Professional Documents
Culture Documents
Calibration: Optimal
Excitation Trajectories
and Experimental
Parameter Estimation
䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇 䢇
which can be determined by applying the reduction of possible trajectories, thus reducing the original
procedure proposed by Gautier and Khalil.4 The infinite-dimensional problem to a more tractable
preliminary aspects related to manipulator model- finitely parameterized optimization problem. This
ing for dynamic calibration are recalled in Section 2. is, of course, done at the expense that the solution
Estimation of the identifiable parameters is usu- will be optimal in the selected class, i.e., suboptimal
ally achieved by application of a least-squares algo- with respect to the original problem. However, what
rithm to the motion or the energy equations of the is of interest in practice is to determine a ‘‘good’’
robot. The validity of the provided estimate strongly Žalbeit not globally optimal. excitation trajectory that
depends on the quality of the available input᎐out- leads to an informative identification experiment. In
put signals and on the choice of the imposed refer- this respect, the above drawback does not seem to
ence trajectories, which must properly ‘‘excite’’ the be critical in an experimental or industrial context,
robot dynamics. The quality of the used input᎐out- as long as the family of allowable trajectories is
put signals can be enhanced by using filters andror chosen to be ‘‘rich enough,’’ according to engineer-
observers Žsee, e.g., refs. 5 and 29.. ing judgment and experimental trials. In this article,
The problem of finding exciting trajectories for in particular, we consider the class of harmonic
the identification has been discussed in several pa- functions with a fixed, finite number of harmonic
pers. The work presented in Armstrong7 is among terms. The introduced parameterization takes into
the oldest and it proposed a functional approach, in account the frequency band constraints on the input
which the calculus of variations is used to solve a excitation, and the number of decision parameters
nonlinear path optimization problem: An optimal of the resulting optimization problem is indepen-
acceleration sequence is determined, and then ve- dent of the number of time samples along the trajec-
locities and positions are computed by numerical tory. Harmonic parameterization also was used in
integration. The weak points of such an approach this context by Swevers et al.11
are the difficulty of including trajectory feasibility The trajectory parameterization reduces the
constraints in the optimization procedure and high problem to a nonlinear parametric optimization that
computational burden due to the large number of is solved here using a genetic algorithm ŽGA.. The
decision variables required when the method is convergence of the GAs to the global optimum is
discretized. not, in general, guaranteed, but in practice such
The methods proposed in refs. 8 and 9 are both algorithms have proved to be particularly efficient
based on the determination of a sequence of optimal and robust in the global search of solutions of non-
joint position᎐velocity couples. A continuous linear and multimodal optimization problems.12,13
smooth ‘‘optimal’’ trajectory is then found by inter- In robotics, genetic algorithms have been used, for
polating them via fixed order polynomials; trajec- instance, for motion planning of mobile manipula-
tory constraints are only checked at the end. The tors14 and for estimation of kinematic parameters.15
number of decision variables and, consequently, the In our case, the use of GAs in conjunction with the
computational effort in this case are dependent on introduced trajectory parameterization allows us to
the number of samples along the trajectory: Gautier take into account the physical constraints on joint
and Khalil9 reported numerical intractability of the positions, velocities, and accelerations directly in
optimization problem in their test cases for a num- the decoding phase of the genotype of the individu-
ber of samples greater than 50. als, i.e., of the strings containing the binary codes of
An alternative solution to the above-described the decision variables, so that the resulting trajecto-
methods, which are gradient based Žand thus re- ries are certainly feasible.
quire time-consuming computations to exactly de- In the experimental part of this article, the com-
termine or estimate the gradient of the objective plete calibration procedure is performed on a
function., is proposed in ref. 10, where a random SCARA two-link manipulator produced by IMI ŽIn-
search approach is used to determine the set of tegrated Motions Inc., Berkeley, CA., starting from
optimal points to be interpolated. trajectories generation according to the proposed
The method proposed in this article is based on method, through issues related to data acquisition
the parameterization of the class of joint position and filtering, to the estimation of the inertial and
time functions. The key idea underlying this method friction parameters of the manipulator and model
is to restrict attention to some parameterized family validation.
Calafiore, Indri, and Bona: Robotic Dynamic Calibration 䢇
57
observers. at n m time instants Žwith n m n ) n p . dur- and the corresponding error covariance matrix is
ing the execution of a task, thus obtaining the equa- given by
tion
y1
P s Ž Ry1 T y1 .
qH R v H . Ž8.
y s H q v Ž 6a.
The above formulas refer to the batch solution of
with y [ w Ž t 1 . ⭈⭈⭈ Ž t n m .x T and the least-squares ŽLS. problem for the accumulated
measurements y. In the experimental tests, the ac-
˙ Ž t1 . , q
D Žq Ž t1 . , q ¨ Ž t1 .. tual estimate will be computed using the following
.. recursive formulation of the LS algorithm Žsee ref.
H[ . , Ž 6b . 18.. Let H i g ᑬ 1, n p and yi g ᑬ represent the ith row
˙ Ž tn m . , q
D ŽqŽ tn m . , q ¨ Ž tn m . . of the regression matrix and the ith measurement,
respectively. Let ˆ i and P , i be defined as the esti-
mate of the parameter vector , given the measure-
where v is assumed to be the zero mean measure- ments up to i, and the relative covariance, respec-
ment noise vector, uncorrelated from , and H is tively. The estimate can then be recursively updated
the data matrix, assumed to be deterministic Žthe using the recursions
effects of uncertainty on the data matrix H are
discussed in ref. 7.. The base parameter vector can
ˆ iq1 s ˆ i q K iq1 Ž yiq1 y H iq1ˆ i . , ˆ 0 s ,
be estimated from the regression equation Ž6a. us-
ing a least-squares algorithm as discussed in the y1
K iq1 s P , i H Tiq1Ž R v q H iq1 P , i H Tiq1 . ,
following subsection.
We note that the energy-based calibration ap- P , iq1 s P , i y K iq1 H iq1 P , i , P , 0 s R .
proach3,17 also leads to a standard least-squares
identification problem, which may be cast in the We recall that the solution obtained from the recur-
form Ž6a.. Therefore, all the subsequent develop- sive algorithm for i s n m coincides with the batch
ments can be applied equally to the energy-based solution, i.e., ˆ
n m s ˆ
; P , n m s P . The recursive ver-
calibration method. sion is, therefore, a computationally efficient way to
solve Ž7., especially when a large number of mea-
surements is to be processed.
2.1. Least-Squares Parameter Estimation
If a priori knowledge is neglected, then R s ⬁,
We recall the regression equation Ž6a., and it follows from Ž8. that the ‘‘size’’ of the estima-
tion error depends on the normal equations matrix
y s H q v, ⌸ s H T Ry1v H. A frequently used scalar measurer
on ⌸ is given by Jd s log detŽ ⌸ .. Maximization of
where y g ᑬ n m is the vector of measurements, g Jd with respect to the input excitation is referred to
ᑬ n p is the parameter vector to be identified, H g as D optimality,19 and it is desirable in order to get
ᑬ n m , n p is the regression matrix Žassumed to be de- estimates with minimal uncertainty bounds. An-
terministic., and v g ᑬ n m is a zero mean random other significant index is derived from the condition
vector representing measurement errors. Vector v is number of ⌸: Jk s condŽ ⌸ .. A small Jk is desirable
assumed to be uncorrelated from , and the auto- to minimize the bias of the estimate due to unmod-
correlation matrix R v ) 0 of v is assumed to be eled dynamics errors Žwhich are correlated with the
known. A priori information about the unknown regressors., to enhance the convergence rate of the
parameter vector is also included in the problem via algorithm, and to minimize the sensitivity of the
the mean value of Žfor instance, may be taken solution to variations of the measurements vector y
as the vector of nominal parameters. and the ‘‘confi- Žsee refs. 19 and 20..
dence’’ on the nominal parameters, expressed by a
given covariance matrix R ) 0. It is then well
known18 that the linear least-mean-squares estima- 3. TRAJECTORY OPTIMIZATION
tor of , given y, is
The previously discussed optimality indices depend
y1 on the kinematic structure of the manipulator, which
ˆ s q ŽRy1 T y1 .
qH R v H H T Ry1
v
Ž y y H . , Ž 7 . is fixed for a given type of manipulator, and on the
Calafiore, Indri, and Bona: Robotic Dynamic Calibration 䢇
59
time history of the joint reference positions, veloci- 3.1. Parameterization of the Joints Trajectories
ties, and accelerations. Denoting with Q a class of
Though the method outlined in this article is con-
continuous functions of the time variable t g ᑬ,
ceptually valid for any parametric family Q Žp., we
defined over the interval T s w t min , t max x , the prob-
treat in detail the class of harmonic functions with a
lem of determining the most ‘‘exciting’’ reference
predefined and fixed number W of harmonic terms.
trajectory q*Ž t . with respect to the generic index J
This class of trajectories was also suggested in ref.
can be formulated as
21 and, independently, in ref. 11.
q* Ž t . s arg min J Ž q Ž t .. Ž 9a . Given W Žtypically a small integer, W s 3, 4, . . . .,
the reference trajectory for the jth joint is assumed
subject to to be
W
qi Ž t . g Q ᭙i, Ž 9b .
qj Ž t . s Ý a j, k sin Ž j, k t . , Ž 10 .
< qi Ž t . < F qi , max ᭙i, t, Ž 9c . ks1
The optimization problem Ž9. specializes now to The constraints Ž12c. and Ž12d. are directly taken
into account in the decoding phase of the genotype
minimize J Ž q Ž t .. Ž 12a. X. Prior to evaluation of the fitness function, each
a j, k , j, k ; ᭙ j, k
element of X is decoded to its real value in the
manner
subject, ᭙ j, k, to the constraints
w A j, k x 1
qj Ž t . g Q , a j, k s 10
y ᭙ j, k ,
q j, max
Ž 12b .
ž 2 n Žj,a.k
y1 2 / 2 a j, max Ž 13.
the second link. The vector p of the identifiable model is expressed as in Ž5., where
inertial parameters is then given by
D 1, 1 s q¨1 ,
p s ⌫1 z q m 2 l 12 m 2 s2 x m 2 s2 y ⌫2 z
T
. D 1, 2 s l 1Ž c 2 Ž 2 q¨1 q q¨2 . y s2 q˙2 Ž 2 q˙1 q q˙2 .. ,
D 1, 3 s l 1Ž ys2 Ž 2 q¨1 q q¨2 . y c 2 q˙2 Ž 2 q˙1 q q˙2 .. ,
Due to the small number of identifiable parame-
D 1, 4 s q¨1 q q¨2 ,
ters for our 2 dof robot, no further reduction of the
parameter vector is needed, while an approximate, D 1, 5 s sgn Ž q˙1 . ,
simplified model Žwhich neglects, if possible, the
effects of some of the identifiable parameters. could D 1, 6 s q˙1 ,
be usefully employed for greater manipulators, as D 1, 7 s D 1, 8 s 0,
suggested in ref. 22. Ž 16 .
D 2, 1 s 0,
In the considered case, only Coulomb and vis-
cous friction are modeled, under the assumption D 2, 2 s l 1Ž c 2 q¨1 q s2 q˙12 . ,
that other components Že.g., hysteresis effects
andror stiction. can be considered as negligible. D 2, 3 s l 1Ž ys2 q¨1 q c 2 q˙12 . ,
The friction torque acting on each joint is then D 2, 4 s q¨1 q q¨2 ,
described as
D 2, 5 s D 2, 6 s 0,
f , i Ž q˙i . s s, i sgn Ž q˙i . q v , i q˙i , Ž 15 . D 2, 7 s sgn Ž q˙2 . ,
D 2, 8 s q̇2
where the parameters s, i and v, i represent
Coulomb dynamic friction and the viscous friction with si [ sin qi and c i [ cos qi .
coefficient, respectively. Note that this model can
correctly describe friction only at ‘‘sufficiently high’’
4.2. Robot Control and Physical Constraints
velocities; otherwise, a specified model for stiction
must be introduced. According to Ž15., the friction The control of the two-link manipulator is per-
parameters of the robot can be collected in a vector formed by an independent joint feedback control
f g ᑬ 4 , which is defined as f [ w s, 1 , v, 1 , s, 2 , law, which is implemented by inserting an observer
v, 2 x T. The complete parameter vector is then for the estimation of the joint velocities, because
given by [ w Tp Tf x T g ᑬ 8. Following the New- only position measurements are available. Unsatis-
ton᎐Euler approach, the planar robot dynamic factory results have been obtained by simply deriv-
62 䢇
Journal of Robotic Systems—2001
ing the joint velocities at discrete time from the 4.3. Data Acquisition and Processing
position measurements. The high-gain observer de-
The main problems to be solved for the acquisition
veloped in ref. 23 were chosen after a comparison
of the data that are required for the calibration
among different solutions proposed in literature
procedures can be summarized as follows:
Že.g., refs. 24᎐27. because it seems to offer, in our
case, a good compromise in terms of tracking error 1. The need to guarantee a good quality of the
and quality of the reconstructed signals Žwith refer- measured or observed signals to be used for
ence in particular to velocities, which must be used the parameter estimation.
both for the control and for the parameter identifica- 2. Computation of the signals not directly avail-
tion procedure.. The implemented controller᎐ob- able: joint accelerations and command
server scheme is then described by torques Žbecause they are not directly mea-
surable in our case..
1 3. Characterization of the torque measurement
˙ˆx 1 s ˆx 2 q H p Žq y ˆx 1 . , Ž 17a. errors to be used in the estimation algorithm.
1 The joint velocities reconstructed by the observer
˙x 2 s
ˆ H v Žq y ˆ
x1 . , Ž 17b . wŽ 17a. and Ž17b.x and directly used for the robot
2
control need to be filtered before the computation of
s K p Žq d y q. q K d Žq
˙ d y ˆx 2 . , Ž 17c. the acceleration signal and the parameter estima-
tion. This filtering should be performed off-line by
means of a noncausal zero phase shift filter to avoid
where the observer state variables ˆ x 1 and ˆ
x 2 , respec- distortion in the dynamic regressor, as discussed in
tively, denote the dynamic estimates of the joint ref. 6. However, due to our hardware limitations
positions and velocities, q is the vector of the joint and conflicting interrupt problems during the mo-
positions measured by the resolvers incorporated in tion of the robot, it was not possible for us to
the motors, q d Ž t . and q ˙ d Ž t . are the assigned refer- acquire samples at the control rate c and then filter
ence position and velocity vectors, obtained by the and decimate the data off-line. Therefore, we pre-
optimization procedure, and is the vector of the ferred to filter the velocity signal on-line using the
applied torques in Ž5.. The observer gain matrices following discrete time filter ŽT s 2r c s 1 ms.
H p s diagŽ h p, i . and H v s diagŽ h v, i . are chosen so that has cutoff frequency f , 14 radrs,
that the spectra of the characteristic polynomials
pi Ž . s 2 q h p, i q h v, i , i s 1, 2, are in the open left ˙ fil Ž kT . s ␣1 ˆx 2 Ž kT . q ␣ 2 q
q ˙ fil ŽŽ k y 1. T .
half-plane, whereas is a small positive parameter.
The controller gain matrices K p and K d are diago- q ␣3 q
˙ fil ŽŽ k y 2. T . ,
nal, positive definite matrices that are selected to
satisfy the desired closed-loop specifications. In the with ␣ 2 s 1.95550858, ␣ 3 s y0.95599748, and ␣1 s 1
tests carried out, the controller᎐observer scheme is y ␣ 2 y ␣ 3 . Note that this filter introduces negligible
designed at continuous time to obtain a closed-loop phase shift for the reference signals used, which
system with a bandwidth of about 60 radrs and a have frequency content below ma x . The joint accel-
maximum peakF 4 dB, and then implemented at erations are then computed on-line at the control
discrete time with sampling time T s 1 ms Žsample rate c from the filtered velocity by means of a
rate c s 2rT , 6280 radrs.. The selected observer central difference algorithm, as discussed in refs. 6
gains are H p s diagŽ7, 6.2., H v s diagŽ2, 4., and s and 28.
0.002. The torques are reconstructed by their expres-
The reference joint trajectories are assumed to sion Ž17c. and then filtered on-line at the control
belong to the class Q of harmonic functions defined rate, using the same discrete time filter employed
in Ž10., with the physical bounds Ž9c. ᎐ Ž9e. on the for the velocities. The filtered torque vector is then
angular positions, velocities, and accelerations given given by
by qma x s 1.5 rad, vmax s 8 radrs, and ␣ max s 5
radrs 2 for both joints. From Ž12d. it follows that the fil Ž kT . s ␣1 Ž kT . q ␣ 2 fil ŽŽ k y 1 . T .
frequency content of the reference signals is always
below ma x , 1.825 radrs for all joints. q ␣ 3 fil ŽŽ k y 2 . T . .
Calafiore, Indri, and Bona: Robotic Dynamic Calibration 䢇
63
For each reference trajectory, the following proce- GA as a n b s 4 n v s 48 digits binary string X. No-
dure is used to estimate the mean-square error tice that the number n v of decision variables is
Žsample variance. of the torque measurements. The independent of the number of time samples n m
given trajectory is repeated by the manipulator for along the trajectory, while it is proportional to the
M s 20 times and n m data samples are collected considered number W of harmonic terms. The length
over each repetition of the trajectory. The sample of the genotypic string X is proportional to n v and
variance of the torque measurements for the ith to the selected resolution Žnumber of binary digits.
joint is computed as to which the variables are encoded. The fitness
function is, in this case, given by yJk or by Jd ,
nm M because the GA is traditionally a maximization al-
1 2
i 2 s Ý Ý Ž i j Ž k . y i Ž k . . , gorithm.
n m Ž M y 1. ks1 js1
ˆ s w 3.445; 0.698; 0.012; 0.107; 0.500; 7.884; Figure 8. Torque reconstruction residuals for the valida-
tion trajectory: first Žtop. and second joints Žbottom.; six
1.023; 0.883x
T
data batches.
66 䢇
Journal of Robotic Systems—2001
nificance, since the relative variance is high com- 5. B. Bona and M. Indri, Experiments on the dynamic
pared to the nominal value of the parameter. calibration of a planar manipulator, Proc 27th Int
Symp on Industrial Robots, Milan, Italy, 1996, pp.
Finally, it must be stressed that the parameter 117᎐122.
estimates obtained using standard LS identification 6. M. Gautier, A comparison of filtered models for dy-
may sometimes lead to physically unfeasible nu- namic identification of robots, Proc 35th CDC, 1996,
merical values of the parameters Ži.e., negative mo- pp. 875᎐880.
ments of inertia, etc.. because no specific constraints 7. B. Armstrong, On finding exciting trajectories for
are imposed in the optimization problem. Future identification experiments involving systems with
non-linear dynamics, Proc IEEE Int Conf on Robotics
work will be devoted to study the effect of the and Automation, 1987, pp. 1131᎐1139.
inclusion of physical constraints on the torque re- 8. F. Caccavale and P. Chiacchio, Identification of dy-
construction residuals. namic parameters and feedforward control for a con-
ventional industrial manipulator, Control Eng Prac-
tice 2 Ž1994., 1039᎐1050.
9. M. Gautier and W. Khalil, Exciting trajectories for the
6. CONCLUSIONS identification of base inertial parameters of robots, Int
J Robotics Res 11 Ž1992., 362᎐375.
In this article, an efficient method for the determina- 10. G.W. Van der Linden and A.J.J. Van der Weiden,
Practical rigid body parameter estimation, Proc 4th
tion of ‘‘exciting’’ trajectories for robot inertial pa- IFAC Symp on Robot Control, 1994, pp. 631᎐636.
rameter estimation has been presented. The method 11. J. Swevers, C. Ganseman, D.B. Tukel, J. de Schutter,
is based on harmonic parameterization of the joint H. Van Brussel, Optimal robot excitation and identifi-
time functions and minimization of different suit- cation, IEEE Trans Robotics Automation 13 Ž1997.,
able measures on the LS normal equation matrix 730᎐740.
performed by means of a binary coded genetic algo- 12. D.E. Goldberg, Genetic algorithms in search, opti-
mization, and machine learning, Addison-Wesley,
rithm. The proposed framework may be applied to Reading, MA, 1989.
both Newton᎐Euler and energy-based parameter 13. Z. Michalewicz, Genetic algorithmsq data structures
estimation models. s evolution programs, Springer-Verlag, New York,
Numerical simulations were carried out using 1992.
the Newton᎐Euler model and showed that good 14. M.W. Chen and A.M.S. Zalzala, Dynamic modelling
trajectories can be found using low parameter reso- and genetic-based trajectory generation for non-holo-
nomic mobile manipulators, Control Eng Practice 5
lution and a small number of harmonic terms. The Ž1997., 39᎐48.
computed trajectories respect the frequency band- 15. H. Zhuang, J. Wu, and W. Huang, Optimal planning
width constraints and the physical limits on joint of robot calibration experiments by genetic algo-
position, velocity, and acceleration. rithms, J Robotic Syst 14 Ž1997., 741᎐752.
The results of the experimental identification of ´
16. B. Armstrong-Helouvry, P.E. Dupont, and C. Canudas
the inertial and friction parameters of a SCARA de Wit, A survey of models, analysis tools and com-
pensation methods for the control of machines with
two-link manipulator, using the optimal input exci- friction, Automatica 30 Ž1994., 1083᎐1138.
tations, also were presented and discussed. 17. S.-Y. Sheu and M.W. Walker, Basis sets for manipula-
tor inertial parameters, Proc IEEE Int Conf on Robotics
and Automation, 1989, pp. 1517᎐1522.
18. T. Kailath, Lectures on Wiener and Kalman filtering,
REFERENCES Springer-Verlag, New York, 1981.
19. L. Ljung, System identification, theory for the user,
1. C.G. Atkeson, C.H. An, and J.M. Hollerbach, Estima- Prentice-Hall, Englewood Cliffs, NJ, 1987.
tion of inertial parameters of manipulator loads and 20. G.H. Golub and C.F. Van Loan, Matrix computations,
links, Int J Robotics Res 5 Ž1986., 101᎐119. The Johns Hopkins University Press, Baltimore, 1989.
2. C.P. Neuman and P.K. Khosla, Identification of robot 21. G. Calafiore and M. Indri, Experiment design for
dynamics: An application of recursive estimation, Proc robot dynamic calibration, Proc 1998 IEEE Int Conf on
4th Yale Workshop on Applications of Adaptive Sys- Robotics and Automation, 1998, pp. 3303᎐3309.
tem Theory, Yale University, 1985, pp. 42᎐49. 22. A. Visioli and G. Legnani, Experiments on model
3. S.-Y. Sheu and M.W. Walker, Estimating the essential identification and control of an industrial robot ma-
parameter space of the robot manipulator dynamics, nipulator, Proc 6th IFAC Symp on Robot Control
Proc 28th Conf on Decision and Control, 1989, pp. ŽSYROCO’00., 2000, Vol. 1, pp. 283᎐288.
2135᎐2140. 23. S. Nicosia, A. Tornambe, ` and P. Valigi, Experimental
4. M. Gautier and W. Khalil, Direct calculation of mini- results in state estimation of industrial robots, Proc
mum set of inertial parameters of serial robots, IEEE 29th IEEE Conf on Decision and Control 1990, pp.
Trans Robotics Automation 6 Ž1990., 368᎐373. 360᎐365.
68 䢇
Journal of Robotic Systems—2001
24. H. Berghuis and H. Nijmeijer, Robust control of robots point-to-point and trajectory controllers for robot ma-
via linear estimated state feedback, IEEE Trans Au- nipulators, Control Eng Practice 4 Ž1996., 991᎐1000.
tomat Control 39 Ž1994., 2159᎐2162. 28. P.R. Belanger, X. Meng, and P. Dobrovolny, Estima-
25. H. Berghuis and H. Nijmeijer, Global regulation of tion of angular velocity and acceleration from
robots using only position measurements, Syst Con- equally-spaced shaft encoder measurements, IEEE Int
trol Lett 21 Ž1993., 289᎐293. Conf on Robotics and Automation, Nice, 1992.
26. R. Kelly, A simple set-point robot controller by using
29. B. Bona and M. Indri, Identification of the dynamic
only position measurements, Proc 12th IFAC World
parameters of a planar manipulator: Simulations and
Congress, 1993, Vol. 6, pp. 173᎐176.
27. M.F. Khelfi, M. Zasadzinski, H. Rafaralahy, E. Richard, experiments, Proc 2nd World Automation Congress
ŽWAC 96., Montpellier, France, 1996.
and M. Darouach, Reduced-order observer-based