On Thu, 27 Mar 2008, Hypnos wrote:

> I am trying to make a realtime inverse kinematics system using kdl in python.

Python and realtime....? maybe that doesn't go together too well :-)

> However i am having a hard time trying to understand how it works.
> Unfortunately the examples and api documentation didnt get me much further.
>
> What i am trying to accomplish:
> - Simulate a human arm
> - 3d visualisation of the arm using cones (with panda3d, panda3d.org)
> - using a 3d position (which i am getting trought a wiimote) to inverse
> calculate the arm bones position/rotation
>
> As i understand, kdl should be able to calculate this for me.
>
> What i currently understand (correct me if im wrong):
> - a chain is built up from segments
> - a segment contains a joint, which handles the rotation, and a frame, which
> defines the distance to the next segment.
> - if i need a joint which rotates in multiple directions, i define multiple
> joints with a frame (0,0,0)
Yes.

> What i dont understand:
> - how do i define a point to which the tip of the chain should move to (and
> thus calculate the pos/rot of the bones)

KDL currently only covers _instantaneous_ kinematics, so no trajectories,
but only velocities.

> - how do i let it calculate the inverse kinematics
There are abstract classes for all instantaneous kinematics routines, which
you have to implement for each particular chain.

> - how do i assign the results to the chain
The "state" of a chain is kept separate from the description of the chain.

> - how do i get the position of the bones out of kdl to import it in another
> system (best would be Vec3-rotation, Vec3-position, relative to the previous
> segement, but could be Mat3x3 or Mat4x4 as well)

Herman

Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm

On Thursday March 27 2008 19:15:05 Hypnos wrote:
> Hello,
>
> I am trying to make a realtime inverse kinematics system using kdl in
> python. However i am having a hard time trying to understand how it works.
> Unfortunately the examples and api documentation didnt get me much further.
>
> What i am trying to accomplish:
> - Simulate a human arm
> - 3d visualisation of the arm using cones (with panda3d, panda3d.org)
> - using a 3d position (which i am getting trought a wiimote) to inverse
> calculate the arm bones position/rotation
>
> As i understand, kdl should be able to calculate this for me.
>
> What i currently understand (correct me if im wrong):
> - a chain is built up from segments
> - a segment contains a joint, which handles the rotation,

or translation for prismatic joints

> and a frame,
> which defines the distance to the next segment. - if i need a joint which
> rotates in multiple directions, i define multiple joints with a frame
> (0,0,0)

correct

> What i dont understand:
> - how do i define a point to which the tip of the chain should move to (and
> thus calculate the pos/rot of the bones) - how do i let it calculate the
> inverse kinematics

KDL has some solvers for generic chains,

Forward position kinematics can be solved using generic recursive solvers:

chainfksolverpos_recursive:
<http://people.mech.kuleuven.be/~psoetens/orocos/devel/kdl/latest/api/html/classKDL_1_1ChainFkSolverPos__recursive.html>
chainfksolvervel_recursive:
<http://people.mech.kuleuven.be/~psoetens/orocos/devel/kdl/latest/api/html/classKDL_1_1ChainFkSolverVel__recursive.html>

for inverse velocity kinematics we have a generic solver:

chainiksolvervel_pinv:
<http://people.mech.kuleuven.be/~psoetens/orocos/devel/kdl/latest/api/html/classKDL_1_1ChainIkSolverVel__pinv.html>
or chainiksolvervel_pinv_boost_givens:
<http://people.mech.kuleuven.be/~psoetens/orocos/devel/kdl/latest/api/html/classKDL_1_1ChainIkSolverVel__pinv__boost__givens.html>

these solvers use the pseudo-inverse of the jacobian to calculate inverse
kinematics. Expect problems near kinematic singularities.

For inverse position kinematics we have a generic solver:

chainiksolverpos_nr:
<http://people.mech.kuleuven.be/~psoetens/orocos/devel/kdl/latest/api/html/classKDL_1_1ChainIkSolverPos__NR.html>

this solver uses a Newton-Rhapson methodi, an inverse velocity solver and a
forward position solver. Expect problems with local minima and near kinematic
singularities.

> - how do i assign the results to the chain

Like Herman said, the chain is stateless, you have keep the results youself

> - how do i get the position of the bones out of kdl to import it in another
> system (best would be Vec3-rotation, Vec3-position, relative to the
> previous segement, but could be Mat3x3 or Mat4x4 as well)

The intermediate bone positions can be calculated using the
chainfksolverpos_recursive and the result is stored in a KDL::Frame (which
contains a 3x3 rotation matrix and a 3x1 position vector), you can treat it
as a Mat4x4 using the operator(i,j).

see:
http://people.mech.kuleuven.be/~psoetens/orocos/devel/kdl/latest/api/html/classKDL_1_1Frame.html

Ruben

Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm