Re: KinematicsInterface

On Wed, 14 Apr 2004, Peter Soetens wrote:

> On Friday 09 April 2004 15:12, Martin LÃŒmkemann wrote:
> > We are currently implementing the kinematics-inteface for a Mitsubishi
> > PA-10 which is a 7R(zxzxzxz)-arm.
> >
> > The KinematicsInterface only allows a Double6D for the joint-variables.
> >
> > For now we omit the last joint in our calculations, but wouldn't it be
> > necessary to parametrise the interface on the number of joints?
> >
> > e.g.: template<unsigned NumberOfJoints> class KinematicsInterface{...}
> 
> Yes, this interface is limited to 6 joints, which is indeed not general 
> enough. Your suggestion is not bad, but others give an extra parameter 
> denoting the number of joints. Your suggestion is type safe, the other allows 
> the same algoritm to do calculations for a variable number of joints (eg when 
> the tool has an extra degree of freedom).
> 
> Thanks to multiple inheritance, we can combine both. 

I also prefer the virtual interface + inheritance approach.

[...]
> class KinematicsNAxisInterface { 
> // pure virtuals with axis nr. 
> virtual bool positionForward( unsigned nbOfJoints, double[] q,
> 	ORO_Geometry::Frame& mp_base, Singularity& s ) const =0;

But I do not like the number of joints to be part of the interface,
nor a "double[]" as argument; I prefer an interface that can be
strongly type-checked and that has no redundant information for which
programmers can easily make mistakes.

Or did I misread your piece of code?

Herman

PS I have adapted the API design document on the Orocos website, to
   reflect this issue:
     <http://www.orocos.org/documentation/deep-shallow-api.html>.
   Please, let me know your comments!

-- 
  K.U.Leuven, Mechanical Eng.,  Mechatronics & Robotics Research Group
    <http://people.mech.kuleuven.ac.be/~bruyninc> Tel: +32 16 322480