Hi Spencer,
In order to have the inverse kinematic compute in function of the
position only, i modified one of the existing solver (i.e. copy paste of
chainiksolverpos_nr.cpp/.hpp).
Mainly, the condition to see whether the IK converges is checked only on
positions, as shown below.
Cheers, Gianni.

------------------------
int ChainIkSolverPos_NR_Pos::CartToJnt(const JntArray& q_init, const
Frame& p_in, JntArray& q_out)
{
// this solver computes the inverse kinematic for the position
only.
q_out = q_init;

unsigned int i;
for(i=0;i<maxiter;i++){
fksolver.JntToCart(q_out,f);

delta_vel = f.p-p_in.p;
//Before was:delta_twist = diff(f,p_in);

iksolver.CartToJnt(q_out,Twist(delta_vel,Vector::Zero()),delta_q);
Add(q_out,delta_q,q_out);
if(Equal(delta_vel,Vector::Zero(),eps))
break;
//before was:
//if(Equal(delta_twist,Twist::Zero(),eps))
// break;
}
if(i!=maxiter)
return 0;
else
return -3;
}
------------------------

Il giorno gio, 22/03/2012 alle 18.35 -0400, Spencer Jackson ha scritto:
> On Thu, Mar 22, 2012 at 05:59:53AM +0100, Herman Bruyninckx wrote:
> > On Thu, 22 Mar 2012, Spencer Jackson wrote:
> >
> > >Hi! I'm doing some experimentation with KDL, and I'm not getting the
> > >results that I'm expecting. My final arm is going to have 4DOF, but as a
> > >prototype, I knocked something up with only 3 joints, looking like this:
> > >
> > > KDL::Chain chain;
> > > chain.addSegment(Segment(Joint(Joint::RotZ),Frame(Vector(0.0,0.0,0.25))));
> > > chain.addSegment(Segment(Joint(Joint::None),
> > >Frame(Vector(0.1,0.0,0.0))));
> > > chain.addSegment(Segment(Joint(Joint::RotX),Frame(Vector(0.0,0.0,0.10))));
> > > chain.addSegment(Segment(Joint(Joint::None),
> > >Frame(Vector(0.1,0.0,0.0))));
> > > chain.addSegment(Segment(Joint(Joint::RotX),Frame(Vector(0.0,0.0,0.10))));
> > >
> > >After having some issues with the IK, I realized I could test the
> > >forward kinematics. I decided to move the end effector to the tip of the
> > >first segment, canceling out the lengths of the last two joints. To do
> > >this, I rotated the last segment by 180 degrees. That came out like
> > >this:
> > >
> > >[sajack@khezef test2]$ ./kdltest
> > >Enter the position of joint 0: 0
> > >Enter the position of joint 1: 0
> > >Enter the position of joint 2: 3.14159265
> > >[[ 1, 0, 0;
> > > 0, -1, 8.74228e-08;
> > > 0,-8.74228e-08, -1]
> > >[ 0.2, 8.74228e-09, 0.25]]
> > >
> > >Why is the y component nonzero?
> >
> > Are you referring to the numbers like "8.74228e-09"? That's probably
> > because your value of Pi is not 100% exact...
> >
> > >Thanks!
> > >
> > >Spencer Jackson
> >
> > Herman
>
> Hm. That's unfortunate. In which case, it's likely something wrong with
> my inverse kinematics stuff. I'm likely misunderstanding something
> trivial. Because my final arm will only have 4DOF, I was trying to
> discard the extra rotational dimensions. I tried creating an arm with a
> single segment Z axis joint, and realized I couldn't simply use
> the Identity rotational matrix. That arm looks like this:
>
> KDL::Chain chain;
> chain.addSegment(Segment(Joint(Joint::RotZ),Frame(Vector(0.0,0.0,0.25))));
> chain.addSegment(Segment(Joint(Joint::None),Frame(Vector(1, 0.0,
> 0.0))));
>
> FK works just about as well as the previous example:
> [sajack@khezef test2]$ ./kdltest
> Enter the position of joint 0: 3.14159265
> [[ -1, 8.74228e-08, 0;
> -8.74228e-08, -1, 0;
> 0, 0, 1]
> [ -1,-8.74228e-08, 0.25]]
>
> However, trying to achieve IK for (-1, 0, 0.25) doesn't work. I'm using
> the Identity Matrix, but with the ChainIkSolverVel_wdls solver. I've
> set the taskspace matrix like so:
>
> Eigen::MatrixXd matrix_Mx = Eigen::MatrixXd::Identity(6,6);
> matrix_Mx(3,3) = 0; matrix_Mx(4,4) = 0; matrix_Mx(5,5) = 0;
> iksolver1v.setWeightTS( matrix_Mx )
>
> Additionally, I'm using the Weighted Error IK position solver, from
> here: http://ros-users.122217.n3.nabble.com/IK-with-KDL-td1867716.html
>
> ChainIkSolverPos_NR_WE
> iksolver1(chain,fksolver,iksolver1v,100,1e-6);//Maximum 100 iterations,
> stop at accuracy 1e-6
> const double weights[6] = {1, 1, 1, 0, 0, 0};
> iksolver1.setWeights(weights);
>
> The Frame I'm navigating to is:
>
> Frame F_dest = Frame(Vector(-1, 0, .25));
>
> I feel as if this is a great deal more simple than the arm I'm going to
> be working with later, and would like to know what it is that I'm doing
> wrong with it. I'm not really sure what I'm missing... Thanks!
>
>
> Spencer Jackson

    KDL::Chain chain;
    chain.addSegment(Segment(Joint(Joint::RotZ),Frame(Vector(0.0,0.0,0.25))));
    chain.addSegment(Segment(Joint(Joint::None),Frame(Vector(1, 0.0,

    Eigen::MatrixXd matrix_Mx = Eigen::MatrixXd::Identity(6,6);
    matrix_Mx(3,3) = 0; matrix_Mx(4,4) = 0; matrix_Mx(5,5) = 0;
    iksolver1v.setWeightTS( matrix_Mx )

    ChainIkSolverPos_NR_WE

    const double weights[6] = {1, 1, 1, 0, 0, 0};
    iksolver1.setWeights(weights);

    Frame F_dest = Frame(Vector(-1, 0, .25));

Hi Spencer,

You cannot use the identity matrix to ignore rotations like that. If you put the identity matrix you are in fact requesting that the orientation of the end effector is the same as the base frame.

I don't think there's an easy solution for this when using the KDL IK algorithms, since all of them assume you are setting a full position and orientation target pose.

One thing that you could do is add a three additional segments with no length and with the configuration of a spherical joint (e.g. Z-Y-Z) in the end of your chain, and then ignore the part of the solution that involves those degrees of freedom. This way no matter the orientation you ask on the IK request, it will be obtainable with those three last DOF's, and you will be effectively calculating a position only IK for your 4 DOF chain.

If there's a more elegant solution to this I hope someone steps up and corrects me :)

Best regards, Miguel.

On Fri, Mar 23, 2012 at 11:11:10AM -0000, miguel [dot] prada [dot] sarasola [..] ... wrote:
>

On Fri, Mar 23, 2012 at 04:17:32PM -0400, Spencer Jackson wrote:
> On Fri, Mar 23, 2012 at 11:11:10AM -0000, miguel [dot] prada [dot] sarasola [..] ... wrote:
> >

On Fri, Dec 21, 2012 at 3:05 PM, <t [dot] t [dot] g [dot] clephas [..] ...> wrote:
> > Debian packages for Orocos Toolchain 2.6 can be fetched using apt-get
> > from the ros debian package repository: apt-get install
> > ros-fuerte-orocos-toolchain ros-fuerte-rtt-ros-integration
> > ros-fuerte-rtt-ros-comm ros-fuerte-rtt-common-msgs
> >
> > You'll still have to build rtt_geometry and soem from source though
> > but it will take most of the compilation time away.
> rtt_geometry is set to "intentionally missing" on
> http://www.ros.org/debbuild/fuerte.html
> Why is that?

No good reason, I issued a prerelease build, if it succeeds I'll
release rtt_geometry 0.3.0 into Fuerte

>
> Tim
> >
> > Ruben
> > Ruben Smits, Phd
> > Chief Technology Officer
> > Intermodalics BVBA
> > +32479511786
> > www.intermodalics.eu
> >
> >
> > On Thu, Dec 20, 2012 at 9:12 PM, Uwe Fechner wrote:
> >> Hello,
> >>
> >> the installation instructions on:
> >> http://www.ros.org/wiki/orocos_toolchain_ros
> >> section "Manual installation" do not work for Orocos 2.6.
> >>
> >> I updated my own installation instructions:
> >> http://www.kieltech.de/uweswiki/ROS
> >>
> >> but this is ridiculously complicated.
> >>
> >> Is there a way to simplify the installation?
> >>
> >> All we want is Orocos + Ethercat, and installing it takes
> >> hundreds of megabytes and hours of compilation.
> >>
> >> Regards:
> >>
> >> Uwe Fechner
> >> --
> >> Orocos-Users mailing list
> >> Orocos-Users [..] ...
> >> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users
> --
> Orocos-Users mailing list
> Orocos-Users [..] ...
> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users