors q1 and q2 obtained from simulation. It is obvious from the gure that, after a transient due to error in initial condition, the position errors tend asymptotically to zero.

0.5

0.5

1 q q (rad) 1 1d q q (m)
2 2d

for all z, q Rn , where we used triangle inequality q q + qd , and property (6). Therefore, function G() is a globally strictly convex funcq tion vanishing at the unique global minimum q = 0. This implies that G() is a globally positive denite function which q holds for any constant qd . Further, positive deniteness of the Hessian H() implies that the gradient F() is a strictly q q increasing vector function with respect to q , i.e. q T F() = q D() 0. q A PPENDIX B This Appendix presents the construction of the Lyapunov function V = V1 + V2 dened by (24) and (25). Taking inner product between (20) and output variable y = q + we get q the following nonlinear differential form q T M (q) + q T C(q, q)q + q T (g(q) g(qd )) + q T T +q KP q + q KD q + q T KI z + +kP
(1)

1.5

2 0
Fig. 2.

1
The position errors.

2
Time (sec)

V. C ONCLUSIONS In this paper a class of globally stable controllers for robot manipulators with mixed revolute and prismatic joints has been presented. The stability criterion in terms of Lyapunovs direct method is proposed to guarantee the global asymptotic stability. The stability criterion provides explicit conditions on the controller gains in terms of a few parameters extracted from the robot dynamics. Of course, some open problems remain to be solved and are currently under research. The rst one is the evaluation of the robustness of the proposed controller in the face of measurement noise and joint exibility. The second problem is the extension of the proposed controller with actuator constraints and velocity estimation through the ltered position measurement, following ideas reported in [11], [5]. Some preliminary results guarantee only semiglobal asymptotic stability. A PPENDIX A This Appendix presents a proof of Property 5 following ideas reported in [2] and [9]. Notice rst that G(0) = 0. The gradient of G() with respect q to q is given by F() = q G() q = kg q + kg2 q q + g( + qd ) g(qd ), (50) q q

q q T q + kD q q
3 (1) kD T T

(1)

+ kD q
(2) kD T

(2)

q
2 T

(1) +kP T

q + q q q+ q q q+ + M (q) + C(q, q)q + (g(q) g(qd )) + q q q q +T KP q + T KD q + T KI z = 0. q q q (53) Some terms in the differential form (53) can be represented in the following way q T M (q) = q q T M (q) = q q T Kj q = q T KI z = q T KI z = q k qT q = d T q M (q)q q T M (q)q q T M (q)q, (54) dt d 1 T 1 q M (q)q q T M (q)q, (55) dt 2 2 d 1 T q Kj q , j = P, D, (56) dt 2 d T q KI z q T KI q , (57) dt d 1 T z KI z , (58) dt 2 1 d q k+2 , k = 1, 2, (59) dt k + 2

where we used z = q in (57) and (58). Further, using denition (2) we get the following identity q T (g(q) g(qd )) = d U (q) U (qd ) q T g(qd ) . (60) dt

where we used (2). The gradient F() vanishes at q = 0, thus q G() has a critical point at q = 0. The Hessian matrix of G() q q with respect to q is given by q qT g( + qd ) q H() = kg I + kg2 q I + kg2 q + . q [ + qd ] q The Hessian is a positive denite matrix because of z T H()z = (kg + kg2 q ) z q kg1 + kg2 q
2

(51)

By inserting expressions (54)-(60) into (53) and separating terms in the form of the time derivatives (rst terms on the right sides of equations (54)-(60)) on the left side of the equality and the rest of the terms on the right side, we get the decomposition (23). ACKNOWLEDGMENT The authors would like to thank the anonymous reviewers for their helpful comments and suggestions.

+ kg2 g(q) q

(z T q )2 g(q) + zT z q q z
2

0,

(52)

R EFERENCES
[1] M. Takegaki and S. Arimoto, A new feedback method for dynamic control of manipulators, ASME J. Dyn. Syst. Meas. Contr., vol. 103, pp. 119125, 1981. [2] P. Tomei, Adaptive PD control for robot manipulators, IEEE Trans. Robot. Autom., vol. 7, no. 4, pp. 565570, 1991. [3] S. Arimoto, T. Naniwa, and H. Suzuki, Asymptotic stability and robustness of PID local feedback for position control of robot manipulators, in Proc. Int. Conf. on Automation, Robotics, and Computer Vision, Singapore, September 1990, pp. 382386. [4] R. Kelly, A tuning procedure for stable PID control of robot manipulators, Robotica, vol. 13, pp. 141148, 1995. [5] A. Loria, E. Lefeber, and H. Nijmeijer, Global asymptotic stability of robot manipulators with linear PID and PI2 D control, Stability and Control: Theory and Applications, vol. 3, no. 2, pp. 138149, 2000. [6] A. A. Pervozvanski and L. B. Freidovich, Robust stabilization of robotic manipulators by PID controllers, Dynamics and Control, vol. 9, no. 3, pp. 203222, 1999. [7] R. Kelly, Comments on: Adaptive PD control of robot manipulators, IEEE Trans. Robot. Autom., vol. 9, no. 1, pp. 117119, 1993. [8] S. Arimoto, A class of quasi-natural potentials and hyper-stable PID servo-loops for nonlinear robotic systems, Trans. Soc. Instrument Contr. Engg., vol. 30, no. 9, pp. 10051012, 1994. [9] R. Kelly, Global positioning of robot manipulators via PD control plus a class of nonlinear integral actions, IEEE Trans. Autom. Control, vol. 43, no. 7, pp. 934938, 1998. [10] J. Alvarez, R. Kelly, and I. Cervantes, Semiglobal stability of saturated linear PID control for robot manipulators, Automatica, vol. 39, pp. 989 995, 2003. [11] R. Gorez, Globally stable PID-like control of mechanical systems, Syst. Control Lett., vol. 38, pp. 6172, 1999. [12] M. J. Luis, V. Santibanez, and M. V. Hernandez, Saturated nonlinear PID global regulator for robot manipulators: Passivity based analysis, in Proc. 16th IFAC World Congress, Prague, Czech Republic, 4-8 July 2005. [13] H. Yazarel, C. C. Cheah, and H. C. Liaw, Adaptive SP-D control of a robotic manipulator in the presence of modeling error in a gravity regressor matrix: theory and experiment, IEEE Trans. Robot. Autom., vol. 18, no. 3, pp. 373379, 2002. [14] W. Dixon, Adaptive regulation of amplitude limited robot manipulators with uncertain kinematics and dynamics, in Proc. Amer. Control Conf., Boston, MA, 4-6 June 2004, pp. 38393844. [15] S. P. Hodgson and D. P. Stoten, Robustness of the minimal control synthesis algorithm to non-linear plant with regard to the position control of manipulators, Int. J. Control, vol. 72, no. 14, pp. 12881298, 1999. [16] R. Gunawardana and F. Ghorbel, On the boundedness of the Hessian of the potential energy of robot manipulators, J. Robotic Syst., vol. 16, no. 11, pp. 613625, 1999. [17] F. Ghorbel and R. Gunawardana, A uniform bound for the Jacobian of the gravitational force vector for a class of robot manipulators, ASME J. Dyn. Syst. Meas. Contr., vol. 119, pp. 110114, 1997. [18] R. Kelly, V. Santibanez, and A. Loria, Control of Robot Manipulators in Joint Space. London: Springer-Verlag, 2005. [19] R. Ortega, A. Loria, P. Nicklasson, and H. Sira-Ramirez, Passivitybased control of Euler-Lagrange Systems: Mechanical, Electrical and Electromechanical Applications. London: Springer-Verlag, 1998. [20] S. Arimoto, Control Theory of Nonlinear Mechanical Systems: A Passivity-Based and Circuit-Theoretic Apprach. Oxford University Press, 1997. [21] J. Kasac, B. Novakovic, D. Majetic, and D. Brezak, Performance tuning for a new class of globally stable controllers for robot manipulators, in Proc. 16th IFAC World Congress, Prague, Czech Republic, 4-8 July 2005. [22] F. Ghorbel and R. Gunawardana, On the uniform boundedness of the derivative of the gravitational force of serial link robot manipulators, in Proc. ASME International Mechanical Engineering Congress and Exposition, San Francisco, California, November 12-17 1995, pp. 2228.