Forearm roll and Wrist flex axes are "backwards" (ros ticket #3123)

[ahendrix]

Refer to: stacks/pr2_common/pr2_defs/defs/arm/arm_defs.urdf.xacro

''forearm roll joint'' {{{

...

}}}

''wrist flex joint'' {{{

...

}}}

Both of these axes are along '''negative''' axes. I was fairly certain that a long time ago (~1 year ago?) we decided that all of these should be along the positive axes.

Do we want to change these? It's definitely a very breaking change.

trac data:

• Owner: hsu
• Reporter: vpradeep
• Reported at: Tue Nov 17 18:37:55 2009
• URL: https://code.ros.org/trac/wg-ros-pkg/ticket/3123

[ahendrix]

[hsu] yes... these were artifacts of trying to match motor rotation to joint rotation. Back in March we talked about changing these back to positive joint axis, but it wasn't deemed a top priority.

• What can we potentially break by changing these joint axis? (even if we are careful updating corresponding joint/safety limits)?
• What do we gain by flipping these joint axis (besides having unexplained negative joint axis and aesthetics)?
• Will hudson catch most of the side effects?

[ahendrix]

[vpradeep] '''Pros:'''

• PR2's Joint axes will be more intuitive for a new user

'''Cons:'''

• People with their own kinematics models will have to update their kinematics. This includes the robot calibration stack and Ben Cohen's planners, and who knows what other code (outside of WG) is making kinematics assumptions.
• Bag files with !JointState messages will have to be freshened (by inverting the some of the joint positions and velocities)

[ahendrix]

[watts] I'd have to change some scripts, but the sooner we make this the better. I'd be a supporter of anything that makes it easier to use the PR2. Now is the time to make the change, before we release Betas and while the Alpha's aren't in use. Dec 1 is the best time for me, I'll be back from IREX.

One possibility is to fix the Beta defs but not Alpha. That would make it much harder to port manipulation code from Alpha to Beta. That code will need to be ported if we make this switch anyway.

[ahendrix]

[berger] Before making this change, we neeed to make a list of all places that will be affected, including

• IK
• tuck-arms
• door code
• Any hardcoded joint angles in internal or external code
• Bag files (these will become wrong, but TF will remain correct)
• Any generated kinematics (TAO, OpenRAVE, etc.)

[ahendrix]

[hsu] added patch for pr2_common stack related to this update

[ahendrix]

[rphilippsen] TAO translation would be easier after the switch. That said, my current code treats -1 and +1 the same in the above cases... so no worries.

[ahendrix]

[hsu] flipper in:

• pr2_defs r27127
• pr2_description r27055

[ahendrix]

[vpradeep] Note that negating these joints also messes up the robot calibration procedure. I think the calibration search velocity and ref positions will both have to be negated as well.

[ahendrix]

[vpradeep] Ok... so I think this is what needs to change to get the robot to boot up correctly:

• Negate the transmission gear reduction in the URDF
• Negate the 'velocity' in pr2_default_controllers/pr2_calibration_controllers.yaml
• Negate the reference position in the URDF

Any thoughts are welcome.

[ahendrix]

[vpradeep] PRF, PRG, and PRE now are running with the flipped joints.

r27270

[ahendrix]

[watts] This isn't working on a Beta wrist. Looking at r27270, it looks like the Beta wrists (in pr2_description) have the opposite flex side as the pr2_defs. I'm changing it, will let you guys know if it works.

This should have been tested on both pr2_defs/pr2_descriptions wrists before closing.

[ahendrix]

[hsu] Kevin, can you point out the offending file? Thanks, John

[ahendrix]

[watts] I just fixed this and tested it on Beta HW using pr2_description. Changed pr2_description, pr2_default_controllers.

Changes in r27323.

[ahendrix]

[vpradeep] Can this ticket now be closed?