Adding potential new feature to pyexotica #519

cmower · 2019-02-21T20:41:36Z

I found it frustrating having to keep stop/starting ros launch in order to tune the rho parameters for task maps. To combat this issue I implemented a script interactive_rho_tuning.py that allows you to tune the rho's without having to constantly stop/start. See here for an example of its use.

If this sort of feature is something you would like in exotica let me know any changes/additions, otherwise feel free to close the pull request.

Checklist

code formatting: run find -name '*.cpp' -o -name '*.h' -o -name '*.hpp' | xargs clang-format-3.9 -style=file -i in the root directory
add unit test(s) (NA)
ensure tests build and check results: run catkin run_tests exotica && catkin_test_results (NA)

VladimirIvan · 2019-02-25T15:08:28Z

exotica_python/src/pyexotica/interactive_rho_tuning.py

@@ -0,0 +1,33 @@
+import Tkinter


New run dependency not included in the package.xml

VladimirIvan · 2019-02-25T15:10:33Z

dynamic_reconfigure was designed precisely for this purpose. Why use custom UI?

cmower · 2019-02-25T15:24:00Z

@VladimirIvan exotica doesn't broadcast Rho values as configurable ros parameters so cannot be picked up by dynamic_reconfigure.

wxmerkt · 2019-02-25T16:30:21Z

I like the idea but haven't been able to test it in particular. I see your issues with dynamic_reconfigure and like the simplicity of this interface.

You can get a callback in dynamic_reconfigure to extract the values and set them to whatever you want - that could help? [I've only used that in C++ before]

cmower · 2019-02-26T11:54:37Z

You can get a callback in dynamic_reconfigure to extract the values and set them to whatever you want - that could help? [I've only used that in C++ before]

This sounds like a good idea to look into. They list in the ros wiki for dynamic_reconfigure that the Python API is stable so I guess should work.

Two questions:

The feature could be generalized to re-configure other parameters on-the-fly (eg. solver params). Would this be worth the effort or should I stick with Rho parameters?
Is there a way to discriminate task maps based on their use as either cost terms or hard constraints? What would be useful is to have a function is_cost_term() (for example). Having this function would allow me to perform the following

In XML:

<Cost>
  <Task Task="ACostTaskMap" Rho="1"/>
</Cost>

<Equality>
  <Task Task="AnEqualityTaskMap"/>
</Equality>

In Python:

task_maps = problem.get_task_maps()
cost_task_maps = []
for k in task_maps.keys():
  task_map = task_maps[k]
  if task_maps[k].is_cost_term(): 
    # True -> task map used in cost function
    cost_task_maps.append(task_map)

# Now cost_task_maps is a list of 1 element
# i.e. the task map with the name "ACostTaskMap"

wxmerkt · 2019-02-26T12:50:59Z

The feature could be generalized to re-configure other parameters on-the-fly (eg. solver params). Would this be worth the effort or should I stick with Rho parameters?

Only implement what you need for your own work right now, it can always be extended later once required.

2. Is there a way to discriminate task maps based on their use as either cost terms or hard constraints? What would be useful is to have a function is_cost_term() (for example). Having this function would allow me to perform the following

The equality, inequality, cost private members (of type Task or TimeIndexedTask) are or should be exposed as read-only for most problem types. You can get the tasks included in them from there - or add a convenience method to Task/TimeIndexedTask.

wxmerkt · 2019-03-27T11:25:43Z

What is the status of this pull request? It seems like after adding the run dependency it could be good?

cmower · 2019-04-17T16:41:38Z

This hasn't been a priority for me for a while so I didn't look at it till now.

I've added a new save feature. See a demo here.
Simply using a try-except solves the error when trying to handle equality/inequality constraints.
Added icon for tk window
Added Tkinter to package.xml

cmower · 2019-04-17T16:57:44Z

Seems Travis CI build failed. I do not know why?

cmower · 2019-04-17T17:42:02Z

Prepend python- to Tkinter in package.xml. Perhaps this solves Travis CI issue...

wxmerkt · 2019-04-17T19:59:17Z

exotica_python/package.xml

@@ -14,4 +14,5 @@
  <depend>shape_msgs</depend>
  <depend>python-matplotlib</depend>
  <depend>python-pyassimp</depend>
+  <depend>python-Tkinter</depend>


No, the correct key is python-tk, cf. https://github.com/ros/rosdistro/blob/61959ac7a9d2d2bc91c0555ce139c22de428a917/rosdep/python.yaml#L4255

resolved, also added python-rospkg as per here

[exotica] added exotica icon [exotica_examples] added an example of rho tuning

cmower · 2019-04-18T12:16:52Z

Travis CI passed all checks
Squashed commits into 1 and force pushed

Waiting for final Travis CI checks to be completed and ready to merge as far as i can tell.

wxmerkt · 2019-04-18T12:28:06Z

exotica_python/src/pyexotica/interactive_rho_tuning.py

+
+__all__=['Tuning']
+
+class Tuning(object):


Can you please select a more descriptive name for this class, e.g., UnconstrainedEndPoseProblemInteractiveCostTuning or InteractiveCostTuning. As far as I can tell it now only supports costs, not constraints - and only for end-pose problems.

wxmerkt · 2019-04-18T12:28:18Z

exotica_python/src/pyexotica/interactive_rho_tuning.py

+import Tkinter as tk
+import rospkg as rp
+
+__all__=['Tuning']


PEP8 formatting (here and in the example file)

wxmerkt · 2019-04-18T12:29:52Z

exotica_python/src/pyexotica/interactive_rho_tuning.py

+        self.master.winfo_toplevel().title("Rho Tuning")
+
+        # Set icon
+        # note, rp.RosPack().get_path('exotica') throws error


Yep, because the core package is now exotica_core and exotica is only a metapackage - which does not work with ROS pack. A metapackage is like a stack, i.e., you can use rosstack, cf. ros/rospack#26. So your solution is:

rp.RosStack().get_path('exotica')

wxmerkt · 2019-04-18T12:32:11Z

exotica_python/src/pyexotica/interactive_rho_tuning.py

+
+        # Set icon
+        # note, rp.RosPack().get_path('exotica') throws error
+        icon = rp.RosPack().get_path('exotica_core')[:-13] + '/exotica/doc/images/EXOTica_icon.png'


I don't like this workaround. It has some assumptions on exotica_core being next to exotica, as in our source builds. This may not hold. I have posted a fix using RosStack above.

wxmerkt · 2019-04-18T12:32:29Z

exotica_python/src/pyexotica/interactive_rho_tuning.py

+        print("Setting Rho parameters:")
+        for k in self.cost_task_map_names:
+            r = float(self.entries[k].get())
+            self.problem.set_rho(k, r)


Note: this only works for end-pose problems.

…interactive_cost_tuning [exotica_python] new features to interactive_cost_tuning * reset button: resets params to the values originally specified in .xml * math feature in entries, you can now write math in entries

cmower · 2019-04-18T13:36:05Z

This will work for unconstrained/constrained problems however the tuning is only for cost terms. It doesn't make sense for equality constraints and will only make sense for inequality constraints if you want to switch the sign and so essentially switching between <= and =>.
New math and reset features added, see demo here.
All other points raised above have been addressed in latest commit.

exotica_python/src/pyexotica/interactive_cost_tuning.py

wxmerkt

Thank you for the quick turnaround - one typo but otherwise good to go.

wxmerkt · 2019-04-18T18:10:13Z

Thank you for adding this feature :-)

Adding potential new feature to pyexotica

cmower requested review from wxmerkt and VladimirIvan February 21, 2019 20:41

cmower changed the title ~~[exotica_python, exotica_example] adding potential new feature~~ Adding potential new feature to pyexotica Feb 21, 2019

VladimirIvan requested changes Feb 25, 2019

View reviewed changes

cmower force-pushed the cem-cost-tuning branch from af81b3a to a8930d7 Compare April 17, 2019 16:35

wxmerkt reviewed Apr 17, 2019

View reviewed changes

[exotica_python] added new interactive_rho_tuning

de383cb

[exotica] added exotica icon [exotica_examples] added an example of rho tuning

cmower force-pushed the cem-cost-tuning branch from c9a541f to de383cb Compare April 18, 2019 12:14

wxmerkt requested changes Apr 18, 2019

View reviewed changes

[exotica_python] some minor mods

237e9a2

wxmerkt reviewed Apr 18, 2019

View reviewed changes

exotica_python/src/pyexotica/interactive_cost_tuning.py Outdated Show resolved Hide resolved

wxmerkt reviewed Apr 18, 2019

View reviewed changes

Fix typo

a82feb4

wxmerkt merged commit 5a28aa7 into master Apr 18, 2019

wxmerkt deleted the cem-cost-tuning branch April 18, 2019 18:10

wxmerkt added a commit that referenced this pull request Jun 22, 2020

Merge pull request #519 from ipab-slmc/cem-cost-tuning

7177943

Adding potential new feature to pyexotica

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Adding potential new feature to pyexotica #519

Adding potential new feature to pyexotica #519

cmower commented Feb 21, 2019

VladimirIvan Feb 25, 2019

VladimirIvan commented Feb 25, 2019

cmower commented Feb 25, 2019

wxmerkt commented Feb 25, 2019

cmower commented Feb 26, 2019

wxmerkt commented Feb 26, 2019

wxmerkt commented Mar 27, 2019

cmower commented Apr 17, 2019 •

edited

Loading

cmower commented Apr 17, 2019

cmower commented Apr 17, 2019

wxmerkt Apr 17, 2019

cmower Apr 17, 2019

cmower commented Apr 18, 2019

wxmerkt Apr 18, 2019

wxmerkt Apr 18, 2019

wxmerkt Apr 18, 2019

wxmerkt Apr 18, 2019

wxmerkt Apr 18, 2019

cmower commented Apr 18, 2019

wxmerkt left a comment

wxmerkt commented Apr 18, 2019


		__all__=['Tuning']

		class Tuning(object):

Adding potential new feature to pyexotica #519

Adding potential new feature to pyexotica #519

Conversation

cmower commented Feb 21, 2019

Checklist

Choose a reason for hiding this comment

VladimirIvan commented Feb 25, 2019

cmower commented Feb 25, 2019

wxmerkt commented Feb 25, 2019

cmower commented Feb 26, 2019

wxmerkt commented Feb 26, 2019

wxmerkt commented Mar 27, 2019

cmower commented Apr 17, 2019 • edited Loading

cmower commented Apr 17, 2019

cmower commented Apr 17, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cmower commented Apr 18, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cmower commented Apr 18, 2019

wxmerkt left a comment

Choose a reason for hiding this comment

wxmerkt commented Apr 18, 2019

cmower commented Apr 17, 2019 •

edited

Loading