Random movement of tip in IK setup

[johnrowlay]

I have imported the Kinova Mico arm into V-REP 3.5, and built a tip-target setup. The IK group uses pseudo-inverse with maximum of 10 iterations, and precisions of 5e-4 m and 1e-1 deg. All joints are in inverse kinematics mode, with 0.01 proportional gain.

Using the Python remote API, I set the target to a few mm in front of the tip. Then, I step the simulation a few times, until the tip's velocity is below a threshold (i.e. the tip has reached the target). Then, I set the target to a few mm in front of the tip again, and the loop continues in this way.

What I am finding, is that for the majority of the path which the tip moves along, things work well and the movement is smooth. However, every now and then (roughly around once every 20 times the target is moved in front of the tip), the tip suddenly moves erratically / randomly around the target, and never actually settles at the target. The tip appears to hover around the target, darting around it and constantly overshooting, without ever coming that close to the target. The kind of movement is similar to the movement observed when the target is outside the reachable zone of the robot, but I have checked that in these cases, the target is certainly within the reachable zone. I have also checked the Jacobian at these points, and it is full rank, so it is also not due to a singularity.

Each time I set the target to a position slightly in front of the tip, it is only about 5 mm in front of it. Therefore, the pseude inverse method should be fine, without requiring the DLS method.

Is there any clear reason for this eratic behaviour?

[fferri]

How do you check that the jacobian is full rank?
Even if it is full rank, it may be near a singularity. I would check its condition number.