difference with KDL or trac_ik

[huiwenzhang]

Hi,
The method looks like very similar to KDL's ik solution, which use the Jacobian to compute a delta q, then iterates to the goal configuration. trac_ik improves the solution quality by handling the joint limts with SQP. This library seems to use the same idea. Any difference there?

[stephane-caron]

I'm not familiar with KDL. Out of curiosity, I searched for its documentation (finding this link), but it doesn't explain the IK algorithm they chose. Next, I went to the code and found the following source files:

• chainiksolvervel_pinv: here pinv likely stands for pseudo-inverse, which is equivalent in terms of solution to solving an unconstrained QP, but less efficient numerically (as one ends up computing the inverse of a matrix rather than solving a linear system).
• chainiksolvervel_wdls: looking at CartToJnt there they start with an SVD (costly but can be useful later on, e.g. if one wants to avoid low singular values) then solve for a single task.

From this, I would guess Pink has the following additional features compared to KDL:

• Multiple tasks
• Joint limits
• End-effector (i.e. Cartesian) limits, for instance box limits
• Collision avoidance with spheres or meshes

You can find examples of each of these features in the examples directory.

What is your experience with KDL? Feel free to correct me if I'm wrong after this quick examination of that library. Do you see some features that KDL has and that Pink doesn't?