Gripper gets stuck during calibrate_pr2 sequence (ros ticket #3321) about pr2_robot

[ahendrix]

[sglaser] Another possibility is to pull the elbow all the way up before calibrating the upper arm.

[ahendrix]

[wim] Eitan modified the calibrate_pr2 script to make it warn when the robot gets stuck. Eitan, does this script verify that the power is up before warning?

[ahendrix]

[gerkey] From #3665:
{{{
Working with PRI (yellow) this morning, I saw calibration get stuck twice. The first time I didn't understand what was wrong and simply restarted everything.

The second time, I noticed that the right gripper was jammed up against the base laser enclosure (see attached picture). I pulled up on the right arm, freeing the gripper, and calibration completed successfully.
}}}

[ahendrix]

[watts] I think Stu owns the calibration sequence at this point. I can help test it out on a Beta robot.

[ahendrix]

[wim] Your picture looks almost exactly the same as the one I originally attached. Did you get a warning message about the calibration being stuck? Was that good enough for you to know how to solve the problem?

It is very difficult to come up with a calibration sequence that will always work, plus when you get stuck it is difficult to do something intelligent because you don't know where the arm is (because it is uncalibrated). Trying to get unstuck autonomously could damage the laser.

[ahendrix]

[gerkey] Replying to [comment:5 wim]:

Your picture looks almost exactly the same as the one I originally attached. Did you get a warning message about the calibration being stuck? Was that good enough for you to know how to solve the problem?

Yes, there were messages somewhere (don't remember if it was in rxconsole or in the shell where I roslaunched pr2.launch) telling me that calibration was taking longer than usual. That was certainly helpful, though it was pretty obvious what was wrong once I looked at (and listened to) the robot

It is very difficult to come up with a calibration sequence that will always work, plus when you get stuck it is difficult to do something intelligent because you don't know where the arm is (because it is uncalibrated). Trying to get unstuck autonomously could damage the laser.

That makes sense to me. Unless there's going to be a concerted effort to fix this in the next day or two, we should punt it past the 1.0 release.

[ahendrix]

[sglaser] I've never touched the calibration sequence.

[ahendrix]

[bhaskara] I can reproduce it consistently on pri by bringing up the robot, tucking the arms, then bringing it down and back up. Lifting the torso higher as in the attached patch fixes it in that case at least.

[ahendrix]

[wim] This is not the Vijay calibration ;-)

[ahendrix]

[wim] For c-turtle, we can solve this by making the spine move up way further. This will take longer, but in c-turtle you won't need to calibrate the robot very often because the joint offsets are stored on the MCB's

[ahendrix]

[berger] I like that as a plan

[ahendrix]

[wim] $ svn ci -m "move torso 25 cm up during calibration"
Sending pr2_bringup/scripts/calibrate_pr2.py
Transmitting file data .
Committed revision r35707.