Port to ROS 2

[clalancette]

This PR is a port of joint_state_publisher to ROS 2. I did the initial port, but got a bunch of help from people along the way (thanks @gonzodepedro, @schornakj, @emersonknapp, @sloretz, and @pbeeson).

Besides the port to ROS 2, this PR does a number of other things to improve the code:

1. It adds a README describing the subscribed topics, published topics, and parameters.
2. If the GUI is in use, and the GUI is closed, this now quits the program (which I think is a little more intuitive).
3. Allows the joint_state_publisher to get its URDF either from the command-line, or from the /robot_description topic. In the latter case, this node will actually do nothing until something is received on /robot_description, at which point it will start publishing. If both the command-line option is specified and the /robot_description topic is available, the command-line option takes precedence.

I've done quite some testing on this to make sure all of the options work. In my testing with a simple URDF, it does seem to work fairly well. There are a few TODOs left:

1. Despite the fact that we allow_undeclared_parameters, there doesn't seem to be a good way to get at undeclared parameters in the rclpy API. Thus, we can't really support dependent_joints and zeros until we figure that out. That will be a PR for another time (maybe with some core rclpy changes as well). I actually found a way to do this, so this should work.
2. While rclpy.Rate is available in Eloquent, I wanted to make sure this was backwards-compatible with Dashing as well. So I left the TODO about switching to Rate in there. This no longer applies since we switched to a timer.

This also fixes #20

[sloretz]

Despite the fact that we allow_undeclared_parameters, there doesn't seem to be a good way to get at undeclared parameters in the rclpy API. Thus, we can't really support dependent_joints and zeros until we figure that out. That will be a PR for another time (maybe with some core rclpy changes as well).

It looks like this can be made to work with automatically_declare_parameters_from_overrides. I'm having trouble understanding the meaning of the options to the parameters API, but it seems automatically_declare_parameters_from_overrides declares any parameter that is passed via yaml (Not just parameters via parameter_overrides), and allow_undeclared_parameters doesn't seem to be necessary.

/**:
  ros__parameters:
    dependent_joints:
      joint_D: {parent: joint_A, factor: 3 }
      joint_E: {parent: joint_B }
      joint_F: {parent: joint_C, factor: -1 }

import rclpy
from rclpy.node import Node


if __name__ == '__main__':
    rclpy.init()
    node = Node('mock_jsp', automatically_declare_parameters_from_overrides=True)
    print(node.get_parameters_by_prefix('dependent_joints'))
    rclpy.shutdown()

$ python3 mockjsp.py --ros-args --params-file ./jsp_params.yaml --
{'joint_D.parent': <rclpy.parameter.Parameter object at 0x7fd4103f9780>, 'joint_D.factor': <rclpy.parameter.Parameter object at 0x7fd4103f9be0>, 'joint_E.parent': <rclpy.parameter.Parameter object at 0x7fd4103a1390>, 'joint_F.parent': <rclpy.parameter.Parameter object at 0x7fd4156cdc88>, 'joint_F.factor': <rclpy.parameter.Parameter object at 0x7fd407c74160>}

[clalancette]

@sloretz You are totally right, that works. Thanks! I've added back in support for dependent joints and zeros in 475e12f

[clalancette]

FYI that this is probably on hold until we decide what to do about #31. Once we've merged that, I'll likely rebase/rewrite this on top of that.

[clalancette]

I've now redone this on top of #31 . Still needs to wait until that is merged, but once that is in we can easily rebase and review this.

[clalancette]

All right, the ROS 1 stuff has been merged into noetic-devel. I've updated the ros2 branch to be the same as noetic-devel, and rebased this branch on top of that. @sloretz This one should be ready for a review.

[clalancette]

@sloretz Thanks for the great review! Things are now updated as you recommend in almost all cases. Please take another look when you get a chance.

[sloretz]

I get an IndexError here when I publish the same URDF twice on the /robot_description topic. It looks like self.joint_list has duplicate joint names, but num_joints was calculated from free_joints and dependent_joints which can't have duplicates because they're dictionaries, so msg.position is half the length. Maybe the variables that are set in __init__ need to be set in self.configure_robot instead so they get cleared when a new description is published?

[clalancette]

This was absolutely a problem. Fixing it turned out to be...complicated. Commit 735f2a7 successfully fixes it, along with making the layout way more dynamic and useful (in my opinion). It also simplifies the code a bit, at the expense of adding a new PyQt layout type (FlowLayout), and putting the Center/Randomize buttons at the top. I think that is a pretty good tradeoff for a dynamic, reflowable UI, but please take a look and let me know what you think.

[clalancette]

@sloretz All of your above points should be fixed now. Ready for another round, thanks!