Appendix 5

Numerical computation of the base

parameters

A5.1. Introduction

The base parameters constitute the minimum set of parameters that characterize
completely a given system. They also represent the identifiable parameters of the system.
They are obtained from the standard parameters by eliminating those that have no effect on
the model and by grouping some others in linear combinations. The determination of the base
inertial parameters has been carried out in Chapters 9 and 10 by the use of straightforward
symbolic methods for serial and tree structured robots. However^ the symbolic approach
cannot give all the base parameters for robots containing closed loops. ITiis problem can be
solved by the use of the numerical method presented in this appendix. In addition, the
numerical n^thod can also be applied to determine the base parameters for the geometric
calibration of robots.
The symbolic approach of computing the base paran^ters is based on determining the
independent elements of the energy functions represented by the row vector h (equation
(9.41]), or by determining the independent columns of the D matrix of the dynamic model
(equation [9.36]). Numerically this problem is equivalent to the study of the space span by
the columns of a matrix W fonmd from h (or D) using r random values of q, q (or q, q, q).
This study can be carried out using the singular value decomposition (SVD) or the QR
decomposition of W [Oautier 91]. In this appendix^ we develop the numerical method that is
based on the QR decomposition of a matrix W, which is derived from the energy functions.
Both cases of tree structured robots and closed loop robots are treated.
418 Modeling* identification and control of robots

A5.2. Base inertial parameters of serial and tree structured robots

The total energy of the system H is linear in terms of the standard inertial parameters. It is
given by the following equation:

H = hK [A5.1]

where:
- K represents the (1 Inxl) vector of the standard inertial parameters of the links and of
the rotors of actuators;
- h(q, q) is the (1x1 In) row vector composed of the energy functions;
- q and q are the (nx 1) joint position and joint velocity vectors respectively.

To determine the base parameters, we construct a matrix W by calculating the energy row
h for r random values of joint positions and velocities such that:

•h(l)-
h(2)
W = [A5.21

Lh(r)J

with h(i) = h[q(iX q(i)). i = 1 r. and r » 1 In.

An inertial parameter has no effect on the dynamic model if the elements of its
corresponding column in W have the same value, i.e. its function in h is constant and
independent of q(i), q(i). By eliminating such parameters and the corresponding columns, the
matrix W is reduced to c columns and r rows.
We note that:
- the number of the base inertial parameters b is equal to the rank of W;
- the base parameters are those corresponding to b independent columns of W;
- the grouping equations are obtained by calculating the relationship between the
independent columns and the dependent columns of W.

The application of the foregoing statements can be achieved by the use of the QR
decomposition of W (§ A4.4), which is given by:

QTW=L -[" 1 [A5.3]

L%-c)xcJ
 Numerical computation of the base parameters 419

where Q Is an (rxr) orthogonal matrix, R is a (cxc) upper-triangular matrix, wd Ojxj is the

(ixj) matrix of zeros.

Theoretically, the non-identifiable parameters are those whose corresponding elements on

the diagonal of the matrix R are zero (Forsythe 77], [Golub 83]. Let x be the numerical zero:

t = r e max(|Rii|) [A5.4]

where |Rii| is the absolute value of Ri|, and e is the computer precision.

Thus, if iRjil < t. then the i*** parameter is not identifiable. On the contrary, if |Rii| > t.
then the corresponding column in W is in^pendent and constitutes a base of the space span
by W. Let the b independent columns be collected in die matrix Wl. and the corresponding
parameters be c^ll^t^ in the vector KL The other columns and parameters are represented
by W2 and K2 respectively, such that:

WK=[W1 W2][^] fA5.5}

The matrix W2 can be written in terms of Wl as follows:

W2 = Wl P [A5.6)

Consequently:

W K = [ Wl W2 ] ([".'J-WK,
U Wl K B IA5.7]

where the base parameter vector KB is given by:

KB = K l ^ - P K l tA5.8]

Thus, the matrix P allows us to obtain the grouping equations of the parameters K2 with
Kl. In order to determine P. we compute the QR decomposition of the matrix [ Wl W2 ],
which is written as:

r Rl R2 1
[ Wl W2 ] = [ Ql Q2 ]l ^ U [ Q l R l Q1R2 ] (A5.9]
L''(r.b)xb v(r.b)x(c.b) J

where Rl is an (rxr) regular upper-trianplar matrix, and R2 is a (bx(c-b)) matrix.

420 Modeling, identification and control of robots

From equation IA5.9], we obtain:

Ql = WlRl-l (A5.10)
W2 = Q1R2 = WlRr*R2 [A5.11]

and finally, using equation [A5.6]:

P = R1'^R2 [A5.12]

A5.3. Base inertial parameters of closed loop robots

The geometric description of closed loop robots is given in Chapter 7. The joint position
vector of the equivalent tree structure is given by:

•a
qar = [A5.13]

where q^ is the (Nxl) vector of the active joint variables, and qp is the ((n - N)xl) vector of
the passive joint variables.

The energy functions of the inertial parameters of the closed loop structure are the same
as those of the equivalent tree structure. This means that we can apply the algorithm of the
tree structured robots to the closed loop structures with the difference that the matrix W is
calculated using random values for the independent active variables q^Ci) and qa(i) for
i s 1,..., r. The corresponding passive variables qp and qp are evaluated from the constraint
equations of the loops.

A5.4. Generality of the numerical method

The numerical method can be used for the determination of the minimum parameters of
other applications such as:
- determination of the identifiable parameters for the geomeuic calibration of the
parameters (Chapter 11) [Khalil 91a];
- calculation of the minimum parameters of flexible structures [Pham 91a].

This numerical method is easy to implement, thanks to a software package such as

SYMOROf- [Khalil 97] for the automatic computation of the symbolic expressions of the
energy functions (to determine the elements of h) and thanks to scientific software packages
of matrix computation such as Matlab and Mathematica.