Proceedings of the 2006 IEEE International Conference on Control Applications Munich, Germany, October 4-6, 2006

FrA08.2

Interconnection and damping assignment passivity-based experimental control of a single-link exible robot arm
Arantza Sanz and Victor Etxebarria
Abstract Energy-based control design methodology (the so-called IDA-PBC, interconnection and damping assignment passivity-based control) is a well established technique that has shown to be very powerful to design robust controllers for physical systems described by Euler-Lagrange equations of motion. Their application potential is particularly important for under-actuated systems. This paper presents the application of this control design methodology to a single-link exible robotic arm. It is shown that the method is well suited to handle this kind of under-actuated devices not only from a theoretical viewpoint but also in practice. A Lyapunov analysis of the closed-loop system stability is given and the design performance is illustrated by means of a set of simulations and laboratory control experiments, comparing the results with those obtained using conventional control schemes for mechanical manipulators.

with hamiltonian structure, where it is wanted the closedloop energy to have a minimum in the desired equilibrium for this assures its stability. The main advantage of this technique based on passivity is that the Lyapunov function is obtained in a natural way by the structure of the system itself. In this paper the application of the IDA-PBC method to the control of a single-link exible robotic arm is presented. Both a theoretical stability analysis and experimental results on a laboratory exible arm are provided, to illustrate the effectiveness of the proposed approach. II. IDA-PBC METHOD In this section a brief outline of the IDA-PBC method is presented, (see [9], [10] for complete details). The goal is to control the position of an under-actuated mechanical system with total energy: 1 T 1 p M (q )p + U (q ) (1) 2 where q n , p n , are the generalized positions and momenta respectively, M (q ) = M T (q ) > 0 is the inertia matrix and U (q ) is the potential energy. If it is assumed that the system has no natural damping, then the equations of motion of such system can be written as: H (q, p) = q p = 0 In In 0 q H p H + 0 G(q ) u (2)

I. I NTRODUCTION Flexible-link robotic manipulators have many advantages with respect to conventional rigid robots. These mechanisms are built using lighter, cheaper materials, which improve the payload to arm weight ratio, thus resulting in an increase of the speed with lower energy consumption. However, for a exible-link manipulator the number of controlled variables is strictly less than the number of mechanical degrees of freedom, so the control aim is double in this case. First, a exible manipulator controller must achieve the same motion objectives as a rigid manipulator. Second, it must also stabilize the vibrations that are naturally excited. The existence of the slow (rigid) and fast (exible) modes allows the application of the singular perturbation theory to exible arms ([5]).Then, a composite control strategy can be adopted, with the controller having slow and fast terms.This combined slow-fast strategy has proved to be a promising control method for robotic applications ([6], [7]). In recent years, control design methods based on the shaping of energy functions have attracted a lot of attention. This fact added to the growing interest in the control of under-actuated mechanical systems, has resulted in a successful development of the so-called IDA-PBC technique, which is a formulation based on passivity very well suited to the problem of stabilization of under-actuated systems ([3], [11]). This energy-based control design method, initially introduced in [8], [9], [10], adds, to the customary modication of the potential energy, the kinetic energy shaping, as an indispensable via to save the obstacles related to the under-actuation. The method produces a closed-loop system
A. Sanz and V. Etxebarria are with the Dept. Electricidad y Electrnica, Facultad de Ciencia y Tecnologa, Universidad del Pas Vasco, Apdo. 644, Bilbao 48080, Spain, wmbsabaa@lg.ehu.es,

In the IDA-PBC method two basic steps of the passivitybased control (see [8]) are followed. The rst of them is energy shaping, where the total energy function of the system is modied so that a predened energy value is assigned to the desired equilibrium. The second step is damping injection, to achieve asymptotic stability. The following form for the desired (closed-loop) energy function is proposed: Hd (q, p) = 1 T 1 p Md (q )p + Ud (q ) 2 (3)

T > 0 is the closed-loop inertia matrix and where Md = Md Ud the potential energy function. It will be required that Ud have an isolated minimum at the equilibrium point q , that is:

q = argminUd (q ) In PBC, a composite control law is dened:

(4)

victor@we.lc.ehu.es

0-7803-9796-7/06/$20.00 2006 IEEE

2504

u = ues (q, p) + udi (q, p)

(5)

where the rst term is designed to achieve the energy shaping and the second one injects the damping. The desired (closed-loop) dynamics can be expressed in the following form: q p where
T = Jd = Jd

Jd (q, p) Rd (q, p)

q H d p Hd M 1 Md J2 (q, p) 0

(6)

Fig. 1.

Photograph of the experimental exible manipulator.

0 Md M 1 0 0

(7) (8)

III. S INGLE - LINK FLEXIBLE ROBOT ARM A. Mathematical model The dynamics of a exible manipulator can be described by means of different methods. One of the more frequently employed combines the modal analysis with Lagranges method (see [1]). The use of this technique provides a dynamical model expressed by ordinary differential equations starting from a representation given by partial differential equations. The object of the study is a exible arm with one grade of freedom that accomplishes the conditions of Euler-Bernoulli. In this case, the elastic deformation of the arm w(x, t) is expressed as a superposition of a innite number of modes where the spatial and time variables are separated:

T Rd = Rd =

0 GKv GT

represent the desired interconnection and damping structures. J2 is a skew-symmetric matrix, and can be used as free parameter in order to achieve the kinetic energy shaping (see [9]). The second term in (5) can be expressed as: udi = Kv GT p Hd
T > 0. where Kv = Kv

(9)

A. Energy shaping To obtain the energy shaping term ues of the controller, (5) and (9) are replaced in (2) and this is equated to (6):
1 p Gues = q H Md M 1 q Hd + J2 Md

w(x, t) =
i=1

i (x)qi (t)

(15)

(10)

In the under-actuated case, G is not invertible, but only full column rank. Thus, multiplying (10) by the left annihilator of G, G G = 0, it is obtained:
1 p} = 0 G {q H Md M 1 q Hd + J2 Md

where i (x) is the mode shape and qi (t) is the mode amplitude. In order to derive a nite-dimensional ODE, a nite number of modes m is considered. Lagrange equations lead to a dynamical system dened by m + 1 second order differential equations: 1 u (0) (16) where q0 and qi are the dynamic variables; q0 (t) is the rigid generalized coordinate, qi (t) is the vector of exible modes, It is the total inertia, Kf is the stiffness matrix that depends on the elasticity of the arm, and it is dened as 2 ), where i is the resonance frequency of each Kf = diag (i mode; are the rst spatial derivatives of i (x) evaluated at the base of the robot. Finally u includes the applied control torques. Dening the incremental variables as q = q qd , where 0d = 0 qd is the desired trajectory for the robot such that q and qid = 0, then the dynamical model is given by: It 0 0 I q0 (t) qi (t) = 1 u (0) (17) The total energy of the mechanical system is obtained as the sum of the kinetic and potential energy: = It 0 0 I q 0 (t) 0 0 + 0 Kf q i (t) q 0 (t) + q0d q i (t) 0 q 0 (t) + 0 q i (t) 0 Kf

(11)

If a solution (Md , Ud ) exists for this differential equation, then ues will be expressed as:
1 p) ues = (GT G)1 GT (q H Md M 1 q Hd + J2 Md (12) The PDE (11) can be separated in terms corresponding to the kinetic and the potential energies:

1 1 p)+2J2 Md p}=0 G {q (pT M 1 p)Md M 1 q (pT Md (13)

G {q U Md M 1 q Ud } = 0

(14)

So the main difculty of the method is in solving the nonlinear PDE (13). Once the closed-loop inertia matrix, Md , is known, then it is easier to obtain Ud of the linear PDE (14).

2505

H ( q , p) = where p = M= 0 It q It 0 i Iq 0 I

1 T 1 1 T p M p+ q Kq 2 2
T

(18)

1 2 2 K1 z ,

Under these restrictions it can be proposed F (z ) = with K1 > 0, which produces the condition: a3 > (0)a2 (25)

and K= 0 0 0 Kf

So now, the term corresponding to the energy shaping in the control input (12) is given as ues = (GT G)1 GT (q U Md M 1 q Ud ) = Kp1 q 0 + Kp2 q 1 (26) where = = It (a1 a3 a2 2) 2 (K1 (a3 + (0)a2 ) + 1 ) (a2 + (0)a1 )2 2 K1 (a1 a3 a2 a1 1 2) (a2 + (0)a1 )

The point of interest is q = (0, 0), which corresponds to zero tracking error in the rigid variable and null deections. B. Controller design The controller design will be made in two steps; rst, we will obtain a feedback of the state that produces energy shaping in closed-loop to stabilize the position globally, then it will be injected the necessary damping to achieve the asymptotic stability by means of the negative feedback of the passive output. The inertia matrix, M , that characterizes to the system, is independent of q , hence it follows that it can be chosen J2 = 0, (see [9]). Then, from (13) it is deduced that the matrix Md should also be constant. The resulting representation capturing the rst exible mode is Md = a1 a2 a2 a3

Kp1 Kp2

The controller design is completed with a second term, corresponding to the damping injection. This is achieved via negative feedback of the passive output GT p Hd , (9). As 1 T q ), and Ud only depends on the q Hd = 1 2 p Md p + Ud ( variable, udi only depends on the appropriate election of Md : 0 + Kv 2 q 1 udi = Kv1 q with Kv 1 Kv 2 = = It (a3 + (0)a2 ) (a1 a3 a2 2) (a2 (0)a1 ) Kv (a1 a3 a2 2) Kv (27)

where the condition of positive deniteness of the inertia matrix, lead to the following inequations: a1 > 0 a1 a3 > a2 2 (19)

Now, considering only the rst exible mode, equation (14) corresponding to the potential energy can be written as: ( a2 (0)a1 Ud Ud 2 ) + (a3 (0)a2 ) =q 1 1 It q 0 q 1 (20)

A simple analysis on the constants Kp1 and Kv1 , with the conditions previously imposed, implies that both should be negative to assure the stability of the system. C. Stability Analysis Proposition 1: Consider the robot manipulator model (17) in closed-loop with the control law 0 + Kv 2 q 1 0 + Kp2 q 1 + Kv1 q u = Kp1 q The closed-loop equation (17), (28) is given by (28)

which it is a trivial linear PDE whose general solution is Ud =

2 2 2 1 (a3 (0)a2) 2 It It 1 q 0 q q 0 1 + F (z) (21) 2 2(a2+ (0)a1) (a2+ (0)a1)

1 z = q 0 + q It (a3 (0)a2 ) (a2 + (0)a1 )

(22)

(23)

where F is an arbitrary differentiable function that we should choose to satisfy condition (4) in the points q . Some simple calculations show that the necessary restriction q Ud (0) = 0 is satised if F (z (0)) = 0, while condition 2 q Ud (0) > 0 is satised if:
2 1 Fz > (a3 (0)a2 )

0 q 1 q (29) 1 q q ( K q + K q + K + K ) p1 0 p2 1 v1 0 v2 1 It 0 + Kv2 q 1 ) 1 + (0)(Kp1 q 0 + Kp2 q 1 + Kv1 q Kf q

q 0 d 1 q 0 = dt q 1 q

(24)

It is simple to check that there exists a unique equilibrium ) = (0, 0) for the closed-loop system. The point point ( q, q ) = (0, 0) is globally asymptotically stable for (17), (28). ( q, q

2506

Proof: To analyze the stability of the closed-loop system we consider the energy-based Lyapunov function candidate ([2], [4]) ) = 1 pT M 1 p + Ud = V ( q, q d 2
2 2 2 2 2 0q 1 a2 + q 2 q 0 a3 2It q 1 It 1 (a3 (0)a2 ) q0 1 It 1 a1 2 a1 a3 a2 2 (a2 + (0)a1 )2 2 2 q 0 q 1 It 1 1 + K1 ( q 0 + q 1 )2 a2 + (0)a1 2

Property Motor inertia, Ih Link length, L Link height, h Link thickness, d Link mass, Mb Linear density, Flexural rigidity, EI TABLE I

Value 0.002kgm2 0.45m 0.02m 0.0008m 0.06kg 0.1333kg/m 0.1621N m2

F LEXIBLE LINK PARAMETERS

(30)

which is globally positive denite, i.e: V (0, 0) = (0, 0) and ) > 0 for every ( ) = (0, 0). V ( q, q q, q The time derivative of (30) along the trajectories of the closed-loop system (29), can be written as 0 (a3 + (0)a2 )+(a2 (0)a1 )q 1 )2 ( )=Kv (It q V q ,q 0 2 (a1 a3 a2 2) (31) is negative semidenite, and ( ) = q, q where Kv > 0, so V (0, 0) is stable (not necessarily asymptotically stable). By using LaSalles invariant set theory, it can be dened =0 the set R as the set of points for which V q 0 q 1 = { 0 q 1 q { q 1 0 , q
1

1 (Kp1 q 0 + Kp2 q 1 ) It 1 + (0)(Kp1 q 0 + Kp2 q 1 ) Kf q whose solutions are of the form q 0 (t) q 1 (t) = 0 0

= =

0 0

In other words, the largest invariant set N is just the ) = (0, 0), so we can conclude that any trajectory origin ( q, q converge to the origin when t , so the equilibrium is in fact asymptotically stable. IV. S IMULATION AND EXPERIMENTAL RESULTS A. Simulations To illustrate the performance of the proposed IDA-PBC controller, in this section we present some simulations. We use the model of a exible robotic arm presented in [1] and the values of Table I which correspond to the real arm displayed in Fig. 1. The results are shown in Figs. 2 to 4. In these examples the values a1 = 1, a2 = 0.01 and a3 = 50 have been used to complete the conditions (19) and (25). In Fig. 2 the parameters are K1 = 10 and Kv = 1000. In Figs. 3 and 4 the effect of modifying the damping constant Kv is demonstrated. With a smaller value of Kv , Kv = 10, the rigid variable follows the desired trajectory reasonably well. For Kv = 1000, the tip position exhibits better tracking of the desired trajectory, as it can be seen comparing Fig. 2(a) and Fig. 3(a). Even more important, it should be noted that the oscillations of elastic modes are now attenuated quickly (compare Fig. 2(b) and Fig. 3(b)). But If we continue increasing the value of Kv , Kv = 100000, the oscillations are attenuated even more quickly (compare Fig. 2(b) and Fig. 4(b)), but the tip position exhibits worse tracking of the desired trajectory. B. Experimental results The effectiveness of the proposed control schemes has been tested by means of real time experiments on a laboratory single exible link. This manipulator arm, fabricated by Quanser Consulting Inc. (Ontario, Canada), is a spring steel bar that moves in the horizontal plane due to the action of a DC motor. A potentiometer measures the angular position of the system, and the arm deections are measured by means of a strain gauge mounted near its base (see Fig. 1).

R =

( ) = 0} = :V q, q 0 + q 1 = 0 } q (32)

&

Following LaSalles principle, given R and dened N as the largest invariant set of R, then all the solutions of the closed loop system asymptotically converge to N when t . Any trajectory in R should verify: 0 + q 1 q = 0 0 K (33) (34) (35)

and therefore it also follows that: q 1 = q 0 + 1 = q 0 + q

Considering the closed-loop system, described by (29) and the conditions described by (34) and (35), the following expression is obtained:
2 1 ( + (0) Kf )K +( + (0)) q0 = 0 (36) It (a2 + (0)a1 )

As a result of the previous equation, it can be concluded 0 (t) = 0, and that q 0 (t) should be constant, in consequence q 1 (t) = 0. Therefore replacing it in (33), it also follows that q q 1 = 0 and replacing these in (29) this leads to the q 0 = following system:

2507

1.2

0.8

0.8 0.6

Rigid variable and ref (rad)

0.4

Rigid variable and ref (rad)

0.6

0.4

0.2

0.2

0.2 0.2 0.4 0.4 0.6 0.6

(a)

0
4

10 Time (s)

12

14

16

18

20

(a)
2 1.5

0
4

10 Time (s)

12

14

16

18

20

x 10

x 10

Flexible deflections (rad)

Flexible deflections (rad)

0.5

0.5

1.5

2 0 2 4 6 8 10 Time (s) 12 14 16 18 20

(b)

(b)

10 Time (s)

12

14

16

18

20

Fig. 2. Simulation results for IDA-PBC control with Kv = 1000: (a) Time evolution of the rigid variable q0 and reference qd ; (b) Time evolution of the exible deections.

Fig. 4. Simulation results for IDA-PBC control with Kv = 100000: (a) Time evolution of the rigid variable q0 and reference qd ; (b) Time evolution of the exible deections.

0.8

0.6

0.4

Rigid variable and ref (rad)

0.2

0.2

0.4

0.6

0.8

(a)
8 6

0
3

10 Time (s)

12

14

16

18

20

x 10

The experimental results are shown on Figs. 5 and 6. In Fig. 5 the control results using a conventional PD rigid control design are displayed. The rigid variable tracks the reference (with a certain error), but the naturally excited exible deections are not well damped (Fig. 5(b)). In Fig. 6 the results using the IDA-PBC design philosophy are displayed. As seen in the graphics, the rigid variable follows the desired trajectory, and moreover the exible modes are now conveniently damped, (compare Fig. 5(b) and Fig. 6(b)). It is shown that vibrations are effectively attenuated in the intervals when qd reaches its upper and lower values which go from 1 to 2 seconds, 3 to 4 s., 5 to 6 s., etc. V. C ONCLUSIONS The IDA-PBC method is a control design tool based on energy concepts. This work has presented a theoretical and experimental study of this method for controlling a laboratory single exible link, valuing in this way the potential of this technique in its application to under-actuated mechanical systems, in particular to exible manipulators. The study is completed with the Lyapunov stability analysis of the closedloop system that is obtained with the proposed control law. Then, as an illustration, a set of simulations and laboratory control experiments have been presented. The experimented scheme, based on an IDA-PBC philosophy, has been shown to achieve good tracking properties on the rigid variable and

Flexible deflections (rad)

(b)

10 Time (s)

12

14

16

18

20

Fig. 3. Simulation results for IDA-PBC control with Kv = 10: (a) Time evolution of the rigid variable q0 and reference qd ; (b) Time evolution of the exible deections.

2508

0.6

0.4

denitely superior damping of the unwanted vibration of the exible variables compared to conventional rigid robot control schemes. ACKNOWLEDGMENT The authors are grateful to UPV/EHU and to the Basque Government for partial support of this work through project 9/UPV 00224.310-15254/2003 and grant BFI04.440, respectively. R EFERENCES
0 2 4 6 8 10 12 14 16 18 20

Rigid variable and ref (rad)

0.2

0.2

0.4

0.6

(a)

Time (s)

0.03

0.02

0.01

0.01

0.02

0.03 0 2 4 6 8 10 12 14 16 18 20

(b)

Time (s)

Fig. 5. Experimental results for a conventional PD rigid controller: (a) Time evolution of the rigid variable q0 and reference qd ; (b) Time evolution of the exible deections.

0.6

0.4

0.2

0.2

[1] C. Canudas de Wit, B. Siciliano, and G. Bastin. Theory of Robot Control. Springer, 1996. [2] V. Etxebarria. Sistemas de control no lineal y robtica (in spanish). Basque Country University Press, 1999. [3] F. Gmez-Estern. Control de sistemas no lineales basado en la estructura hamiltoniana (in spanish). PhD thesis, University of Seville, 2002. [4] R. Kelly and R. Campa. Control basado en IDA-PBC del pndulo con rueda inercial: Anlisis en formulacin lagrangiana (in spanish). Revista Iberoamericana de Automtica e Informtica Industrial, 2(1):36 42, Enero 2005. [5] P. Kokotovic, H. K. Khalil, and J.OReilly. Singular perturbation methods in control. SIAM Press, 1999. [6] Y. Li, B. Tang, Z. Zhi, and Y. Lu. Experimental study for trajectory tracking of a two-link exible manipulator. International Journal of Systems Science, 31(1):39, 2000. [7] I. Lizarraga and V. Etxebarria. Combined PD-H approach to control of exible manipulators using only directly measurable variables. Cybernetics and Systems, 34(1):1932, 2003. [8] R. Ortega and M. Spong. Stabilization of underactuated mechanical systems via interconnection and damping assignment. In Proceedings of the 1st IFAC Workshop on Lagrangian and Hamiltonian methods in nonlinear systems, pages 7479, Princeton, NJ,USA, March 2000. [9] R. Ortega, M. Spong, F. Gmez, and G. Blankenstein. Stabilization of underactuated mechanical systems via interconnection and damping assignment. IEEE Transactions on Automatic Control, AC47(8):12181233, August 2002. [10] R. Ortega, A. van der Schaft, B. Maschke, and G. Escobar. Interconnection and damping assignment passivity-based control of portcontrolled hamiltonian systems. Automatica, 38:585596, 2002. [11] S. Yoshiki, K. Ogata, and Y. Hayakawa. Control of manipulator with 2 exible-links via energy modication method by using modal properties. In Proceedings of the 2003 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM 2003), pages 14351441, 2003.

Rigid variable and ref (rad)

Flexible deflections (rad)

0.4

0.6 0 2 4 6 8 10 12 14 16 18 20

(a)

Time (s)

0.03

0.02

Flexible deflections (rad)

0.01

0.01

0.02

0.03 0 2 4 6 8 10 12 14 16 18 20

(b)

Time (s)

Fig. 6. Experimental results for IDA-PBC control: (a) Time evolution of the rigid variable q0 and reference qd ; (b) Time evolution of the exible deections.

2509