IKFast solution for IRB2400 misses 'best' solution near all-zero position

[GoogleCodeExporter]

The IKFast solution for the ABB IRB_2400 robot doesn't give the expected value 
at the "all zeros" position.  IK should return an all-zero vector, but instead 
returns an alternative solution.  Note that FK works correctly, but a 
round-trip FK > IK does not return the original all-zeros vector.

To reproduce graphically:
  1) Start an rviz planning session for the IRB2400
        - either arm_nav or moveit works, but moveit is easier to control.
  2) Note that the robot starts in the "all-zeros" configuration
  3) Select "Interact" mode, and barely nudge the end-effector
  4) Observe that the arm "flips" to a radically different config
  5) Moving the EEF around can restore the "normal" orientation, but the "flipping" behavior is very reproducible near the all-zeros position.

To reproduce numerically:
  1) Compile the IKFast test program as described on the wiki page (http://ros.org/wiki/Industrial/Tutorials/Create_a_Fast_IK_Solution#Testing_output_from_IKFast).
  2) Run FK
       ./ikfastdemo fk 0 0 0 0 0 0
  3) Observe the correct end-pose (0.94,0,1.455  /  1,0,0,0)
  4) Run IK
       ./ikfastdemo ik 0.9 0 1.455 1 0 0 0 
  5) Review the list of solutions

I will forward the IRB2400 DAE to the IKFast developers, to see if they have 
any suggestions.

Original issue reported on code.google.com by jz...@swri.org on 20 Mar 2013 at 9:30